This PDF is a selection from an out-of-print volume from the National Bureau of EconomicResearch

Volume Title: Interaction in Economic Research

Volume Author/Editor: NBER

Volume Publisher: NBER

Volume URL: http://www.nber.org/books/unkn77-1

Publication Date: 1977

Chapter Title: Staff Reports on Research Under Way

Chapter Author: Various

Chapter URL: http://www.nber.org/chapters/c4192

Chapter pages in book: (p. 11 - 104)

II

Staff Reports onResearch Under Way

1. MEASUREMENT OF ECONOMIC AND SOCIAL PERFORMANCE

IntroductionThis project, supported by the National

Science Foundation, is involved in extend-ing the national economic accounts toprovide better measurements of economicand social performance and a framework forthe integration of both macro and micro eco-nomic and social data. At the macro level,balance sheets and wealth, nonmarketactivities, the environmental factors, andtotal income measurement (including re-valuations) are being introduced into the na-tional accounts in order to make them morecomprehensive. At the micro level, data setsfor households, enterprises, and govern-ments are being developed in a form whichwill fit into and be aligned with the sectorsestablished for macroeconomic data. Suchmicrodata sets make it possible to introducesocial, demographic, and locational in-formation into the extended national ac-counting system. The MESP project hasbeen organized into subprojects, each withits own principal investigators, as follows:

1. National Economic Accountsa. The extended system of national ac-

counts—Richard Ruggles andCharlotte Boschan

b. National wealth and balancesheets—Raymond W. Goldsmith

c. Total income measurement—RobertEisner

d. National accounting and the environ-ment—Henry M. Peskin and LeonardP. Gianessi

2. Microdata Sets for the National Accountsa. Household microdata sets—Richard

Auggles and Edward Wolff

11

b. Lifetime income patterns of indi-viduals—Milton Moss

c. Enterprise microdata sets—RobertLipsey and Michael Gort

d. Government microdata sets—JohnQuigley

Richard Rug gles

The Extended System of NationalAccou nts

The current system of United Statestional income accounts is being extended toinclude nonmarket activity and balancesheet information at a subsector level, forwhich microdata sets are being developed.At the macro level the national income dataof the Bureau of Economic Analysis and theflow-of-funds data of the Federal ReserveBoard are being augmented with data pro-vided by John W. Kendrick, Raymond W.Goldsmith, Henry M. Peskin, and RobertEisner. The nonmarket activity is beingseparated sharply from market transactionsdata in order to obtain a better understand-ing of the operation of both market and non-market activities and their interrelationship.In some cases microdata sets forhouseholds are being used to generate esti-mates of nonmarket activity.

A simple version of NEAT (computerizedsystem for national economic accounting ta-bles), the project's set of programs tofacilitate the designing and redesigning ofeconomic accounting structures, has beencompleted. It is operational in the sense thatdata can be extracted from our variouscollections of basic data, new relationshipscan be specified, and tabular output can be

designed—all in an interactive mode. Somenew modules have been programmed butare not yet operational (e.g., providing auto-matic footnotes which may be attached toentire time series or individual observations,facilitating interactive revision, and updat-ing of time series); others are in the pro-gramming stage (e.g., providing more ele-gant tables, cross-referencing relationshiand providing period averages); still othersare in the planning stage (e.g., compilingperpetual inventories, making programseasier to use).

The major data bases available to thesystem include the NBER time series databank, the FRB flow-of-funds tape, most ofthe national income and product accountsof the Bureau of Economic Analysis, andsome of BEA's fixed capital series aspublished in its "Fixed Nonresidential Busi-ness and Residential Capital in the UnitedStates, 1925—75." We are in the process ofadding and revising the last of the nationalincome and product accounts. We are alsosetting up series titles to go with the shortcode names. These titles are necessary inorder to identify individual series, and theywill automatically be used as stubs in the ta-bles produced by NEAT, unless specificstubs are provided by the user. Also, sincedata collections are incorporated into theproject's data base with as little change inoriginal code names as possible, and sincesome series appear several times withseveral different codes, we need a list ofequivalent codes, so that the system knowsthat NBER's series called GNP is the sameas FRB's 86901005 and BEA's Table 1.1 line1 (or 010101). As always, the initial settingup of specifications and other tables takestime but will be very helpful later when thesystem is used more intensively.

Richard Rug glesCharlotte Boschan

National and Sectoral BalanceSheets

The goal of this project is to develop an-nual balance sheets for the nation and forseveral dozen sectors and subsectors for

12

the period 1947 through 1975, Work so farhas been concentrated on tangible assets,and this part of the project is expected to becompleted shortly. Preparatory work hasbeen done on financial assets, which will bethe center of attention during the comingmonths.

Raymond W. Goldsmith

Total Income MeasurementRevisions have been completed on

"Capital Gains and Income: Real Changesin the Value of Capital in the United States,1946—75," which was presented to the To-ronto meetings of the Conference on Re-search in Income and Wealth in October1976.

The paper presents capital gains or "netrevaluations" as the difference betweenmarket value or replacement cost, as esti-mated from price deflators for the relevantitem or category of capital, and the moneyvalue of capital necessary to keep its realvalue intact. Capital gains or net revalua-tions are thus increases in the value of exist-ing capital over and above the amount cor-responding to general inflation.

Net revaluations are estimated by sec-tor—household, farm and noncorporatenonfarm, corporate nonfinancial, allfinancial, and government—for tangibleassets and financial assets and liabilities,item by item, and for total net worth. Esti-mates of total capital accumulation and re-valuations are presented for both humanand tangible nonhuman capital.

Net revaluations or capital gains turn outon balance to be substantially positive,despite considerable fluctuation, particu-larly in certain financial assets and lia-bilities. Capital gains over a thirty-year pe-riod, even after allowance for generalinflation effects, have not been completelyoffset by capital losses.

On tangible assets alone, mean net re-valuations were approximately $28 billion.Mean tangible net investment aside from re-valuations was $68 billion. These mag-nitudes relate to all tangible nonhumancapital in government and households and

in business, and include both land and re-producible assets. By comparison, meannet private domestic investment in theconventional income and product accountswas $40 billion over these years. Total tan-gible capital accumulation from 1946through 1975 measured $2,870 billion. Totalnet private domestic investment reported bythe Bureau of Economic Analysis was$1,197 billion,

In 1969 dollars and including allhousehold assets, mean household non-human capital accumulation came to $64billion for the period from 1946 to 1969; $19billion stemmed from net revaluations. Meanhuman capital formation for this period wasestimated at $103 billion, of which $8billion related to net revaluations.

There were major capital gains by govern-ment, substantial gains for nonfinancialbusiness, both corporate and noncorporate,and some losses on balance in the financialsectors. Households (including personaltrusts and nonprofit institutions) were onbalance gainers, with substantial dif-ferences, however, by category of assetsand liabilities. Owners of land and owners ofhomes (because of the gains on mortgagedebt liabilities), owners of unincorporatedbusiness, and, until recently, owners of cor-porate equity tended to do well, Householdsnot relatively well situated in terms of thesekinds of investments tended to be losers be-cause of the substantial negative net re-valuations on currency, demand and timedeposits, and life insurance and pensionfunds.

Our major sources included publishedand unpublished material from the Flow ofFunds Section of the Federal ReserveBoard, the Bureau of Economic Analysis ofthe Department of Commerce, and JohnKendrick's The Formation and Stocks ofTotal Capital (NBER, 1976). An appendixdescribing sources and methods in detailhas been prepared by Marsha McClure andRene Moreno,

Work in completing and revising esti-mates for the total income system of ac-counts is proceeding. A paper embodyingthe results will be presented to the meeting

13

of the International Association for Researchin Income and Wealth in York, England, inAugust 1977. Also, John Graham and RoyWebb will be presenting an updated versionof their "Present-Value Estimates of HumanCapital Stocks" (described in last year's An-nual Report).

Graham plans to complete his study,initiated last year, on the determinants andcomposition of total household saving. Hedefines saving as change in net worth andincludes not only the accumulation of alltypes of financial assets, but also of tangibleassets such as household durables andresidential structures, and intangible humancapital. His theoretical framework stressesinterrelationships of the different types ofasset accumulation, with particular em-phasis upon the human capital decision.The theory generates hypotheses concern-ing the timing, magnitude, and mix of assetaccumulation which are subjected to em-pirical verification using a cross-sectionalpanel study, "Consumer Durables and In-stallment Debt," provided by the MichiganSurvey Research Center.

Webb is continuing his study of costs ofinflation, including those stemming from re-lated increases in the variation of relativeprices and in the time devoted by buyers tomarket search. Theoretical analysis sug-gests, however, that the costs are not relatedto steady inflation but to changes in the rateof inflation. Preliminary empirical work sup-ports this suggestion. An interesting by-product of Webb's analysis is the indicationthat in markets with imperfect informationand random shocks to supply and demand,sellers may 'have kinked demand curves,even with many actively competing sellers,quite aside from an oligopolist's concernwith reactions by rivals to price changes.

Ronald Rost has estimated productionfunctions for two-digit industries and moreaggregated sectors of the U.S. economywhich suggest that the slowdown in laborproductivity growth can be attributed in partto the replacement of capacity-expandinginvestment by pollution abatement invest-ment. Currently he is estimating the socialcost of the stock of pollution abatement

capital as the present value of the conven-tionally defined output that was notproduced because of the resultant slowergrowth of productivity. Available data permitsuch an estimate over the period 1967—1976, and will indicate the extent to whichthe social cost of pollution abatementcapital has deviated from its market value.In any analysis of the net benefit resultingfrom pollution abatement investments, thissocial cost must of course be weighedagainst the corresponding social benefits.

Augustine Fosu is working on estimates ofhuman capital that will reflect the presentvalue of expected future earnings in the formof supplements or fringe benefits as well aswages and salaries. It would thus relate toall of the expected marginal product oflabor, Aggregative data on fringe benefitsare available at the firm, establishment, orindustry level but not for individuals cate-gorized by personal characteristics such asage and sex. Fosu hypothesizes, however,that employers minimize their labor costs byfeasible combinations of fringe benefitsand wages and salaries which fit thepreferences of their employees. Hence,since employee characteristics differ fromindustry to industry, fringe benefits will differas well.

Fosu is utilizing Bureau of Labor Statisticstapes containing data on fringe benefits,direct earnings, hours, and other details ofemployment by establishment for 1971. Healso has census data on individual charac-teristics, which can be mapped onto indus-tries. The relationship between fringebenefits, earnings, age, and sex may thusbe estimated. Further, by filtering out in-dustry effects unrelated to employeepreferences, it should be possible to es-tablish a relationship between individualcharacteristics and preferences for fringebenefits.

Since fringe benefits constitute asubstantial portion of total labor compensa-tion (19.6 percent of wages and salaries in1975, according to estimates of the U.S.Chamber of Commerce), their considerationmay prove of substantial importance incomprehensive estimates of human capital

as well as in various aspects of labor marketbehavior.

Robert Eisner

National Accounting and theEnvironment

The purpose of this project is to includeimputed measures of the services of envi-ronmental assets and associated pollutiondamage in an expanded national accounts

14

framework.Major emphasis during the past year has

been placed on the development of datathat will permit estimation of the dollar valueof water pollution damage in geographicaldetail. In this connection a new data set hasbeen developed that permits a linkage of in-dustrial and household activity at the countylevel with the physical characteristics ofthose water bodies that constitute thecounty's drainage system.

Several research papers which use theenvironmental accounts as a data basehave been completed or are nearing com-pletion. One is a study of alternative esti-mates of industrial waste generation anddischarge. In that paper it was shown thatthe source of often vast differences in theestimates of different investigators was acombination of different implicit assump-tions and different interpretations of officialgovernment data.

A second study is a theoretical and empir-ical investigation of the effects of extendingthe environmental asset accounting systemto other nonmarket goods and services,particularly in-kind governmental subsidies.This paper (being prepared with theassistance of Janice Peskin of the U.S. De-partment of Health, Education and Welfare)is to be presented at the fifteenth conferenceof the International Association for Researchin Income and Wealth in York, England,August 1977.

A third paper contains analyses of perfamily net benefits (benefits minus costs) ofthe Clean Air Act of 1970 by region, incomeclass, and race. The principal result is thatnet benefits are positive for only about one-third of the population, generally those liv-

ing in urban areas and having low incomes,large families, and relatively low per capitaautomobile ownership. However, theaverage gain of a family with positive netbenefits far exceeds the average loss of afamily with negative net benefits.

Finally, a report on the project waspresented at a conference on environmentalindicators held by the International Institutefor Environment and Society in West Berlin,Germany, December 1976.

Henry M. PeskinLeonard P. ,Gianessi

The Construction of HouseholdBalance Sheets

Work on the construction of householdbalance sheets was divided into twophases. The first involved the constructionof a micro data base containing demo-graphic data, consumer durable holdings,and income flow data from financial andother income-earning assets for house-holds. The data base was formed by a statis-tical match of the 15 percent and 5 percent1/1 000 Public Use Samples of 1970 (the twosamples differ in their inventory of consumerdurables) and the 1969 Internal RevenueService (IRS) Tax Model. This phase hasbeen completed. Estimates of income flowtotals made from this data base agree veryclosely with corresponding entries in theNational Income and Product Accounts for1969.

The second phase involves the imputationof values to assets and liabilities ofhouseholds. The value of owner-occupiedhousing and the holdings of consumer dura-bles such as automobiles and televisionsets are provided in the Public Use Sam-ples. To value these durables, we ranregressions, using the 1961 Bureau of LaborStatistics Consumer Expenditure Survey, ofthe purchase price of durables on family in-come and demographic characteristics ofthe family. Moreover, we computed thelikelihood of purchase of each durable byincome and demographic characteristic.Using a random error generator, we will im-pute to each household the current value of

15

each durable held and reconcile these esti-mates with specially obtained Bureau ofEconomic Analysis estimates. Income flowssuch as interest, dividends, farm income,pension income, and small-business profit(or loss) are provided in the IRS file. Theseflows will be capitalized using averageyields for each asset (and each liabilitysuch as mortgages and consumer debt) andreconciled with flow of funds estimates. Thiswork is currently in progress.

A Reliability Test of the StatisticalMatching Procedure

Since an important question in our con-struction of household balance sheets is thereliability of the matching technique, wehave designed a test of its reliability. Thecentral problem is to determine how closelythe joint distribution of actual and imputedvariables corresponds to the actual jointdistribution. In most cases the actual jointdistribution is unknown. However, in thespecial case of the two Public Use Samples,the actual joint distributions of most vari-ables are known. By regressing imputedvariables on actual ones and actual vari-ables on actual ones and comparing theregression results by using a Chow test, wecan then determine whether the joint dis-tributions are statistically different.Preliminary results suggest little statisticaldifference between the imputed joint dis-tributions and the actual ones.

Determinants of Earnings Inequality

A paper entitled Age, Education, andOccupational Earnings Inequality" waspresented at the Econometric Society meet-ings in Atlantic City in September 1976. Us-ing data from the 1960 and 1970 CensusPublic Use Sample, we arrived atfour majorconclusions: (1) The earnings function—therelation between earnings, schooling,experience, and demographic characteris-tics—varied considerably among occupa-tional groups. (2) The earnings functionshifted between 1960 and 1970 becausethere were demographic shifts and a stableoccupational distribution. (3) The earningsfunction was not independent of the occupa-

tional structure. (4) The results broadly sup-ported the hypothesis that the size distribu-tion of earnings was generated by the occu-pational (and by implication the industrial)structure of the economy, though individualearnings were associated with individualcharacteristics. This paper is currently be-ing revised.

Screening and Earnings

A draft of a paper, 'Educational Screen-ing and Occupational Earnings," wasrecently completed and the resultspresented at a seminar at the Palo Altooffice of the National Bureau. Using datafrom the 1970 Public Use Sample, we corn-pared schooling and earnings for self-em-ployed and salaried members of thirty-fouroccupations. In occupations with customercontact, we predicted that self-employedworkers would acquire more schooling thansalaried workers to signal prospectiveclients of their productive capability. In

other occupations, we predicted that sincethat need to signal would not exist, self-em-ployed workers would acquire less school-ing than salaried workers. The resultsgenerally supported our hypotheses.

Life-Cycle Expenditure Patterns

A draft of a paper, "The Life-Cyclers," wasrecently completed. Using data from the1961 Consumer Expenditure Survey, weexamined life-cycle earning and expendi-ture patterns by demographic group to de-termine whether they conformed to thewidely accepted model of dissaving in earlyyears followed by saving in the middle yearsand dissaving in the years after retirement.We found that this pattern held for whiteurban families headed by a married, white-collar man. For other groups, the patterndeviated considerably, with some showingno saving at all over the life cycle.

Edward Wolff

Lifetime In come Patterns ofIndividuals

In this study time series of lifetime in-comes were constructed for several birth

16

cohorts of men and women, coveringextended periods of their adult lives sinceWorld War Il. It is believed that this data setand the analyses undertaken with themprovide useful perspectives on a variety ofincome distribution questions.

This progress statement outlines the mainparts which are now being processed for in-clusion in our final report on the lifetime in-come subproject. In the introduction to thefinal report, this subproject is related to thegeneral project on the Measurement of Eco-nomic and Social Performance. The objec-tive of the general project is to improve thedata base for the national economic ac-counts to permit a wider range of analysesof welfare problems, of changes in eco-nomic growth, and of income distribution.The purpose of the lifetime income sub-project also is to improve the data base foranalysis of income distribution, particularlyanalysis of changes over time rather than ofdifferences at one point in time.

The first main part is a historical view ofthe average age-income profile for U.S. menfor the past twenty-five to thirty years.Averages are presented of both cross-sec-tional differences at a point in time andcohort changes over time. The differencebetween these patterns provides a basis foran estimate of the contribution of economy-wide changes to the growth of lifetime in-come for representative cohorts. Economy-wide changes include both the general risein real wage rates resulting from produc-tivity gains and changes in the distributionof jobs among the different wage ratebrackets, The growth in output per man-hourof approximately 3 percent per year in theperiod from 1947 to 1972 seems to accountfor most of the difference between cohortand cross-sectional patterns. The reportcontains discussions of a number of meth-odological issues associated with suchcomparisons as well as a discussion ofproblems and promises of cohort and longi-tudinal data generally.

The second major set of results relates tothe analysis of differences among groups inthe level and time shape of lifetime income.The differences are analyzed both within

and among cohorts. Intercohort differencesare presented by comparing the time shapeof income of older and younger cohorts atcomparable stages of life. For most of thepostwar period, except for short periods ofrecession, the time shape, measured as arate of change with age, appears to haveremained fairly uniform for white men. Forother groups and more recent periods weare still working through numerous dataproblems to evaluate evidence of changesover time in the shape of lifetime incomepatterns. This aspect is particularly im-portant to those who entered the labor forceduring the last decade or so—the so-calledbaby boom cohorts.

Intracohort differences are also presentedin the report. The general pattern of increasein inequality of income for a given cohort asit ages appears to be true for white men butnot for nonwhites or women.

A third major part of the study isconcerned with lifetime income patterns bysource of income, An exploration is made ofthe different time shapes of earnings,property income, and other income and theirrelative contributions to total lifetime in-come.

A final section contains a methodologicaldiscussion of cohort analysis—its problemsand promises and avenues for major dataimprovement both in federal and privatesurvey statistics.

Milton Moss

Lifetime Indicators of Well-BeingResearch on Goals Accounting, a project

funded by the National Science Foundation,is to be carried out by a consortium whichincludes the National Planning Association(coordinator), the Carnegie-MellonUniversity, the Conference Board, theUniversity of Michigan, the University ofTexas at Austin, and the National Bureau.

A subproject, Lifetime Indicators of Well-Being, is to be undertaken at NBER. A one-year feasibility effort by the consortium isprovided for by NSF. The purpose of thissubproject is to extend the lifetime ap-

17

proach beyond lifetime income (an ongoingNBER project) to other domains of life suchas lifetime health and employment. In theone-year feasibility study by the consortium,however, the extension is confined tolifetime employment or labor force partici-pation. Should the project be funded on alonger-term basis, lifetime indicators wouldbe developed for other domains of life aswell. Our long-term aim with the use oflifetime indicators is to map social changesin such a way as to be able to comparechanges in well-being of successivegenerations in their major "walks of life."

Lifetime employment or labor force par-ticipation was chosen for the one-year feasi-bility phase of the project for three reasons:First, this important domain of life couldprovide a good illustration of a lifetimeperspective to the study of social change.There are particularly pronounced life-cyclepatterns in labor force participation, andthese patterns are undergoing dramaticchanges.

The second reason is that the consortiumplans to make a modest effort in the first yearto analyze a significant policy area—a seg-ment of the Social Security system—in orderto provide an example of a policy applica-tion of the goals-accounting project.Changes in the size and composition of thelabor force resulting from changes in theage distribution of the population and inage-specific rates Of labor force participa-tion and their effects on the "dependencyratio" are central questions concerning thefuture of the system.

A third reason is that future growth of thelabor force is one of the major resourceconstraints which must be taken into ac-count in assessing the capabilities of theUnited States for improving national well-being. Changes in the manner in whichhousehold members apportion their timebetween market work and other activitiesover their lifetimes have important implica-tions for the size and composition of thelabor force, the economy's productivity andrate of growth, and the distribution of in-come.

Milton Moss

Enterprise Microdata Sets

The purpose of this projeàt has been tocreate new sets of data for enterprises bycombining information from a variety ofpublic and private sources. set is to bea historical record for firms, including thosethat have disappeared, by merger or bank-ruptcy, and the other is to be a cross sec-tion, for a period in about 1972, of firms andtheir individual establishments.

Our fullest historical record for firms willcover about three hundred companies. Forthese we have combined the standard typesof financial accounts, such as balancesheets and income statements, with a va-riety of detailed data on the investmentprocess, including breakdowns of capitalexpenditure by type and extensive informa-tion about research and development ex-pend itures.

For a larger group of some two to threethousand companies we have not expandedon the financial statements but have addedinformation on hundreds of companies nolonger in existence. The result is a data setthat covers considerably more of the corpo-rate universe of, say, 1950 than a currentfinancial data tape such as Standard andPoor's Corn pustat. Our restoration of defunctcompanies is not complete, but does covermost of the disappearances of the late1960s and early 1970s and indicates that amore thorough job could be done with addi-tional time and resources.

A greater part of our effort has gone intothe construction of microdata sets for es-tablishments as of about 1972. These are toproduce both 'a geographical and a com-pany cross section for that year which willcontain a wide variety of characteristics thatcan only be derived from information aboutthe establishments of each company.

For the most part, the establishment datacover manufacturing, The basic originaldata sets we use are a Dun & Bradstreet fileof well over 350,000 establishments and anEconomic Information Systems (EIS) file ofover 100,000 In both casesthe entries are identified by parent com-pany, location (county, address, and zip

18

code), and industry, and include a count ofthe number of employees. A third originaldata source, an employment characteristicsfile of about 140,000 establishments in1970, reports the composition of employ-ment in each establishment by occupation,race, and sex.

Of the three basic data sets, the D&B ismost complete. The manufacturing employ-ment reported is about 99 percent of theCensus of Manufactures total for 1972 instates other than Alaska and Hawaii, whilethat in the EIS tape is about 86 percent andin the employment characteristics tape,considerably less. In individual states, theD&B totals fall within 10 percent of theCensus aggregates in 36 out of 49 casesand are more than 15 percent away in onlytwo. However, the cross-classification bystate and industry indicates much largerdiscrepancies with Census figures, about athird diverging by 20 percent or more. Al-though we can already produce a break-down of employment by detailed industry bystate and county we hope to improve thisbreakdown by using some of the EIS data,which seem to be more exact about industryclassifications.

The identification of ownership links be-tween enterprises and establishments isproving to be a more difficult problem thanthe geographical distributions. Although allthree tapes purport to show such links wefind that there are many discrepancies andomissions. We are working to improve allthree data sets in this respect but we are notsure that we will be able to produce aunified, complete data set. We will thereforebegin by developing the company-es-tablishment links for the 400 firms with themost complete time series data, go on to dothe same with all of the firms for which wehave financial data, and only then try to en-compass large numbers of the other firms.

Since one of the main plans for the currentyear is to incorporate the employmentcharacteristics data into our larger datacollection we carried out an experiment on asample of firms to merge these data withone of the other establishment data sets andwith information outside our system. The

sample consisted of 130 companies in nineindustries, their 794 establishments in theprimary industries of these companies, andthe counties in which the establishmentswere located. The work on the sampleconfirmed our fears that the links betweencompany and plant on the employmentcharacteristics tape were seriously incom-plete. However, we were encouraged tolearn that between 70 and 90 percent of theplants on the EIS tape could be identified onthe employment characteristics tape andthat about 75 percent of the employmentreported in the Census of Manufactures wascovered. We now have, for the companies inthe sample, the combination of EIS and em-ployment characteristics data on eachcovered establishment; the race, sex, andoccupational distribution of each county inwhich a plant is located; and financial datafor the firm as a whole. While we do not yetsee a way of merging the whole employmentcharacteristics file with the others, we arenow optimistic about its usefulness indescribing labor force composition for fairlylarge samples of firms.

Stanley Lewis carried out the detailedmatching and computer manipulation offiles for this study.

Michael GortRobert E. Lipsey

Government Microdata SetsThis project has been directed toward

producing geographically disaggregatedestimates of the public-sector activity in-cluded in the national income and productaccounts. Detailed revenue and expendi-ture accounts for state and local govern-ments have been prepared which are

Introduction

While prices and productivity have tradi-tionally been major foci of research at theNational Bureau, recent developments,

consistent with the aggregate accountingstructure. Less detailed accounts were pre-pared for the composition of and changes inpublic debt. The accounting structure hasbeen applied to the creation of two datasets—one providing income and productaccount information separately for each ofthe fifty states and the other providingsimilar estimates for 'county aggregates" oflocal governments. The county aggregatesinclude a separate revenue, expenditure,and debt account for each of the 3,118counties in the United States by aggregat-ing (and netting where appropriate) theactivities of all local governments withineach county boundary.

The results of this analysis have beensubjected to a number of consistencychecks and compared with available ag-gregate data sources on public-sectoractivity. In general, it appears that the ac-counting structure is quite satisfactory andis in close agreement with the aggregatepublic-sector accounts.

A detailed working paper outlining thesources, methodology, and assumptions ofthe analysis has been prepared, togetherwith the results of a series of internal andexternal tests for consistency. Data sets forthe benchmark year of 1972 and the sup-porting documentation are available tointerested users on request.

Currently under way is a further disag-gregation of these accounts for urbancounties. The extended data set will presentseparate estimates for each central city andfor the remainder of each metropolitancounty. It is anticipated that these additionalestimates will be available before the end of1977.

19

John M. Quigley

particularly inflation, have served to elevategeneral interest in these subjects. Workunder way, as well as new initiatives, havebeen complemented both by research oninternational aspects of inflation and by two

2. PRICES AND PRODUCTIVITY

Income and Wealth conferences, one onprice behavior and the other on productivity,cited elsewhere in this report.

The research program covers bothmeasurement and analysis. A majoranalytical focus is on the development of amodel of the economy by stage of process,which traces and quantifies the propagationof inflation domestically. A paper describ-ing the model was presented by Joel Popkinat the 1976 meetings of the American Eco-nomic Association. As part of this project,Avram Kisselgoff continues his analysis ofhow changes in crude oil prices aretransmitted to the prices of refined petro-leum products purchased by end-users. Anew project has been undertaken which hasas its objective the analysis of productivitychange by stage-of-process sector, of howsuch change affects sector prices, and ofthe routes by which such effects ultimatelyimpact on the overall inflation rate. M. lshaqNadiri is examining the impact of expendi-tures for research and development onproductivity at the firm level, Some pilotwork is also under way by Wayne Vroman onthe behavior of wage rates.

The episode in the early 1970s of double-digit inflation underscored the need toimprove the measurement of prices and thevariables that determine them. Richard Rug-gles has completed a review and evaluationof the U.S. wholesale price index and itscomponents, a price index system of majorimportance as an early indicator of changesin the overall rate of inflation.and as adeflator of nominal output on an industrybasis, Robert J. Gordon has been studyingthe components of the wholesale priceindex covering producers' durable goodsand other price series. He is currently com-pleting his manuscript on the measurementof durable goods prices. The problem ofmeasuring inventory change, an importantelement in short-run changes in economicactivity and in price determination, hasbeen exacerbated by high rates of inflation.A study of the measurement of business in-ventories and recommendations for itsimprovement are being prepared for sub-mission this year by Murray Foss, Gary

20

Fromm, and Irving Rottenberg. RobertEisner's manuscript, 'Factors in BusinessInvestment," is scheduled for review by theBoard of Directors prior to publication.

This program of research began undergrants from the Sloan Foundation and theAlex C. Walker Educational and CharitableFoundation. Current work is being financedby the National Science Foundation, theCouncil on Wage and Price Stability, andthe Census Bureau.

Inflation Study

Joel Popkin

During the year work neared completionon the development of an integrated modelof final and intermediate demand by stageof process. The model, which will soon un-dergo thorough testing, will be used toanalyze the sources of U.S. inflation, 1965—1974. The project has been funded by theNational Science Foundation.

In the model, the final demand and inter-mediate commodity sectors interact in thefollowing way: For any grouping of finalgoods—consumer home goods, for ex-ample—their prices, expected demand,and any inventory disequilibrium determineorders placed with manufacturers. Thelevels of manufacturers' production,finished goods inventories, and unfilledorders are then determined by using modelsbased on linear decision rules for the firm.These decisions in turn affect capacityutilization rates and the change in unfilledorders and finished goods inventories.These three, together with hourly labor costsand materials input prices, enter the equa-tion for manufacturers' prices. This process,which is depicted in Figure 2.1, is repeatedbackward through the stages of process.

Work on the subset of price equations ismost advanced. Equations have been esti-mated on the basis of several assumptionsabout price equations for each of the sixteenmanufacturing sectors and five retail tradesectors. When the results for the twenty-onesectors are viewed as a whole, they appearto support the general conclusion that theinfluence of demand on the relationship of

FIGURE 2.1

INTEGRATED MODEL OF FINAL AND INTERMEDIATE DEMAND BY STAGE OF PROCESS

Note.' K signifies intersectoral influences within the trade bloc, and L indicates a lagged effect.

output to input prices takes place in theprimary and semifinished goods industries,rather than in the finished-goods-producingand -distributing industries. This isparticularly apparent for consumer goods.The results for the manufacturing sectorshave been used in simulations to assess theimpact on final prices of changes in wagesand prices at earlier stages of process. Thiswork has been supported by the Council onWage and Price Stability.

A related, newly initiated project, sup-ported by the National Science Foundation,concerns the relationship between produc-tivity change and price change. The mainobjective of this research is to provide in-formation needed to determine the sectorsof the U.S. economy in which productivityimprovement—arising with respect to

21

capital, labor, or materials—can be ex-pected to result in the largest reductions inoverall inflationary pressures.

The study is designed to shed light on thefollowing questions: (1) What has been themeasured change in the productivity oflabor, of capital, and of materials in eachstage-of-process sector from 1954 to 1975?(2) What has been the change in total factorproductivity in each stage-of-process sectorduring the same period? (3) How have thechanges in individual and total factorproductivity in each sector corresponded tothe impact that price changes in each sectorhave had on the price of final output? (4) Towhat extent, in the long run, has productivitychange been reflected in price behavior ineach sector and in the overall rate ofinflation? The analysis will be done for

II COMMODITY

INPUTS

th PRIMARY

MANUFACTURINGSECIOP

FINISHEDMAN UFAC TURING

SECTOR

TRADE

SECTOR

VARIABLES FROMELSEWHERE IN

MODEL

thirty-one private commodity- and service-producing sectors in the intermediate sectorof the overall model described above.

Joel Popkin

Research and Development andFirm Productivity

The purpose of this research is to find therelationship between R&D activities andproductivity at the level of the individualfirm. A dynamic model, described in lastyear's reports, of the determinants of invest-ment, employment, and R&D has beendeveloped and estimated. In estimating themodel we have compiled the necessarydata for 114 companies for the period 1964—1974. The appropriate variables have beenassembled by merging data from varioussources, including supplemental data oncosts of R&D personnel, prices of capitalgoods, and relative input prices obtainedfrom the U.S. Department of Commerce. In-formation on some alternative indexes of in-novative activities such as patents has alsobeen assembled. The model was first fittedto the data for a subsample of 62 firms forthe period 1965—1972, and the results sup-ported the specifications of the model. Apaper based on these results will bepublished in New Developments in Produc-tivity Measurement, a volume of theConference on Research in Income andWealth. A larger manuscript describing insome detail the formulation of the model, thenature of the data, and the results of severalspecifications of the basic model, using thedata on a sample of 114 firms, has also beencompleted. The main findings of our re-search are:

1. The firm's employment, capital accu-mulation, and research and developmentdecisions are closely intertwined, and adynamic interaction process seems tounderlie these decisions.

2. The research and development activitiesof the firm, like its demand for labor andcapital, are influenced significantly byboth output and relative input prices andby degree of capacity utilization.

3. The long-run output elasticities of the

22

inputs, especially those of labor and re-search and development, are quitesimilar and suggest a constant return toscale in production.

4. Both labor productivity and investmentdemand of the firms are affectedsignificantly by their research anddevelopment expenditures, and laborproductivity is significantly affected bycapital accumulation.

5. Demand for each of the three inputs isquite stable when firms are stratified bysize of their assets; however, there is evi-dence of cross-sectional differencesamong firms and some changes overtime in their input decisions.

6. The model is a good predictor of invest-ment, employment, and R&D investmentof individual firms. The parameters esti-mated from the pooled cross-sectionaltime series of the firms in the sample areused to generate predicted values of thedependent variables for each individualfirm.

This study is being financed by a grantfrom the National Science Foundation.

M. lshaq Nadiri

The Wholesale Price IndexOn the basis of comments of January 15,

1977, of an NBER advisory committee andadditional comments provided by the Bu-reau of Labor Statistics, a draft of "TheReview and Evaluation of .the WholesalePrice Index" was revised and given to theCouncil on Wage and Price Stability onMarch 15, 1977.

In the report it was concluded that thewholesale price index and the price data onwhich it is based are widely used andconstitute a centrally important set of eco-nomic data. It was further concluded,however, that there are a number of majordeficiencies which seriously impair the use-fulness of the index. These are as follows:

1. The scope and coverage of thewholesale price index is inadequate for theuses made of it. The index currently coversagriculture, mining, and manufacturingprices. Most of the agricultural prices are

provided by the Department of Agriculture,and only about 25 percent of the four-digitSIC industries (constituting about half oftotal production) are adequately covered.The wholesale price index should bechanged to be an industrial-sector priceindex covering only mining and manufactur-ing, and the collection of price data shouldbe expanded to provide adequatecoverage.

2. The classification system used for thewholesale price index is inappropriate.Rather than using a unique classificationsystem for the wholesale price index whichcan only be transformed into the Standard,Industrial Classification System with somedifficulty and ambiguity, the Bureau of LaborStatistics (a) should move directly to the SICSystem and (b) should develop a stage-of-processing sectoring based on input-outputrelationshiPS.

3. The lack of integration betweenwholesale price data and other economicdata, such as production, employment,orders, and inventories, impairs theanalytical usefulness of these relatedbodies of data published by the variousagencies in the federal statistics system. Asystematic effort is needed to integratethese various series at the micro and ag-gregate level.

4. The present intermingling of shipmentand order prices results in data which aredifficult to interpret, As is shown in the tablebelow, which is for mid-1976 and is basedon a 1-in-14 sample drawn from WPIobservations, some of the price data refer toorder prices and some to shipment pricesand the order-to-shipment lags varyconsiderably:

Time Lag between Ordersand Shipments

This suggests that the Bureau of LaborStatistics needs to give greater attention toascertaining whether it is collecting orderprices or shipment prices and additional in-formation is also needed on order-to-ship-ment lags. It is not correct to deflate currentshipment values by current order pricesand, conversely, shipment prices may notserve as early warnings of price inflation ifthere is a long order-to-shipment lag and noescalation provision.

5. In some instances wholesale price datafail to capture changes in transactionprices. As a result in some cases thewholesale price index does not show thesoftening of price increases which may oc-cur in a period of oversupply or decliningeconomic activity and, conversely, it mayunderestimate the increase in price withincreasing economic activity and tighteningup of supplies. Although no single source ofinformation can be relied upon to providetransaction prices, monitoring a number ofdifferent types of price information (i.e.,purchasers' prices, net realized prices, unitvalues, etc.) can help keep track of the di-rection and magnitude of the bias involved.

6. The current use of specification pricingand the manner in which new products arebrought into the wholesale price index resultin measures that deal inadequately withquality change takihg place in various com-modities. Hedonic price measurements andthe use of functional specifications may incertain cases be helpful, but as in the caseof transaction price indexes, what is re-quired is the exploration of a number of dif-ferent methods for developing adjustment

Number of Series

OrderPrices ShipmentPrices Unknown Total

Spot or shelfOne week to one monthOne month to three monthsThree months to six monthsSix months to one yearOne year or moreUnknown

Total 285

23

8 31

7 14433 24911 57

o 34o 12

34 102

251 93 629

16

51

113423412

17

7

86103

4

0

051

factors. In view of the growing use of priceindexes for escalation of contracts andwages, an understatement of quality changeand the consequent overstatement of pricechange may have serious effects on the be-havior of the economic system.

7. The BLS system of weighting wholesaleprice data results in general price indexesthat are misleading. The use of value-of-shipment weights to combine commodityprice data into aggregate price indexesresults in overweighting of raw materialsinput and intermediate outputs. As a conse-quence, the behavior of the generalwholesale price index unduly reflects thebehavior of these commodities. In Order todevelop appropriate price indexes for dif-ferent levels of aggregation, e.g., seven-digit product, five-digit product class, four-digit industry, and two-digit industry, asystem of net output weights should be usedthat would weight the commodity price databy the amount of the commodity sold out-side of the grouping for which the priceindex is being compiled. In addition toindexes based on net output weighting itwould also be. useful to compile value-added price indexes from input prices andoutput prices at a fairly detailed industrylevel.

8. The use by the Bureau of Labor Statis-tics of judgmental data collection ratherthan probability sampling prevents statis-tical estimation of sampling errors andreduces the efficiency of the collection ef-fort. A sampling frame is needed as a basisfor developing a probability sample; this al-ready exists in other statistical agencies ofthe federal government and it should be ob-tained through cooperation with them.

Richard Rug gles

The Measurement of BusinessInventories

The study of the measurement of businessinventories that the National Bureau has un-dertaken for the Census Bureau is now in thedrafting stage and was scheduled to besubmitted to the sponsor this summer.

Some of the recommendations that will

24

appear in our report have already beenadopted by the Census Bureau. Forexample, all known LIFO firms are now be-ing asked for the first time to report LIFO andnon-LIFO inventories separately on themonthly reporting form and to report theirLIFO reserves. Firms that make bona fideLIFO calculations through the year whenprices are rising should show variations intheir LIFO reserves, but this is not always thecase. The month-by-month reporting of anunchanging LIFO reserve is symptomatic ofimproper interim LIFO reporting and will re-quire that such firms be treated by theCensus Bureau as FIFO respondents. Also,companies that value their inventories on astandard cost basis are requested to reportinventories on a comparable valuation basisfor adjacent months when a change in thecost standard occurs. Previously, changesin inventories that reflected changed coststandards went undetected.

A major recommendation of our report willcall for dual reporting requirements basedon the size of the reporting unit. Small firms,representing mainly single-plant compa-nies, will report as in the past. Large mul-tidivisional firms will be asked to report totaldomestic inventories with divisional break-downs each month. Year-end figures of suchfirms, with special information pertaining tothe year-end closing, will constitute the an-nual benchmark for this group. In the past,plant totals provided through the AnnualSurvey of Manufactures constituted thebenchmark even though inventory valuationmethods at the plant and divisional levelswere sometimes quite different. The CensusBureau has already embodied this recom-mendation in their latest budget request. Itis of interest that our recommendations re-garding divisional reporting units coincidewith the announcement of new standards bythe Financial Accounting Standards Boardfor financial reporting by diversified com-panies. The new standards pertaining to"segments" of a business enterprise, orwhat could be called rules for disaggrega-tion, should not only make corporationfinancial reports more meaningful to inves-tors and the general public but should also

be of assistance in the Census Bureau'sstatistical programs.

Other aspects of our study have beenmentioned at earlier dates. Since the lastprogress report to NBER we agreed toexamine the Census Bureau's statistics on

IntroductionBusiness cycle developments received

much public attention during 1976, and thework of the National Bureau in this field, pastand present, seemed particularly germane.The so-called pause in the recovery duringthe late summer and early autumn was oneof the big events of the season, both duringand after the political campaign. To some itrecalled the pauses in earlier recoveries,during the 1950s and 1960s as well asbefore World War II. Indeed, to a few, amongthem Arthur Burns, it recalled WesleyMitchell's observation that a pause in mid-expansion was a common characteristic ofthe business cycles that he had devoted alifetime of study to. To most it was a brand-new experience, a demonstration of theweakness and fragile nature of the expan-sion itself, a problem that required specialattention and an extra dose of governmentstimulation.

The index of leading indicators,published currently by the Department ofCommerce but a product of earlier NBER re-search on the factors that generate businesscycles, was partly responsible for the elec-tion-year excitement about the pause, Thedeclines in the preliminary figures for thisindex in August and September, and itssluggishness in October, made front-pagenews. Unfortunately, virtually no attentionwas paid to the fact that preliminary figuresare subject to revision. Even less excusablewas disregard of previous experience that asmall decline or a leveling off in this indexafter about a year of recovery is par for thecourse. It has been connected, in the past,with the unsustainable upsurge in inventoryinvestment that usually occurs early in a

25

unfilled orders in order to improve currentbenchmarking procedures.

Murray FossGary Fromm

Irving Rottenberg

recovery. It was so connected in thisinstance, since final sales (that is, GNPexcluding the change in inventories)showed no pause in its rate of expansion.

The subsequent revisions in the leadingindex almost erased the decline, leavingonly a small dip (less than one-half of 1

percent from July to September), with newhighs being reached in October, November,and December (see Chart 3.1). Theexperience underlines the need for betterand more promptly reported data (requiringless extensive revision) and for closer atten-tion to historical studies of the kind the Na-tional Bureau has specialized in.

This experience and others like it makeone think it might be desirable to establish acontinuing system for monitoring the statis-tical quality of the more important economicindicators issued by the federal govern-ment. A small research staff, knowledgeablein the history, behavior, and methodologyunderlying these figures, could exploreproblems they find or that are brought totheir attention, consider and propose solu-tions, and prepare reports on their findings.Some of these reports might be technical,others for wide circulation. To ensure ob-jectivity and appropriate attention topriorities an advisory committee of datausers from business, labor, government,and universities could be set up to guide thework and review the results. We areconsidering plans for an enterprise of thatsort.

Some problems of the kind that might beattacked in such a project were brought tolight during the year. It was found, forexample, that the widely used monthly

3. BUSINESS CYCLE RESEARCH

0II0

CHART 3.1

RECOVERY PATTERN: INDEX OF TWELVE LEADING INDICATORS

Exp!anation of Chart: Changes in the index from the business cycle trough to the current month are expressed aspercents of the preceding business cycle peak. The peak and trough levels are centered three-month averages. Forthe current recovery, the preceding business cycle peak level, November 1973, is 130.1, and the trough, March1975, is 107.6 (1967 = 100). The five earlier recoveries are identified by the year in which the trough month oc-curred: October 1949, May 1954, February 1961, and November 1970.

series on real spendable earnings of aworker with three dependents understatesthe average level of that worker's earnings,as reported in other government data, by

• about 35 to 40 percent, and also fails toreflect the rising trend in his real earningsover the past ten years (see my article in theMorgan Guaranty Survey, October 1 976). Aquite different problem besets the ratio ofprices to unit labor cost in manufacturing,one of the leading indicators published inBusiness Conditions Digest. This ratio sug-gests that prices rose about 25 percentfaster than costs during the past ten years,but is contradicted by other governmentdata that show prices rising at about thesame pace as unit labor costs. The latterresult is much more consistent than theformer with the trend of corporate profitmargins during the same period. These arebut two of the many problems that could wellreceive more attention if a program for moni-toring the quality of government economicstatistics were organized.

The need for continuing studies of busi-ness cycle experience and the developmentof data for observing this experience cur-rently and comparing it with the past also

26

were demonstrated by our study of interna-tional economic indicators, A simple tablewas constructed, showing the number ofcountries (of the seven covered in our studyso far) experiencing recession or recoverymonth by month during 1973—1976. Theresults are depicted in Chart 3.2. It showshow the worldwide recession of 1974—1975could be observed to spread among thoseseven major industrial countries and that theprocess could be seen a few months earlierin the leading indicators than in the coin-cident ones. Similarly, the recovery spreadgradually from country to country and couldalso be observed a few months earlier in theleading indexes.

Now the fact is that most of the indicatorsthat are reported widely on an internationalscale are coincident ones (industrialproduction, shown separately in the chart, isamong them). Leading indicators, like neworders received, construction work started,or the average workweek, are as a rule eithernot reported by international sources at allor are not seasonally adjusted or are notdeflated for price changes not up todate. Furthermore, the of acquiringcomparable monthly data historically, so

-, recession trougF

—6 —3 0 +3 +6 +9 +12 +15 +18 +21 +24 +27 +30 +33 +369/74 12/74 3/75 6/75 9/75 12/75 3/76 6/76 9/76 12/76 3/77 6/77 9/77 12/77 3/78

CHART 3.2

NUMBER OF COUNTRIES WITH RISING INDEXES OF ECONOMIC PERFORMANCE, 1973—1 976

(includes seven countries: Canada, the United Kingdom, West Germany, France, Italy,Japan, and the United States)

1973 1974 1975 1976

No.of

that the behavior of these indicators duringprevious business cycles can be analyzedand understood, is a time-consuming chore.With the limited resources we havemanaged to obtain, enough has been ac-complished to demonstrate the practicalvalue of the results (for further details seethe report by Moore and Klein below). TheOrganization for Economic Cooperation andDevelopment (OECD) has recently ex-pressed interest in collecting the necessarydata from member countries and exchang-ing them for the composite indexes andother measures we derive from them. Wehope that this year will see the beginning ofa current NBER international indicators databank that will make these products widelyaccessible,

The questions to which business cyclestudies are addressed are of a continuingnature, requiring continuing investigation.Business publications and governmentalreports make wide use of such results ofbusiness cycle research as the NBERchronology of peaks and troughs, compara-tive patterns of current and previousrecoveries or recessions, lists of leadingand lagging indicators and correspondingrecords of performance, and the like. Duringthe past year, for example, the device ofcomparing the current recovery with pre-vious recoveries, which was developed inNational Bureau studies years ago, wasused in virtually every business and govern-ment publication dealing with the businessoutlook. The comparisons pertained toproduction, prices, income, consumerspending, housing starts, capital invest-ment, bank loans, interest rates, employ-ment, unemployment, and profits.1 This sug-

gests not only that the results of businesscycle research are of practical value overlong periods of time, but also that it isessential to re-examine previous findingsfrom time to time in the light of a changingeconomic environment.

Without this continuous re-examination ageneral deterioration of the scientific qualityof the inferences drawn from earlier work islikely to occur. How many, for example, areaware that the common view that the 1974—1975 recession was the most serious in theUnited States since the 1930s is subject toan important qualification, namely, that it isnot true of employment? Measured by thedecline in employment, the 1974—1975recession was one of the shortest andmildest of the seven recessions since the1 930s. How many know why this is so, why itis true of employment but not of unemploy-ment, and what it probably implies as to thenature of future recessions? Business cycleresearch is designed to uncover such ques-tions and track down the answers (see theforthcoming NBER report, "The Recessionand Recovery of 1973—76," by VictorZarnowitz and me).

To take another example, it seems to 'methat the development and use of a

chronology of the rate of inflation, analogousto what has been done with the businesscycle chronology, is a timely subject for re-search. It has been documented in someinitial studies that the rate of inflation asmeasured by the consumer price index hastypically begun to rise at about the time thebusiness cycle reached its lowest ebb. Atsuch times unemployment is still high andcapacity utilization rates are still low, butneither the one nor the other has prevented

1. Among the publications using this method ofanalyzing the current recovery that have come to my at-tention are the following: Business Conditions Digest,Appendix 0, monthly; The Conference Board StatisticalBulletin, monthly; The Current Business Picture,Bankers Trust Company, Fall 1975; Economic Report ofthe President, January 1977; Business Week, McGraw-Hill Company, April 19, 1976; Business Conditions,U.S.A., Irving Trust Company, Spring 1976, Spring1977; Quarterly Business Conditions Analysis, Manu-facturers Hanover Trust Company, June 1976; The 1976

28

Midyear Review of the Economy, Report of the JointEconomic Committee of the Congress of the UnitedStates, 94th Cong., 2d sess., September 29, 1976; Busi-ness in Brief, Chase Manhattan Bank, October 1976;Economic Forecast, Prudential Insurance Company,November 26, 1976; Business and Money Review andOutlook, 1977, Harris Trust and Savings Bank,November 29, 1976; The Pocket Chartroom, Goldman,Sachs & Company, December 1 976—January 1977; An-nual Report, Federal Reserve Bank of New York, 1976;WafiStreet Journal, March 14,1977.

the rise in demand from spreading to moreand more prices and raising the rate ofinflation in the process. As Arthur Burnswrote in 1951, viewing the records compiledat the National Bureau before World War II,"Inflation does not wait for full employ-ment."2 How many, besides Burns, have be-come aware that every cyclical recoverysince World War II has supported this state-ment? How many were surprised, during thecurrent recovery, when the inflation rateclimbed from its low point in the spring of1976 to an uncomfortably high level a yearlater, despite unemployment rates that hadrisen to about 7 to 8 percent and capacityutilization rates that were down to about 80percent?

The lessons of experience in this regardare simply not well known. If they were,fewer would regard the unemployment rateas if it were a leading indicator of theinflation rate—i.e., that it must come downto a low level before the rate of inflation willrise, It is more nearly a coincident indicator,and not a very good one at that. With widerknowledge, through research, of what arethe more reliable leading indicators ofinflation—and what are not—perhaps thosewhose difficult task it is to judge what is go-ing to happen to the general price level andwhat policies are best suited to deal with itmight be able to do a better job.

All this leads me to note that the financialsupport that business cycle research hasbeen getting from business, from govern-ment, and from foundations has not beenkeeping pace with the need for it. This mightbe understandable if it appeared that thebusiness cycle itself, and the inflation thatgoes with it, had been conquered, that weknew all the answers and had the matterfirmly in control. Obviously that is not thecase. It would also be understandable ifother approaches to the problems posed bybusiness cycles had proved themselves su-perior in every respect and adequate to thetask. I do not believe that is the case either.

2. ntroduction to Wes'ey C. MitcheU's What Hap-pens During Business Cycles: A Progress Report (NewYork: NBER, 1951), p. xxi.

29

The irony is that just when public interest inthe subject has climbed to new heights, andpublic use of the findings that the NationalBureau has contributed over the years is sowidespread, we should experience moredifficulty than ever before in obtaining thewherewithal to continue the work.

Geoffrey H. Moore

Monetary Trends in the UnitedStates and the United Kingdom,1867—1973

During the past year we revised Chapter10 of our draft, dealing with money andinterest rates, and completed Chapter 11,on long swings. The work that remains to bedone is (1) a final revision of the existingdraft, which numbers roughly sevenhundred typed pages, (2) preparation of afinal summary chapter and preface, (3) re-working of the many statistical tables toassure their consistency with our latestrevised series, (4) the same reworking of ourmany charts, and (5) seeing the manuscriptthrough the Bureau refereeing process(which we plan to begin in the next fewmonths) and then through the press.

Partial support for the study for the yearbeginning April 1, 1977, has been providedby the Ford Foundation and the J. HowardPew Freedom Trust.

Milton FriedmanAnna J. Schwartz

International Economic IndicatorsAn early objective of the International

Economic Indicators (tEl) project was to ap-ply well-tried NBER techniques to the datingof business cycles in a number of market-oriented economies. A further objective in-volved experimenting with the 1966 NBERlist of leading, lagging, and roughly coin-cident indicators for the United States tosee whether corresponding series wouldbehave in the expected way in other coun-tries. To accomplish these objectives, wehave largely duplicated the twenty-six indi-cators for Canada, the United Kingdom,West Germany, France, Italy, and Japan,starting in the 1950s.

The results of this work have beensufficiently promising to induce the OECDand the national statistical offices of thecountries li$ted above to agree to cooperatewith the National Bureau on an experimentalbasis in setting up a current data bank forthe indicator series and related measures.The plan includes the construction by NBERof composite leading, coincident, and lag-ging indexes and other summary measuresuseful in analysis of business cycledevelopments in each country. We hope that

the data bank can be in operation this year.The cooperating organizations, as well asother governmental and private users, willthen have access to the current and his-torical data, It is our hope that it will greatlyfacilitate the accretion of knowledge oninternational instability and the manner of itstransmission.

During 1976—1977 the lEl project madeprogress on several fronts. During the sum-mer of 1976 and again in April 1977 the indi-cator data for the countries mentioned

CHART 3.3

30

130

120

110

100

90

(continued)

COMPOSITE LEADING AND COINCIDENT INDEXES, SEVEN COUNTRIES, 1970—1976

RatioScale

- 140

CHART 3.3 (concluded)

RatioScale

- 140

130

120

110

100

90

above were brought up to date, i.e., beyondour original stopping point in 1973. Wethought it was important to trace the indica-tor record through the recession suffered byall seven of the listed countries in 1973—1975 and to examine the extent of theirrecovery as well, At the same time we under-took to convert to a constant price basis allthe indicators that are expressed in currentprices. This change was introduceddomestically as well by the Department ofCommerce when the 1966 U.S. list of indica-tors was revised in May 1975.

We utilized the deflated series, along withupdated versions of the other indicators, to

31

construct composite indexes of the leadingand coincident indicators for each of theseven countries and for all countriescombined. During the relatively severerecession of 1973—1975 and the subsequentrecovery the indicators generally behavedin a manner consistent with their earlierperformance (Chart 3.3). We found the newmeasures were useful in showing the widevariation in rates of recovery experienced bythese countries. We are persuaded anew ofthe importance of monitoring cyclicaldevelopments through a variety of economictime series rather than relying upon anysingle measure.

1970 1971 1972 1973 1974 1975 1976

Another direction in which we haveprogressed recently has been in further ex-perimenting with the use of leading indica-tors to forecast changes in U.S. trade flows.Because imports by market-oriented econo-mies are closely associated with theirdomestic cycles, an exporting country canget a reasonably clear assessment of likelysubsequent cyclical changes in its exportsby examining the leading indicators of thecountry's trading partners. Chart 3.4 sug-gests the possibilities of this sort of ap-proach. As a forecasting instrument we uti-lize the leading indexes for the UnitedKingdom, West Germany, France, and Italy,combining them by weighting each by therelative magnitude of the countries' GNP in1970. The percent changes in this

combined 'Western Europe" leading indexthrough December preceding the year to beforecast are then correlated with the sub-sequent year-to-year percent changes inU.S. exports to this region. This assumes im-plicitly that the leading index is a sensitive,early-warning measure of the demand forimports, and is reflected in actual importssix months hence. The chart shows both theforecast and actual changes in the physicalvolume of U.S. manufactured goods exportsto Western Europe, 1964—1976. It is clearthat the export volume as predicted by thefour-country leading index conforms quitewell to what the subsequent fluctuation inexport volume proved to be. The correlationcoefficient (R2), reflecting the associationbetween actual and forecast changes is

0)/0

CHART 3.4

aUnited Kingdom, West Germany, France, and taly, weighted by their GNP in 1970. R2 = 0.73.

32

FORECAST AND ACTUAL PERCENT CHANGES IN U.S. EXPORTS OF MANUFACTURED GOODS TOWESTERN EUROPE, QUANTITY, 1964—1 976, USING LEADING INDEX FOR FOUR COUNTRIES8

30.

-15.01964 1969 1974 1979

0.73, i.e., about three-fourths of the fluctua-tion in the year-to-year change is accountedfor. Some preliminary attempts to use thisapproach with quarterly rather than annualdata have shown promising results.

Plans for an extension of the IEI researchand development program have been drawnup. They include:

1. Organizing and maintaining a computerdata bank for current leading and otherindicators for a number of industrial,market-oriented countries;

2. Investigating new indicators and improv-ing the quality of the international indica-tor system;

3. Investigating the synchronization of cy-clical movements in different countriesand the related tendency for theeconomy of the industrial world toexperience waves of rapid growth andinflation alternating with slow growth orrecession;

4. Testing and improving the trade forecast-ing method developed in our initialstudies in terms of its application tomonthly or quarterly data on imports,exports, and the trade balance;

5. Making comparative internationalstudies of the cyclical behavior of closelyrelated types of indicators, such as thosepertaining to investment, to financialmarkets, to labor markets, and to com-modity prices.

An ambitious program of this kind cannot,however, be undertaken unless the researchfunds available to the IEI project aresubstantially increased. During the pastyear the work has been supported by the de-partments of State, Treasury, Commerce,Labor, and Agriculture, the Federal ReserveSystem, and a number of American corpora-tions. Reports published and in preparationare as follows:

Published

Moore. "The State of the International Busi-ness Cycle." Business Economics, vol. 9,no. 4, September 1974.

—. "Economic Indicators and Econometric

33

Models." Business Economics, vol. 10,

no. 4, September 1 975.U.S. Department of Commerce, Office of

Corn petitfve Assessment. "New Interna-tional Business Cycle Indicators System:Applying the NBER Indicator Approach toInternational Growth Cycles." September1975.

"Impact on U.S. Trade Performanceof Business Cycle Movements in Major In-dustrialized Countries." October 1975.

Moore. "Booms and Slumps: The Impact onU.S. Trade." Morgan Guaranty Survey,November 1 975.

Klein. "Postwar Growth Cycles in the UnitedKingdom: An Interim Report." Explora-tions in Economic Research, NBER, vol. 3,no.1, Winter 1976.

"Business Indicators: Growth andClassical Cycles." Euromoney, London,March 1976, pp. 42—45.

"Recent Work in International Eco-nomic Indicators: Problems and Pros-pects." Industri konjunkturen, SwedishFederation of Industries, Spring 1976, pp.115—128.

Business Cycles in the PostwarWorld: Some Reflections on Recent Re-search. Domestic Affairs Study 42. Wash-ington, D.C.: American Enterprise In-stitute, 1 976.

Moore. "Forecasting Foreign Trade Flowswith Leading Indicators." The BusinessEconomist, Society of Business Econo-mists (Watford, Herts, England), vol. 8, no.1, Summer 1976.

Moore and Klein. "Recovery—and Then?"Across the Board. The Conference Board,October 1976.

Moore. "Economic Indicators NeededOverseas." New York Times, January 16,1977.

"Leading Indicators: A New Tool forForecasting World Trade." Institutional/n-vestor (International Edition), InstitutionalInvestors Systems, Inc., March 1977.

Klein and Moore. "A New Index for the Eco-nomic Summit." New York Times, May 1,1977.

Moore and Klein. "Summit Nations' Eco-

nomic Strength Building." Los AngelesTimes, May 8, 1977.

In Preparation

Moore and Klein. "Monitoring BusinessCycles at Home and Abroad." Draft, 242pages.

______

"The Development and Use ofInternational Economic Indicators." Draft,97 pages.

"Forecasting Foreign Trade Flowswith Leading Indicators." Invited paper forthirteenth CIRET (Centre for InternationalResearch on Economic Tendency Sur-veys) conference, Munich, September1977.

Geoffrey H. MoorePhilip A. Klein

long-term trends. The reference chronologyof growth cycles has been established by aclose examination of time series represent-ing national output, employment and unem-ployment, income, sales, and industrialproduction. The series were obtained by di-viding the successive observations for eachindicator by the corresponding trend values,the trend having been estimated in such away as to cut through, and contain nosignificant elements of, the short-termcyclical movements in the series.1

Chart 3.5 shows a monthly compositeindex, 1948—1976, constructed from elevenindicators expressed in the form of such de-viations from trend, All components areseries in constant dollars or physical units

Cyclical IndicatorsFurther progress has been made in the

analysis of U.S. growth cycles, that is, move-ments in aggregate economic activitydefined by the consensus of fluctuations incomprehensive indicators adjusted for their

1. Specific turning points in the ratios of monthly(quarterly) data to centered 75-month (25-quarter) mov-ing averages are used to break up the original seriesinto segments, and the trend is then interpolatedbetween the centered values of the averages of thedata within these segments. Percentage deviations ofthe data from this trend form the final trend-adjustedseries.

CHART 3.5

INDEX OF DEvIATIoNS FROM TREND, ELEVEN INDICATORS IN PHYSICAL UNITS ORCONSTANT DOLLARS, 1947—1 976

Note: U and D are growth cycle upturns and downturns, respectively. Crosses (x) denote upturns and downturns ofthe composite index,

34

Co M P0 SITE

U DUD

48 53 58 63 68 73 76

TABLE 3.1

BUSINESS CYCLE EXPANSIONS AND CONTRACTIONS AND HIGH- AND LOW-GROWTHPHASES IN THE UNITED STATES, 1948—1975

Business CycleReference Dates

Duration in Months Growth CycleReference Dates

Duration in Months

Contraction(PtoT)

Expansion(TtoP)

Low-GrowthPhase(DtoU)

High-GrowthPhase(UtoD)Trough(T) Peak(P) Upturn(U) Downturn(D)

(1) (2) (3) (4) (5) (6) (7) (8)

Nov.1948 July1948Oct. 1949 11 Oct.1949 March 1951 15 17

July 1953 45 July 1952 March 1953 16 8May1954 Aug.1957 10 39 Aug.1954 Feb.1957 17 30April 1958 April 1960 8 24 April 1958 Feb. 1960 14 . 22Feb.1961 10 Feb.1961 May1962

Oct. 1964 June 196612

2915

20Dec.1969 106 Oct.1967 March1969 16 17

Nov.1970 Nov.1973 11 36 Nov.1970 March 1973 20 28March 1975 16 March 1975 24

Average 11 50 Average 18 20

Source: National Bureau of Economic Research, Inc.

or quantity index numbers.2 The solidvertical lines denote upturns (U), the brokenvertical lines downturns (D), of U.S. growthcycles. It is apparent that these phases areon the average of similar length, unlike busi-ness cycles where expansions tend to bemuch longer than contractions because therising trend of activity prolongs expansionsand shortens contractions.

Table 3.1 compares business cycles andgrowth cycles in terms of their timing andduration. Since 1948, five complete busi-ness cycle expansions and six contractionsare identified, with average durations of 50and 11 months, respectively; also, eightcomplete high-growth and nine low-growthphases, with average durations of 20 and 18months. A growth cycle downturn (D) pre-ceded each of the six business cycle peaks(P) of the 1948—1973 period by intervalsranging from 2 to 9 months, and averaging5.5 months. Clearly, each recessionfollowed the onset of a low-growth phase.On the other hand, all but one of the six busi-ness cycle troughs (T) coincided with

growth cycle upturns (U). (In 1954, U laggedbehind T by 3 months.) Thus, recovery wasapproximately synchronous with the onset ofa high-growth phase. On three occasionssince 1948—in 1951—1952, 1962—1964,and 1966—1967—low-growth phases inter-rupted business expansions but did notdegenerate into business contractions.

Ten out of 15 dates in the NBER growthcycle chronology for 1948—1969 areidentical with those derived earlier by IlseMintz,3 and all but one of the others areclose, the exception being the 1964 upturn.This similarity of results obtained inde-pendently in studies using somewhat dif-ferent data and techniques is reassuring. Inanother test of the robustness of the presentapproach the method of trend adjustmentoutlined above was applied directly to thecomposite indexes of leading, roughly coin-cident, and lagging indicators currentlypublished in Business Conditions Digest(BCD) of the Bureau of Economic Analysis. Itshould be noted that the NBER referencecycle dating is based on an examination of

2. Other composite and diffusion indexes were usedas well, including some with seven additional series incurrent dollars. The results generally resemble eachother closely.

35

3. "Dating United States Growth Cycles," Explora-(ions in Economic Research (Summer 1974). The com-parison refers to the dates based on the set of indica-tors in real terms used in Mintzs study.

CHART 3.6

TREND-ADJUSTED COMPOSITE INDEXES OF U.S. LEADING, ROUGHLY COINCIDENT,ANDLAGGING INDICATORS, 1948—1 976

Note: U and D are growth cycle upturns and downturns, respechvely. Arrows denote leads and lags. Crossesdenote upturns and downturns of the composite indexes.

many more series than the four that com-prise the new revised index of roughly Coin-cident indicators; also, that we are using thetrend-adjusted indexes, not indexescomputed from trend-adj usted component

36

series. Because of these and other dif-ferences in the composition and construc-tion of the indexes used, there was nothingin our procedure that would necessarilyassure the transformed BCD indexes of a

49 52 55 58 61 64 67 70 73 76

good record of conformity and timing in thegrowth cycle context. As shown in Chart 3.6and Table 3.2, however, the record is

reasonably good. The leading index shows16 leads and 2 coincidences at the 18 turnscovered; the coincident index, 12 coinci-dences1 4 leads and 2 lags; and the laggingindex, 15 lags. The timing averages are allin the proper ranges. Evidently the seriesthat tend to lead at business cycle peaksand troughs also tend to lead at growthcycle downturns and upturns, andanalogously for the roughly coincident andlagging indicators. The timing measureslisted show considerable dispersion but thisis also true of the timing of the originalindexes at business cycle turns. The main

problems are with the weakest growthcycles whose dating has been difficult: thelagging index does not conform well to the1951—1 952 low-growth phase and the lead-ing index has an especially early upturn inthe 1962—1 964 phase.

In accordance with what would be ex-pected from the trend elimination, there areno marked asymmetries between the leadsand lags at growth cycle upturns anddownturns (unlike in the business cycles ofthe same period, where long leads are ob-served mainly at peaks and long lagsmainly at troughs). Because of the similarduration of high-growth and low-growthphases, one of the desirable characteristicsof lagging indicators—their lead relative to

TABLE 3.2

TIMING OF TREND-ADJUSTED COMPOSITE INDEXES AT GROWTH CYCLETURNING POINTS, 1948—1975

Composite Indexes of Trend-adjusted Indicators;

Growth CycleReference Dates

Leads (—) or Lags (+) in Months

Leading

U D

RoughlyCoincident

U D

Lagging

U D

InvertedLagging

Upturn Downturn(U) (D) U D

July1948 0 +1Oct. 1949 —4 0 +5 —14

March1951 —7 —2 b —12

July1952 —11 0 b b

March1953 0 0 +4 b

Aug. 1954 —7 0 +8 —13

Feb.1957 —17 0 +7 —22

April 1958 —3 0 +7 —7

Feb.1960 —10 —1 +4 —15

Feb. 1961 —2 0 +9 —8

May1962 —3 —1 +6 —6

Oct.1964 —24 —7 +11 —23

June1966 —3 0 +2 —9

Oct. 1967 —9 0 +5 —14

March 1969 —2 +7 +7 —12

Nov.1970 0 0 +16 —13

March1973 —1 +8 +18 —12

March 1975 —1 0 C —6

Median —4 —3 0 —1 +8 +5 —13 —12

Mean —6.8 —5.4 —0.8 +1.2 +8.7 +6.1 —12.2 —12.6Standard deviation 7.4 5.4 2.3 3.6 3.9 5.3 5.4 5.0

a Measured from the highest value for the available data which begin in January 1948.b No specific cycle (no turns corresponding to the indicated reference dates can be clearly identified).C Not identified.

37

the opposite turn of the leaders—is muchmore consistent for the trend-adjustedindexes than for the original ones. As can beseen by comparing the first two columns ofTable 3.2 with the last two, the turns of thetrend-adjusted laggers lead the oppositeturns in the trend-adjusted leaders at allturns except the upturn, where theleaders lead by 24 months and the invertedlaggers by 23. The mean leads of the in-verted lagging index are 6 months at peaksand 6.4 months at troughs of the leadingindex, and the corresponding standard de-viations are 4.1 and 2.9 months.

A new revised edition of BCD was in-troduced in November 1976 as a result of acomprehensive study of business cycle in-dicators conducted by the Bureau of Eco-nomic Analysis in collaboration with the re-search staff of the National Bureau. A Hand-book of Cyclical Indicators will be publishedshortly as a supplement to BCD.

Victor ZarnowitzCharlotte Boschan

Regularity of Growth CyclesWhile at the Bureau as a research fellow

under the program supported by the SloanFoundation, I have been extending myearlier work (Economic Inquiry, September1975) on what Irving Fisher called the MonteCarlo cycle in business activity. In myearlier work the expansions and contrac-tions implied by the standard NBERchronology were examined for evidence ofnonrandomness, with negative results. I

have extended this to the newer Mintzgrowth cycle chronology, again with nega-tive results, However, these negative resultsare not very conclusive, given the presentshort duration (only since World War II) ofthe growth cycle chronology. GeoffreyMoore has called to my attention the longertrend-removed series on aggregate activityof Frickey, Ayres, Persons-Barrons, and theAmerican Telephone and Telegraph Com-pany. In the near future, I plan to investigatethe regularity of the expansion and contrac-tion durations implied by these series.

J. Huston Mc Cu/loch

The Recession and Recovery of1973—1976

This study has two major objectives: (1) toreview the evidence concerning the factorsinvolved and the timing of the 1973—1976sequence of slowdown, recession, andrecovery; and (2) to measure the severity ofthe recession and the vigor of the recoveryin comparison with earlier cyclical contrac-tions and expansions. During the past yearwe completed a report and circulated it forreview by the staff and others. A revised ver-sion will soon be submitted to the Board ofDirectors.

The expansion that began near the closeof 1970 moved in the spring of 1973 into alow-growth phase during which some sec-tors of the economy declined or weakenedwhile others continued to enjoy strong de-mand and in some cases encounteredshortages of the required materials. Thebalance tipped toward a decline in ag-gregate output in the last quarter of 1973.

Both the decline in output and the rise inunemployment were considerably larger inthe 1973—1975 recession than in any of thefive earlier recessions of the period sinceWorld War II, but the reduction in total em-ployment was relatively mild. The sharpincrease in the rate of inflation during thefirst half of 1974 had no precedent in earlierU.S. business contractions. The pecu-liarities of this episode derive in the firstplace from the impact of the unexpected ac-celeration of the general price rise andshortages of materials; later on, from con-tinuing fears of a recurrence of such-phenomena and the consequent caution inp'rivate and public decision making.Consumer outlays on housing, automobiles,and other durable goods, expressed inconstant prices, declined substantially afterthe first quarter of 1973, but business invest-ment in structures rose, albeit slowly, foranother two quarters, and investment inequipment rose for another year. Output ofnondurable consumer goods did not turndown decisively before the fall of 1974, andthe volume of services purchased byconsumers did not decline at all during the

38

recession. It is not easy to explain crosscur-rents or lags of this type and size, either intheoretical or statistical terms. Some of thedata inspire skepticism, and there may belarge systematic errors because of the ex-traordinary rise in prices and the difficultiesof measurement, especially of inventories,that this occasioned. The problem is madeevident by the large revisions of the datathat have been and are still being made.

Restrictive actions on the supply side,which tend to produce rising prices alongwith falling output, played a prominent rolein 1973—1974, probably more so than inmost business cycles of the past. However,weaknesses on the demand side, whichcontribute to output reductions but maycounteract price rises, developed early inhousing and later became more and moreapparent generally. This resulted in sharpinventory movements of a familiar cyclicaltype, which marked the rapid transition ofthe economy from the severe downswing ofthe fall and winter of 1974—1975 to theinitially vigorous upswing beginning inspring 1975.

The recovery during 1975—1976 was onthe whole moderate—closely resemblingthe average performance of total output andemployment during previous recoveries. It

was accompanied by a relatively weak re-vival of business capital formation and slowreduction of unemployment. At the sametime, the declines in the rate of inflation andin interest rates, which are associated witheach other as well as with the pace of therecovery, were unusually large by historicalstandards.

Victor ZarnowitzGeoffrey H. Moore

Short-Term Economic ForecastingAboUt 160 individuals have participated

at one time or another in the quarterlyASA/NBER Survey of the Economic Outlook,which was initiated late in 1968; 67 of thesehave been regular members in the sensethat they submitted forecasts on at leasttwelve occasions (out of a total of twenty-nine surveys taken through 1975), Charlotte

39

Boschan and I are studying the distributionsof these individual forecasts for real GNP,the implicit price deflator, industrial produc-tion; and the unemployment rate. Forecasterrors are measured in their several dimen-sions by using alternative summary statis-tics and decomposition methods on predic-tions of both levels and relative changes.The factors to which the observed forecast-ing behavior and performance are being re-lated include the characteristics of thejump-off and target periods, the span offorecast, and the methods reported by theforecasters. Error statistics are computed forforecasts of each individual covered and forselected subsets of the sample, cross-classified by variable, period, span, andprincipal method used. This will enable usto compare the individual with the cor-responding group forecasts (means andmedians) from the same surveys in an ap-propriately standardized form.

There is no presumption that individualparticipants are consistently as accurate asthe average survey forecast; on the contrary,research so far indicates that averagingoften helps to offset errors in componentpredictions. Our previous work has centeredon the global results of the surveys, whichhave also attracted much attention of others,both at the National Bureau and elsewhere(V. and J. Su, Stephen McNees, ArthurOkun, Herman Stekler, Economic OutlookUSA, and the Conference Board StatisticalBulletin, to mention some authors andsources). However, it is the individualforecasts that are the primary data and ofmajor interest in their own right. We wish tofind out which of them are relatively suc-cessful, when, and why.

For the period beginning in 1970, a recordof forecasts based on econometric modelshas been assembled, including theWharton, Bureau of Economic Analysis,Chase, and Data Resources, Inc., models.These forecasts combine the models withjudgmental adjustments and (often severalalternative) assumptions about exogenousfactors. Forecasts of the MAPCAST group atthe General Electric Company, which useindividual econometric relationships in an

iterative process, are also available on tape.All forecasts in this set are generatedquarterly for sequences of up to eightquarters ahead. They cover real andnominal GNP, seven major expenditurecomponents of GNP, the implicit pricedeflator, and the unemployment rate. Al-though comprehensive and informativestudies of the overall accuracy of theseforecasts exist, much remains to be doneto broaden and deepen the analysis ofthese and related materials both over timeand cross sectionally. Forecasts differ bysource, type, model, role of judgmental andother elements, frequency, se-lection of exogenous variables, and com-plexity; the predicted variables, by eco-nomic significance and statistical prop-erties; and the periods covered, by the na-ture of concurrent exogenous events andendogenous developments. These separatedistinctions have been employed to formu-late the kinds of comparisons to be madeand questions to be asked.

The comparisons involve the individualand group forecasts from the ASA/NBERsurveys; the forecasts with econometricmodels; predictions obtained with the aid ofregressions of changes in real GNP and in-dustrial production on lagged changes inthe composite index of leading indicators;and extrapolative benchmark models of thegeneral autoregressive—moving average va-riety. Many interesting questions need to beraised or reconsidered. Are some variables

systematically more difficult to predict thanothers and, if so, what are their distinctivecharacteristics? Have some periodspresented the forecasters generally withgreater problems than others and, if so,why? Are there relationships between thetypes of forecasting errors (underestimation,overestimation, turning point errors) and thenature of the variables and periods in-volved? Stated more broadly, what if any arethe time patterns of forecasts and their er-rors? How do forecasts vary betweenperiods distinguished in terms of businesscycle phases, degrees of inflation, andintensity of outside shocks? Are the reportedforecasts consistent with the concepts orhypotheses of "rational expectations" and"efficient markets"—have they in effectutilized all predictive information from pastdata and models of economic theory? Howdo forecasts vary in these different respects,by type, variable, span, etc.?

Previous research provides some evi-dence on some of these questions, but aclose analysis of multiperiod quarterlyforecasts now available should be able toadd much to our as yet very fragmentaryknowledge of these matters. Even thoughthe data are still limited to a rather shortslice of history, they cover a variety ofsources, methods, and experiences, andare in this sense quite rich. A grant from theEarhart Foundation is helping to support ourstudies of these data.

Victor Zarnowitz

4. FINANCIAL INSTITUTIONS AND INDUSTRIAL ORGANIZATIONS

Financial Institutions andProcessesIntroduction

During the past year, National Bureau re-search on financial institutions and pro-cesses has continued to examine problemsof inflation in the investment sphere and infinancial markets and to study portfolio be-havior as reflected in the actions of indi-viduals and institutions.

As reported by Phillip Cagan and John

40

Lintner, research on the impact of inflationon financial institutions and markets is in theprocess of publication. Included are studieson specific types of decisions and theiroverall impacts. Benjamin Friedman isconstructing a model of the corporate bondmarket that combines equations basedupon the actions of both the borrowers andthe major purchasers of bonds.

Other research is concerned with themeaning for institutional and individual be-havior of developments in portfolio theory

and knowledge. The work on capital ade-quacy draws both on modern concepts ofportfolio risk and on detailed informationabout the way individual institutions have al-tered their distribution of assets and lia-bilities by major categories as the economyhas moved through cycles of interest rates,prices, investment, and output. It is hopedthat the results will enable those concernedwith institutions to measure differences inrisks associated with such portfolio deci-sions.

Other research is concerned with thetheory, measurement, and analysis of indi-vidual portfolio aOtions in relation to marketmovements. The statement by Wilbur G.Lewellen furnishes more complete informa-tion on a variety of studies that have been orare about to be published.

Sherman J. Maisel

Regulations, Risks, and Capital Adequacyin Financial Institutions

Data show that over the past thirty years adramatic shift has occurred in the composi-tion of the assets and liabilities of banks.Both the percent of riskier, nonliquid assetsand the ratio of purchased funds, particu-larly among large banks, have increasedsharply, At the same time, the share ofequity has fallen. There has been a generalrecognition that these changes haveincreased the risk of individual banks, butviolent debates have taken place as towhether the added risk has caused asignificant deterioration in the safety of indi-vidual banks and, if so, what to do about it.

Information from bank stock prices showsthat the market has greatly increased itsestimate of the risks involved in large banks.Whereas from 1945 through 1960, the beta,or relationship of changes in the price ofbank stocks to the market average (Stan-dard and Poors 500-Stock Index), was under0.5, in recent years it has been well above1.0.

We are attempting to model the balancesheets of different types of banks and otherfinancial institutions to see how their networth may be expected to respond to

41

changes in macroeconomic variables. Suchchanges alter an institution's cash flow.They affect the returns to type of assets, thecosts of liabilities, and the spread. Move-ments in the term structure of interest ratescause differing movements in the presentvalue of future flows, depending upon theperiod in which they are expected. TheNBER's 56th Annual Report (September1976) gives a more detailed explanation ofthe theory underlining this work.

We have specified a model of a wholesalebank to measure how its net worth changesas a result of such movements. We haveestimated the equations with data from asingle large bank. We have since expandedthe specifications to banks in general andare in the process of estimating the modelwith micro data for a variety of bank types.We hope these models will show howspecific differences in balance sheets relatedirectly to capital requirements.

We are examining in detail the types ofchanges that have occurred among differentbanks. We are estimating the impact ofthese changes on movements in bankreturns as measured by both book andmarket values. We also are examiningtheory and practice concerning the costs ofbank capital and their relationship to theform of insurance of deposits made avail-able by the Federal Deposit Insurance Cor-poration and the Federal Savings and LoanInsurance Corporation.

This research is funded by NSF—ResearchApplied to National Needs under grantAPR76-0251 1. We expect to have our resultsavailable by the end of fiscal 1978.

Sherman J. MaiselDavid Lane

David H. PyleBarr Rosenberg

William F. Sharpe

The Effects of Inflation on FinancialInstitutions and Markets

Several reports which represent the finalphase of this project are nearing comple-tion. Three studies of the effects of inflationand interest rate expectations on invest-

ment decisions of financial institutions,scheduled for publication this year, arereported on below. A study of the effects ofinflation on the various components ofhousehold saving, by Paul Wachtel, hasbeen submitted to the Board of Directors, Anoverall review of recent research andexperience of the effects of inflation on sav-ing and investment by the different sectorsof the economy and the consequences forfinancial markets has been prepared byPhillip Cagan and Robert Lipsey and willsoon be ready for circulation.

This project has been supported by theAmerican Council of Life Insurance.

Phillip Cagan

Investment Policies of Major FinancialInstitutions under Inflationary Conditions

Three of our studies have been revised inthe light of staff and Board reviews and havebeen published or are in the process ofpublication in Explorations in Economic Re-search. These are "Income Participations onMortgage Loans by Major Financial Institu-tions," by Piper, and "Interest Rate Expecta-tions and Optimal Forward Commitments forInstitutional Investors" by Lintner, both involume 3, number 4, and "Warrants andConvertible Debt as Financing Vehicles inthe Private Placement Markets" by Piperand Arnold, in volume 4, number 3. Our jointwork on "Forward Commitment Decisions onBonds and Mortgages by Life InsuranceCompanies" has been updated and revisedpreparatory to Board review.

The other paper described in previousreports, on testing the log-normality of secu-rities returns over holding periods of dif-ferent lengths, and developing the implica-tions of log-normality of returns for portfolioselection and market equilibrium, will besubmitted for publication shortly.

John LintnerThomas PiperPeter Fortune

Modeling the Corporate Bond andGovernment Securities Markets

This project is proceeding along threeprincipal avenues: First, I am continuing torefine the structural model of interest rate

42

determination in the corporate bond market.One recent result in this regard indicatesthat the past histories of both bond and shortyields, which may perhaps be interpreted asreflecting expectations of future yields, havea significant and sizable influence on bor-rowers' supply of and lenders' demand forbonds, and hence on the current bond yieldas well. Another recent result, supported byseveral different elements of evidence, indi-cates that bond investors' portfolio behavioris characterized by risk aversion of such na-ture and degree as to render bonds andother assets highly imperfect substitutes foreach other. One part of the evidence for thisfinding is that bond investors' portfolioresponses to changes in the (uncertain) ex-pected capital-gain-or-loss component ofthe holding period yield on bonds aresignificantly smaller than their responses toequivalent changes in the (certain in theabsence of default risk) coupon componentof the holding period yield. A second part ofthe evidence is that the respective yieldvariances on bonds and on competingassets significantly influence investors'portfolio allocations in much the way sug-gested by the capital-asset-pricing model.

Secondly, Vance Roley is constructing astructural model of interest rate determina-tion in. the U.S. government securitiesmarket, disaggregated by four maturitygroups. Thus far the model is complete forsecurities of two to five years and five to tenyears; these two submodels perform wellempirically,.providing further support for theuse of structural models of interest rate de-termination. An especially interesting ele-ment of this model is an asymmetrical treat-ment of commercial banks' management oftheir securities portfolios according towhich increases and decreases in the sizeof the portfolio have differential effects onthe portfolio allocation.

Thirdly, David Jones and I are estimatingthe various "bridge" relationships which willfacilitate putting the corporate bond marketmodel into the MPS model as a replacementfor the MPS term-structure equation. Thiswork is about half completed.

Benjamin M. Friedman

Misintermediation and BusinessFluctuations

Misintermediation is the mismatching ofthe maturity stwctures of an intermediary'sassets and liabilities. In NBER WorkingPaper 160, I argue that misintermediationpermits the intertemporal mismatching ofconsumption and production plans, result-ing in business fluctuations. During my yearas an NBER fellow lam, among other things,continuing research into empirical implica-tions of this model in two directions.

One direction is to measure as preciselyas possible unanticipated movements ininterest rates associated with historicalbusiness fluctuations. I have reportedelsewhere on modifications I made while atthe Bureau to my program for spline curvefitting the term structure of interest rates. I

have also developed a formula for a forwardlong-term bond yield, which can be calcu-lated very easily from the current short-terminterest rate and long-term bond yield. Thisformula has a precise relation to instan-taneous forward rates as the long bond ap-proaches a perpetuity. I have written a shortpaper on this forward bond yield for journalpublication.

The second direction is to quantify his-torical misintermediation prior to actualbusiness fluctuations. Thrift institutionsshould be particularly important, as they arebased on long-term assets and virtually de-mand liabilities. A complication of mypreliminary analysis of thrift institutions hasbeen that the relevant liability maturity is nottheir contractual maturities, but rather the(unobservable) planned withdrawal dates.Inferences about these plans from sub-sequent behavior is clouded by real interestrate changes, by unanticipated inflation,and by generational turnover. I am still at thestage here of developing concepts.

J. Huston Mc Cu/loch

Individual Investor Portfolio Performance

A concern with possible institutionaldominance of the market for equity secu-rities has prompted a number of studies ofthe trading behavior and portfolio policies ofthose entities. Direct data on the cor-

43

responding dimensions of individual inves-tor activities, at the micro level, have notgenerally been available, however. Throughthe cooperation of a large national retail bro-kerage house, I have been able to compilesuch data for a large sample covering theperiod 1964 through 1976, and have supple-mented it with questionnaire surveys of thesample group. The approach, scope of thedata, and some of the early findings wereoutlined in previous reports and are sum-marized in papers published in the Journalof Finance, May 1974 and May 1975, andthe Financial Analysts Journal, September—October1976.

The past year has been spent in extend-ing the analysis in those papers, which dealt'primarily with the questionnaire responses,to the measurement of investment perfor-mance by the sample, to updating the data,and to investigating a range of ancillary be-havioral issues. The results are contained inseveral additional papers either scheduledfor 1977 publication or currently underreview.

Of these, perhaps the major set of find-ings will be presented in two papers to bepublished in the Journal of Business, July1977 and October 1977: "Patterns of Invest-ment Strategy and Behavior Among Indi-vidual Investors" and "Realized Returns onCommon Stock Investments: The Exper-ience of Individual Investors." These dealwith the linkages between attitudes, at-tributes, and behavior and with investmentperformance from 1964 through 1970, forwhich transactions data are presentlyavailable for the sample. Surprisingly fa-vorable investment results emerge in thefigures portrayed, suggesting some reason-able acumen in security selection on thepart of the group. Other papers due to bepublished are "Stock Exchange Listingsand Securities Returns," in the Journal of Fi-nancial and Quantitative Analysis; "SomeEvidence on the Patterns and Causes ofMutual Fund Ownership," in the Journal ofEconomics and Business, and "An Empiri-cal Analysis of Brokerage House SecuritiesRecommendations," in the FinancialAnalysts Journal. All are offshoots of the

main study, and are made possible by thedata base compiled for it. Several othermanuscripts are in the working paper stage.

The coming year will be devoted to bring-ing the data base on securities prices andtransactions up to date through December1976, to executing a new questionnairesurvey of the sample, and to examining al-ternative investment performance measuresand perspectives. Moreover, work hasbegun on the possibility of developing asurvey instrument that will yield an "investorsentiment index" comparable to the SurveyResearch Center's Consumer SentimentIndex, which might act as a leading indica-tor both of market trading volume and secu-rities price movements. Exploratory effortsin this connection have been initiated with asample to be provided by the cooperatingbrokerage house. Grants from it and fromthe Investment Company Institute are themajor sources of project financial support.

Wilbur G. Lewellen

Industrial OrganizationsIntroduction

The Bureau's current studies in this areaare focused on the process of innovation.One, reported on below by Michael Gort,deals with the way in which a product inno-vation is diffused throughout an industry.Another, described by Henry C. Grabowskiand John M. Vernon, analyses the effects ofregulation on the costs and rate of innova-tion in the drug industry. Also of interest tothose concerned with the determinants ofthe rate of technical progress is M. IshaqNadiri's study, Research and Developmentand Firm Productivity, discussed in section2, above.

The program of research on organizationof markets, now almost completed, hasdealt mainly with advertising as a form of in-formation and as an influence on demandand market shares and with the more tradi-tional industrial organization subjects of theeffects of concentration and productdiversification. Michael Gort reports on thisset of studies below.

44

Aside from the recent and forthcomingpapers listed in the reports in this section,another one on a related topic is John C.Hause's 'The Measurement of ConcentratedIndustrial Structure and the Size Distributionof Firms," in Annals of Economic and SocialMeasurement, volume 6, number 1 (Winter1977). The Bureau has also been develop-ing some enterprise data sets that should beuseful in future studies of industrial struc-ture. These are reported on in section 1,above.

Organization of Markets

Robert E. Lipsey

This research program on the organiza-tion of markets was financed jointly by theEducational Foundation of the AmericanAssociation of Advertising Agencies and theNational Bureau of Economic Research. It isapproaching completion with two studiespublished in Explorations in Economic Re-search, one by Henry Grabowski on interin-dustry effects of advertising on demand andthe other by Rao Singamsetti and me onconcentration and profit rates. Two addi-tional studies have passed the staff reviewstage and are being revised for publication,one by Grabowski on the impact of advertis-ing on market shares in five industries andthe other by Phillip Nelson on advertisingand information. A fifth study on trends, de-terminants, and consequences of productdiversification (jointly with Grabowski andR. McGuckin) is being prepared for staffreview.

A summary of the results of the fivestudies was prepared by Hedy D. Jellinekand has been published by the AAAA.

Michael Gort

Diffusion of Product Innovations

This study, financed by the NationalScience Foundation, is approaching com-pletion. One paper, "Time Paths in the Diffu-sion of Product Innovations," was presentedjointly with Steven Klepper at the 1976 meet-ings of the Institute of Management Scienceand the Operations Research Society ofAmerica. It is being revised. It traces the dif-

fusion of product innovations, defined as theincrease in the number of producers of theproduct, over five phases. The phases, theboundaries of which are defined by dis-criminant analysis, consist of (1) the initialstage following commercial introduction ofthe product and preceding rapid 'take-off,"(2) the period of high growth in sales andhigh entry, (3) the phase of declining rate ofentry as the peak in number of producers isapproached, (4) the phase in which exit offirms exceeds entry, and (5) the stageassociated with maturity and a plateau innumber of producers.

The analysis, which was based on datadeveloped for forty-six product innovations,resulted in a number of strong conclusions.First, contrary to the conclusions of mostearlier work on entry, the number ofproducers reaches a peak and óommencesa decline while demand is still growing—indeed, frequently growing rapidly. Second,the rate of patenting is not closely related toother (and superior) indicators of the outputof innovative activity. The rate of patentingmoreover lags behind the rate of growth indemand by too long an interval to allow oneto infer a simple causal connection betweenthe two. Innovative activity appears to reacha peak in Stage 2, while patenting does notreach a peak rate until Stage 4 or 5.

A second paper, "Deriving the PersonalityProfile of a Firm from Published Panel Data,"(jointly with Rao Singamsetti) has beencompleted and is scheduled for presenta-tion at the Symposium on the Econometricsof Panel Data scheduled for August 1977 inParis and sponsored by the Institut Nationalde Ia Statistique et des Etudes Econo-miques. This paper, using factor analysis inconjunction with behavioral and managerialvariables, successfully identifies three cor-porate personality profiles, namely, ag-gressive, risk averse, and resistant tochange. The first two appear to be related toinnovative behavior. Specifically, firms that

introduce product innovations are shown tohave more aggressive and less risk-averseprofiles.

Michael Gort

The Effects of Product Quality Regulationon Innovation: The Case of the U.S.Pharmaceutical Industry

Our paper, "Consumer Protection inEthical Drugs," was presented to the 1976meetings of the American EconomicAssociation. It was published in the Amer-ican Economic Review, Papers andProceedings, February 1977. In the first partof our paper, we summarize the evidence forthe proposition that more stringent regula-tion of pharmaceuticals by the Federal DrugAdministration over recent years has been amajor factor in the industry, leading tohigher costs, time lags, and risks in pharma-ceutical innovation. In the second part, weshow that a significant shift has occurred inthe structure of innovation in the industry: In-novation has become increasingly concen-trated in the large, multinational drug firms,apparently because they are better able tobear the additional costs and risks of inno-vation and also because they can shiftresources on a worldwide basis to offsetsome of the adverse impacts of regulationsin this country. Some evidence concerninginternational transfers of technologyassociated with regulatory differencesacross countries is presented in the finalpart of the paper.

Our paper with Lacy Thomas, "Estimatingthe Effects of Regulation on Innovation: AnInternational Comparative Analysis of thePharmaceutical Industry," discussed in the1976 Annual Report, has now been revisedand submitted for publication.

Support for this work has been providedby a grant from the National Science Foun-dation.

Henry G. GrabowskiJohn M. Vernon

45

5. HUMAN BEHAVIOR AND SOCIAL INSTITUTIONS

IntroductionThe NBER's research on human behavior

and social institutions is carried on throughfive programs: health, income distributionand labor markets, law and economics,population and the family, and taxation andsocial insurance. The emphasis has beenon applying economic theory and empiricalanalysis to a wide variety of market and non-market activities that have important im-plications for human welfare. In the past thebulk of the research has focused on explain-ing differences across individuals andfamilies in earnings, fertility, health, andother variables, using large-scale microdatasets. Recently, there has been a moderateshift in emphasis, with some additional at-tention being given to the analysis ofchanges over time, using aggregate data.This approach has been used in some of theresearch on crime and health, but is mostevident in the work on taxation and social in-surance under Michael Boskin and in thefamily project, especially Robert Michael'sstudy of divorce.

The new family project, which waslaunched with two-year support from the Al-fred P. Sloan Foundation and the Lilly En-dowment, will try to deal with some of themost pressing and fundamental problemsfacing our society, namely, the changes inthe size, structure, character, and functionsof American families. The research plandeveloped by Gary Becker and RobertMichael, the project directors, calls for afive-year effort which will include studies ofmarriage and divorce, parent-child rela-tionships, and other aspects of householdliving arrangements. The goal is to increaseunderstanding of both the causes and theconsequences of recent changes in familyorganization. It is hoped that the earlyresults of this project and increasing publicrecognition of the importance of the problemwill result in additional research support.

During the past year findings from thefive programs of the Center for EconomicAnalysis of Human Behavior and Social In-

46

stitutions have been reported in twenty-onejournal articles (published or in press) andfifteen NBER working papers. In addition,studies by several staff members appear inThe Distribution of Economic Well-Being, anIncome and Wealth volume edited by F.Thomas Juster, published this summer andin "Research on Taxation," a Journal ofPolitical Economy supplement edited byMichael Boskin. Details on current research,new projects, and sources of support areprovided in the following program reports.

VictorR. Fuchs

Income Distribution and LaborMarkets

Introduction

More than ever before our society isconcerned with variations in income andwelfare among individuals and families. Theresearch program in this area seeks toincrease understanding of thesephenomena by a multifaceted attack onmany related factors such as labor supply,job mobility, migration, nonmarket work,and on-the-job training, as well as bydetailed examination of actual earnings pat-terns.

The underlying theoretical models em-phasize the role of human capital, broadlydefined, and seek to explain behavior in alife-cycle context. The empirical workmakes extensive use of large longitudinaldata sets with repeated observations on thesame individuals. Some of the studies giveparticular emphasis to the role of women inboth market and nonmarket production.Others highlight black-white differentials.Some focus on mature adults; others onyoung people at the early stages of theircareers. All contribute substantially to ourknowledge of labor markets and income dis-tn bution.

Research output from this program areaduring the past year includes seven NBERworking papers:

Ann P. Bartel and George J. Borjas. 'Mid-

die-Age Job Mobility: its Determinantsand Consequences." Working Paper 161,January 1977.

George J. Borjas and Jacob Mincer. "TheDistribution of Earnings Profiles in Longi-tudinal Data." Working Paper 143, August1976.

Reuben Gronau. "Who Is the Family's MainBreadwinner?—The Wife's Contributionto Full Income." Working Paper 148, Sep-tember 1976.

Edward Lazear. "Family Background andOptimal Schooling Decisions." WorkingPaper 141, July 1976.

"Intergenerational Externalities."Working Paper 145, August 1976.

"Wage Differentials Are Larger ThanYou Think." Working Paper 168, March1977.

Lee Lillard and Yoram Weiss. "Experience,Vintage and Time Effects in the Growth ofEarnings: American Scientists, 1960—70."Working Paper 138, May 1976.

In addition, the following papers have been,will be, published:

Reuben Gronau. "Leisure, Home Productionand Work—The Theory of the Allocation ofTime Revisited." Journal of PoliticalEconomy, October 1977. (Working Paper137.)

Edward Lazear. "Age, Experience, andWage Growth." American EconomicReview, September 1976. (Working Paper151.)

"Schooling As a Wage Depressant."Journal of Human Resources, Spring1977. (Working Paper 92.)

______•

"Academic Achievement and JobPerformance." American EconomicReview, March 1977.

• "Education: Consumption orProduction ?" Journal of PoliticalEconomy, June 1977. (Working Paper104.)

Lee Lillard. "Inequality: Earnings Vs. HumanWealth." American Economic Review,March 1 977.

Lee Lillard and Robert J. Willis. "DynamicAspects of Earnings Mobility."

47

Econometrica (forthcoming). (WorkingPaper 150.)

The center's research on income distribu-tion and labor markets is currently sup-ported by the National Science Foundationand the U.S. Department of Labor. Somesupport in previous years was provided bythe Rockefeller Foundation.

Labor Mobility and Earnings

VictorR. Fuchs

In our study of the determinants of earn-ings in longitudinal data we moved to anintensive analysis of local job mobility andof migration in the National LongitudinalSurvey (NLS) of men over 45 years of age.We also completed an analysis of the dis-tribution of earnings profiles in theColeman-Rossi Sample of men who wereless than 40 years old at about the sametime period (the late 1 960s).

I. Local Job Mobility. Though less thanat younger ages, substantial amounts of jobmobility take place at older ages. In the NLSsample of men aged 45 to 59 in the initial(1966) survey, 22 percent of respondentschanged employers in the five-year period1966—1971. Indeed, mobility appeared toincrease somewhat in the years near retire-me nt.

The separations involving job changewere about equally divided between quitsand layoffs. In studying job separations wenot only distinguished quits from layoffs, butalso job-related quits from other types ofquits (e.g., "personal reasons"). We esti-mated probabilities of separating from thejob with and without information on jobtenure and within classes of short and longtenure.

Some of the findings were:1. Wage levels prior to separation are in-

versely related to quits but positively tolayoffs, while the opposite appears to holdfor education. All effects are weaker in "per-sonal" quits, and the distinction betweenquit and layoff holds mainly in longer-termjobs (tenure of over three years). It is impor-tant to note that most separations occur inearly tenure, and their nature (quit, layoff)

does not depend on the type of prior separa-tion. These findings are consistent withhypotheses of job-sorting informationalactivities of workers and firms in early tenureand increasing worker specificity later on.

2. The positive relation between wagesand layoff appears to be largely an industryeffect. Industries with unstable employmentapparently compensate workers by payingthem higher wage rates.

3. Employment, earnings, and educationof wives are negatively related to husbands'labor turnover. Intrafamily substitution sug-gests that cause and effect here work in bothdirections.

4. Job change resulting from quits occursusually without intervening unemployment.However, about half of exogenous ("per-sonal") quits and almost all layoffs involveunemployment. More than half of layoffs didnot result in job change. These temporary(recall) layoffs affect primarily experiencedworkers, while permanent (involving jobchange) layoffs are largely restricted to newworkers, except, of course, in the case ofshutdowns. Workers temporarily laid offhave personal characteristics (tenure,wage, education) much more similar to thecharacteristics of those who remain em-ployed than to the people who change jobs.Temporary layoff appears to be a device tomaintain a stable experienced work force inthe face of unstable labor demand.

5. Persons who quit without interveningunemployment had larger immediate wagegains than stayers and than all other jobmovers. Quitting did not affect the sub-sequent wage growth on the new job. Thosewho quit because they were dissatisfiedwith their previous job had negative or zerogains in wages (relative to stayers), but theyreported significant gains in job satisfac-tion.

II. Migration. One of the novelties of ourresearch on migration is its emphasis on therelationship between job mobility andgeographic mobility. The availability ofdetailed longitudinal data sets enables us toshow how the determinants and conse-quences of an individual's decision to mi-grate depend on whether he quit his pre-

48

vious job, was laid off, or remained with thesame employer. Joint and conditionalprobabilities of migration and separationhave been estimated for the NLS MatureMen and the Coleman-Rossi Sample. Takentogether, the results from several data setswill provide a life-cycle picture of the deci-sion to migrate.

Some of the more important resultsderived from the NLS Mature Men data setare as follows: (1) The likelihood of an indi-vidual's migrating following a quit ortransfer is positively related to the level ofhis education, but the likelihood of his mi-grating following a layoff is not related. (2)There is a positive relationship between thewife's level of education and her husband'slikelihood of a job separation, but a negativeone between her education and his likeli-hood of migrating. Hence, her level ofeducation shows no relationship to hiscombined likelihood of migrating andseparating. (3) The probability of migratinggiven a quit is negatively related to thewife's education and the presence ofschool-age children. These two variableshave no effect, however, on the probabilityof migrating given a layoff. (4) Among thestrongest determinants of the decision to mi-grate are the length of stay in the current lo-cation, and the length of job tenure.

Ill. The Distribution of Earnings Profiles inLongitudinal Data. The longitudinal earn-ings profiles in the Coleman-Rossi Samplecovered the first sixteen years of workexperience. On average, the estimatedprofiles showed the usual concavity and agreat deal of variation in levels, slopes, andcurvatures.

Estimates indicated that if slopes and cur-vatures of individual trajectories wereavailable to analysts, an additional 20percent of the relative variance in monthlyearnings could be explained beyond theusual power provided by the cross-sectionalearnings function approach. The decom-position of the slope and curvaturecoefficients into investment ratios and rate-of-return parameters provided a smallerincrease in explanatory power because oferrors introduced by the procedure.

Estimates of individual capacities withinschooling groups were obtained from theresidual of the schooling regression at the"overtaking stage." The results indicatedthat individuals with greater investmentratios grew more rapidly than others, and—holding earnings capacity constant—hadlower initial earnings. It was also found thatvariation in earnings capacity, which is apersistent characteristic of individuals, is atleast as important as variation in slopes andcurvatures of earnings profiles in terms ofexplanatory power.

Further analysis indicated that the indi-vidual coefficients and parameters of theearnings profile are weakly, if at all,associated with individual, personal andbackground characteristics. The role ofpost-school investment parameters in earn-ings remains strong even after all the avail-able personal information is utilized. Al-though the background variables stronglyaffect educational attainment, they show lit-tle or no relation to the personal accumula-tion of post-school human capital.

We are currently extending our work to theyounger NLS cohorts in order to complete alife-cycle analysis. We are also resumingour work on the integration of labor supplyfunctions into the analyses of earnings func-tions.

Ann P. BartelGeorge J. Borjas

Jacob Mincer

The Allocation of Nonmarket Time

Only a relatively small fraction of people'stime is spent in theform of market work. Inrecent years there has been an increasinginterest in the way people spend the rest oftheir time and in the economic considera-tions affecting the allocation of nonmarkettime.

In "Leisure, Home Production and Work—The Theory of the Allocation of TimeRevisited" (due to appear in the October1977 issue of the Journal of PoliticalEconomy) I try to formalize the trichotomy ofwork in the market, work at home, andleisure. Time is used at home to producehome goods that are perfect substitutes for

49

market goods, where home production issubject to diminishing marginal produc-tivity. An increase in the market wage rate isexpected to reduce work at home but the ef-fect of the increase on leisure and work inthe market is indeterminate. An increase inincome increases leisure, reduces work inthe market, and leaves work at home un-changed. These conclusions are supportedboth by empirical tests based on theMichigan Income Dynamics data and byprevious time budget studies. Further im-plications for labor supply, fertility, gainsfrom marriage, demand for child care, andthe measurement of home output are alsoinvestigated.

This work is extended in "Who is theFamily's Main Breadwinner?—The Wife'sContribution to Full Income" (NBER WorkingPaper 148) where I try to estimate the valueof the wife's home production, allowing fordiminishing marginal productivity of work athome. Preliminary findings indicate that thevalue of home production of the Americanwife is about four times her earnings.Adding this value to her market earnings,she contributes, regardless of education, al-most as much as her husband to the family'sfull income. The wife's share in the family'sfull income is about 45 percent, and thisshare comes very close to one-half when thefamily has young children, Needless to say,since the value of home production is nottaxed, the wife's contribution to net full in-come (full income after taxes) exceeds thatof everybody else in the family. Employedwives contribute a greater share than the notemployed, but at least in the case wherethere are young children in the family, thedifference in the contribution is small.Whether or not she is employed, the wife'sshare peaks early in life (by the age of 35)and drops significantly thereafter (though itmay pick up somewhat in late middle age).

Reuben Gronau

Dynamics of Female Labor Supply andWages

This research analyzes three principalproblems. The first problem analyzed is thespecification of dynamic labor force partici-

pation functions for women. This workgeneralizes previous work by Robert J.

Willis and me to consider the effects ofhuman capital and job search investmentson the labor market turnover of women. Themodel also allows for change in explanatoryvariables and the effect of macroeconomicvariables on labor supply. A new statisticalmodel is proposed and estimated by whichapparent state duration effects on participa-tion due to population heterogeneity can bedistinguished from the effect of humancapital (or true state duration) on labormarket turnover. The methodology is ap-plicable to a wide variety of related dynamicproblems of the labor force such as the ef-fect of unemployment duration on theprobability of employment. This researchwill be presented at the December 1977meetings of the American EconomicAssociation. A copy of the paper is avail-able on request to the author.

The second problem analyzed is thespecification of labor supply functions forwomen. A full life-cycle model of hours ofwork, participation, and wage growth isformulated and estimated.

The third problem analyzed is thespecification of the wage dynamics ofwomen, A more careful analysis of the wagegrowth of women reveals that previous worksubstantially overstates the effect of labormarket experience in reconciling male-fe-male wage differentials.

Family Considerations in theDetermination of Earnings

James J. I-Ieckman

This year was spent working primarily onthree papers: "Family Background andOptimal Schooling Decisions," "In-tergenerational Externalities," and WageDifferentials Are Larger Than You Think"(NBER Working Papers 141, 145, and 168,respectively).

The first paper is an attempt to determinewhether or not capital markets for the financ-ing of schooling are perfect and invariantacross groups. Because attained levels ofschooling are positively correlated with in-come, some researchers have suggested

50

that the ability to finance the acquisition ofeducation internally tends to perpetuateschooling and income inequalities acrossgenerations. Capital cost differences, it isargued, may be a crucial variable in theexplanation of education and income varia-tions. Using the model presented in thispaper, capital cost differences can betested across income groups. Other thingsconstant, no evidence of any difference wasfound.

The second issue dealt with in the familybackground paper relates to optimization.Even if capital costs do not differ across in-come groups, information about costs andreturns to education may. The differences ininformation among individuals maythemselves well be optimal, given dif-ferential costs and returns to the information.This will appear as a larger variance aroundthe 'perfect information optimum" level ofeducation for less-informed groups. That is,all things considered (including informa-tion), observed variations in schoolinglevels should be optimal. It is interesting toask, however, how variations in schoolinglevels would look in a setup where allpossessed the same amount of ex-post in-formation on costs and returns to schooling.The model provides an answer to this ques-tion as well. It is found that groups for whichthe marginal returns to schooling are lowalso have larger variances in the differencebetween actual and predicted optimallevels of schooling. This is consistent withthe observation that low-wage groups investa smaller amount in information than high-wage ones. The finding therefore makessense. Individuals should optimally acquiremore education when both the returns to be-ing correct and the costs of making a mis-take are high.

The model is constructed to allow estima-tion of returns to education for each indi-vidual, based on the assumption that all in-dividuals face the same borrowing rates.Given costs and returns, an optimal wealth-maximizing level of education can be ob-tained for each individual. Differencesbetween actually acquired and wealth-maximizing levels of education can then be

calculated and it can be determinedwhether or not these residuals aresystematically related to background vari-ables. If, for example, the estimated wealth-maximizing level of education of low-in-come individuals is consistently larger thanthe actual level, it could be concludedeither that returns to schooling differbetween groups or that capital market dif-ferences exist. The model allows these twoexplanations to be distinguished. Since dif-ferential returns are caused by wage dif-ferences among groups, the wealth-maximizing level can take these labormarket variations into account.

Data from the National LongitudinalSurvey (1966—1969) on young men whowere no longer in school have been used totest the theoretical predictions of the model.Empirical findings are:

1. It appears that there is no discriminationamong particular groups with given en-dowments either via capital marketmechanisms or governmental provisionof schooling.

2. The average marginal cost and return toa year of schooling is about $7,000 incurrent terms. For nonwhites, it is onlyabout $5,700.

3. The value of schooling is more elasticwith respect to measured 10 than tomother's education.

4. The elasticity of schooling with respect tomarginal cost is, at the point of means,—1.8. It would require a subsidy of $318to induce the mean individual to attendschool for one more year.

5. Estimated wealth-maximizing and actuallevels of education vary positively withmother's education and 10 and nega-tively with family size. Other thingsconstant, actual and predicted wealth-maximizing levels of education arehigher for nonwhites than for whites.

6. If nonwhites faced the same labor marketconditions as whites, their attainedschooling levels would be lower thanthey actually are because the averagelevels of other endowment variables arerelatively lower for nonwhites.

51

In "Intergenerational Externalities," a newdimension is brought into the analysis. Up tothis point it has been assumed that personaloptimization is the only relevant issue; thatis, human capital models and the type ofmodel used in "Family Background andOptimal Schooling Decisions" are designedto maximize the wealth of the individual. Inthis second paper, the analysis is extendedby recognizing that the individual is amember of a family and that his wealth de-pends in large part on what his parents dofor themselves and for him directly.Optimization is considered from a socialpoint of view, and an investment is said tobe socially optimal if all parties involvedreap returns which exceed the cost of thatinvestment. Specifically, the problem takenup is that parents sometimes underinvest intheir children's health, education, andgeneral welfare as the result of an in-tergenerational externality. If the parentdoes not fully take his child's preferencesinto account, and if the child cannot,through contract, insure repayment to hisparents, then it is likely there will be a devia-tion between the socially optimal amount ofinvestment in that child and the privatelyoptimal amount. Through a model of generalhuman capital stock which maximizes thediscounted value of wealth over all genera-tions, it can be shown that the sociallyoptimal level of parents' investment in. thehuman capital of their children is greaterthan the privately optimal one, Thetheoretical framework is used to estimatethe size of these intergenerationalexternalities.

The data used in this analysis come fromthe Michigan Income Dynamics Panel Studyon 1,455 heads of households. From thesedata, one can ascertain the extent to whichparents underinvest in their own educationbecause they fail to consider spillovers totheir children. The results show that al-though all groups may tend to underinvestfrom the socially optimal viewpoint, the dis-tribution of underinvestment is not uniformacross groups. Low-income, and especiallyblack, men are more likely to underinvest intheir education from a social point of view

than are whites and wealthier individuals.Note the difference between this finding andthe one in the first paper, where it is arguedthat private optimization across individualsdoes not result in there being different levelsof schooling as the result of different borrow-ing costs. In the second paper it is assumedthat private markets are efficient. The resultof the first paper is taken for granted, but it isthen argued that, given that efficient market,deviations between social and privateoptima vary across groups. It is estimatedthat a subsidy of $9,600 would be requiredto induce the average black man to invest inthe socially optimal level of education ifnone of the intergenerational spillover ef-fects from parents are considered.

The rest of the past year was spent on ananalysis which takes into account the effectof family background on the measurement ofwage rates. Specifically, this research,reported in NBER Working Paper 168, inves-tigates wage differentials between blackand white workers, where wages are definedto include the on-the-job trainingcomponent of earnings. This unobservedcomponent can be estimated by looking atthe capitalized value of wage growth.

Using data from the National LongitudinalSurvey on young men aged 14—24 in 1966, itwas found that the normally measured wagedifferential was 58 cents per hour. The 'true"differential, which includes the capitalizedvalue of wage growth, was almost threetimes as large. About two-thirds of the unob-served and one-half of the total wage dif-ferential could be eliminated by bringingthe average level of blacks' schooling, cur-rently at about 1 0 years, to that for whites, atabout 11.5 years. This result is obtained pri-marily because on-the-job training seems tobe complementary with level of schoolingattainment. Hence, other things equal,highly schooled individuals have steeperage-earnings profiles.

By using the National Longitudinal Surveyof the high school class of 1972, true wagedifferentials in 1972 can be compared withthose obtained from the 1966 survey.. Themost important finding is that the narrowingof black-white wage differentials in

pecuniary terms which occurred between1966 and 1972 was almost exactly offset byan increased differential in on-the-job train-ing. Thus it appears that employers haveresponded to pressure to narrow wage dif-ferentials by increasing differentials in aform of payment that is more difficult to ob-serve.

Edward P. Lazear

Earnings and Occupational Prestige overthe Life Cycle

This research considers the lifetime pat-tern of earnings and occupational prestige.It examines both the effect on lifetime pat-terns of schooling, ability, and family back-ground differences and the degree to whichindividuals differ permanently, over a 25-year period, in earnings and occupationalprestige. It also considers the implicationsof these sources of earnings variation forinequality in human wealth, defined as thepresent value of lifetime earnings. Many ofthe results for earnings are summarized in"Inequality: Earnings Versus HumanWealth." A more complete analysis is pro-vided in "An Essay On Human Wealth," andthe results were summarized in last year'sreport. Some new results are summarized in"Earnings and Occupational Prestige overthe Life Cycle," in which several novelissues can be addressed owing to theavailability of longitudinal data, First, the ex-post lifetime patterns of occupationalprestige and earnings are compared formen from the U.S. birth cohorts of 1917—1925. The lifetime patterns of each areinfluenced in different ways by years ofschooling completed, measured 10-typeability, and early family background.Second, individual earnings and occupa-tional prestige are affected by some charac-teristics the researcher is unable to observe.I estimate permanent, transitory, and seriallycorrelated components of the variation ofearnings and prestige, each net of thepredicted lifetime profiles. Individuals whoare observationally identical are found tohave important unmeasured permanent andserially correlated residual components inprestige as well as earnings. To put the

52

issue slightly differently, I address the issueof the strength and sign of the correlation ofearnings and prestige residuals among in-dividuals observed repeatedly over a periodas long as twenty-five years. As expected,the correlation for both earnings andprestige is a declining function of the timedistance between two observed years orpoints in the life cycle. However, the correla-tion is always strongly positive, even aftertwenty-five years.

The existence of these important indi-vidual differences among observationallyidentical individuals raises the followingthird issue. Are those individual earningsand prestige differences positively or nega-tively correlated? That is, if an individual isobserved as permanently above the lifetimeearnings profile predicted for individualsobservationally like him, is he likely to bepermanently above or below the cor-responding lifetime prestige profile? I findsubstantial positive correlation. Hence, de-viations from estimated profiles are not com-pensatory.

A similar issue relates to transitory earn-ings and prestige deviation, which arethemselves serially correlated. If an indi-vidual is transitorily shocked above (below)his own projected profile, is his occupa-'tional prestige more likely to be shocked inthe same direction or in the opposite direc-tion? Again I find a positive correlation.These across-equation results must,however, be interpreted with caution, sinceoccupation may signify more than simplyprestige. These are two reduced form equa-tions among many which might bedeveloped with respect to occupations.

Lee A. Li/lard•

A Study of Earnings Dynamics

In the past year our research has focusedon the development and implementation ofan econometric methodology that may serveas a link between theory and panel data onseveral dynamic aspects of labor market be-havior. The issues we address jointly and in-dividually in various subject areas include(1) dynamic aspects of earnings distribu-tions and poverty analysis, (2) simultaneous

53

estimation of long- and short-run labor sup-ply elasticities, (3) the role of hours andwages in earnings variation over time, and(4) covariation over time in the earnings ofhusbands and wives in two-earner families.We will discuss these in turn.

In the first study, "Dynamic Aspects ofEarnings Mobility" (NBER Working Paper150), the econometric methodology for theanalysis of earnings mobility is used to de-termine whether poverty is a transitory orpermanent condition of individuals. It hasthe advantage of providing a direct linkbetween traditional human capital theory,earnings functions, and Markov chain typemodels. The methodology is fairly simple.First, we estimate an earnings function withmale earnings as the dependent variable,using the seven years of panel data. Weestimate permanent, transitory, and seriallycorrelated components of earnings due toboth measured and unmeasured variables.Measured variables are represented bycomponents of variance. The estimatedearnings function and the estimatedcomponents are then used to compute theprobability that an individual's earnings willfall into a particular, but arbitrary time se-quence of discrete earnings classes. Forsimplicity, we focus on the probability thatan individual's earnings fall below an arbi-trary poverty line defined as one-half ofmedian U.S. male earnings each year,1 967—1 973. With these techniques, indexesof income immobility or the degree of"persistence" in individual income 'evelsare developed for various subgroups of thepopulation (e.g., by race, experience, andeducation). Such measures include the pro-portion of earnings variation (gross variation

variation around a predicted level)among individuals, which represents a

permanent difference; the degree to whichearnings are correlated from year to yearforthe same individual; and the rate of decay ofcorrelation around the permanent level.

Analysis of the variance components ofthe log of annual earnings of men (in 1970dollars) indicates that 73 percent of totalvariance in log earnings representspermanent earnings differences. The

remaining transitory earnings variation is

serially correlated, with a coefficient of 0,4.The magnitudes of these earnings variancecomponents have important implications forthe study of poverty and the stability of thepoverty population.

These points are illustrated by consider-ing the poverty dynamics of blacks. Usingour definition of poverty, about 14 percent ofblacks were predicted to be in poverty in1967. Of those who were in poverty in 1967,65 percent were expected to be in poverty in1968, while of those not in poverty in 1967,only 4 percent were predicted to be inpoverty in 1968. The importance ofpermanent earnings components isillustrated by the persistence of these ef-fects. For example, of those in poverty in1967, the probability of poverty in 1973 is 44percent, while of those not in poverty in1967, the probability of poverty in 1973 isless than 3 percent.

The second issue is the development ofthe previous earnings model from an un-derlying dynamic labor supply function inwhich permanent, transitory, and seriallycorrelated variations in an individual's (log)wages induce long- and short-run labor sup-ply responses. An analysis of covariationover time of (log) annual hours and (log)earnings in panel data yields estimates ofboth the long- and short-run labor supplyelasticities.

This earnings model in turn provides thebasis for studying the third issue, the role ofhours and wages in the variation of earningsover time. For example, not only can we de-compose cross-sectional earnings varianceinto permanent and transitory parts, but wecan also determine the extent to whichexogenous wage variation, exogenoushours variation, and induced hours variationare responsible for permanent and transitoryvariation in earnings. The intertemporal dis-tribution of earnings will be examined bycalculating predicted and actual transitionprobabilities among discrete earningsclasses defined both in relative andabsolute terms.

The above earnings analysis is used forboth husbands and wives in the final study

topic, in which the emphasis is on a com-parison of permanent versus single-periodearnings inequality among families and therole of within-family variation for two-earnerfamilies. The objective is to discoverwhether husband's and wife's earnings arecompensatory or not. Given that both ofthem work throughout the period, weanalyze the correlation between both thepermanent and transitory earnings of each.If the aim of couples is a combined familyincome of a certain level, then permanentearnings of pairs will be negatively relatedand compensating and family incomes willbe less unequally distributed than, say, hus-band's earnings. If the two permanent earn-ings components are positively related, theninequality among family incomes will beeven greater than among husband's earn-ings. Also transitory variation in the earningsof one spouse may induce a response by theother. Hence, the dynamic labor supplymodel can be developed to include bothhusband and wife.

Lee A. Li/lardRobert J. Willis

The Effect of Minimum Wage Legislationon Income Equality

While at the Bureau I have corrected,extended, and rewritten for journal publica-tion an earlier paper of mine on thetheoretical effect of a minimum wage law onincome equality under a variety of assump-tions. A one-sector model of general equi-librium is used to analyze a universally ap-plicable minimum wage, and a two-sectormodel is used to analyze a minimum wagethat is only applied to certain industries. Inboth cases I find that a minimum wage maywell lower equality (as computed by the Giniindex) if we consider reasonable values forthe parameters of these two models. In theabsence of unemployment compensation,equality can increase only if the elasticity ofsubstitution in production is quite low. In theone-sector case, however, equalitynecessarily rises if unemployment com-pensation is present and sufficientlygenerous.

54

J. Huston McCulloch

Population and Family EconomicsFamily Economics

During the past year much of the Center'swork on family economics has beenrefocused under a new several-year project,"Change and Stability in the AmericanFamily." The co-directors are Gary Beckerand Robert Michael and the staff includesVictor Fuchs, Reuben Gronau, and EdwardLazear. Larry Kenny, from the University ofFlorida, will join the project as a visitingscholar for the 1977—1978 year and otheradditions to the staff are expected. Theproject is funded by grants from the AlfredP. Sloan Foundation and the Lilly Endow-ment, Inc.

The project is motivated in part by the pro-found changes in recent years in severalaspects of family behavior—the well-pub-licized decline in fertility, the rise inwomen's labor force participation, the rise inage at marriage, and the increase in divorcerates. It is also motivated in part by ourassessment that it is an appropriate time toapply to time series behavior the insightsobtained from cross-sectional studies offamily behavior. In addition to extending theanalytical work on the family conducted atNBER and elsewhere in recent years, we ex-pect the project to enrich descriptive in-formation about the nature of recentchanges in the family.

One objective of the project will be toassess the advantages to adults of living inthe context of a family. Despite the rise in di-vorce rates and median age at first mar-riage, it appears that a vast majority of indi-viduals still choose to spend a dominantportion of their adulthood in a family. Pre-sumably this form of living arrangement ischosen because of its value to the individ-ual, and we intend to estimate the mag-nitude of some of the benefits of living in afamily and consider how these benefits maybe changing in the context of developmentsin social customs and government policy. Inone study Kenny will attempt to measure theimpact of family status on the labor marketproductivity of men, using longitudinal data.In another, Lazear and Michael are attempt-

55

ing to measure the economies of scale inconsumption attributable to family size, Dif-ferences in health status and perhaps nutri-tional adequacy of diets also appear toreflect an advantage of family membership,and the project may involve studies of thesefacets as well.

The concept of "family," the relationshipbetween it and other multiperson living ar-rangements, the various social functionsperformed by it, and the changes in thesefunctions over time and across societies willbe explored by Becker, The bearing andraising of children represent one of the im-portant aspects of family behavior, and weexpect to include several studies onchildren in this project. For example, Lazearis beginning a study on differential expendi-tures on children within families, designedin part to indicate whether parents tend toreinforce or compensate for differencesamong their children in ability, health, andthe like.

A second focus of the project will be onthe relationship between the family and thedistribution of income. In this area thestudies we expect to undertake include thework on economies of scale, mentionedabove, the interaction between the laborsupply of husbands and wives, and thechanges over time and differences betweenthe distributions of income when definedamong earners or among all individuals andwhen defined among families or amongfamilies and "unrelated individuals."

The staff will engage in severalessentially descriptive analyses, attemptingto document better the changes andstability in postwar family behavior and theircauses and repercussions. One such effort,for example, involves the construction fromvital statistics data of the fractions of 18-year-olds in each year who were first-bornchildren, second-born children, etc.—a setof time series which are, of course, affectedby the baby boom but which contain a quitedifferent time pattern. Another time series isan estimate of the fraction of adult lifetimespent living outside any family. We hopethat the set of series under construction willprovide more economically meaningful in-

formation about relevant changes takingplace and will also be usable in subsequentstatistical time series analyses. Fuchs andMichael are currently beginning this work on'family accounting."

By way of an update on the previousyear's reports on the study of divorce, twoNBER working papers were completed thispast year: Michael's "Factors Affecting Di-vorce: A Study of the Terman Sample" (no.147, August 1976) and Becker, Landes, andMichael's "Economics of Marital Insta-bility" (no. 153, October 1976). Michaelis currently engaged in a time seriesanalysis of U.S. divorce rates over the pe-riod 1920—1975. The findings to date sug-gest that about two-thirds of the rise in di-vorce in the 1960s was among couples inwhich the woman was in her twenties. Thechange in contraceptive technology ap-pears to be the single most important"cause" of the recent rise in divorce, but therise in women's labor force participationdoes not seem to be a major determinant ofthat increase.

Population Economics

Robert T. Michael

The program in population economics,under my direction, includes a number of re-search activities that are complementary tothe family economics program. These corn-plementarities are especially apparent in re-search supported by the Ford Foundationand the U.S. Agency for InternationalDevelopment (USAID) concerning eco-nomic and demographic aspects of familybehavior in less developed countries. In

theoretical workin progress, I am investigat-ing the role of the family as an economic in-stitution in situations in which other privateand governmental institutions are poorlydeveloped. A major goal of this work is toinvestigate the implications of thehypothesis that children are "the poor man'scapital" (Mahmood Mamdani, The Myth ofPopulation Control.' Family, Caste and Classin an Indian Village, New York, MonthlyReview Press, 1972)—a source of labor, ofsecurity, and of old-age support—for fertilitybehavior, saving, investment, and other

aspects of household behavior, and for thedevelopment of economic activities in or-ganized markets.

Several empirical studies of householdeconomic and demographic behavior usingmicrosurvey data from several developingcountries are under way or planned for thecoming year. A preliminary draft of an empi-rical study, "Sharecropping and Family Sizein the Brazilian Northeast," has been com-pleted by Anna Luiza Ozorio de Almeida,who was a visiting scholar at NBER in PaloAlto from the National Institute of Economicand Social Research (INPES-IPEA) in Rio deJaneiro, Reuben Gronau is working out aprocedure for measuring the value of non-market time in developing countries, and Iam planning a study of life-cycle aspects offertility and female allocation of time, usingsurvey data from Guatemala.

Law and EconomicsIntroduction

Robert J. Willis

The research activity during the past yearincluded the following studies: IsaacEhrlich's continued cross-sectional andtime series analysis of deterrence andcapital punishment in the United States;Kenneth Wolpin's study of crime in Englandand Wales from 1894 to 1967; WilliamLandes's work on the deterrence of skyjack-ing; Landes and Richard Posner's continuedand expanded work on legal precedent andtheir new study of the law and economics ofrescue; Posner's study of the economics ofcontracts; B. Peter Pashigian's work on themarket for lawyers and his new study of theeffects of occupational licensing; and SamPeltzman's study of industrial concentrationand profitability and his new study of thesize of governments. Eight journal articleswere published this year (including one byPosner on the legal rights of creditors andanother by Peltzman on regulation that weredescribed in detail in last year's report) orwill be published shortly—Journal of Law &Economics (3), Journal of Legal Studies (2),American Economic Review (1), Journal of

56

Political Economy (1), and University ofChicago Law Review (1).

The law and economics program is sup-ported by grants from the National ScienceFoundation and the American Bar Founda-tion, which supports Pashigians researchon the legal profession.

William M. Landes

Studies on Punishment and Deterrence

During the past year I completed thefollowing papers, which were subsequentlypublished:

The Deterrent Effect of Capital Punish-ment—A Reply." American EconomicReview, June 1977.

"On the Measurement of the Deterrent Effectof Capital Punishment and the Theory ofDeterrence." Journal of Legal Studies,January 1977.

'Capital Punishment and Deterrence; SomeFurther Thoughts and Additional Evi-dence." Journal of Political Economy,August 1977.

My monograph, tentatively titled "A Matterof Life and Death: Essays in the Economicsof Crime and Deterrence," is progressing.The book will contain five chapters plus anintroduction and a concluding chapter. Inthe introduction I view the work on deter-rence in a historical perspective, withreferences to previous work by classicaleconomists and the utilitarian school, and Ianalyze the challenge to the original work ofeconomists advanced by other socialsciences, starting in the late nineteenthcentury. In Chapter 1, I review past literatureon deterrence by criminologists who placeda major emphasis on the deterrent effect ofcapital punishment. Chapter 2 contains myearlier paper, "Participation in IllegitimateActivities. An Economic Analysis," (origi-nally published in the Journal of PoliticalEconomy), and Chapter 3 contains "TheDeterrent Effect of Capital Punishment—AQuestion of Life and Death" (published inthe American Economic Review). Chapter 4includes my new cross-sectional study ofcapital punishment and deterrence, whichis scheduled for publication intheJPE, sup-

57

plemented by additional theoretical work onoptimal enforcement of sanctions and a dis-cussion of trends in the application of thedeath penalty. Chapter 5 contains most ofthe work still in progress. It deals with thepooling of time series and cross-sectionaldata on murder and other crimes, the si-multaneity of offense and defense, a study ofthe effect of killings through police interven-tion, and analysis of demographic factors.

Isaac Ehrlich

Skyjacking and Deterrence

In this study I attempt to estimate theresponsiveness of airline hijackings tovarious deterrence measures (e.g., theprobability, of apprehension, the conditionalprobability of conviction, and the length ofthe sentence) and economic variables (e.g.,the unemployment rate) from time seriesdata on commercial airline hijackings in theUnites States and foreign countries since1961. I also seek to estimate the factors af-fecting the probability of apprehension andto assess the costs and benefits fromspecific procedures such as the skymarshal program in 1970—1973 and themandatory search of passengers that wasbegun in 1973. The data used for the studyare individual hijackings that include in-formation on the hijacking (time, place, typeof plane, destination), the characteristics ofthe hijackers (age, sex, race, number ofpersons involved), and the outcome, includ-ing information on the disposition of eachdefendant. Preliminary results showsignificant deterrent effects on the fre-quency of hijackings of the various deter-rence variables, and a positive effect onhijackings of unemployment. These findingsare based on both a quarterly time seriesregression analysis of the U.S. data and adisaggregated analysis in which the fre-quency of hijackings is estimated by thetime interval between their successive oc-currences.

William M. Landes

Economic Analysis of Legal Precedent

A legal precedent is a previous decisionthat is relied upon as a basis for deciding a

subsequent case, From an economicstandpoint, the body of precedents createdby judicial decisions in prior periods is astock of capital that yields a flow of informa-tion services which depreciates over time asnew conditions arise that were not foreseenby the framers of the existing precedents.New capital is created by investment in theproduction of precedents. A capital theoryapproach to legal precedent yields testableimplications on the relationships among in-vestment, depreciation, and the capitalstock, and provides a basic framework forcollecting and organizing a great deal ofdata on precedent. The basic data are casecitations appearing in judicial opinions andestimates of investment based on publishedjudicial opinions in various substantiveareas of the law. These data make it possi-ble, to estimate depreciation rates, invest-ment, and the capital stock of precedents invarious areas. These estimates, in turn, canbe combined with other legal, social, andeconomic data and used to examine a widerange of questions, as illustrated by thefollowing examples.

1. Has the role of precedent in legal deci-sion-making declined, as many ob-servers believe, over the last 100 years?

2. Did judicial activism, associated with the'Warren Court" in the 1960s, lead to afaster rate of depreciation and a destruc-tion of legal capital?

3. Do areas of the law with relatively largeamounts of statutory activity have higherdepreciation rates and lower amounts ofjudicial capital?

4. Does a new statute lead to a depreciationof legal capital?

5. What is the relationship between the age-earnings profiles of lawyers and the de-preciation rate of legal capital?

6. What is the relationship between theamount of lower-court litigation, the rateof investment precedent, and depre-ciation?

7. the size of the capital stock and de-betwee'n

higher and Federalstate courts? Federal administrativeagencies and federal courts?

58

Although these issues have received littlesystematic analysis, their clarification iscentral to an understanding of the legalsystem and, ultimately, the relationshipbetween the legal and economics systems.We have completed the first of severalplanned studies on legal precedent. Ourinitial study, which was described in detailin last year's annual report and published inthe Journal of Law & Economics (August1976), was based on a sample of 658federal court of appeals decisions in 1974—1975 and a sample of 156 U.S. SupremeCourt cases during the 1974 term. Our on-going study is based on a sample of aboutseven thousand federal court of appeals de-cisions from 1892 to 1976, and a sample ofabout a thousand Illinois appellate courtcases from 1970 to 1976. In addition, we areplanning in future studies to develop dataon precedent utilization in a federaladministrative agency and possibly for aforeign country.

William M. Lan desRichard A. Posner

The Law and Economics of Rescue

We use the term "rescue" to cover all at-tempts to preserve the safety of a person orproperty, other than pursuant to a contract. Asynonym for rescue in this sense is "unbar-gained assistance." Some examples willillustrate the variety of settings in which aquestion of rescue may arise.

1. A physician renders medical assistanceto a person lying unconscious in thestreet. Can the physician collect for theservices rendered?The fnaster of a ship, in order to prevent it

sinking in a storm, jettisons some ofthe Is the owner of the jettisoned

entitled to any reimbursement byshipowner or other owners of cargo?

A ship tows a stranded ship to safety. Isthe rescuer entitled to a reward for his ef-forts from the owners of the ship andcargo he saved?

4. A drowning man cries for help, a strongswimmer ignores his cries, and hedrowns. Is the decedent's estate entitledto damages from the swimmer?

5. A man loses a wallet and posts a reward.Someone finds the wallet, returns it, andclaims the reward, but the owner refusesto pay. Has the finder a cause of actionagainst the owner?

The central feature of all these examplesis the existence of high transaction coststhat precludes the voluntary negotiation ofrescue contracts. Thus, in each of theseexamples (except 4), if the rescuer weresuccessful and later claimed a reward, hisclaim could not be rejected on the groundthat, had the rescue been a value-maximiz-ing exchange, the victim and rescuer wouldhave agreed in advance on the terms of thecompensation. In the first part of our studywe develop a model of a competitive marketin rescues where transaction costs areassumed to be zero. Although our primaryinterest is rescue situations where transac-tion costs are prohibitive, the competitivemodel can be used to predict what legalrules will emerge (assuming the legalsystem in this area is concerned withpromoting efficiency), and as a benchmarkfor evaluating the observed legal rules. Themodel adds a new element to the economicanalysis of the legal system by incorporat-ing the possibility of altruistic action, an im-portant factor in rescues and one that maysignificantly affect the choice of legal doc-trines to maximize efficiency in this area.

In the second part of the study we will ap-ply the model to a variety of specific legalproblems along the lines suggested by theabove examples. The principal doctrines weplan to analyze are the admiralty doctrinesof general average, salvage, and abandon-ment; the principles of the law of restitutiongoverning the conferral of benefits in casesof emergency or incapacity; common lawand statutory regulation of finders of lost ormislaid property; the Good Samaritan (no-duty) doctrine and its exceptions; and thelegal liability of rescuers.

We have found extensive published dataon salvage awards, from the eighteenthcentury to the present, made in admiraltycourts in the United States and Great Britain.The data include the size of awards, the

59

value of property saved, the capital involvedin the rescue, the amount of time involved inthe rescue, the type of rescue operations,etc. We plan to use these data to test thehypothesis that the actual setting of salvageawards is an attempt to simulate the condi-tions and outcomes of a competitive market.

William M. LandesRichard A. Posner

The Demand and Supply of Lawyers

During the past year, I have expanded thetime series study of the demand and supplyof lawyers. During the last fifty years, thelegal profession has experienced twoextended periods when the rate of return tolegal education was relatively high. The firstperiod covers the 1 920s and the early 1 930sand the second extends from the 1960s intothe first half of the 1 970s. Some tentativeexplanations of the length of these pros-perous periods are investigated. The timeseries evidence indicates real nationalproduct is the major determinant of thenumber of lawyers. Other less important de-terminants are the divorce rate and thenumber of court cases. As expected, thenumber of lawyers appears to decline withincreases in alternative earnings and oppor-tunities.

Historical trends on the lawyer-populationratios in Canada, England, and the UnitedStates show higher lawyer intensity in theUnited States than in Canada and Englandthroughout the period from 1920 to 1970 andcast doubt on the importance of regulationas a determinant of demand. While thenumber of lawyers is directly related to thenumber of federal district court cases com-menced, it is not related to the combinedbudgets of fifteen regulatory agencies.

The results of the study are reported inNBER Working Paper 158 and will appear ina forthcoming issue of the Journal of Law &Economics.

B. Peter Pashigian

The Effect of Occupational Licensing ofProfessionals on Interstate Mobility

I have begun a study on the effects of li-censing and reciprocity on interstate mo-

bility of professionals with particular em-phasis on the legal profession. Theinterstate mobility rate of lawyers may below because of (1) the bar examination andlimitations on reciprocity or (2) investmentsin state law and procedure undertaken dur-ing law school or during practice. Somepreliminary results have been obtained in astudy of thirty-four professional occupationscomposed primarily of male workers withhigh educational attainment. Occupationswithout licensing are compared withlicensed occupations with reciprocity andthose with little reciprocity. Licensed occu-pations with reciprocity have lower mobilityrates than unlicensed ones. Limitations onreciprocity reduce interstate mobility ofhuman capital still more.

If state-specific law is taught in lawschools, it should be reflected in a lowinterstate mobility rate for law professors.However, the interstate mobility rate of lawprofessors is relatively high and similar tothe rates of professors in other fields, If in-vestments in state-specific law are un-dertaken through practice, the interstate mo-bility rate of lawyers should decline morerapidly with age than the rates of those inother occupations. However, it does not ap-pear to do so after controlling for the effectsof private practice.

B. Peter Pashigian

The Gains and Losses from IndustrialConcentration

The positive relationship between in-dustry profitability and concentration is oneof the more durable empirical regularities inthe industrial organization and antitrustliterature. Most analysts interpret the rela-tionshi p as implying supercompetitiveprices in concentrated industries; othersargue that it reflects the costs of unusuallyefficient firms which come to dominateconcentrated industries. In my study, I try tosort out these contentions empirically. I

examine changes in prices, costs, and con-centration in 165 four-digit SIC industries. Amodel is developed in which price changesreflect cost changes and concentrationchanges, but in which cost changes can

60

also be affected by concentration changes.The principal empirical finding is that hold-ing cost changes constant, prices rise morewhere concentration increases, but thatcosts fall where concentration increases.The second effect dominates the first;hence, on balance, prices fall whenconcentration increases.

These findings imply that legal hostilitytoward concentration may have beenmisguided by the traditional view thatconcentration is solely an index of collusioncosts. The magnitudes of the effects I haveestimated imply that a vigorous anti-concentration policy can be expected tohave a net resource cost of several billiondollars per year. This paper is available asan NBER Working Paper 163 and will bepublished in the Journal of Law & Eco-nomics,

The Size of Government

Sam Peltzman

For most of the twentieth century, anincreasing share of national income hasbeen channeled through governmentspending almost everywhere in the world. Inaddition, government policies (e.g., regula-tion) appear to have had an increasing im-pact on nongovernment spending, thoughthis impact is difficult to measure satisfac-torily. To some extent, this growth of govern-ment may reflect growth of demand forpublic goods. However, this is unlikely to bethe whole story. The rapid growth in directtransfers and subsidies for private goodssuggests that distributive goals havesignificantly affected the political process.This project will attempt to establish whatthese distributive goals are and how far theytake us in explaining the growth and size ofgovernment.

The underlying assumption of the study isthat the political process will be pushedtoward establishing a "politically efficient"distribution of wealth, i.e., one that eithermaximizes the power and security of currentofficeholders, or, more simply, dominates inthe competition for office. The hypothesis tobe tested is that government's role in theeconomy is larger when more wealth has to

be redistributed to attain the politicallyefficient distribution. This leads to an ex-pected relationship between the actual dis-tribution of income and the growth ofgovernment, with government spending mo-tivated by discrepancies between the actualand efficient distribution.

The study is at a preliminary stage inwhich the model is being developed and themost easily accessible income distributionand government spending data (states ofthe United States) are being processed. Themodel has so far distinguished two ways inwhich existing income distributions can af-fect the growth of government. First, if a fewindividuals get a large fraction of total in-come, there will be political gains to taxingthem. But less obvious, government maythrive where some forms of equality areincreased. For example, the development ofa numerically large group with similar in-comes makes it attractive to spread thescope of government and also its size.

I intend to develop the model more fully toaccount for the role of education (andinequality in education) and, if the pre-liminary empirical work is encouraging, I

will apply the analysis to a variety of his-torical and cross-sectional data.

Economic Analysis of Contracts

Sam Peltzman

During this past year, I completed a paperwith Andrew Rosenfield that analyzes im-possibility and related doctrines in the lawof contracts from an economic standpoint.The paper appeared in the Journal of LegalStudies, January 1977. The defense of im-possibility" allows a contract promisor, insome circumstances, to excuse his failure tocomplete performance of the contract.Death and war are examples of circum-stances that have been used to excusecontract performance on this ground.

Our study examines judicial doctrinesand the outcomes of reported cases to de-termine whether the courts excuse contractperformance where (and only where) theresult is efficiency-enhancing because thepromisee is the superior risk bearer in the(economic) sense in which we define this

61

concept. We find that on the whole thecourts do employ an implicit economiclogic in deciding these cases.

We are now collecting form contracts(printed in all states) in order to conduct amore systematic empirical analysis of theeconomic properties of the impossibilitydoctrine and other doctrines in contract law.Contracting parties can by expressagreement normally override a judicial ruleof contract law. Accordingly, where statesdiffer in the judicial rule, we would expectcontracts in states that have rejected theefficiency rule to waive its application. Thisis. the hypothesis that we hope to test bycollecting and examining the form contractsprinted in the various states.

I am also working on a study of so-calledgratuitous promises. I seek to give an eco-nomic explanation for such promises and todetermine whether the courts enforce themwhen economic theory indicates thatenforcement would enhance efficiency.

Richard A. Posner

An Analysis of Crime in England andWales: 1894—1 967

I have recently completed a manuscripton crime and punishment in England andWales, spanning the period 1894—1967. Thepurpose of this research has been to gainfurther understanding of criminal behaviorand sanctions. Using a model of optimal lawenforcement, I attempt to explain the move-ments over time, both within and betweenspecific types of offenses, of variables thatare under societal control. The time seriespattern that emerges is as follows: The pro-portion of crimes cleared and the proportionof convictions leading to imprisonment havefallen, the latter quite dramatically; actualtime served has remained roughly constant;and conviction rates have risen. As betweencrimes, the time series generally show thatthe more serious crimes entail greater risk ofcapture and more severe penalties. Most ofthese observations are shown to beconsistent with the economic framework.

In the second part of the paper, theresponse elasticities of offenses to changesin the law enforcement variables are esti-

mated in a regression framework. After con-trolling for several demographic and eco-nomic characteristics (e.g., unemploymentarid urbanization), negative relationshipsare found between the law enforcementvariables described above and offenserates. An attempt is made to isolate thedeterrent from the incapacitation effect ofimprisonment. One approach is to comparethe effect on crime of a noncustodial penalty(e.g., recognizance) relative to imprison-ment. Since the former does not inca-pacitate, its sole effect must be as a deter-rent and since it is a less severe penalty, itmust deter less than imprisonment. As alower bound, it is found that, dependingupon the specific crimes considered, thedeterrent effect of imprisonment is as largeas 70 percent of the total effect.

An alternative approach to separatingdeterrence from incapacitation is based onthe proposition that unanticipated changesin law enforcement variables can only haveincapacitation effects. Assuming that theforecast of a variable is based only on thepast behavior of that variable, estimates ofunanticipated changes are obtained usingtime series methods. Preliminary resultshave been somewhat ambiguous, some-times yielding smaller deterrent effects thanthe first method, I am now making partialcorrections for the omission of other vari-ables that may be relevant to the formationof anticipations.

Kenneth I. Wo/pin

Taxation and Social InsuranceIntroduction

The program on the economics of taxationand social insurance is now entering itssecond year. Research on taxationcontinues to be funded by a grant from theU.S. Treasury Department. A major effort inthe study of many aspects of social in-surance began in July with the award of acontract from the Department of Health,Education and Welfare. The award sup-ported a workshop on social insurance,which was held at the Palo Alto office of theBureau in January and was attended by fifty

62

economists working in this field. Eightprogress reports were presented for dis-cussion, in addition to informal discussionof their current work by most participants.

Taxation

Work in the taxation field included the(forthcoming) publication of my paperTaxation, Saving and the Rate of Interest,"

in the Journal of Political Economy this yearand the presentation of my paper, "OptimalTax Theory, Econometric Evidence, and TaxPolicy" (NBER Working Paper 152) at theInternational Economics Association Con-ference on Contributions of Econometrics toPublic Policy, held in September 1976. Inthe latter paper, attention is drawn to the im-plications for tax structure of different laborsupply elasticities among primary andsecondary earners in households and of theinterest elasticity of saving.

Joseph Stiglitz and I presented "SomeLessons from the New Public Finance" atthe American Economic Associationconvention in September, 1976; the paperwas published in the American EconomicReview, Papers and Proceedings, February1977. In that paper we summarized recentadvances in public finance and their rela-tionship to traditional theory, including theway it has been misapplied in second-bestsituations. We showed the application of anoptimal taxation framework to the analysis oftax credits and deductions.

Eytan Sheshinski and I are working on thetax treatment of the family, analyzing a va-riety of efficiency and equity aspects of al-ternate ways of treating the family. The effectof taxation on the labor supply of husbandsand wives is one of these. Another aspect isthe design of optimal tax treatment when thebivariate distribution of the ability of hus-bands and wives is taken into account. Weare also looking at the design oftax exemptions and credits for family sizewith regard to equivalent consumptionscales,

In the study of the effects of taxation ofhuman capital, I am now finishing twopieces of work. One is a theoretical analysisof the optimal tax treatment of physical

capital under alternate constraints on the taxtreatment of human capital. The other is anempirical examination of the effects of taxeson human investment, using U.S. time seriesdata.

Social Insurance

In the social insurance area, the data fromthe first four years of the Social SecurityAdministration's Retirement History Surveyhave just been received; we are now work-ing on a wide-ranging study of the economicbehavior of the aged. One major questionbeing examined with the help of these datais the effect of social security on the laborsupply of the elderly. Another topic understudy is the pattern of income and expendi-tures of the elderly. The aim of this examina-tion is to obtain a more accurate and com-plete picture of their economic welfare.Michael Hurd, Martin Feldstein, SherwinRosen, Lee Lillard, and I are among thoseworking on aspects of this study.

Michael Hurd and I are working on themeasurement of the labor supply response.We are using cross-sectional and paneldata for preretirement and retirement age in-dividuals to measure this and have de-lineated the sources of potential bias arisingfrom each kind of data. My earlier paper,"Social Security and Retirement Decisions,"in which I develop a suitable regressiontechnique and apply itto a subsample of theUniversity of Michigan's Panel Study of in-come Dynamics, has now been published inEconomic Inquiry, January 1977. Our cur-rent work on early retirement embodies bothmethodological and data improvements.The analytical method focuses on theprobability of early retirement between theyears 1969 and 1971 asafunction of severalvariables, including parameters of theSocial Security system.

Eytan Sheshinski, Jerry Green, JohnShaven, and I are developing an appro-priately realistic model to explore implica-tions for the economy of changing popula-tion patterns, with special reference to theireffects on public income maintenance andsocial insurance programs. Among the fea-tures important for such a model are mdi-

63

vidual lifetimes composed of four distinctand variable-length periods, a nonhomo-geneous labor supply, and the existence ofseveral public programs which variously taxworkers and pay benefits to nonworkers.

A variety of other, smaller projects are un-derway on topics ranging from unemploy-ment insurance to private pensions. Somehave been commissioned on a pieceworkbasis; others are being studied by regularNBER staff members. For example, SherwinRosen is studying unemployment insuranceand employment-income risk. He is devel-oping a market equilibrium analysis of thisproblem and will test it empirically nextyear. Martin Feldstein has been working onbasic conceptual issues in analyzing publicfinancing of programs across generations.Lee Lillard is working on short- versus long-term adjustments in labor markets, particu-larly labor supply, a subject generallyneglected in analyses of labor markets butobviously important in understanding howlabor markets for the elderly (who, by defini-tion, have a short adjustment horizon) differfrom those of the young and middle-aged.

Michael J. Bos kin

Economics of HealthIntroduction

The research program in the economics ofhealth is supported by grants from theRobert Wood Johnson Foundation, the Na-tional Center for Health Services Research(HEW), the Henry J. Kaiser Family Founda-tion, and the Ford Foundation, Victor Fuchsand Michael Grossman are the co-directorsof the program. Daniel Graham, a researchfellow from Duke University, and JerryGreen, a visiting scholar from HarvardUniversity, joined the program during thepast year.

Michael Grossman, Dov Chernichovsky,Douglas Coate, Ann Colle, Linda Edwards,Jacob Gesthalter, and Fred Goldman arestudying economic aspects of children'shealth. Benefits of improvements inchildren's health might be substantial.Preliminary and scattered evidence sug-gests that health at early stages in the life

cycle might be an important determinant ofhealth at later stages and of intelligence,years of formal schooling completed,market wage rates, and hours of work. Giventhis evidence and the policy goal to improvethe health of certain children, it is crucial toprovide a theoretical and empirical frame-work that will promote an efficient allocationof scarce resources by government policy-makers. The child health project is attempt-ing to contribute to such a framework. Theobjectives of the project are to understand(1) the determinants of variations in

children's health, (2) the factors that affectthe demand for their medical care servicesand nutrition, and (3) the consequences ofvariations in children's health with specialreference to cognitive development.

Claire Bombardier, M.D.; Victor Fuchs;Daniel Graham; Jerry Green; Lee Lillard;and Kenneth Warner are studying de-terminants of the cost of medical care andthe health of adults. Bombardier, Fuchs,Lillard, and Warner are analyzing socioeco-nomic factors affecting surgical utilization.Graham is working on theoretical issues in-volved in the valuation of human life. Greenis working on the relationship betweenhealth and schooling and on the role of thephysician in the demand for medical care.

The following papers appeared last year,are in press, or are available in preliminaryform.

Claire Bombardier and J. Fries. "VariationsAmong Physicians in the Costs Generatedfor the Treatment of Rheumatoid Arthritis."Presented at the meeting of the AmericanRheumatology Association, Chicago,June 1976.

Claire Bombardier, Victor R. Fuchs, Lee A.Lillard, and Kenneth E. Warner. "Socio-Economic Factors Affecting the Utilizationof Surgery." Presented at the annual meet-ing of the American Public HealthAssociation, 1976, and accepted forpublication by the New England Journalof Medicine.

Linda N. Edwards and Michael Grossman."An Economic Analysis of Children's

64

Health and Intellectual Development."Presented at a conference on "The Familyand the Subsequent Development of theChild" sponsored by the MathematicalSocial Science Board, Stanford, March1977. Also presented at a Health Eco-nomics Research Organization (HERO)session at the Allied Social ScienceAssociations meetings, Atlantic City, Sep-tember 1976.

Victor R. Fuchs. "Health and Economics."Encyclopaedia Britannica 1976 Annual.

_____

"An Economist Looks at HealthCare." Hospital Forum, vol. 18, no. 12,March 1976.

"A More Effective, Efficient andEquitable System." Western Journal qfMedicine, vol. 125, no. 1, July 1976.

Kenneth E. Warner. "Treatment DecisionMaking in Catastrophic Illness." MedicalCare (forthcoming).

"The Effects of the AntismokingCampaign on Cigarette Consumption."Presented at the annual meeting of theAmerican Public Health Association,1976.

Surgical Utilization bySocioeconomic Class

Victor R. FuchsMichael Grossman

This study measures the relationshipbetween socioeconomic factors and theutilization of in-hospital surgical care in1970. We also analyze changes in surgicalutilization between 1963 and 1970, a periodwhich spans the introduction of Medicareand Medicaid, the spread of poverty healthclinics, and other efforts designed toimprove access for disadvantaged groups.We examine differentials in surgical utiliza-tion and the nature of the differences. If non-whites or rural residents or low-incomefamilies have less surgery than do others,are the differentials fairly uniform by type ofsurgery, or do they vary by complexity, ur-gency, or necessity?

The data source is the Health InterviewSurvey (for 1963 and 1970) conducted by

TABLE 5.1

INDEXES OF COMPLEXITY, URGENCY, AND NECESSITY; ELEVEN SURGICAL PROCEDURES

Complexity

Valuea

Urgency Necessity

Valueb Valueb

(CRVS (percent) (percent)Operation units) Rank Rank Rank

Appendectomy 1.12 6 99 1 99 1

Cataract . 2.12 3 2 11 96 2.5Hernia repair 1.00 8 24 5 96 2.5Prostatectomy 2.21 2 47 4 92 4

Cholecystectomy 1.79 5 48 3 88 5

D&C(excludingabortions) 0.52 10 53 2 77 6Hemorrhoidectomy 0.71 9 12 7 65 7

Varicoseveinstripping . 1.06 7 4 10 63 8

Lumbar laminectomy (for disc) 2.88 1 22 6 38 9

Hysterectomy 1.92 4 8 8 28 10

Tonsillectomy 0.45 11 5 9 24 11

Weighted mean values for theeleven procedures(1970) 1.16 28.7 63.9

Note: Operations are listed by their ranking on the necessity index.a California Relative Value Scale (CRVS) Study(1969) expressed in hernia equivalents (hernia repair = 1).

Derived from questionnaires mailed to California physicians, 93 replies (50 percent response rate).

TABLE 5.2

UTILIZATION RATES FOR ALL SURGERY (EXCLUDING OBSTETRICAL)IN THE UNITED STATES, 1970, BY SOCIOECONOMIC CHARACTERISTICS

(operations per 1,000 population)

Standardized for

Age, Sex, Race, Age, Sex, Race,Residence, Residence,

Unstandardized Age, Sex Education Education, Income(1) (2) (3) (4)

All 55.5 55.5 55.5 55.5Race

Whites 57.7' 57.3' 57.4' 56.2Nonwhites 42.1 41.8 50.2

Residence ,

In SMSA 57.2 57.1' 57.3' 56.9Out of SMSA 52.3 52.5 52.2 52.9

Education of head (yrs.)54.5 50.Of 51.8t 58.3

9—12 57.5 59.1 59.0 58.513—14 56.6 58.2 57.0 56.615—16 49.4 50.4 48.8 43.717+ 50.9 52.3 50.4 36.0

SexMales 51.0' 51.9' 51.8' 51.0'Females 59.6 58.9 58.9 59.7

Age (yrs.)0—9 38.2f 38.4f 38.6t 36.3t

10—19 34.7 34.9 35.0 34.220—34 57.1 57.0 56.6 53.635—49 71.8 71.8 71.4 67.550—64 69.6 69.5 69.4 73.965+ 81.2 80.8 81.8 95.5

'Difference significant at .05 level using t test.tDiflerence significant at .05 level using F test.

65

the National Center for Health Statistics. The1970 interviews were conducted with ap-proximately 37,000 households containingabout 116,000 individuals; and the 1963 in-terviews, with 42,000 households containing134,000 individuals. We also calculateindexes of "complexity," "urgency," and"necessity" for each of eleven surgicalprocedures that occur frequently and arewell defined. The complexity index is basedon the California relative value scale; the ur-gency and necessity indexes are based on93 replies (50 percent response rate) by

physicians to our mailed questionnaire (seeTable 5.1).

The original unit of observation in thesurvey is the individual person, but in ouranalysis individuals have been aggregatedinto cells by year of observation, sex, race(white, nonwhite), residence (inside SMSA,outside SMSA), age (six classes), andeducation of the head of the household inwhich the individual resides (five classes).We apply a linear regression model, withcell surgery rates being a function of cellcharacteristics. Each cell is weighted by the

TABLE 5.3INDEXES OF TYPE OF SuRGERY,a UNITED STATES, 1970,

BY SOCIOECONOMIC CHARACTERISTICS

Complexity IndexStandardized for

Urgency IndexStandardized for

Necessity IndexStandardized for

Age, Sex, Age, Sex, Age, Sex,Race, Race, Race,

Residence, Residence, Residence,Education, Education, Education,

Age, Sex Income Age, Sex Income Age, Sex Income

All 100 100 100 100 100 100Race S

Whites 100 100 101 101 100 100Nonwhites 99 99 91 87 99 99

ResidenceInSMSA 99 99 97 102 99 100Out of SMSA 101 101 106 97 102 100

Education of head (yrs.)102t 102t 101 94 103 103

9—12 101 101 98 98 98 99

• 13—14 95 94 99 103 100 10015—16 92 90 97 105 98 97

17+ 105 109 119 125 104 104

Sex

Males 104* 104* 102 102 111* 111*

Females 97 97 98 98 89 89

Age (yrs.)0—9 56t 58t 48f 43f 63t 62t

10—19 69 68 129 128 90 8820—34 101 102 130 118 109 10535—49 128 127 95 98 103 103

50—64 129 126 99 110 124 12665+ 155 159 93 111 137 146

*Difference significant at .05 level using t test.tDifference significant at .05 level using F test.a Based on eleven selected procedures, see Table 5.1.

66

square root of the number of persons in thecell. In addition to the dui-nmy variables cor-responding to the characteristics used toclassify individuals into cells, we alsoregress on cell family income in several al-ternative specifications, but we do not groupby income.

The results for all surgery (exceptobstetrical) for 1970 are presented in Table5.2 in the form of number of operations perthousand population per year. The overallrate is 55.5 per thousand. The unstan-dardized rate (column 1) is higher forwhites, residents of standard metropolitanstatistical areas, women, and the aged thanfor nonwhites, those living outside of stan-dard metropolitan statistical areas, men,and persons under 65. There are also somesignificant variations in utilization depend-ing upon education of head of family.

Table 5.3 reports the results of regressionanalyses when the indexes of type of sur-

C0

0.00.000a-a)0.U)C0

a-a)0.0

CHART 5.1

Note: Dashed lines represent predictions outside the90 percent range of mean cell income in each year.

67

gery are the dependent variables. In

general, we find very little variation in theseindexes across the various socioeconomicgroups, even when there are significant dif-ferences in the surgical rate. It has beenspeculated that groups that have above-average rates of surgery (or above-averagerates of increase in surgery) would have amix of operations which on average werelower (or which decreased) on the necessityor urgency scales. The data we haveexamined do not support this view. Forinstance, the surgical utilization rate ofurban nonwhites rose 49 percent between1963 and 1970, but the average necessityscore remained unchanged. Surgical ratesfor persons 65 and over with eight years ofschooling or less also rose substantially, butthe average necessity score actuallyincreased slightly during the same period.

In general there was a weakening of therelationship between surgical utilizationand family income between 1963 and 1970,as evidenced in Chart 5.1. Additional workon the relationship between surgical utiliza-tion and income is being undertaken.

Claire BombardierVictor R. Fuchs

Lee A. Li/lardKenneth Warner

Doctor and Patient Components inTreatment Cost: Rheumatoid Arthritis

There is growing evidence of wide varia-tion in medical care utilization and expendi-tures. Previous studies have documentedwide variation in medical care costs amongdoctors. This paper is concerned with thedeterminants of expenditures for the treat-ment of the chronic disease rheumatoidarthritis. We consider separately expendi-tures for medication and expenditures fordiagnostic laboratory tests and X rays, aswell as the degree of correlation betweenthem, We focus on the contribution of doctorand patient differences to variation in thesetwo aspects of treatment cost.

Data are rarely available for full adjust-ment for patients' characteristics. Our dataprovide the opportunity to study the relative

PREDICTED SURGICAL UTILIZATIONSTANDARDIZED FOR AGE AND SEX AS A

FUNCTION OF FAMILY INCOME, 1963 AND1970

4 6 8 10 12141618Family income (thousands of 1970 dollars)

contribution of physicians and patients tothe cost of a clinic visit for the treatment ofrheumatoid arthritis. In the clinic studied,patients are often seen by different clinicphysicians on successive visits. This re-plication allows the separation of patientand physician components by simpleanalysis-of-variance techniques. Weconsider separately the determinants ofmedication and laboratory costs. The de-terminants are (1) the physician; (2)measured patients characteristics, includ-ing the medical condition of the patient asroutinely measured by the physicians, thedistance traveled for treatment by thepatient, and the type of insurance coverage;and (3) other important but unmeasuredpatient characteristics.

The results of this study confirm previousreports of significant variations among doc-tors in laboratory and medication costs.Year and doctor differences alone explain10.4 percent of medication cost variationand 2.8 percent of laboratory cost variation,and the net contribution of doctors' dif-ferences remains quite significant, evenafter controlling for patients' measuredmedical and socioeconomic variables, Themost striking result of this study is obtainedafter we fully adjust for the very significantunmeasured patient differences. The netcontribution of doctor differences isreduced to 1.6 percent for medication(MED) costs and becomes insignificant forlaboratory (LAB) costs. It appears that mostof the variation in cost among doctors is dueto the costs attributable to the patients theysee. The patient component remainedsubstantial and accounted for 18.6 percentof LAB cost and 28.5 percent of MED cost. Alarge proportion of the variation in MED costwas accounted for by gold therapy; howeverit remained significant even after adjustingfor gold therapy. The results of this studysuggest some caution in interpreting reportsof wide variations among doctors' treatmentcosts where full adjustment for differencesin patients is not possible.

Claire BombardierLee A. Li/lard

68

The Choice of Diet by the Householdand Its Effect on the Growthof Young Children

Interest in the nutritional status of youngAmerican children has heightenedconsiderably in the last decade. Much of theconcern has resulted from allegations of un-dernourishment among young children inlow-income families and from recent re-search suggesting that intellectual develop-ment can be permanently impaired if dietsare deficient in the first year or two of life.Interest in child nutrition has not beenconfined solely to the issue of undernourish-ment. Recent evidence suggests thatovernourishment in the first year of life is animportant determinant of adult obesity,perhaps the number one health problem inthe United States today. The primary pur-pose of our research is to analyze thehousehold's choice of diet for its youngchildren (up to 36 months old) and its rela-tion to the children's growth.

As a point of departure, we assume thatthe utility of household decision makers ispositively related to children's growth withinthe bounds of perceived norms. Wehypothesize that nutrient intakes are im-portant inputs in the production of children'sgrowth, although their influence on growthlevels is tempered by genetic factors, whichare measured or approximated by the heightand weight of the mother and sex, race, andbirth weight of the child. We furtherhypothesize that the household demand forthe children's nutrient intakes is a function ofchildren's growth, deviations of it fromperceived norms, and socioeconomic fac-tors. The endogenous variables in the modelare measures of children's growth—height,weight, and head circumference—and nu-trient intakes—protein and calories.

The data base we utilize is derived fromthe Ten State Nutrition Survey, 1968—1970.In this survey dietary and other data werecollected by interviews, clinical evalua-tions, and physical examinations from 24,-000 families residing in low-income areas inthe United States. Preliminary results indi-cate a close association between children's

growth and their diet during their first threeyears of life. The quality of this diet,however, does not seem to besystematically correlated with household in-come or mother's education. Hence, theresults conflict with the view that low-in-come households in the United States lackthe necessary education or income toprovide adequate diets for their children.We do find, however, that protein and ca-loric intakes are higher in urban areas, inwhite families, and in small families. Al-though further study is necessary, thesefindings do not appear to result from incomeand education effects. The results also indi-cate that the allocation of nutrients tochildren in the household is based on thegrowth of the children relative to norms. In-takes of nutrients are increased byhousehold decision makers if a child has alow birth weight or is light for his/her age.Similarly, nutrient intakes are decreased if achild is heavy for his/her age. These adjust-ments are substantial. Low-birth-weightchildren, for example, have protein and ca-loric intakes about 20 percent higher thannormal-birth-weight children in householdswith the same socioeconomic characteris-tics.

Dov ChernichovskyDoug/as Coate

Determinants of Pediatric Care Utilization

The purpose of our research is to under-stand the determinants of utilization ofpediatric care—care rendered to childrenby all physicians. Appropriate pediatriccare is an obvious and natural vehicle formaintaining children's health and an objectof government policy via programs such asMedicaid and the neighborhood healthcenter program. Bills introduced inCongress in 1976 by Senator Jacob K. Ja-vits of New York and by CongressmanJames H. Scheuer of New York wouldexpand the scope of government financingby providing national health insurancecoverage for maternal and pediatric care.Our study will shed light on the most effec-tive means of increasing utilization bycertain groups of children and will contain

69

estimates of the effects of national health in-surance on number of pediatric visits, com-position of visits between specialists(mainly pediatricians) and other providers(mainly general practitioners), and the totalcost of care.

Our data source will be a 1970 healthsurvey conducted by the National OpinionResearch Center (NORC) and the Center forHealth Administration Studies of theUniversity of Chicago. It has information onhealth, medical care utilization, health in-surance, and socioeconomic and demo-graphic characteristics of 11,822 adults andchildren from 3,880 families. In addition,characteristics (age, specialty, boardcertification status, et cetera) of all phy-sicians in private practice who renderedcare to sample families were obtained fromthe American Medical Association Direc-tory. The NORC sample is a particularly richsource for calculating the "full" price ofpediatric care on the basis of family's traveltime to and from its usual source of care,waiting time in the physician's office,parents' wage rates, and various parametersof the family's health insurance policy (coin-surance rate, deductible, upper limit, etcetera).

Our work on estimation of demand curveswill be guided in part by Linda Edwards andMichael Grossman's study of children'shealth and intellectual development (seetheir report below). An interesting implica-tion of their study is a two-regime specifica-tion of demand functions for health inputssuch as pediatric care. Membership in agiven regime is determined by whether ornot parents make a financial transfer to theirchildren. The greater the parents' incomethe greater is the probability that a financialtransfer will be made. By dividing oursample into high- and low-income subsam-pIes, we can allow for differences inparameters of demand curves for pediatriccare between income classes.

In the context of the two-regime specifica-tion, we will focus on estimating parametersassociated with variables such as family in-come, parents' wage rates, parents' school-ing, travel time, waiting time, private health

insurance coverage, Medicaid coverage,number of children in the family, andmother's knowledge of appropriate healthpractices. Pediatric care will be measuredby number of physician visits and alterna-tively by visits adjusted for the 'quality' ofthe physician. Here the assumption is thatthe higher average fees of visits to spe-cialists represent a greater quantity ofservices. Finally, we will examine the de-terminants of visits to specific kinds of phy-sicians such as board-certified pedia-tricians.

Ann Colle will devote most of her time toan analysis of the behavior of low-incomefamilies. Michael Grossman will focus onthe effects of mother's schooling and theextent to which its effects operate via wagerates and knowledge of appropriate healthpractices. In addition he will estimate childhealth productio.n functions in the NORCsurvey.

Ann D. ColleMichael Grossman

An Economic Analysis of Children'sHealth and Intellectual Development

The basic purpose of our research is tocontribute to an understanding of the jointdetermination of children's cognitivedevelopment, which has received a gooddeal of attention in the literature, andchildren's health, which has not. As anatural by-product of this overall objective,we explore interrelationships betweenvarious aspects of children's physicalhealth and their intellectual development. Inparticular, we seek concrete empirical evi-dence concerning the widespread beliefthat poor health can pose a threat to thecognitive development of children.

During the past year we devoted most ofour time to an empirical investigation of thedeterminants of cognitive development andhealth of children aged six through eleven.We focused on the roles of home environ-ment variables and proxies for the endowed(initial or inherited) level of health in the cur-rent health and development functions so asto uncover similarities in or differencesbetween health and development effects.

70

The empirical work was guided by aninsight provided by a theoretical model ofintergenerational transfers of human andnonhuman wealth, as follows: To understandthe behavior of parents regarding theirchildren's health and development, it is im-portant to distinguish low-income from high-income families. Clearly, this is a policy-relevant insight, for public policy often isaimed at low-income groups. Our results in-dicate it would be inappropriate to buildpolicies directed at improving the welfare ofchildren in low-income families on the basisof empirical results derived from an exami-nation of the population at large.

In our general analytical framework weassumed that parents make decisions overtwo periods or stages in their life cycle. Inthe initial period their children are com-pletely dependent upon them for financialsupport, while in the second, the childrenbecome financially independent. Parentscan affect their children's income or eco-nomic well-being as adults in two ways:They can make a financial transfer (a be-quest) at the beginning of the period of inde-pendence, and they can invest in theirchildren's human capital in the period of de-pendence. In turn, investments in humancapital have two forms: investments iiihealth capital and investments inknowledge or cognitive developmentcapital.

As is usual in models of this kind, we im-posed the solvency constraint that parentscannot leave net debts to their children orthat the financial transfer cannot be smallerthan zero. This implies a two-regimespecification of demand functions forchildren's cognitive development andhealth. Parents who do not make financialtransfers are members of Regime 1, whileparents who do make positive financialtransfers are members of Regime 2. Thehigher parents' income, the more likely it isthat they are members of Regime 2.

We show that the demand functions in thetwo regimes have very different properties.In particular, parents' income has a positiveeffect on health or cognitive development inRegime 1 but no effect in Regime 2. The

marginal cost of investment in humancapital enters the demand functions in bothregimes. A change in the marginal cost hasthe same direction of eftect in each regime,but the magnitude of the effect differs. Themarginal cost of investment depends on theprices of inputs such as medical care andparents' time in the production of healthyand intelligent children. It also depends onhome environmental variables, such asparents' schooling, that shape the efficiencyof the production process and on thechildren's endowments of health and cogni-tive development. To summarize, our modelhas an important implication for the estima-tion of demand curves for children's healthand intellectual development. Besides sug-gesting the relevant explanatory variables, itcalls attention to the necessity of allowingfor interactions between parents' income,which is an important determinant of therelevant regime, and determinants of themarginal cost of investment.

Our data source for the estimation ofhealth and development functions is Cycle IIof the U.S. Health Examination Surveyconducted by the National Centerfor HealthStatistics. Cycle II is a nationally representa-tive sample of 7,119 children aged six toeleven years, examined over the period1963—1965. This sample is an exceptionallyrich source of information about children'shealth and intellectual development andcharacteristics of their families. Morespecifically, the data comprise completemedical and developmental histories of thechild, which are provided by the parent; in-formation on family socioeconomic charac-teristics; birth certificate inf.ormation; and aschool report with data on school perfor-mance and classroom behavior, which areprovided by teachers or other schoolofficials. Most imp6rtant, there are objectivemeasures of health derived from detailedphysical examinations and scores onpsychological (including vocabulary andachievement) tests. The physical examina-tions and psychological tests wereadministered by the Public Health Service.

The empirical analysis is restricted towhite children who reside with both of their

71

parents. We use three measures of in-tellectual development and four measuresof current health as alternative dependentvariables in an ordinary least squaresmultiple regression analysis. The in-tellectual development variables are an 10measure, derived from two subtests from theWechsler Intelligence Scale for Children,and reading and arithmetic test scores onthe Wide Range Achievement Test. Thehealth measures are height, the peridontalindex, the number of decayed primary andpermanent teeth, and parental assessmentof the child's current state of health. Heightis a standard indicator of children's nutri-tional status, and good nutrition is an ob-vious and natural vehicle for maintainingchildren's health. The peridontal index andthe number of decayed teeth are measuresof oral health and also reflect nutritionalstatus. Parental assessment of the child'scurrent health is employed as a dependentvariable to show how results differ whenhealth is measured subjectively by parentsas opposed to objectively by physicians. Toestimate the two-regime model, we dividethe sample into two income subsamples:children whose families have an annual in-come under $7,000 and those whose familyincome is $7,000 or more.

Our major findings are:1. The prediction of two distinct regimes

or two different relationships between eachof our health and development variablesand the set of explanatory variables isgenerally supported by our results. Statis-tically significant differences in the sets ofcoefficients for the two income classes arereported for five of seven dependent vari-ables. Although these results cannot becharacterized as "unanimous" support forthe basic structure of our model, they doconstitute stronger verification than may beinitially apparent. The two income classesused here are unlikely to coincide com-pletely with the two regimes specified by themodel. The resulting misclassification ofobservations will tend to bias the coefficientin the two incomeclàsses toward equality,making it more difficult to obtainsignificantly different coefficients in the two

income classes even though such dif-ferences do exist in the two regimes. We ob-serve significant differences in coefficientsfor five of our seven variables despite thisbias toward finding no such difference.

2. The prediction that income will have apositive effect on health or development forfamilies in Regime 1 and no effect inRegime 2 receives weaker support in ourresults. For the two achievement measures,income has a positive, significant impact forlower income families and a nonsignificantimpact for upper income families. For theother dependent variables, family income iseither statistically significant in both incomeclasses or in neither class. One likely expla-nation for the significant coefficients in theupper income class is the previously men-tioned bias resulting from the misclassifica-tion of observations. In particular, it is likelythat members of Regime 1 are erroneouslyincluded in the over-$7,000 income class,causing an upward bias in the coefficient ofincome for that class.

3. When height is the measure of health,parents' schooling, mother's work status,and family size are significant predictorvariables in the low-income sample but notin the high-income one. On the other hand,these three variables tend to be as important(or more important) predictors of intellectualdevelopment in the high-income sample asin the low one.

4. Health endowment and investmentmeasures have significant, positive effectson cognitive development. In particular,cognitive development scores are higherwhen children weighed more than fivepounds at birth, when they were breast-fed,when their current hearing is normal, andwhen abnormal vision is corrected by theuse of eyeglasses. These findings suggestthat prenatal and pediatric care programsthat could identify high-risk mothers andchildren at modest cost would have rela-tively high expected benefits.

Our findings highlight at least two fruitfulareas for research in the coming year. Oneis an investigation of the extent to whichendogenous current health measures affectintellectual development. The second is an

72

investigation of health and developmentrelationships at later stages in the child'slife cycle. Both of these will contributefurther to our understanding of how healthand development interact and will providemore refined measures of benefits from in-vestments in children's health.

Linda N. EdwardsMichael Grossman

Effects of Mother's Schooling onChildren's Health

The aims of this research are to examinethe role of mother's schooling in the produc-tion of children's health and in the demandfor health inputs such as curative pediatriccare visits, preventive pediatric care visits,and parents' time. Health production func-tions and input demand functions will beestimated in a special sample of New YorkCity residents conducted by the New YorkCity Department of Health (the Mindlin-Densen Survey). This is a longitudinal andcross-sectional survey of infants andpreschool children, covering twenty-onemonths of the period 1965—1966. It containsa variety of health indexes, includingmother's evaluation of children's health,number and type of acute and chronic con-ditions, and number of restricted-activitydays due to illness (total and per condition).These measures and health inputs areavailable monthly or bimonthly over a one-year period. Therefore, the effects ofpediatric care and other variables on childhealth outcomes can be traced, with proxiesfor initial state of health or seriousness of ill-ness held constant.

Typically, the mother is the main providerof child health care in the family. Con-sequently, differences in her schoolingmight affect the overall efficiency of theproduction process (the amount of outputobtained from a given set of medical careand time inputs) as well as the relativeproductivity of various inputs. By estimatingproduction functions for various conditions,we allow the effects of mother's schooling todiffer by type of condition, In this contexttwoplausible hypotheses are that the schoolingeffect and the interaction between schooling

and pediatric care are larger for those con-ditions in which health care makes a dif-ference in health outcomes.

If schooling affects absolute and relativeprod uctivities of inputs, it can alter the com-position of inputs by means of income andsubstitution effects, Estimates of demandfunctions for preventive pediatric visits, cu-rative pediatric visits, and the mother's owntime input will quantify the partial effects ofmother's schooling in the input demandfunctions. In addition, these estimates willenable us to explore the hypothesis that anincrease in the relative productivity of aninput because of schooling will increase thequantity demanded of that input.

Gesthalter is devoting most of his time tothe above research. Goldman is studyingtrade-offs between public and privatepediatric care in the production and de-mand for children's health.

Willingness to Pay

Jacob GesthalterFred Goldman

This research aims at clarifying certainconceptual problems associated with usingwillingness to pay" measures in evaluating

public-sector projects. The analytical foun-dation of this valuation method—the 'com-pensating variation" of Hicks and the'hypothetical compensation" test of Kaldor,Hicks, and Scitovsky—is built upon a time-less and certain world with completemarkets. There are significant difficulties,both positive and normative, in extendingthis analytical basis to the temporal and un-certain world in which projects must typi-cally be evaluated.

The positive difficulties can perhaps bestbe described by first summarizing "willing-ness to pay" in a static, certain world of ap-ples, oranges, and public parks. Here thereare a variety of combinations of apples andoranges that an individual would willinglysacrifice for an additional unit of parks. Theouter boundary of this "willingness-to-sacrif ice" set would appear (under conven-tional assumptions) as a smooth curve,strictly concave to the origin in a two-dimen-sional graph with payments in oranges and

73

payments in apples plotted along the axesThe existence of markets for apples andoranges in which the individual is a price-taker transforms this willingness-to-sacrificecurve into a line whose slope reflects therelative price of apples and oranges andwhich is tangent to the willingness-to-sacrifice curve. Along this line all combina-tions of apples and oranges have the samemarket value, and this market value isidentified as "willingness-to-pay." What isimportant here is the realization that it is theexistence of markets for all other goods thatprovides the "common denominator" bywhich the multidimensional willingness-to-sacrif ice curve is transformed into the scalarmagnitude willingness-to-pay.

Consider now a simple illustration from atemporal and uncertain world. At issue is theappropriate value to attach to airbags, apassive restraint system for automobiles.Here the potential user suffers 'from un-certainty regarding which of two possibleevents will occur: wreck or no wreck. Mak-ing use of the notion of contingent com-modities we may again consider a two-di-mensional willingness-to-sacrifice curve ina coordinate system with payments in"dollars, if wreck occurs" plotted along oneaxis and payments in "dollars, if no wreckoccurs" plotted along the other. If marketsexist in which financial claims against thesecontingencies can be bought and sold atgiven prices, this willingness-to-sacrificecurve would again be transformed into a linealong which market value is constant and ascalar magnitude representing willingnessto pay would again emerge.

The problem, of course, is that unlikemarkets for sure deliveries of apples andoranges, these contingent claims marketstypically do not exist. One obvious reasonfor their absence is the tremendous numberof such markets required for completecoverage. If, for example, there were sixpeople in a society and each could findhimself in one of ten "individual states"(alive, dead of heart attack, dead of automo-bile wreck, etc.) then there would be 106social states of the form: all six survive; 1

survives and 2 through 6 die of heart attack,

etc. In short, 1 million contingent claimsmarkets would be required for completecoverage. Under restrictive circumstances(all individuals alike in endowments, tastes,and probabilities with independent risks)the required number of markets wouldcollapse to ten—a market for annuities plusnine insurance markets, one for each causeof death. Even in these restrictive circum-stances one market for life insurance is notsufficient for complete coverage because itrequires that the buyer purchase identicalamounts of nine inherently different types ofinsurance.

In the absence of complete markets, whatmagnitude should be identified as willing-ness-to-pay? The one most frequently em-ployed is that associated with requiring thesame payment in all states. In the airbagexample, the intersection of the forty-five-degree line with the willingness-to-sacrificecurve, yields this maximum 'sure payment."Another measure frequently mentioned is"expected consumer surplus," the expectedvalues of willingness to pay if "wreck" werecertain and if "no wreck" were certain. (Thiscombination of payments is also on the will-ingness-to-sacrifice curve.) In partially com-pleted research, the relationship betweenthese two magnitudes is clarified, and athird measure, "expected willingness topay," is suggested as the conceptually cor-rect measure of value for many public policydecisions. In this approach, that combina-tion of payments along the willingness-to-sacrifice curve is selected which has thegreatest expected value. Here an airbagbuyer would contract for one payment if"wreck," and another, possibly differentpayment, if "no wreck." A seller entering intoa sufficiently large number of these

74

contracts (assuming independent risks)would collect the sum of these expectedvalues with virtual certainty (with an arbit-rarily small variance), and no alternativepayments mechanism could increase thesereceipts. By virtue of this paymentsmechanism, the seller becomes a jointprovider of airbags and wreck insurance.

In this context, useful insights weregained regarding the value of reductions inthe probability of loss of life or other goodsfor which there are no perfect market sub-stitutes. New perspectives were gained onissues ranging from an individual's willing-ness to buckle his seat belt to his willing-ness to pay for flight insurance. Similarly,the correct "option value" could beidentified of preserving irreplaceablenatural resources such as the Hell's Canyonregion of the Snake River. Finally, the extentto which observable behavior in existingcontingent claims markets yields informa-tion pertinent to valuing the good whoseloss characterizes the contingency wasexamined.

In separate but related research, I haveinvestigated the normative aspects of will-ingness-to-pay in a temporal oran uncertaincontext or both. The difficulties here cameabout because one person becomes a mul-tiplicity of recognizable interests. Willing-ness to pay based upon today's preferencesmay or may not be compatible with calcula-tions based upon tomorrow's preferences.Similarly, if bets are placed in contingentclaims markets it is possible to be a loser inparticular states. Hence, I examined theextent to which current willingness to paycould be justified in terms of hypotheticalcorn pen sat ion.

Daniel Graham

6. INTERNATIONAL STUDIES

IntroductionThe international program has continued

to grow during the past year, with severalnew studies begun and plans made for ad-ditional projects to start in the coming year.The Bureau's international activity includes,aside from research, an international ex-•change of scholars, a fellowship program,and a conference series; the last is

described below in this section by M. lshaqNadiri; the report on fellowship programs,by Douglas H. Eldridge, is in section 9. Boththe research and the other internationalactivities have involved the participation ofeconomists from many countries andcooperation with foreign research institu-tions.

Among the studies begun during the yearwas the one by Schwartz, Darby, Klein, andLothian on the international transmission ofinflation through the world monetary system.That study together with Cagan's researchon the effect of world commodity prices onU.S. manufacturing prices and the Kravis-Lipsey study of foreign inflation and ex-change-rate influences on several countries'export and domestic prices represent awide range of approaches, using com-pletely different data, to the analysis of thetransmission of price impulses from onecountry to another.

The continued expansion of the Bureau'sinvolvement with economic problems of lessdeveloped countries was reflected inseveral projects. These include NarongchaiAkrasanee's new study of trade policies ofThailand; the initiation of work on a model oftrade and development in the Pacific basin,by Bert Hickman and Lawrence Lau; thecontinuation of Anne Krueger's majorcollaborative project on alternative tradestrategies and employment; and the LatinAmerican workshop series. The mostrecently planned of these workshops, to beheld in the coming year, will deal with com-modity markets, models, and policies inLatin America. Another notable activity wasthe conference of the Universities—National

75

Bureau Committee on Population Growthand Economic Change in Less DevelopedCountries. One of the projects planned forthe coming year, the study by Kravis,Lipsey, and Nadiri on the indexing of com-modity prices, also addresses a questionthat has been of particular concern to lessdeveloped countries in the last few years.

Some other new and proposed projects inthe area of multinational firms and theprogress of the U.S.—U.S.S.R. Scientific andTechnical Program of Cooperation, directedby McMains, Meyer, and Smith, aredescribed in the separate reports below.

Publication of volumes on Colombia, byCarlos DIaz-Alejandro, and on Chile, byJere Behrman, completed the countrystudies in the Bhagwati-Krueger project,Foreign Trade Regimes and EconomicDevelopment. Anne Krueger's synthesisvolume has been approved by the Bureau'sBoard of Directors and completion of thecompanion volume by Jagdish Bhagwati isexpected very shortly. Also among theinternational publications for the currentyear are Money, Financial Flows, andCredit in the Soviet Union, by George Garvy,and a collection of papers from theConference on Indexation, edited by M.lshaq Nadiri and Affonso C. Pastore, invoJume 4, number 1 of Explorations in Eco-nomic Research.

Robert E. Lipsey

The International Transmission ofInflation through the WorldMonetary System

We have begun research focusing on therole of the world monetary system in theinternational transmission of inflation. Theprincipal thrust of the study will be thedevelopment and estimation of a unifiedquarterly model to investigate major issuesin both the short-run transmission processand the long-run equilibrium and stability ofthe system. The model will comprise threeelements: (1) reaction functions explaining

government monetary, fiscal, and exchangerate policy as responsive to national andinternational economic conditions; (2) na-tional macroeconomic equations explainingthe behavior of key national macroeconomicvariables of concern to policymakers,namely, nominal income, real income, theprice level; and unemployment; (3) equa-tions explaining balances of payments andexchange rates as simultaneously de-termined and describing the nature of theforces that produce differences amongcountries in their rates of inflation.

At this stage of the study, only the secondof the three elements has been specified—the part of the model designed to explainmovements over the period 1955—1976 inoutput and the price level of individualcountries in our sample (the United States,Canada, Japan, Germany, the Netherlands,France, Italy, the United Kingdom). Themovement of the economy in the model isviewed as reflecting two types of forces. Thefirst are macroeconomic shocks whicheither change the moving equilibrium of theeconomy or move the economy away froman unchanged equilibrium; the second,equilibrating forces which move theeconomy toward equilibrium, though notnecessarily smoothly or rapidly, when theeconomy is away from equilibrium. Themonetari st-Keynesian debate has beenlargely due to differences of opinion as tothe variances and impacts of different typesof shocks and to the rapidity of the adjust-ment toward equilibrium. The model hasbeen formulated so as to allow the data(rather than assumptions) to answer theseempirical questions.

In the model, two behavioral equationsand an identity will determine nominal in-come, real income, and the price level, Onthe basis of work to date, it appears that thebehavioral equations are best formulated asa nominal income equation and a real in-come equation with the price level de-termined by the identity. This approachpermits relatively simple implicit modelingof complex price adjustment behavior notpermitted by the standard price level equa-tions but required for long-run equilibrium.

76

For example, lags in the adjustment of theprice level in early stages of increasedmoney growth imply that at a later periodrates of inflation will increase temporarily bymore than the increase in the growth rate ofthe money supply in order to restore equilib-rium. This catch-up in the price level isreadily implied by the adjustment of realand nominal income toward their respectiveequilibrium values. An explicit price equa-tion incorporating the somewhat differentshocks affecting nominal and real incomeappearsawkward.

The nominal income equation will com-bine a tendency toward equilibrium ofmoney supply and money demand withmeasures of monetary, fiscal, and interna-tional stimulus. The real income equationwill combine the tendency toward a natural-employment real income with the short-runeffects of unexpected changes in nominalincome and of domestic and internationalsupply shocks.

Expected values are important variablesin our model. The basic theoretical ap-proach underlying the model can be viewedas combining elements of the monetarist, ra-tional expectations, and neoclassicalgrowth literature. The rational expectationsapproach divides changes in variables intothose which could be rationally predictedby extracting the information in the pastvalues of the variable and related series andthe residual or unexpected changes. Muchemphasis has been placed on the proposi-tion that expected changes in money will bereflected in price changes while unex-pected changes affect real income and em-ployment. Similar distinctions between theeffects of expected and unexpectedchanges in such other variables as govern-ment spending and exchange rates can befruitfully made.

We initially indicated that we wouldconstruct expected values for the moneysupply or high-powered money by estimat-ing central bank reaction functions. We havesince deferred the estimation of the reactionfunctions until the data base is substantiallycomplete, and in the meantime we shall useestimates of expected magnitudes derived

from the Box-Jenkins ARIMA technique thatextracts all the information in the past his-tory of a series. At a later point we shall com-pare these estimates with those derivedfrom central bank reaction functions andother government reaction functions. It is aconvenience to use Box-Jenkins estimatesat this juncture in the trial runs of the model,although the estimates will be of interest inthemselves in comparing them with reactionfunction results.

The central banks' reaction functions willindicate the response of monetary au-thorities to economic developments in thedomestic and international spheres in

manipulating domestic credit relative toforeign reserves. The response will figure inthe transmission mechanism, not merely inthe derivation of expected values for moneyor high-powered money. Since during mostof our period the international monetarysystem consisted of a dominant-dollarpegged exchange rate, one variable weshall test in reaction functions for moneysupply in countries other than the UnitedStates is changes in central bank holdingsof official reserves. This should enable us totest the hypothesis that the United States ex-ported inflation to the rest of the world dur-ing the 1960s. We should also be able totest if the balance of payments served at allas a constraint on U.S. monetary policy, orwhether the United States acted as an inde-pendent dominant money supplier.

We may also find it useful to break up thetime period when estimating reaction func-tions if we can determine that a significantchange occurred in U.S. and foreign centralbank behavior over time. In addition, it willbe instructive to use changes in purchasingpower parity as an alternative measure offoreign influence on domestic monetarypolicy, since changes in official reservesmay not represent a disequilibriumphenomenon but merely an increased de-mand by the foreign central bank for reserveholdings.

Another possible international trans-mission mechanism we plan to investigateconcerns substitution between monies onthe demand side of the market. There may

77

be competition between alternative moneysupply issuers in the international moneymarket. This would imply that even under asystem of flexible exchange rates, an indi-vidual country will not be completely insu-lated from money supply changes abroad,since such foreign changes could influencethe demand for the country's money throughinternational substitution (e.g., if individualsin England increased their demand formarks and reduced their demand forpounds in reaction to a U.K. inflation). Aninitial crude test would entail includingforeign interest rates in each money de-mand equation. Since limited degrees offreedom are available, it may be difficult tocarry out this test.

We have not yet begun the modeling ofthe international transmission element of thefull model, and that will be our major taskduring the next few months of our work. Afterwe estimate our complete model we intendto compare the differential reaction of realincome and prices to monetary shocksamong the different countries to see if thefraction of the shock that affects real mag-nitudes (the coefficients on the effects ofunexpected changes in the real incomeequation) is positively related to price un-certainty over the last two decades. A usefulstarting point for such a measure of averageprice uncertainty would be the standard er-ror of estimate of an ARIMA process on therate of inflation for each country.

The statistical estimation of the unifiedmodel presents the formidable data require-ments of a postwar quarterly data base foreach of the eight major countries studied.International data are uneven in availability,consistency, and accuracy.

This study is being financed by moneygrants from the National Science Founda-tion—Research Applied to National Needs,the Scaife Family Charitable Trusts, the AlexC. Walker Educational and Charitable Foun-dation, the Reim Foundation, and a grant ofpart of Lothian's time by Citibank.

Anna J. SchwartzMichael R. Daiby

Benjamin KleinJames R. Lot hian

The Effect of'World CommodityPrices on U.S. ManufacturingPrices

This study, which is part of a broaderexamination of short-term price determina-tion and particularly of the relation of outputprices to input prices, is focused on the dis-tinction between inputs for which prices aredetermined in international markets andthose for which prices are affected mainlyby U.S. developments. It asks how much ofoutput price change can be attributed toprices of traded inputs and whether theresponse of output prices to prices of tradedinputs differs from the response to prices ofdomestic inputs.

Price indexes were constructed foroutputs and inputs in 54 U.S. manufacturingindustries covering 79 percent of the manu-facturing sector, A matching of input andoutput price indexes had not been availablefor detailed industries before. The inputprice indexes were divided intotraded and domestic products, with theformer defined by the fraction of domesticsupply imported and exported. The seriesso far compiled run from 1968 to May 1975.The price indexes are supplemented byseries for each industry on straight-timeaverage hourly earnings of productionworkers.

The first step in the analysis was to esti-mate the time pattern of effects of inputprices on output prices. These estimateswere derived from regressions of quarterlychanges in output prices on changes ininput prices and earnings. It was assumedthat changes in materials costs and wageswere passed through to output prices overthe period of a year and that demand condi-tions did not appreciably affect the patternof the pass through. Lag patterns of inputprice effects covering the concurrent andtwo lagged quarters were estimated foreach industry. There is no indication thatdomestic and world-traded inputs had dif-ferent lag patterns.

The second step was to trace pricechanges of world-traded goods through the

78

manufacturing sector from one industry tothe next. Feedback of output prices in sell-ing industries on the input prices in pur-chasing industries was taken into account inthis analysis, but not feedback on wages.For example, prices of grains and soybeansrose 110 percent from third-quarter 1972 tothird-quarter 1973. Together they accountfor 1 7.5 percent of the materials inputs to theprocessed foods sector. The effect on anaverage of all 54 industries weighted bygross outputs was to raise the averageoutput price 2.2 percent by third-quarter1973 and 3.2 percent by second-quarter1974. The actual increase in the averageover this entire period was 26.3 percent. Wecan therefore state that these inputs ac-counted for (3.2/26.3 =) 12 percent of theincrease in manufacturing prices. Such cal-culations will be carried out for all world-traded goods.

Phi/lip Cagan

Price-Quantity Relations in U.S.Trade

This project is part of a series growing outof our earlier work on international pricemeasurement and the analysis of responsesto price changes. Much of the work duringthe past year has been on differencesbetween export and domestic price move-ments, the role of export prices in the trans-mission of inflation, and the relationshipbetween prices and exchange rates.

Several papers have been completed orpublished during the year. 'Export Pricesand the Transmission of Inflation" waspublished in the American EconomicReview, Papers and Proceedings, February1977. The results reported there indicatedthat commodity markets for manufacturedgoods are sufficiently tied together that arise in an important country's domesticprices is reflected in other country's prices.However, the adjustment is often incom-plete, is delayed a year or so, and is fasterand larger for the affected country's exportprices than for its domestic prices. There isthus a flexibility in the links between na-

tional price systems that is not always takeninto account.

Three working papers have also beendrafted: Export and Domestic Prices UnderInflation and Exchange Rate Movements"(NBER Working Paper 176), "Price Behaviorin the Light of Balance of PaymentsTheories" (NBEA Working Paper 181), and"Export Prices and Exchange Rates," withEliot A. J. Kalter (NBER Working Paper 182).The first is an expanded version of the AERpaper, and is a study of the response of ex-port and domestic prices in three coun-tries—the United States, Germany, andJapan—to foreign prices and exchangerates, The second, a comparison of pricemovements with those that would be ex-pected under various theories of balance ofpayments adjustment, was presented at aconference on Purchasing Power Parity inAthens, Greece, and will be published in theJournal of International Economics, May1978. The third is a study of the pass-through of exchange rate changes to U.S.export prices.

The price data available for these studies,extending through 1974 or 1975, now in-clude U.S., U.K., German, and Japanesedomestic price indexes and German andJapanese export price indexes for all manu-factured goods. In addition, there are U.S.export price indexes for machinery andtransport equipment.

Further work is under way, using ourinternational price competitiveness indexesfor machinery and transport equipment toexplain changes in the volume of exports.We are extending some of our earlieranalyses through 1975 to cover the period ofmore flexible exchange rates and are alsoexamining exports to particular destinationsand for more detailed commodity groupsthan in earlier studies.

The financing for this research was origi-nally provided by the National ScienceFoundation and has been supplemented bythe Office of Competitive Assessment of theU.S. Department of Commerce and by theU.S. Department of State. Mary Boger,Marianne Rey, and Judy Rosenzweig were

79

responsible for data collection andprogramming during the past year.

Irving B. Kra v/sRobert E. Lipsey

Studies of Multinational FirmsThe program of research on multinational

firms is a set of related studies, most ofwhich make use of the National Bureau'sdata base on U.S. firms and the reports on•U.S. companies and their foreign affiliatescollected by the Bureau of EconomicAnalysis of the U.S. Department of Corn-me rc e.

The study of the impact of multinationalfirms on technology and trade flows wascompleted during the year with a final revi-sion of the paper by Arthur Lake, "ForeignCompetition and the U.K. PharmaceuticalIndustry" (NBER Working Paper 155,November 1976). Plans are now being madefor a follow-up study which will concentrateon some of the less technologically ad-vanced industries, in contrast to the phar-maceutical and semiconductor industrieswhich were the subjects of special studiesin the first project. Transfers of technology inthe form of production methods or newproducts and the effects of such transfers ontrade flows would again be the focus of theresearch, but with greater attention to U.S.imports than in the earlier work. We alsoplan to carry out some studies of the opera-tions of U.S. firms in several countries orareas, comparing them to local firms withrespect to size, factor proportions,efficiency, and other characteristics, andinvestigating the impact of the activity ofU.S. firms on locally owned companies.

Some exploratory work was done duringthe year on a study of the factors determin-ing the location of manufacturing opera-tions, and particularly manufacturing for ex-port, by U.S. companies. We have been try-ing to develop appropriate models ofoverseas activities of U.S. firms and alsodata on costs of labor in different countries,adjusted for the wide variation in educationand other elements of labor quality. We also

have been experimenting with measures ofthe costs of fixed capital, making use of thecapital equipment and construction pricesdeveloped in the United Nations project onreal income comparisons. We hope to com-plete a preliminary working paper on thistopic shortly and are seeking support forfurther research.

A new project on the relation betweendomestic and overseas capital investmentby individual U.S. firms, in which GuyStevens of the Federal Reserve Board willbe cooperating with Robert Lipsey, is justgetting under way. The first step in the studyis the assembling of some unusual datawhich will include, for over a hundred indi-vidual firms, information on sales, salesforecasts, investment plans, actual invest-ment, and other variables, divided betweenactivities in the United States and activitiesoverseas, The study will try to assess theextent, if any, to which investment overseasis a substitute for investment in the UnitedStates by the same firm or, possibly, itscompetitors.

Another current study, part of the projecton alternative trade strategies and employ-ment, is concerned with factor use by U.S.multinational firms in less developed coun-tries. We find, in virtually all industry groups,higher capital-labor ratios in U.S. parentcompanies than in their affiliates in

developed countries and higher ratios in

those affiliates than in the ones in lessdeveloped countries. We are trying to learnhow much of this difference is explained byindustry mix within industry groups, howmuch by the use of different technologies,and how much by adaptation within giventechnologies. We are hoping to use for thispurpose, in addition to the U.S. data, similarinformation on the foreign operations ofSwedish firms, collected by the IndustriensUtredningsinstitut. Another subject of studyis the differences in factor proportionsbetween U.S. affiliates and other firms inhost countries. Romualdo Roldan of theUniversity of Pennsylvania is working withus on this topic.

The National Bureau's studies of multina-tional firms were begun under grants from

80

the National Science Foundation and theFord Foundation, and the more recent workhas received financing from the Agency forInternational Development, the U.S. Depart-ment of Labor, and the Treasury Depart-ment, as well as assistance in kind from theFederal Reserve Board. In many of thesestudies we are indebted to the Bureau ofEconomic Analysis of the U.S. Departmentof Commerce for the use of their data and toArnold Gilbert and Michael Liliestedt of theBEA for programming and advice in usingthe data,

Irving B. KravisRobert E. Lipsey

Alternative Trade Strategies andEmployment

The past year has been devoted to gather-ing and analyzing data for individualcountry studies. In late August, all projectparticipants assembled for a second work-ing meeting in which preliminary findingswere presented and research strategies dis-cussed. It is expected that all data will havebeen gathered by the time this report ispublished. There are already first drafts ofsome country studies, and it is hoped thatall country studies will be completed by thesummer of 1977.

Among the preliminary findings ofinterest, perhaps the most striking is the uni-formity with which country authors are find-ing that exports to other LDC's are far morecapital using and less labor using than theirexports to developed countries. In a numberof studies skill intensity has been found tobe even more significant than capitalintensity in distinguishing between exportand import-competing industries, but thoseresults are less uniform across countriesand require further analysis.

Anne 0. Krueger

Selected Issues in Trade Policies ofThailand and Trade Strategies forEmployment Growth

Since the beginning of my fellowship, inJuly 1976, I have been engaged in two

major areas of research. One is selectedissues in trade policies of Thailand and theother is trade strategy for employmentgrowth. Two issues of trade policy wereselected: rice export policy and cooperationin trade policies with ASEAN countries. Thestudy on trade and employment follows theoutline of the NBER project of the same title.

The study on rice export policy hasproduced three papers, which will bepublished in the forthcoming Stanford FoodResearch Institute Studies, Volume 25,Number 2: "Comparative Advantage in RiceProduction: A Methodological Introduction"(written with Scott A. Pearson and GeraldNelson), "Comparative Advantage in RiceProduction in Thailand" (written with At-chana Wattananukit), and "Comparative Ad-vantage in Four Rice Producing Countries"(written with Eric Monke and Scott R.

Pearson).The first paper contains a discussion of

the concepts and methodology to be usedfor policy analysis of a commodity such asrice. These include private and socialprofitability, nominal and effective protec-tion, and domestic resource cost. In thesecond paper, these concepts were used toassess the rice export policy of Thailand. Itwas found that Thailand had a very strongcomparative advantage in rice export, as in-dicated by social profitability and domesticresource cost. Yet the results from nominaland effective protective rates showed veryclearly that government policy discouragedrice export. In the third paper, the results forThailand were compared with those for theUnited States, Taiwan, and the Philippines.Among the four countries, Thailanddemonstrated the strongest comparative ad-vantage and the Philippines the least. Theresults were used to analyze the ricepolicies of these countries.

The study on ASEAN produced one paperentitled "Development of Trade Policies inThailand and Prospects for Trade Coopera-tion with ASEAN," which will be publishedby the UnitedNations Asian DevelopmentInstitute in Bangkok. The paper outlines thedevelopment of trade policies from the endof the Second World War up to 1976. It

81

points out that the policies have been quiteliberal, with attempts at import substitutionin the 1 960s and export expansion from theearly 1970s up to the present. At the nextphase of import substitution, economy-of-scale considerations call for a largermarket, and this could be accomplishedthrough cooperation with ASEAN countries.Cooperation is thought to be viable becauseof some complementarity of Thai exportproducts with ASEAN countries.

The project on trade strategies for em-ployment growth has been divided into fourparts: analysis of trade pOlicy, factor propor-tions of trade, distortions in factor pricesand factor markets, and trade and employ-ment. I had written a paper on the first partbefore coming to the National Bureau. Thepaper was revised during my stay at the Bu-reau, and this version will be published inthe Papers and Proceedings of the EighthPacific Trade and DevelopmentConference, to be distributed by theUniversity of Hawaii Press in mid-1977. Inthe paper I present results on nominal andeffective protection as indicators of tradepolicies, and analyze employment implica-tions of these policies. I conclude thatpolicies in the early 1970s encouraged thesubstitution of capital for labor.

Second drafts of papers on the secondand third parts have been written. The firstpaper is a report on factor proportions intrade. Import-competing industries werefound to be more capital intensive than ex-port industries, both in 1971 and 1973. Fac-tor intensities were also consistent with thefactor proportion theorem in terms of the di-rection of trade. Finally, it is noted that fac-tor-intensity differentials were very large.

The second paper is on wage differentialsin the manufacturing sector. It is shown thatthere were substantial wage differentials inthe sector and that the most importantexplanatory factors were the male-femaleratio and size of the firm. The tentative con-clusion is that wage distortions and dif-ferentials exist in the Thai manufacturingsector.

Research results described above werepresented at seminars at various

universities and institutions. Results in rela-tion to the trade and employment projectwore presented at the University of Minne-sota, Yale University, the Agency for Interna-tional Development, the World Bank, andthe University of Hawaii. The paper onASEAN was presented at HarvardUniversity, Simmons College, and the AsiaSociety in New York. The rice export policypaper was presented at Boston University,the University of Minnesota, and the Na-tional Bureau in New York.

I am currently completing my research ontrade and employment. I am revising thework on factor proportions and factor pricedistortions and analyzing the implications oftrade policies on employment.

Narongchai Akrasanee

A Model of Trade and Developmentin the Pacific Basin

The purpose of this project, which beganJuly 1, 1976, is to study the interde-pendencies between developed anddeveloping economies in the Pacific Basinand to assess the effects of introducing newsystems or rules for international economictransactions in that region. The PacificBasin regional economy includes all the na-tional economies that border on the PacificOcean, except the Latin Americaneconomies. The latter are excluded on thegrounds that their trade and investment tiesare primarily with the United States orEurope and almost negligible with the restof the Pacific Basin. The Pacific Basin re-gional economy has been one of the fastestgrowing regions in the world in terms ofinternational trade and foreign investment.

The project will be concerned with eco-nomic policy issues which have interna-tional implications, such as the impact ofgrowth recessions in the developedeconomies, changes in export and importprices due to proposed preferential tariffschemes and to the formation of an ASEANcustoms union, and exogenous changes inforeign aid or net capital flows.

The approach is based on building andoperating an econometric model of interde-

82

pendencies among the developed anddeveloping economies in the Pacific Basin.Individual national models for the variouscountries will be linked through a centralmodel of the trading relationships amongthe Pacific Basin countries and betweenthem and other regions of the world. Thecomplete system will be used to analyze theproblems of international transmission ofeconomic disturbances and the organiza-tion of alternate trade systems.

The countries and regions that will be in-dividual ly distinguished are Australia,Canada, the Republic of China, the People'sRepublic of China, Hong Kong, Indonesia,Japan, the Republic of Korea, Malaysia,New Zealand, the Philippines, Singapore,Thailand, the United States, the U.S.S.R.,Western Europe, oil-producing countries ofthe Middle East, other Middle Eastern coun-tries, oil-producing countries of Africa, otherAfrican countries, oil-producing countries ofthe Western Hemisphere, other WesternHemisphere countries, Socialist countries ofAsia, other Asian countries, Socialist coun-tries of the rest of the world, and rest ofOceania.

The goals for the initial year include thedevelopment of the basic trade and nationalincome data files and the construction of thetrade model. The basic trade data file is nowcomplete. Trade matrices representing theflows of trade within the Pacific Basin andbetween it and the other regions have beenassembled for each year from 1948 through1975. From these trade matrices, completebilateral trade balance matrices have alsobeen constructed, giving the trade surplus(or deficit) of any one country or region withany other. Trade shar2 matrices based onimport and export shares have also beenconstructed for each year to reveal thechanges in the structure of imports by originand exports by destination of each countryand region over time. Work has also beenstarted on the compilation of capital flowmatrices for the Pacific Basin for selectedrecent years.

In parallel with the trade-modeling effort,a complete national income accounts datafile will be set up for each of the countries

and regions included in the study. This datafile will be used in the construction of indi-vidual-country econometric models. Work isunderway on the specifications of a pro-totype national model incorporating therelevant features for the proposed regionalsystem.

Bert G. HickmanLawrence J. Lau

Latin American ComputerWorkshops

During October 1976 we had planned tohold a large conference on commoditymarkets, models, and policies in LatinAmerica, sponsored jointly by the NationalBureau, the Coffee Federation, and theBanco de Ia of Colombia, inBogota, Colombia. However, the conferencewas postponed at the request of the Bancode Ia Repiiblica and the Coffee Federation.

The discussion was to be organizedaround the following subjects: (1) primarycommodity exports of Latin America: inter-relationship with international and domesticmarkets; what can Latin America do ininternational commodity planning?; (2)long-term projections: forecasting the metalmarkets for today's investment decision;world sugar economy—an econometricanalysis of production and policies;programming approach to intercrop plan-ning; transnational commodity policies andnational interests; (3) international com-modity stabilization: compensatory financemechanisms; and (4) Latin America: anagenda for action in research and policy im-plementation.

Two other conferences have been tenta-tively scheduled for 1977. One, jointlysponsored with ECIEL (Programa de Estu-dios Conjuntos sobre lntegracibn Eco-nbmica Latinoamericana), is on educationand economic development in LatinAmerica. The other planned workshop willdeal with economic aspects of the role ofgovernment in the Latin Americaneconomies. It will be jointly sponsored withCIEPLAN (Corporaciôn de InvestigacionesEconômicas para Latinoamerica).

83

A series of articles presented at ourconference on indexation in Sao Paulo,Brazil, February 26—28, 1975, waspublished in Explorations in Economic Re-search, volume 4, number 1. Papers givenat the conference on planning and short-term macroeconomic policy in LatinAmerica, in Isla Contadora, Panama,October 31—November 2, 1975, are alsoscheduled to be published.

These workshops were made possible bya grant from the IBM World Trade Corpora-tion.

M. lshaq Nadiri

U.S-U.S.S.R. Scientific andTechnical Program of Cooperation

During 1976 the National Bureau of Eco-nomic Research continued to plan and coor-dinate a program of cooperation ineconometric modeling (Topic 1) and model-ing of large-scale systems (Topic 2) underthe U.S-U.S.S.R. Scientific and TechnicalProgram of Cooperation in the Field of Ap-plication of Computers to Management.There are presently five topics in the area ofapplication of computers to managementwhich have the approval of the joint U.S.-U.S.S.R. commission appointed to monitorand authorize these activities by the U.S.and Soviet governments. The chairman ofthe U.S. side of this area of cooperation isDonald Aufenkamp of the National ScienceFoundation.

The Bureau has been involved in theprogram for the last three years but will endits participation sometime in 1977, and ispresently arranging to transfer theseactivities to such other institutions in theUnited States as may be designated by theNational Science Foundation.

Progress was made during the year indevelopment of joint longer-term work onselected research topics by scholars of theUnited States and the U.S.S.R. The first ofthese longer-term exchanges took place inMay 1977, with Alfred Dale (University ofTexas-Austin) and Edward Sibley(University of Maryland) working in theU.S.S.R. for one month on data structure

models (Topic 1). This activity was piannedand contemplated in the protocols signedover the past three years. Additional long-term visits for subtopics under Topic 2 areexpected to take place later in 1977. Thefirst one contemplated is a visit by DonaldAufenkamp (NSF) to the U.S.S.R. for twomonths for research on information flows incomplex systems. A return visit to the UnitedStates by Soviet experts on computer in-formation systems is planned for late tall.

During the year the following technicalseminars were held:

Econometric Modeling of Various Aspects ofthe Economy—Moscow

Computers Applied to Planning andManagement of Large Agro-IndustrialComplexes—Moscow, Riga, andK i shin ev

United States:Management Information Systems—DenverData Base Management—AustinEnterprise and Decision Models—Wash-

ington, D.C.In addition, Alfred Dale (University of

Texas), Judy Thornton (University of Wash-ington), Holland Hunter (Swarthmore),

Robert Leone (Harvard University), andDavid Kresge (NBER) visited the SovietUnion during 1976 to discuss longer-termresearch exchanges between U.S. andU.S.S.R. scholars in the areas of transporta-tion, energy, enterprise, and decisionmodels, as well as data base management.

Plans for publishing the proceedings ofthe Econometric Modeling, Transportation,Management Information Systems, and Ag-ricultural seminars are complete and thesepublications will be available sometime inthe latter part of 1977.

During the year the Bureau's collection ofU.S.S.R. publications obtained over ourthree years of involvement in this programwas transferred to the University of Chicagoas requested by the National Science Foun-dation. This material will be maintained aspart of their larger collection of specializedSoviet publications on mathematics andcomputer-related sciences. These publica-tions will be translated and made availableto interested scholars through a programfunded by the National Science Foundation.

Harvey J. McMa insJohn R. Meyer

E. K. Smith

7. THE NBER URBAN SIMULATION MODEL AND EVALUATION OFTHE MARKET EFFECTS OF HOUSING ALLOWANCES

Completion of a three-volume final reportduring the past year fulfilled NBER's obliga-tions to the Department of Housing andUrban Development under its 1972 contractwith HUD to "improve the NBER urbansimulation model and to analyze housingmarket dynamics and abandonment." TheNBER Urban Simulation Modeling project,which received initial support from HUD inspring 1968, has been an ongoing project atNBER for nearly nine years. A final report toHUD in the fall of 1971 and a subsequentNBER book describe the progress madeunder the first HUD grant.1

1. Gregory K. Ingram and J. Royce Ginn, H. JamesBrown, and John F. Kain, "The NBER Urban SimulationModel,' vol. I, The Model Description," processed

84

In November 1972 the Urban Studiesgroup was awarded a second contract toimprove the exisiting NBER Model and touse it to analyze housing abandonment, asource of great concern to HUD personnel atthat time; our preliminary research efforts todevelop a model version suitable for analyz-ing abandonment are described in a Sep-tember 1973 interim report to HUD.2

In response to changes in HUD'spriorities, we modified our research objec-

(1968); and Ingram, Kain, and Ginn, The Detroit Pro-totype of the NBER Urban Simulation Model (1972).

2. John F. Kain, et al., "First Interim Report onContract to Improve the NBER Urban Simulation Modeland to Use It to Analyze Housing Market Dynamics andAbandonment," processed (September 1973).

tives on at least two occasions with theresult that the scope and content of the com-pleted work is far different from that outHnedin our original proposal to HUD.Specifically, in the spring of 1974 HUDrequested that we modify and extend ourexisting research contract and redirect ourmodeling efforts toward the analysis of themarket impacts of housing allowance pro-grams. While much of our previous efforts toadapt the NBER Model to the study of hous-ing abandonment was useful to our newassignment, the evaluation of the market im-pacts of housing allowances presented uswith a completely new and far more difficultset of conceptual and estimation problems.The Experimental Housing AllowanceProgram (EHAP) is principally concernedwith short-term market impacts and particu-larly with the nature and extent of rentinflation that might result from implementinga large-scale housing allowance. Housingmarket analysts generally agreed that themaximum rent increases would occur soonafter the program was introduced and thatallowance-induced rent inflation would tendto dissipate as housing suppliersresponded by constructing new units and byincreasing the services they produced fromexisting structures. Unfortunately, existingand proposed versions of the NBER Modelwere poorly suited to the study of short-termdynamics, a capability which was, however,not essential for the proposed analyses ofabandonment. Undertaking the housingallowance study thus required us to re-consider several aspects of model designalready incorporated into Pittsburgh II, theversion being developed to analyzeabandonment. Major changes involvedenrichment of the model's supply sector,greater elaboration of the marketsubmodels, and a much more precise dis-tinction between the production of housingcapital and housing services. The resultingimprovements are evident from the detaileddescription of model specificationpresented in Volume I of the final report.3

3. John F, Kain, William C. Apgar, Jr., and J. RoyceGinn, "Simulation of the Market Effects of Housing

85

A futher complication in revising our re-search objectives was HUD's urgent needfor quick answers. Our amended contract in-cluded two goals that were completely com-plementary in the long run, but somewhatcompetitive in the short run: (1) to refine theNBER Simulation Model for the purpose ofanalyzing housing allowances and housingabandonment; (2) to provide HUD withanalyses of the probable market impacts ofhousing allowances for its Fall 1974 internalevaluation of housing allowances. The im-probability of completing both tasks in thetime allotted was recognized, and in April1974, HUD and NBER agreed to pursue anadaptive strategy. NBER would give priorityto model development in the hope that theseefforts would produce a model suitable forsimulations of the market impacts of hous-ing allowances by Fall 1974. It was furtheragreed that we should jointly evaluate theseefforts during the summer and adjust ourwork program to reflect our progress inmodel development and HUD's priorities. Atthe July 1974 meeting at HUD we agreed todefer model development and toconcentrate our efforts on interim analysesof housing allowances. These interimanalyses were presented first in a series ofinformal working papers and then includedin a December 1974 report to HUD.4

Section I of the December 1974 reportsummarized a series of housing allowancesimulations carried out using the HousingAllowance Demand Simulator (HADS)Model, adapted expressly for this purposefrom the demand sector of the NBER UrbanSimulation Model. The simulations weredesigned to evaluate the effects of alterna-tive allowance programs on the demand forhousing attributes and bundles in the Pitts-burgh SMSA and to provide some indicationof the likely size of allowance-inducedmarket impacts, including possible priceinflation among specific housing sub-

Allowances, Vol. I: Description of the NBER UrbanSimulation Model," processed (August 1976).

4. John F. Kain, ed. "Progress Report on theDevelopment of the NBER Urban Simulation Model andInterim Analyses of Housing Allowance Programs,"processed (December 1974).

markets. Section II of the interim reportcontained a similar analysis of the impacton household demand of a housingallowance program in the San Francisco—Oakland metropolitan area. This analysis,based on an econometric study of the SanFrancisco housing market, employed quitedifferent estimation methods and evaluateda single housing allowance program thatbore only a superficial resemblance to thoseexamined with the HADS Model. An analysisof household mobility, program participa-tion, and the housing choices of elderlyhouseholds was presented in section Ill;section IV includes a series of simulations ofsupply-side responses to an allowance us-ing a stand-alone supply submodel; andfinally, section V described the progress todate in developing a version of the fullNBER Model that would be suitable foranalyzing the market impacts of housingallowances.

Following delivery of the report, the EHAPstaff agreed that NBER should devote fulltime to model development, calibration, andexecution of the previously agreed uponbaseline and housing allowance policysimulations for Pittsburgh and Chicago. Be-cause of the time and resources alreadyexpended on the interim policy analyses,however, this task required working beyond

IntroductionNew activities at the Computer Research

Center indicate a shift in emphasis towardenergy research and increased attention tomodel validation and reliability, Work hasalso continued in several applied researchareas as well as in the development ofanal.ytical methods, both theoretical andsoftware.

Energy research has various aspects. Themodel of the Alaskan economy, on whichwork being done concurrently at theUniversity of Alaska and here under DavidKresge's direction, pivots on the fiscal im-pact of North Slope petroleum. A newproject supported by the Electric Power Re-

the original contract completion date at noadditional cost to the government. Duringthis period, HUD accepted our proposal todevote resources previously earmarked forthe abandonment analyses to furtherimprovements in model calibration, andrequested only a brief working paper outlin-ing how the model could be adapted toanalyzing housing abandonment.

The major analyses completed under thiscontract are presented in the three-volumefinal report, "Simulation of the Market Effectsof Housing Allowances." Volume 1,

"Description of the NBER Urban SimulationModel," provides a detailed description ofthe current version of the NBER Model andits operation. Volume 2, "Baseline andPolicy Simulations for Pittsburgh andChicago," reports the results of baselineand housing allowance simulations for thetwo SMSAs, and "Development of the Sup-ply Sector of the NBER Model," the lastvolume of the final report, includes a seriesof simulation experiments using two stand-alone versions of the model's sector;analyses of the NBER landlord surveys inPittsburgh and Chicago; and analyses ofchanges in the location of new constructionin those two SMSAs.

86

John F. Kain

search Institute is examining the structuraland computer requirements for linking dif-ferent models and is exploring options in thedemands for basic energy resources, al-ternative conversion technologies, and im-pacts on the national economy. Design andimplementation of new computer software toexpedite complex energy modeling is nowcoming to fruition. The Computer Centerplans to expand energy research further inthe coming year, in collaboration with theMIT Energy Laboratory, by assessing the re-liability of several major energy models,also with Electric Power Research Institutesupport.

Work on numerical methods that are fun-

8. THE NBER COMPUTER RESEARCH CENTER

damental to valid estimation has continued.A unique interdisciplinary enterprise thatcombines statistics and numerical analysisinvolves Virginia Klema at the Computer Re-search Center and Gene Golub of Stanford.David Gay, in collaboration with John Den-nis of Cornell and Roy Welsch, has imple-mented semiportable, nonlinear algorithmsthat are at the forefront of computationalprocedures required for econometric ap-p11 c ati ons.

Mathematical programming research hastwo facets. Work on integer programming,with Jerry Shapiro and Bill. Northup, incorpo-rates number-theoretic considerations withmore standard methods for solving mixedinteger programs. This research capitalizeson interactive computing possibilitiesavailable at the Center. The other project, amodular LP system being created by RoyMarsten and Mike Harrison, provides avaluable modeling-language capability; themodularity of this system will make it anatural test bed for algorithmic research.Both systems have significant potential inenergy research.

An interest in research on model reli-ability is expected to gain momentum. Thecontinuing Bureau tradition of combining astrong data orientation with analysis gearedto data strengths and weaknesses has tosome extent been at odds with theincreased mathematical and econometriccomplexity of much contemporary research.Research on sensible ways of probing un-derlying data structure and its influence onestimated parameters began last summer.Roy Welsch and I found that perturbing datainputs of a regression problem, for instance,by deleting one row of data at a time (whichcan be done at slight additional cost)reveals unsuspected instabilities incoefficient estimates and can locate datasegments that exert a disproportionateinfluence on those estimates. It turns out that"good" regression results often depend toan uncomfortable extent on extreme, or highleverage, points. It seems possible, in thespirit of diagnostic inquiry, to obtain a betterunderstanding of the empirical foundationsof multivariate processes than is feasible

87

with standard procedures. David Belsleyhas also contributed to diagnostic researchby devising, readily applicable ways ofevaluating the sources and severity of col-linearity in linear regression. While still ex-perimental, these methods appear morepromising than competing alternatives.Douglas Martin has been working on statis-tical tests that have greater sensitivity to ashift in regression structure over time thanpresently available methods.

In summary, then, more effort has beendevoted to various facets of energy researchthan before; research on modeling funda-mentals continues to be a strong thrust; andattention is more than ever focused on diag-nostic procedures in regression that make itpossible to relate data directly toeconometric analysis.

Energy and EnvironmentalResearch

Man in the Arctic Program

Edwin Kuh

The development of policy planningmodels of the Alaskan economy is continu-ing as a joint project of the National Bureauand the University of Alaska's Institute ofSocial and Economic Research (ISER). Thiswork is part of the Man in the Arctic Program(MAP) funded by the National Science Foun-dation. A major component of the current re-search is to extend the existing MAP modelsto analyze the effects of petroleum develop-ment on the distribution of income in Alaska.The income distribution model will bebased on the results of recently completedMAP studies dealing with Alaska's man-power demands and supplies. Earnings dif-ferences among occupation classes, ethnicgroups, and regions will be combined withemployment estimates to project potentialchanges in Alaska's income distribution.Dan Seiver, who is based in Anchorage,spent several weeks at the Computer Re-search Center developing a preliminary ver-sion of the income distribution model.Seiver will continue his work on this modelduring the coming year, which he will spendin Cambridge as an NBER Faculty Fellow.

The income distribution model will alsodeal explicitly with the incomes of AlaskaNatives. In doing so, it will consider the ef-fects of possible changes in Native laborforce participation rates and the impacts ofthe land and money coming from the AlaskaNative Claims Settlement Act (ANCSA). Apreliminary version of the necessary model-ing has been completed and implementedusing aggregative data to evaluate the im-pacts of ANCSA on the economic status ofAlaska Natives. Work is also under way todevelop a model of Native populationgrowth, incorporating reasonably detaileddemographic factors. The early results com-ing out of the Native Claims studies indicatethat the impact of the ANCSA cash settle-ment will have minimal effect on the gapbetween Native and non-Native incomes; itis the land settlement portion of ANCSA thatwill provide the bulk of the long-range in-come gains. The results also indicate thatprograms designed to bring Natives into theprojected employment structure through jobtraining and education can be far more ef-fective than direct cash payments in improv-ing their economic status.

Wayne Christian has provided the TROLLprogramming support for all aspects of MAPactivities. Bernice Jee has been responsiblefor coordinating the data collection andsimulation analysis in the Native Claimsstudies. Jee has also assisted in thepreparation of a book-length manuscriptpresenting the MAP research results to date,which was submitted to publish'ers thisspring.

Edward Porter has constructed a first-pass model to estimate the economic im-pacts of petroleum development on Alaska'sOuter Continental Shelf (OCS). The modelfocuses on the state and local governmentrevenues generated by OCS activities andon the increased government expendituresrequired by the associated expansion ofAlaska's population. The OCS developmentprocess itself is modeled using thetechnological information from the environ-mental impact statements produced by thefederal agencies involved. Throughlinkages with the MAP economic-demo-

graphic models, the OCS impact studytraces out the induced changes in income,employment, and population throughout thestate. Although the OCS model is beingdeveloped in the Alaskan context, it is beingdesigned to be generally applicable so thatit can be extended to analyze OCS develop-ments in other areas as well.

88

David Kresge

Energy Model Integration and Assessment

Two projects have been undertaken dur-ing the past year dealing with the designand evaluation of computer models forenergy-systems analysis, Both projects arefunded by the Electric Power Research In-stitute (EPRI). The primary objective of thefirst, the energy-model integration project, isto design a framework for linking togethermodels of individual energy sectors so as todeal with the energy system as a whole. Inaddition, the modeling framework would in-corporate a model of the U.S. economy sothat the energy analysis explicitly recog-nizes the key energy-economy rela-tionships. The project is to draw, to themaximum feasible extent, on existingstudies, data bases, and methodologies indesigning the integrative framework.

The National Bureau's primary responsi-bility within the energy-model integrationproject, which is being conducted jointlywith three other research organizations, is todesign an appropriate set of interfaces tolink existing models, and then to define thecomputer environment and operatingsystems required to implement such in-tegrative energy models. Royce Ginn hasoutlined the characteristics of computersystems that would be appropriate forenergy modeling at several different levelsof complexity. In doing this, he has drawn onthe expertise and software being developedin the National Bureau's energy-modelingresearch.

An appropriate modeling system wouldfirst need to be capable of maintaining andmanipulating the data bases required todrive the major energy models now in use.Next, the modeling framework would needto provide the interfaces to handle the in-

formation flows required to operate thevarious models together in a common envi-ronment. Finally, the modeling systemwould need to carry out the analytical com-putations and process the inputs andoutputs required by substantive energy-policy analysis. The final report on thisproject, which was submitted to EPRI in the.spring of 1977, contains specific recom-mendations on the energy models whichmight be included within such an integrativeframework and on the procedures requiredto implement the framework for energy-policy analysis.

The objective of the second energyproject, which began in March, is to developprocedures for assessing the validity ofenergy models. A number of recentlydeveloped energy-system models are start-ing to play an important role in the processof energy-policy evaluation. However,detailed knowledge of these models isavailable only through the documentationthat the individual modeler releases, andthat documentation is often incomplete orinadequate for a full understanding of thestrengths and weaknesses of the model. Toprovide adequate assessments of large,complex models, they must be run andtested by someone other than the originalmodelers. In this project, the National Bu-reau has been asked to develop a prototype,third-party energy-model evaluation andassessment capability. This work will be un-dertaken jointly with the M.I.T. Energy Labo-ratory.

The model which has been selected asthe principal test case for the assessmentfacility is the Baughman-Joskow model ofthe electric utility industry. Royce Ginn,Wayne Christian, and Rick Wilk will put themodel up on the TROLL system. NationalBureau and M.I.T. Energy Laboratorypersonnel will then carry out detailedsensitivity analyses to pinpoint the model'scritical points. In addition to the in-depthassessment of the Baughman-Joskowmodel, an overview assessment (not involv-ing "hands on" operation of the model) willbe made of the Wharton annual energymodel. Particular attention will be given to

89

the model's potential usefulness as a frame-work for linking together other, moredetailed energy models. The energy-modelassessment project is closely related to thework being conducted as part of the energy-model integration project.

David Kresge

Energy Modeling Software

The Computer Center has completed thecontrol software for its energy-modelingsystem. The overall purpose of the system isto provide a comprehensive environment foreconomic analysis of contemporary energyproblems. The system has been designedso that system-control software is separatefrom the applications (problem-solving)software; thus, the effort of implementingnew applications is reduced. The controlsoftware includes the operating systemACOS (Application Control System) and thespecial-purpose programming languagesACOL (Application Control Language) andDASEL. The initial application software in-cludes subsystems for creating, estimating,and simulating econometric models, andthe AMPS subsystem for creating and solv-ing mathematical programming models.

ACOS has been completed and hasproved to be a flexible, reliable environmentfor the applications. ACOL has also beencompleted and has been used extensivelyas the command and control language forthe Center's new applications. The docu-mentation for ACOS/ACOL has been revisedand is ready for outside users of the system.Walter Oney is the principal implementor ofACOS, and Annette Somers, with the aid ofMark Gelfand, Maurice Herlihy, and DavidRice, has been maintaining and expandingACOL.

DASEL, an interactive programming lan-guage for mathematical algorithms, hasbeen used in a number of applications,including experimental regression andsimulation programs, a simple linearprogramming system, and statisticalanalysis research. The DASEL referencemanual has been substantially expanded toinclude the newest features of the language,including time series data. The functional

capabilities of DASEL have been extendedby allowing FORTRAN and PL/I proceduresto be called by DASEL via a user-controlla-ble interfaáe. A new utility allows transfer ofdata between DASEL and TROLL, and thisfacilitates both the verification of applica-tion results and the combined use of the twosystems to extend existing applications.David Rice is responsible for maintainingDASEL, Maurice Herlihy and Richard Wilkhave designed and coded the additions tothe language, and Wayne Zafft is the prin-cipal author of the reference manual.

In the near future, the ACOS environmentwill be combined with that of TROLL so thatthe application subsystems of TROLL willbe available in the energy-modelingsystem. Once this is done, we will have asingle environment incorporating all thefunctional capabilities of TROLL, the cOm-mand and control processing of ACOL, andthe algorithm development of DASEL.

As interest in modeling energy systemshas grown, the analytic and theoreticalmethodologies used to analyze thosesystems have also developed. Some of themethods involve well-understood tech-niques used in other research for manyyears; others model specific energy ap-plications; some use new analytic tools cur-rently being researched; and finally, otherslink up some combination of these tech-niques, models, and tools.

The ACOS environment is designed to ex-periment with the implementation of energymodeling. Two well-understood techniquesare linear programming and regression.Mike Harrison has programmed the linearprogramming model editor (XME) andproblem solver (XLP). An experimentalmodel translator (matrix generator) andreporter is being coded with the aid of othersupport (XMP). An experimental regression-model editor and estimator has beenprogrammed by Mark Gelfand and MauriceHerlihy. The system solves difficultnonlinear least squares problems usingJohn Dennis' optimizer, MINOPT. The mostchallenging nonlinear problems will soonbe solved by the state-of-the-art optimizer,DMINIMIZE, being developed at the Bureau.

A host of exploratory data analysis tech-niques advocated by John Tukey, RoyWelsch, and Dave Hoaglin are growing inimportance. Some of these techniques havebeen implemented as DASEL functions byStephen C. Peters.

Industrial Demand for Energy

Mark Ge/fandDavid Rice

Richard Wi/k

Our research has been focused on fourtopics: interfuel substitution in two-digit in-dustries; substitution among energy,capital, and labor; technical change inenergy use; and the dynamic structure ofenergy demand.

Interfuel substitution is examined by esti-mating cost-share equations derived fromtranslog unit cost functions with cross-sec-tional state data for 1971, 1962, and 1958.The results indicate considerable variationin energy substitution both across industriesand across types of energy. Estimated own-price elasticities for all types of energy ex-cept electrical are generally substantiallygreater than unity.

Substitution among energy, capital, andlabor is examined with data from the 1958Census of Manufactures. Preliminaryestimates of cross-price elasticities indicateconsiderable variation in substitutionbetween different types of energy andnonenergy inputs. The results also indicatethat energy inputs are not separable fromcapital and labor for all industries.

The dynamic structure of energy demandis analyzed with the Nadiri-Rosen disequi-librium model of demand for factors ofproduction. Results include estimates ofshort- and long-run price elasticities as wellas the time path of changes in the responseto price. The response of demand for eachinput to excess demands for other inputs isalso estimated.

Technical change in energy use isexamined with time series data for primarymetal industries. The results indicate thattechnical change has occurred through fac-tor augmentation at unequal rates.

Research results have been reported at

90

the 1975 and 1976 meetings of theEconometric Society, the 1976 meeting ofthe Western Economic Association, and the1976 Atlantic Economic Conference."Energy Substitution in U.S. Manufacturing"is forthcoming in the Review of Economicsand Statistics and "A Dynamic Model ofEnergy Demand in U.S. Manufacturing" hasbeen submitted for publication.

This research, which has been conductedthrough the NBER office in Palo Alto and,separately, from the Computer ResearchCenter, has been supported by a grant fromthe National Science Foundation-ResearchApplied to National Needs (RANN). Thefinal report for the project is available as adiscussion paper.

Robert Ha/vorsen

Regulation of the Pufp and Paper Industry

With the aid of a computer simulationmodel of the U.S. pulp and paper industry,we are examining several political and eco-nomic consequences of water pollution con-trol regulation. Our research is based on twogeneral premises.

First, environmental control policiesdevelop in an evolutionary fashion; thus, anyregulatory outcome is the product of boththe legislative strategy followed byCongress in establishing policy and theadministrative strategy of the EnvironmentalProtection Agency in implementing policy.

Second, there are both transitory andlong-run consequences of pollution controlregulation for industries and regions. The in-traindustry and interregional distribution ofthese consequences—both costs andbenefits—are major inputs to this evolu-tionary process.

The research has required the estimationof the costs the pulp and paper industry willincur to comply with the various require-ments of the 1972 amendments to theFederal Water Pollution Control Act. Currentefforts are directed at the refinement of acomputer simulation model that permitsthese marginal compliance costs, disaggre-gated by major product line, to interact withdemand to yield higher sustainable pricelevels. These prices, in turn, generate cash

91

flows and investments, which yield timepaths of adjustment for prices, output, ca-pacity, and investment for various marketstructure scenarios, e.g., pure competitionand various oligopolistic pricing scenarios.Plant-level impacts, measured both bychanges in cost and net worth, will becomeinputs to the political analysis.

The political analysis will utilize simplemodels of self-interest and more complex"log-rolling" models to identify the likelyresponse of individual legislators to aspectsof the regulatory package: timing, coverage,stringency, etc. Congressional behavior willbe modeled as a function of constituents'demand for clean water, the economic im-pact of the regulations on the districts, thelobbying efforts of industry and environ-mentalists, and the institutional structure ofCongress. In modeling the industry position,"perceived" self-interest (measured in termsof compliance costs) will be distinguishedfrom "actual" self-interest (measured bysimulated changes in net worth). The dy-namic element enters through considerationof the behavior of industry and Congress onsubsequent legislation in which standardsare adjusted and agency funds appro-priated.

The research results will be useful to twodistinct groups: policymakers, who willbenefit from an improved understanding ofregulatory impacts and the evolutionaryprocess of regulatory development, andprivate-sector decision makers responsiblefor developing corporate strategies regard-ing capacity decisions in increasingly regu-lated industries.

Robert A. LeoneJohn E. Jackson

Economic Modeling, Estimation,and Analytical MethodsPrice Expectations and Adaptation inAmerican Agriculture

In this study, alternative ways of estimat-ing directly or modeling unobservable priceexpectations are explored. These methodsare then applied to models of U.S. agricul-

ture in order to investigate the process of ex-pectation formation and revision. Oneprimary goal is to explore the process of ex-pectation formation over long periods oftime in order to examine the impact ofexogenous events, such as changes ingovernment policy, on the expectationprocess.

A major activity of this project to date hasbeen the preparation of the data to be used.In order to study the impact of exogenousevents on the formation of expectations, anextensive historical data bank is required.We have compiled cross-sectional timeseries data for every major American agri-cultural crop; the sources are published andunpublished reports of the U.S. Departmentof Agriculture. The data include observa-tions by state, from 1 867 to 1970, on prices,acreage planted, acreage harvested, yield,and production.

In addition to the data on field crops, wehave obtained comparable data onlivestock from Stephen DeCanlo of YaleUniversity. DeCanio and I are also compil-ing disaggregated temperature and rainfalldata from various sources from the mid-1880s to the present. We are also compilingas much data as possible on factor pricesand inputs, although these are more sparsethan we had hoped.

The crop and livestock data are scatteredthrough the publications and archives of theU.S. Department of Agriculture, and havenever before been collected into a singleunified sample. The entire data set will beavailable both on magnetic tape and as partof a TROLL data archive. Details of the com-pilation of each series, reconciliation of dataanomalies, and sources will be contained inan NBER working paper which is in prepara-tion.

The second phase of this study is thecomparison of methods for the estimation ofunobservable price expectations. One suchmethod is described in my paper, "A State-Space Approach to the Estimation of PriceExpectations," Proceedings of the IEEEConference on Decision and ControlTheory, December 1976. Alternativemethods using the unobservable

92

component approach of Joreskog are nowbeing explored.

The third phase of this study is the ap-plication of those methods to the large databank that has been compiled. That phase isjust beginning; results are expected nextyear.

Thomas Cooley

Model Validation and Reliability

The funding last summer of a two-yearNSF grant has allowed us to make rapidstrides in the development of sensitive, sen-sible tools to detect potential problems inmodel estimation that may be caused by theexcessive influence of a few observations orby a departure from error-model assump-tions.

Three techniques have been used todevise diagnostic tools: differences, de-compositions, and derivatives. The first istypified by row deletion. One useful way todiscover infLuential data points is toexamine / — /3(/)where is the ordinary leastsquares (OLS) estimate for the standardlinear regression model, and is the sameexcept that the ith observation has beenremoved from the data, These differencescan be computed at little additional cost.

An example of the second technique isthe decomposition of the covariance matrix(XTX)_1 into parts related to each observa-tion, namely (XTX)_1 X. TX1 (XTX)1, wherex, is the ith row of X. These quantitiesprovide another way of finding influentialpoints and ascertaining how they affect theestimated coefficient variances.

The third technique, differentiation, canbe applied in many contexts. A disad-vantage of the Durbin-Watson statistic isthat it does not indicate how first-order auto-correlation might affect the individualcoefficient estimates. Differentiating thegeneralized least squares (GLS) estimates(based on a first-order autoborrelation ma-trix) with respect to p provides a way ofmeasuring the sensitivity of the coefficientestimates to different amounts of autocor-relation.

This year our work has focused on single-equation linear models, but in the second

year we envision work on simultaneousequation and nonlinear models. A TROLLsubsystem called SENSSYS, which containsa variety of diagnostic tools, has beendeveloped and documented by Steve Petersand David Jones.

We have summarized a portion of ourwork on single-equation models in "LinearRegression Diagnostics" (NBER WorkingPaper 173),

Collinearity Analysis

Roy E. WelschEdwin Kuh

A new procedure can diagnose the pres-ence of multicollinearity in data to be em-ployed in linear regression and can assessthe potential damage that such collinearitycauses in the resulting least squares esti-mates. This diagnostic tool is now incor-porated in the SENSSYS system as yARD-COM. The procedure and its theoreticalfoundations are described in my "Multi-collinearity; Diagnosing Its Presenceand Assessing the Potential Damage It

Causes Least Squares Estimation" (NBERWorking Paper 154). Research continues onextending these procedures to simul-taneous equation and nonlinear estimators.

David A. Beisley

Simultaneous Equation Estimation

The nonlinear full-information maximumlikelihood (NLFIML) and nonlinear three-stage least squares (NL3SLS) estimators,recent additions to the GREMLIN subsystemof TROLL, are being subjected to compara-tive testing. Preliminary results for NLFIMLare encouraging; Kent D. Wall and I reporton them in "Estimation of EconometricModels Using Nonlinear Full-InformationMaximum Likelihood: Preliminary ComputerResults" (NBER Working Paper 142). In

essence, the results indicate, contrary toconventional wisdom, that NLFIML is corn-putationally competitive with NL3SLS forsmall models (5 to 10 equations, 20 to 25parameters) that are not too nonlinear.Further testing is being conducted to addnonlinear full-information-efficient instru-mental variable (NLFIVE) estimation to the

93

GREMLIN arsenal and to examine its com-parative computational advantages.

David A. Beisley

Bayesian Estimation

Beginning in the summer of 1976, JosephB. Kadane (Carnegie-Mellon) and James M.Dickey (University of Wales) have been ex-ploring a Bayesian approach to predictionbased on the standard linear regressionmodel. The Bayesian approach to dataanalysis has attracted increasing attentionfrom economists; see, for example, thecontributions in S. Fienberg and A. Zeilner,Studies in Bayesian Econometrics andStatistics (North-Holland, 1975). Kadaneand Dickey have identified several interest-ing families of prior distributions and havedeveloped practical procedures for assess-ing these distributions for multidimensionalproblems.

I designed and coded the TROLLprogram BAYESREG as an initial experi-mental implementation of their computa-tional techniques. BAYESREG examines alinear model and then, in an interviewformat, determines the user's prior un-certainty concerning model parametersby eliciting assessments of the distributionof values of the dependent variable atselected values of the explanatory vari-ables. This information is combined with thesample likelihood to deliver posterior pre-dictive inferences.

We demonstrated these techniques at aBayesian econometrics conference in Juneand have been assimilating feedbackreceived there. Future work will extend therange of prior distributions that BAYESREGcan handle and will refine the interviewprocedure.

Stephen C. Peters

Structural Shifts in Time Series Models

Work is now well under way on the detec-tion and estimation of structural shifts intime series models. One problem is thedevelopment of statistics that will be sensi-tive to a type of structural shift which wouldbe expected frequently in practice. Anotheroroblem is related to the estimation of time-

varying parameter models of the Kalmanfilter variety.

A number of test statistics can be used forthe detection of structural shifts ,in timeseries regression models. Among the betterknown approaches are those based onChow's F tests and the Brown-Durbin-Evansforward cusum and cusum-squared statis-tics. More recently, Schweder has es-tablished certain optimality properties aswell as heuristic justifications for backwardcusum and cusum-squared statistics. Adrawback to these approaches is that theyare relatively insensitive to all but ratherlarge structural shifts. In order to overcomethis shortcoming a relatively simple statistichas been developed which is optimal for de-tecting small shifts having specified cor-relation structures. The full development ofthis approach required the following addi-tional work: (1) useful confidence intervalsbased on approximations or numericalinversion of characteristic functions, and (2)determination of the power of the test over arange of misspecified correlation structures.

The maximum likelihood estimation ofeconomic time series models of the Kalmanfilter variety has recently received some at-tention from econometricians. The approachis useful for estimating regression modelswith (randomly) time-varying coefficientswhich would account for structural change.Although the procedure has been used invarious forms, almost nothing is knownabout its statistical properties. Weconcentrated on evaluating Cramér-Rao—type lower bounds on the parameter esti-mates; evaluating the performance of theKF1 algorithm (a Computer Research Centeralgorithm) relative to the lower bounds viaMonte Carlo; and studying the likelihoodsurface for some simple cases. Manageableforms for the bounds have beenobtained, and the bounds may now beeasily evaluated. This work is being donewith Peter Caines of Harvard University.

Douglas Martin

Causality in Economic Models

In this study the basic theoremcharacterizing causal events and existing

94

testing methods were surveyed. Two al-ternative testing procedures weredeveloped to provide a unified approachtoward the problem of adjustment for serialdependance. The various methods were ap-plied to money stocks (Ml and M2) and cur-rent measures of GNP to detect possiblecausal relationships between money and in-come. A causal spectral analysis was usedto examine the sources of variation in thedata and to explore their lead-lag relation.

Cheng Hsiao

Errors-in-Variables Model

P. M. Robinson, of Harvard University,and I are studying the estimation ofparameters in a dynamic simultaneousequation model with stationary disturbancesunder the assumption that variables aresubject to random measurement errors. Anasymptotically efficient frequency-domainclass of instrumental variables estimatorwas developed. The procedure consists oftwo basic steps. First, the model istransformed in such a way that the observedexogenous variables are asymptotically or-thogonal to the residual terms. The secondstep is an analogue of the three-stage leastsquares estimator except that we take ex-plicit account of the nonlinearity in theparameters and nonuniform property of thespectral density matrix of the disturbanceterm. A report on this research is forthcom-ing in the International Economic Review.

Cheng Hsiao

Robust Regression

In collaboration with Edwin Kuh, RoyWelsch, and Douglas Martin, I have beenstudying the behavior of linear regressionestimators in situations where the carriers(independent variables) contain outliers. Arow of the design matrix is said to be an out-her if its removal could cause large changesin the least squares estimates of thecoefficients. If an individual observation ofany carrier is sufficiently large, the cor-responding row of the design matrix will bean outlier, but some outliers cannot be soreadily detected (that is, they are notassociated with the large entries of an mdi-

vidual carrier). We have developed andstudied new methods for detecting outliersin the carriers, and we systematicallystudied several estimators that are intendedto be robust even in the presence of outliersin the carriers. The most successful detec-tion procedures are based on the singularvalue decomposition of the design matrix,as are the most successful estimators.

The estimators studied include those sug-gested by Mallows and by Andrews. All aregeneralizations of the robust regression M-estimator. Although all have low efficiencieswith respect to least squares, if the randomerrors in the linear model are normally dis-tributed, we conclude that some of them canbe useful in exploratory data analysis. Wefind that the set of estimators studied can bedivided into three classes. The first classconsists of unsatisfactory estimators, andthose in the second and third classesexhibit, qualitatively different behavior whenapplied to certain data configurations. Thisqualitative difference can be attributed todiffering assumptions used to solveregression problems that are nearlyunidentified. That is, if there is a single verylarge outlier in the carriers, then thecoefficient estimates obtained by usingmethods that downweight this point are verydifferent from those obtained by methodsthat do not downweight. Yet in certaincases, the data alone are insufficient to de-termine whether the outlier should bedownweighted. Thus, different methods fordownweighting the outliers in the carrierscan result in very different sets of coefficientestimates.

Robust Statistics

Richard Hill

Interdisciplinary research in numerical al-gebra, robust statistics, and reliablesoftware is in progress. Mathematicalsoftware has been designed and developedand now serves as a vehicle of communica-tion among participating numerical analystsand data analysts. The design and in-linedocumentation of the software has made itsimmediate use possible by numericalanalysts, statisticians, and econometricians

95

who were not involved with the research onthe software or the numerical or statisticalalgorithms that were employed. The vehi-cles for testing the reliability of the softwarefor ROSEPACK (RObust Statistics Estima-tion PACkage) were used in testing thesoftware for nonlinear least squares. The re-search on nonlinear least squares was doneby John Dennis, David Gay, and RoyWelsch, with the programming andtechnical assistance of David Coleman,Sandra Moriarty, Steve Peters, and ScottSmolka.

The research on software for robust statis-tical estimators has been particularlystrengthened by the collaboration with Den-nis, Gay, and Welsch. Collectively, we havebeen able to meld the vocabularies of nu-merical algebra, data analysis, and robuststatistics to focus on the difficult problemsin robust statistics that have to do with nu-merical and computational stability.

Our research project has been influencedand strengthened by ties between the NBERComputer Research Center and Gene Golubof Stanford University. The software hasbeen strengthened by feedback on its use ina different computing environment by JohnSununu of Tufts University. Helpful com-ments and several computer runs byRichard Bartels of Johns Hopkins Universityhave helped to motivate some modificationsfor portability of input and output.

Next year we expect to add to our informa-tion on robust estimation. At present very lit-tle is known about perturbation theory for theLi norm. Golub has devised a means of ex-perimentally computing an Li conditionnumber that involves a mathematicalprogramming problem with a comple-mentarity condition.

In connection with the work on thisproject, a conference on numerical andstatistical computing was held at StanfordUniversity June 20—21, 1977. This meetingimmediately preceded the June 1977 meet-ing of the Institute of Mathematical Statis-tics. We discussed research in progress thatis common to numerical analysis and statis-tical computing with emphasis on robuststatistics. Attention was focused on ill-

posed problems, unconstrained optimiza-tion, sensitivity analysis, and selection.

Virginia Kiema

Nonlinear Algorithms

During the past year we conducted exten-sive theoretical and computational researchinto the solution of nonlinear least squaresand nonlinear robust regression problems;we are now completing work on NL2SOL, amodular, portable, state-of-the-art computercode for the nonlinear least squaresproblem, which reflects the conclusions ofour research. In consultation with John Den-nis and Roy WeIs'ch, and with the aid of NeilKaden and Scott Smolka, I have tested anumber of alternatives to various computa-tional details in the basic algorithm thatJohn Dennis designed and Mark Gelfandimplemented as the TROLL segmentDMINIMIZ early last year. These detailsrelate to how we update, and when wedecide to use the part of the least squaresHessian that we approximate and to how wedeal with negative curvature in our Hessianapproximation. While we intend to explorecertain questions (such as the latter) further,especially in the more general context ofnonlinear robust regression, by the fall of1976 we felt that we had made sufficientprogress to warrant preparing the NL2SOLpackage mentioned above. Stephen Peters,with some assistance from Scott Smolka andmyself and some helpful advice from Vir-ginia Klema, has done an excellent job ofpreparing this package in such a way that itshould be portable (easily transferred toother computers) and readily publishable.Thus, we look forward to getting feedbackfrom users in and outside the NBER.

David M. Gay

Experimental Mathematical ProgrammingSystem

The overall system on which we havebeen working is now called AMPS (Adapt-able Mathematical Programming System).The two component parts are XMP (Experi-mental Mathematical Programming) andXML (Experimental Modeling Language).

XMP is an expandable mathematicalprogramming system that is designed to be

96

used in experimental and subservientmodes. It is being written according to thestructured programming methodology toensure modularity at all levels; The first andprimary component of XMP is the linearprogramming system XLP. Michael Harrisonhas written a working version of XLP in FL/I,and David Gay has installed routines forconstructing and updating an LU (lowertriangular times upper triangular) factoriza-tion of the basis matrix. This LU factorizationuses the latest techniques for exploitingsparsity and enhancing numerical stability,and enables XLP to solve large problems.Roy Marsten is translating the XLP routinesinto FORTRAN to make them available to awider range of algorithmic researchers.

XML is an innovative modeling languagefor specifying linear programming modelsin a form that is close to common journalusage. We believe that XML will make cur-rent "matrix generators" obsolete by beingmore convenient to use and more efficient inoperation. The main components of XML arethe model editor (XME), the data editor(XDE), the model translator (XMT), and thesolution reporter (XSR). Michael Harrisonhas completed a working version of themodel editor. Michael Harrison and WilliamNorthup have been working on the design ofthe model translator. Robert Fourer is work-ing on the design of the solution reporterand on the ACOL interface for the modeleditor.

The major accomplishments this yearhave been the design and implementationof XLP and the design of the languagespecifications for XML. Next year XLP willbe improved and the language processorsfor XML will be designed and built. Further-more, XLP and XML will be integrated withother Computer Research Center systemsso that combined mathematical program-ming and econometric models can besolved.

Michael HarrisonRoy Marsten

Mixed Integer Programming

Our activities of the past year in mixed in-teger programming have centered aroundresearch and computer experimentation

with dual methods. The theoretical founda-tions of these methods were extended byDavid Bell, William Northup, and JeremyShapiro in the development of a completeduality theory for pure integer programmingproblems and its adaptation for use inmixed integer programming. The methodshave been implemented by William Northupand J, S. D'Aversa and combined with moretraditional branch-and-bound methods. Theimplemented codes are being used as a testbed for further experimentation to achieve amore effective blending of the different al-gorithmic approaches to the dual problems.To this end, Bruno Simeone experimentedwith a new method for solving the groupoptimization problem underlying the dualtheory. Experimenting continues with thesynthesis of the dual methods and branch-and-bound, using interactive direction totest strategies for a general code, A numberof applications provide the test problems,including a water resources investmentproblem being studied by Dorothy Elliott,and minicomputer configuration problemsby Jonathan Wexler.

William NorthupJeremy Shapiro

Other Computer-oriented ActivitiesNBER Computer Service

The National Bureau Computer Service(also identified as the Computer OperationsActivity) is responsible for the Bureau'scentral computer facilities and fordisseminating software developed by theComputer Research Center for Economicsand Management Science (CRC). Responsi-bility for the central computer facilities in-cludes coordinating the computer and data-communications requirements of all NBERoffices; managing the purchase of computertime; and establishing standard rates forcomputer, communication, and terminaluse.

In spring 1975, the Bureau contractedwith Cornell University to buy a block of timeon Cornell's IBM 370/168 computer. Sincethen, this machine has served the comput-ing needs of all NBER research projects.

97

Data-communications between Cornell andthe Bureau's offices have been provided bythe TYMNET network. Maurice Herlihy andCharles Byron manage day-to-day opera-tions of the Bureau's Cornell and TYMNETfacilities.

The Computer Service disseminatesTROLL and other products of the CRC's al-gorithmic research, An interactive system,TROLL had as its original purpose the esti-mation and simulation of econometricmodels. To these base capabilities wereadded modules for exploratory dataanalysis and innovative econometric ap-plications (e.g., spectral regression, logitand probit analysis, nonlinear three-stageleast squares estimation). Other CRCproducts that the Computer Servicedisseminates include FIXPOINT, a systemfor simplicial approximation of economicequilibriums.

The National Bureau disseminates TROLLto selected nonprofit institutions, which ac-cess the Cornell system via the network.Such dissemination has created a user com-munity and has given the CRC valuablefeedback on new TROLL modules as theyemerge from various research projects.

Dissemination in the form of externalusage of the Cornell machine has continuedover the past year, but on a smaller scalethan in the previous year. The larger outsideusers have included Carnegie-MellonUniversity, the Conference Board, theFederal Energy Administration, variousFederal Reserve banks, the MassachusettsInstitute of Technology, the University ofHouston, the Urban Institute, and the U.S.Export-Import Bank. As a matter of NBERpolicy, no commercial organizations areallowed to use the Bureau's Cornell facility.

Another form of dissemination, licensing,began during the past year. Under licensecontracts, the NBER installs TROLL on com-puters of other organizations and providesongoing maintenance support. In return,licensees pay fees, which cover the cost ofsystem maintenance. This arrangement pro-motes dissemination of the CRC's com-puter-system research while reducing theBureau's commitment to direct support of a

large user community. It also permitsdissemination to commercial organizations(which, as noted above, were not allowed touse the Cornell facility). Further, it allowsgovernment agencies that are very largeconsumers of TROLL to obtain system usefrom government-approved vendors.

A standard license contract wasdeveloped over the past year, together witha fee structure that differentiates betweencommercial and nonprofit licensees.Licenses were negotiated with the Norwe-gian School of Economics, PrincetonUniversity, de Ap-pliquées, Inc. (Montreal), Technical Uni-versity Malaysia, TYMSHARE, Inc., andWirtschaftsuniversität Wien (Austria). Amajor maintenance task in support oflicensees was the preparation of TROLL-re-quired modifications of a new release ofVM/370, the operating system under whichTROLL runs. Paul Bernard developed thesemodifications and assisted licensees withtechnical problems of installation andmaintenance, Another maintenance taskwas the development of a version of TROLLthat requires no modifications of VM/370.

John Kirsch

Time Series Data Bank

The year 1976 could be called the year ofrevisions. The Bureau of Economic Analysisrevision of the national income and productaccounts started a chain of revisions—Bu-reau of Labor Statistics wholesale priceindexes, Federal Reserve System consumercredit, industrial production, and capacityutilization, Bureau of Labor Statistics em-ployment figures, and Business ConditionsDigest cyclical indicators. Although mosteconomic time series have always beenrevised about once a year, these revisionsall involved much more than the usualroutine seasonal and benchmark changes.In some cases statistical changes wentback to 1946; in some, concepts werechanged and new concepts added. Fromthe point of view of the data bank adminis-trator, the use of computers by statisticalagencies is at an unfortunate stage. Mostagencies are capable of using computers to

insure the consistency and facilitate thehandling of large and sweeping revisions ofsets of data, but they are not yet sufficientlycomputerized to release the revised data inmachine-readable form. Some agencies, doprovide tapes but with a considerable lagafter their printed releases.

As a consequence of all this we havebeen kept busy revising data and docu-mentation. A partially revised directory wasissued in September 1976 and was wellreceived by users in and out of the Bureau.However, we find that conceptual and otherrevisions proceed so rapidly that 'supple-ments" consisting of new or revised pagesto be inserted in the existing directory areneeded almost continuously. We issued onein January and another in the spring.

In order to cope with the increasedworkload caused partly by the many data re-visions and partly by the addition of anothertime-sharing vendor of daily updates, wehired Jonathan Sheer, on a part-time basis,to help Wan-Lee Hsu and Constance Lim inthe updating operations. In addition, we arereceiving some help in contacting issuingagencies from Joslin De Puys, of the Bu-reau's Washington office. This arrangementhas turned out to be very useful, since it en-ables us to receive important materialssooner and with greater accuracy than wedid previously.

Charlotte BoschanAnn Wood

Machine-Readable Data Files

The National Bureau of Economic Re-search has maintained a tape library for thepast three years. There are approximatelytwo hundred reels of magnetic taperepresenting fifteen master or original ma-chine-readable data files. The data include,for example, the extensive 1960 and 1970Public Use Samples published by Bu-reau of the Census, the United Nations com-modity export files, and the comprehensiveThorndike sample file with various modifica-tions. These files are documented,classified, and catalogued according to asystem outlined in the 1975 Annual Report.Most of these data are available to NBER re-

98

searchers and, for a fee, to outside re-searchers.

During 1976 the following files wereadded to the tape library.

1. National Longitudinal Survey (Parnes) foryoung men from 1966—1973.

2. National Longitudinal Survey (Parnes) foryoung women 1966—1973.

3. United Nations. Latin American Growthin Industry, a data set acquired for Pat-ricio Meller's study, 'International Com-parisons of Industrial Concentration inLatin America." This contains raw datafrom 1963 to 1973, collected by theUnited Nations for publication in Growthof World Industry, vol. 1, Latin America,by industry and country. The programsused by Mellertoaccess and manipulatethe U.N. data are also available as a ma-chine-readable data file.

4. Although not new or recently purchased,the wholesale price index valuespublished by the Bureau of Labor Statis-tics from 1962 to 1975 in machine-read-able form have been added to the library.

5. The Board of Governors of the FederalReserve System published revised in-dustrial production data for January 1954to March 1976. Both the machine-read-able form and the paper copy are avail-able in the library.

During July 1976 Ann Wood participatedin the workshop sponsored by the Universityof Michigan, Institute for Social Research, atAnn Arbor, Michigan. The workshop,designed . to help librarians and datamanagers deal with collections of machine-

readable data sets, was a follow-up of theworkshop and seminar we sponsored for theConference on the Computer in Economicand Social Research, which was held atNew York University in April 1974. Effortsare being made to develop logical docu-mentation techniques acceptable to usersand libraries for the establishment of a unioncatalog for machine-readable data files. TheBureau will continue to participate in effortsdirected toward this goal.

Improved Term Structure Program

Ann Wood

During my year as an NBER fellow I havesucceeded in modifying the program I wrotein 1973 for the U.S. Treasury for splinecurve-fitting of the term structure of interestrates to take into account the fact that bondcoupons arrive semiannually instead of in acontinuous stream. This modification tooklonger than anticipated, partly because ofthe complicated taxation of the first coupon.However, it is now complete. I plan to makethe new program available to business andgovernment. In particular, I have suggestedto the U.S. Treasury Office of Debt Analysisthat my precise 'tax adjusted par-bondyield curve" be adopted in place of the im-precise hand-fitted yield curves shown inthe monthly Treasury Bulletin. I am now us-ing the new program to generate time seriesof term structure measures of a quality pre-viously unavailable. These series will be in-corporated into the NBER tape library,where they will be available to other re-searchers.

J. Huston McCulloch

9. CONFERENCES, WORKSHOPS, AND OTHER PROGRAMS

Conference on Research in Incomeand Wealth

The proceedings of the conferences onthe Distribution of Economic Well-Being(May 1974), on Price Behavior (November1974), and on the Economics of ResidentialLocation and Urban Housing Markets (May1975) are in press and publication is ex-

99

pected by fall of this year. Papers presentedat the conferences on New Developments inProductivity Measurement (November 1 975)and on the Measurement of Capital (October1 976) have been revised by the authors andare being edited.

James D. Smith of Pennsylvania StateUniversity is serving as chairman for theConference on Modeling the Distribution

and Intergenerational Transmission ofWealth, to be held in December 1977. Thefollowing papers are anticipated:

"Estimates of Nineteenth Century WealthDistri butions"—Jeffrey G. Williamson,University of Wisconsin, and PeterLindert, University of California, Davis

"Estimates of the Distribution of Wealth inTennessee"—William Perry, University ofTennessee

"Taxation and Distribution of Wealth"—Nelson 0. McClung, U.S. Treasury De-partment

"Simulating Intergenerational Wealth Trans-mission"—Guy H. Orcutt, Yale University,and James D. Smith, Pennsylvania StateUniversity

"The Bequest Process: A SimulationAnalysis for Canada—Michael Wolfson,Treasury Board of Canada

"Changing Patterns of Inheritance"—William H. Newell, Miami University

"Personal Sector Balance Sheets"—EdwardWolff, New York University

"Wealth between Generations—MichaelAllen, University of Washington

"Intergenerational Wealth Transmission"—Paul Menchik, University of Wisconsin

As possibilities for a future conference,J. A. Sawyer of the University of Toronto isexploring the topic "foreign trade statistics,"and Allan H. Young of the Bureau of Eco-nomic Analysis is considering a review ofthe U.S. national income and product ac-counts.

The conference program has beenassisted by grants from the NationalScience Foundation. The Executive Commit-tee consists of Clopper Almon (chairman),Laurits R. Christensen, Stanley Lebergott,Milton Moss, Joel Popkin, J. A. Sawyer, JackE. Triplett, Dorothy Walters, Burton A.

Weisbrod, Allan H. Young, and Mildred E.Courtney (secretary).

Universities—National BureauCommittee for Economic Research

The Universities—National Bureau Com-mittee held a conference on population and

economic change in less developed coun-tries at the University of Pennsylvania onSeptember30 through October 2, 1976. Theproceedings are being prepared forpublication. The planning committee for theconference consisted of Richard Easterlin,Robert Willis, Allen Kelley, and T. PaulSchultz. Richard Easterlin is the conferenceeditor.

A conference on low-income labormarkets will be held in the fall of 1977. Theplanning committee members are SherwinRosen (chairman), OrleyAshenfelter, JamesHeckman, and Finis Welch.

At its annual meeting on October 1, 1976,the Universities—National Bureau Commit-tee approved the recommendation of the ex-ploratory committee to hold a conference onthe economics of information and un-certainty. George Stigler was the chairmanof the exploratory committee, and John Mc-Call is the chairman of the planning commit-tee.

Two topics were selected for possible fu-ture conferences: (1) natural resources and(2) sources of growth in developing coun-tries. The respective chairmen of the two ex-ploratory committees are Harold Barnett andIrma Adelman. There is also an exploratorycommittee on taxation and household be-havior which was appointed earlier, withMartin Feldstein as chairman.

One conference volume has beenpublished this past summer: Education asan Industry, edited by Joseph 1. Froomkin,Dean T. Jamison, and Roy Radner.

The Universities—National Bureau Com-mittee accepted the University of Californiaat San Diego as a new member. There arenow forty-one universities, together with theNational Bureau, represented on the com-mittee. The participating universities are:

Mildred E. Courtney

BrownCalifornia, BerkeleyCalifornia, Los

AngelesCalifornia, San

DiegoCarnegie-MellonChicagoColumbia

New YorkNew York State,

BuffaloNorth CarolinaNorthwesternOhio StatePennsylvaniaPittsburghPrinceton

100

RochesterStanfordTexasTorontoVanderbiltV rg in i a

Washington(St. Louis)

Washington(Seattle)

WisconsinYale

Other members of the U-NB Committee,elected as members-at-large for a four-yearterm, July 1, 1974—June 30, 1978, are IrmaAdelman, Bela A. Balassa, Carl E. Beigie,Daniel Creamer, Frank de Leeuw, Walter S.Salant, and George J. Stigler. Robert E.Lipsey is the representative from the Na-tional Bureau of Economic Research.

The members of the Executive Committeeare Edwin S. Mills (chairman), Leonard W.Weiss (vice chairman), Irma Adelman,Richard Bird, Robert Eisner, Robert E.

Lipsey, and Dudley C. Luckett. ChristineMortensen is secretary.

The conference program is assisted by agrant from the National Science Foundation.

Universities interested in membership onthe Universities—National Bureau Commit-tee for Economic Research should get intouch with the chairman of the Committee.The criterion for acceptance of a universityas a member of the Committee is the extentand quality of economic research carried onat that university.

Universities wishing to be informed aboutproposals for future Universities—NationalBureau Committee research conferencesand about the availability of conferencepapers before publication should communi-cate with the secretary of the Committee.

Christine Mortensen

Economic and Social ResearchIn 1976 a workshop on stochastic control

was organized by David Kendrick of theUniversity of Texas at Austin and Edison Tseof Stanford University and took place onMay 26—28, at Stanford University in PaloAlto, California. About thirty economists andengineers participated in the workshop, andthe major topics considered were the useful-ness of imperfect models in settingmacroeconomic stabilization policies,monetary policy under uncertainty,identification conditions and time-varyingmodels, the international energy model, theproject independence evaluation system,recent results in least squares estimationtheory, minimax control of stochasticsystems, comparison of optimization tech-niques for a monetary model of France,optimal pricing models for exhaustibleresource cartels, identification and estima-tion of linear models with measurement er-ror, the use of optimal control techniques tomeasure economic performance, policymodels of agricultural commodity trades,generalized least squares applied to time-varying parameter models, and stochasticcontrol for linear, discrete time, distributedlag models. Some of the papers presentedat this conference are scheduled to bepublished in a future issue of the Annals ofEconomic and Social Measurement. Theworkshop was made possible by a grantfrom the National Science Foundation.

Two workshops have been scheduled for1977. A stochastic control theory workshopwas organized by David Kendrick of theUniversity of Texas at Austin and Edison Tseof Stanford University and was held at YaleUniversity, May 25—27, 1977. A Social Se-curity research files workshop is scheduledto take place in late 1977, in Williamsburg,Virginia. The workshop will be organized byFrederick Scheuren of the Social SecurityAdministration and Richard C. Taeuber ofthe Commission on Federal Paperwork.

M. lshaq Nadiri

PurdueQueen's

Conference on the Computer inCornellDukeHarvardIllinoisIndianaIowa StateJohns HopkinsMarylandMassachusetts Institute

of TechnologyMcGillMichiganMichigan StateMinnesotaNew School for

Social Research

101

Conference on Econometrics andMathematical Economics (CEME)

This conference, organized in 1970,continues its program to stimulate dis-cussion and research on the frontiers of thestate-of-the-art of econometric andmathematical economic theory and meth-odology and their application in empiricaleconomic studies. The conference is or-ganized in seminar groups which meet pe-riodically at leading universities and re-search centers throughout the United States.The leaders and participants in theseminars consist of a changing set of seniorand young scholars, depending on topicand research being conducted in the field.

The seminars currently active are GeneralEquilibrium Models, Evaluation ofEconometric Models, Comparison ofEconometric Models, Decision Rules andUncertainty, Optimal Economic Growth andNatural Resources, Bayesian Inference inEdonometrics, Quantitative Studies in In-dustrial Organization, Public Economicsand Nonmarket Decisions, DecentralizedEconomic Planning and Programming, andGlobal Modeling. 1n addition, seminars inDistributed Lags and Time Series Analysisand in Analysis of Panel Microdata, whichheld meetings in prior years,.are expectedto be reactivated during the next academicyear.

Seminar groups have held seventysessions, produced more than 240 workingpapers, and have published several booksand numerous articles in professionaljournals. These have been widely circulatedand have made the results of seminarswidely

Besides its regular seminar program, theconference periodically holds special meet-ings on selected subjects. The next meetingof this kind, on aggregation andeconometric models, is expected to be heldin March 1978.

Support for the Conference since its in-ception has been provided by the NationalScience Foundation.

Gary Fromm

Seasonal Analysis of EconomicTime Series

A conference sponsored by the U.S. Bu-reau of the Census and the NBER onSeasonal Analysis of Economic Time Serieswas held in Washington, D.C., September9—10, 1976. It was attended by experts fromthe academic, business, and governmentsectors. Fourteen papers were presentedand discussed. In addition, a keynote ad-dress was delivered by Julius Shishkin, theoriginator of the X-1 1 program now widelyused for seasonal adjustment of economicdata. Shishkin has long been associatedwith the NBER and the Census Bureau, andis currently Commissioner of Labor Statis-tics.

The full program was outlined in NBER's1976 Annual Report. Proceedings of theconference are to be published by the Bu-reau of the Census and are expected to beissued in the fall of 1977.

Gary FrommArnold Zeilner

Conference on Public RegulationThe National Bureau of Economic Re-

search, under a grant from the NationalScience Foundation-Research Applied toNational Needs (NSF-RANN), hasscheduled a conference on public regula-tion, to be held in Washington, D.C.,December 15—17, 1977. The purpose is toadvance both the theory and practice ofregulation. A call for papers was circulatedin January 1977 to universities, researchcenters, government agencies, and leadingscholars. The announcement stated thatpapers should make significant newcontributions which further the formulation,analysis, and evaluation of regulatorypolicies; should be relevant to contempo-rary public policy issues; and should besuitable for discussion with outstandingexperts from academic institutions, busi-ness, consumer groups, and variousbranches of government. Some results maybe immediately applicable to regulatoryproblems while others may suggest promis-ing paths for additional research.

102

The steering committee for the conferencereviewed 117 abstracts submitted inresponse to the call for papers and selectedthe following for presentation and dis-cussion:

"Theory and Practice in Public Regulation:A Current Overview"—Roger Noll, Stan-ford University

"The Public Interest Component of the Sup-ply and Demand for Regulation"—Richard 0. Zerbe, Jr., University of Wash-ington

"Income Distributional Concerns in Regula-tory Policy Making"—Robert D. Willig andElizabeth E. Bailey, Bell Laboratories

"Regulation of the Casualty Insurance In-dustry: Consumer Protection against In-solvencies"—Patricia Munch and DennisSmallwood, Rand

"Regulatory Performance in Surface FreightTransportation in Australia, Canada,Great Britain and the United States"—James C. Nelson, Washington StateUniversity

"Regulation, Barriers to Exit, and RailroadInvestment Behavior"—Richard C. Levin,Yale University

"A Political Economy of Business Regula-tion: The Case of Water Pollution"—Robert Leone and John Jackson, HarvardUniversity

"Enforcing Administrative Rules"—PaulDowning, Virginia Polytechnic Institute—State University

"Regulation of the Multiproduct Firm: TheCase of Telecommunications inCanada"—Melvyn A. Fuss and LeonardWave rman, University of Toronto

"Open Entry and Cross Subsidy in Regu-lated Markets"—Kenneth C. Baseman,U.S. Department of Justice

"Regulatory Entry Into the Electric Utility In-dustry"—Gregg A, Jarrell, University ofChicago

"The Transition to Uncontrolled Crude OilPri ces"—W. David Montgomery, Cal-ifornia Institute of Technology

Two formal discussants are to be selectedto review each paper. It is planned that the

103

proceedings of the conference will bepublished.

Research Fellowships

Gary Fromm

The National Bureau's research fellow-ship program is intended to provide addi-tional educational opportunities to scholarsof outstanding promise, generally at anearly postdoctoral stage of their careers.The fellows devote full time, usually for ayear, to their research interests with accessto Bureau facilities and in association withmembers of the research staff who aregenerally familiar with the types of problemsbeing investigated. The nature of thefellow's research interests usually de-termines which of the National Bureauoffices is most suitable for his work.

One phase of the program is the awardingof one or more Faculty Research Fellow-ships to scholars selected primarily fromuniversities in the United States. The AlfredP. Sloan Foundation is currently providingfinancial assistance for some of thesefellowships. The second phase, which hashad support from the Rockefeller Founda-tion, has brought to the Bureau Foreign Re-search Fellows from universities in Africa,Asia, and Latin America.

Faculty Research Fellows for 1976—1977were Daniel A. Graham of Duke University,Cheng Hsaio of the University of California(Berkeley), and J. Huston McCulloch ofBoston College. Graham spent his fellow-ship period at the Bureau's office in NewYork; Hsaio, at the Computer ResearchCenter in Cambridge; and McCulloch, atNBER in Palo Alto. The Foreign ResearchFellow in 1976—1977 was Narongchai Ak-rasanee of Thammasat University, Bangkok,Thailand, who worked in the New Yorkoffice. Comments by these respectivefellows on their work at the National Bureauare included with the staff progress reportsin the present volume.

For 1977—1978 the Faculty ResearchFellows are Michael D. Hurd of StanfordUniversity, Mieko Nishimizu of PrincetonUniversity, and Daniel A. Seiver, University

of Alaska. Hurd will pursue his interests ineconometric investigations of labor marketdynamics at the Center for EconomicAnalysis of Human Behavior and Social In-stitutions at the NBER offices in Palo Alto.Nishimizu will work in the Bureaus Wash-ington office on a comparison of sectoralchanges in productivity in the United Statesand Japan. Seiver will be at the Cambridgeoffice, continuing his research in econo-

104

metric models of Alaska with a view toprojecting the demands and supplies ofmanpower and the income distributionsamong demographic subgroups whichwould result from various growth paths forthe Alaskan economy.

Appointment of a Foreign ResearchFellow for 1977—1978 is not expected.

Doug/as H. E/dridge