classic and contemporary - umass amherstcourses.umass.edu/econ753/berndt.poe/ch01.pdf · 2019. 9....
Post on 15-Mar-2021
1 Views
Preview:
TRANSCRIPT
>
The Practiceof Econometrics:
Classic and
Contemporary
ERNST R. BERNDT
Massachusetts Institue of Technology andthe National Bureau ofEconomic Research
Avv
ADDISON-WESLEY PUBLISHING COMPANY
Reading, Massachusetts - Menlo Park, California . New YorkDon Mills, Ontarlo - Wokingham,England - Amsterdam - 8onnSydney « Singapore - Tokyo . Madrid . San Juan
ADDITIONALCREDITS:
Arrow: Reprinted with permission of the and Navasky, The Experts Speak: The Donn:Econometrica Society. Nordhaus: Re- itive Compendium ofAuthoritative Misinfor.printedwith permission of The Brookings mation, NY: Pantheon Books, 1984, p. 209Institution. Klein; Reprinted with permis- Reprinted with permission. Ein-sion of the author. Schultz: Reprimicd _slcin: NY: Crown Publishers, 1984, 274.
Musgrave: Reprinted Mulkiel. NY: W.WW. Norwnand ne,i Ohta, Griliches, Terlee 19RS, p. 152. Reprinted with permission,
Reprinted with permission, Ashen- Spence: Copyright 1981. Reprinted from
printed with permission. Stigler: TheBelt Journal ofEconomics, with permis-
Reprintedwith permission. Schultz: Re- sion of The RAND Corporation. Theprinted with permission. Watson: From Prive Statistics of the Federat Government.
Morgan and Langford, Facts and Fatlacies, (Credits continue on p. 680, whichExeter, England: Webb & Bower, 1981, p. constitutes a continuation of the44, Reprinted with permission. From Ceri copyright page.)
[BM and [BM PC areregistered trademarks ofInternational Business Machines Cor-poration. Lotus, 1-2-3, and Symphonyare registered trademarks of Lotus Develop-meat Corporation. MicroTSPis a trademark of Quantitative Micro Software. TimeSeries Processor and TSP arc trademarks of TSP International. MINITABis a regis-tered trademark of Minitab. Inc. Other products mentioned as examples may also beUrademarked by their manufacturers.
Addison-Wesley makes no warranties, cither expressed or implied, regarding theenclosed computer software, its merchantability or fitness for any particular purpose.‘The caclusion ofimplied warranties is not permitted by some states. The above exclusion may not apply to you. This warranty provides you with specitic legal rights. Theremaybe other rights thal you may have which vary (fom state to state.
It isa violation of copyright law to make a copy ofthis software except for backup:
purposes to guard against accidental loss or damage.
Library of Congress Cataloging-in-Publication DutaBerndt, Ernst R.
The practice of econometrics : classic and contemporary / Ernst R.Berndt.
. cm,Includes bibliographical references and index.ISBN 0-201-!7628-9.1. Econometrics. I. Title.
HBt39.B47_ 1990.330°.01/5195-4dc20. 9046332
CIP
Jopyright « 1991 by Addison-Wesley Publishing Cumpans, Inc.MH rights reserved. No part ofthis publication may be reproduced, stored in a retrievalystem, of transmitted, in any form or by any me elecironig, me: photoopying, recording, or otherwise, without theprior written permissionofthe publisher.‘rinted in the United States of America,
sackage ISBN 0-201-5 1489-3ok ISBN 0-201-17628-9
Jisk ISBN 0-201-5 1488-5
45678910 -MA- 95 94 93 92
|
\|
Preface
The term “econometrics” was apparently first used by Pawel Ciompain 1910
in a somewhat obscure book published in Germany. To Ciompa,the goals of
“gekonometrie” were to describe economic dataseries mathematically and to
display them geometrically and graphically. According to the Nobel Laureate
Ragnar Frisch, however, Ciompa’s view of econometrics was too narrow,
since it emphasized only the descriptive side of econometrics.’ Writing as
founding editor in the inaugural issue of Econometrica in 1933, Frisch
defined econometrics in more general terms:
Econometrics is by no means the same as cconomicstatistics. Nor is it
identical with what we call genera! economic theory, although a consid-
erable portion ofthis theoryhas a definitely quamtitative character. Nor
should econometrics be tiken as synonomous(sic) withthe application of
mathematics to economics. Experience has shown that eachof these three
view-points, that ofstatistics, economictheory, and mathematics,is a nec-
essary, but notbyitself a sufficient, condition for a real understanding of
the quantitative relationsin modern economiclife. It is the unificationof
all three that is powerful. And it is this unification that constitutes
econometrics,”
To Frisch, econometrics embodies 2 creative tension between theory
and observation:
Theory, in formulating its abstract quantitative notions, must be inspired
toa farge extent by the techniqueof observatiun. And fresh stati and
otherfactual studies must be the healthy clementofdisturbance that con-
stantlythreatens and disquiets the theorist and prevents him from comingto rest on someinherited, obsolete set ofassumptions.’
Thefield ofeconometrics has developed and matured enormouslysince
Frisch wrote in 1933. lLis mybelief, however, that the way in which we typ-
ically teach econometrics todayoften loses sight of the underlying goals iden-
tified by Frisch. In paniicular, in the customary classroom teaching environ-
“Fora brief discussion, see Ragnar Frisch, “Note on the Term“Econumetries," Econamettica,3:1, 1936. p95.2 RagnarFrisch, “Editorial,” Econometeica, 1:1, January 1933. p. 2.* Ibid.
orakeeaeetne ta focus Primarily on econometric theory and to lose sight‘ing forces ind econometrics, nam: i‘ nd ¢ 4 ely, to increase our “1
understanding of the quantitative relations in modern economiciife.” alvane grbis ba0k focuses exclusively on such quantitative relations—impor.fant ppl ications and empirical implementations of econometrics, It is not a‘xtbook in econometric theory but is complementary to such a text.I ? . :: havestructured and written this text for primary use in four types of
1 aaSupplementary textbookin
a
first-year graduate econometrics anding course for students in econamics or busi: imiusiness or
2 course atthe advanced undergraduate level. fora similar7 As the Primary textbook in an applied econometrics course that focuses
Q portant classic and contemporary applications and empirical3 implementations of econometric techniques,
. As2 ae. extbook for students who wish to learn, on their ownseminars, how to ii intei is
vechnina implementandinterpret econometric
- As a reference material for students doing term paper exercises or theses.ina varietyofapplied areas, for example, labor economics,
Computation and Econometrics
peusefulness ofthis text is due in large part to recent developments in com:ion. The continued steady drop in costs i ‘I :
> car of mainframe com,greater accessibility to students of mainfr veite diefon ata ‘i rame computers, as well as the dif-
ly affordable personal computers and iremoved most previous obstacles and n ii ible fortashaveF ved mosi iow makeit feasible for studin their waining to becomeseriously i init ig empirical,
’ o ly involved in implementi: iticgthe theory taught in econometrics and forecasting comes nag empuresiy
In this textbook [ exploit these and other recent computa evel.ther computational devel-
8 for2 umberof importanttopical applications this text provides stu-dents“i hrea ble surweys on the underlying economictheory. major: measurement and econometric implementati he pricipal empirical findings obtained to date. e ation: and the prin:2, ristext makesaccessible to students a variety of classic and contem-porary lata bases, thereby encouraging replication and extension of thesx iB empirical literature. These data bases are provided on a flo;cickaccompanying this textbook. Since the data are in files written ing
conventions, th icompute i data should be readable by virtually all
3. Forindividuals who have access to IBM or compatible personal cam-puters or to Apple Macintoshes, this text provides information on avail.
able sofware sources, including free sofware and inexpensive student
editions. This information enables students and instructors to choose
software and then to implement and experiment with, al minimalcost.
a good portion of the econometric techniques used by empirical
researchers today.
Organization of the Book
The book consists of an introductory chapterfollowed by ten applied topics
of interest to economics and business students, emphasizing oneset of tools
per chapter. The orderofthe chapters correspondswith the sequenceoftopics
that is typically taught in econometrics and forecasting courses, beginning
with the bivariate regression model (Chapter2), moving on to multiple regres-
sion andeffects of omitted variables (Chapters 3 and 4), dummy variables and
specification issues (Chapters 4 and 5), then to generalized least squares and
distributed lags (Chapters 6 and 7), forecasting with generalized least squares
and timeseries analysis (Chapter 7), causality and simultaneity (Chapter 8),
systems of equations (Chapter 9), simultaneous equations (Chapter 10), and
discrete dependent variable models (Chapter11).
The amountof material in this text is substantial, and instructors may
well choose to work with only a subset of the chapters in a one-scmester or
even a year-long course. Two chapters, “Explaining and Forecasting Aggre-
gate Investment Expenditures” (Chapter 6), and “Causality and Simultaneity
between Advertising and Sales" (Chapter8) are considerably tonger than the
other chapters; these two chapters have been designed for instructors in eco-
nomics or business who wish to have their students focus on one particular
topicfor a substantial timeperiod, such as a quarter or a semester, Alternative
waysofusing these and other chapters in this text are discussed further in the
Instructor's Resource Guide.Each chapter begins with a discussion of the economic theory underly-
ing the application,outlines issues in econometric implementation (including,
measurementissues), summarizes important empiricalfindings to date, and
then engages the studentin a carefully designed set ofeight 10 ten empirical
exercises involving replication and extension of classic and contemporary
empirical findings.
Applications and ‘Hands-on’ Exercises
Several aspectsofthese exercises are worth noting.First, this applicationstexi-
book will be most helpful to users if its exercises are coordinated with their
econometric theory textbook. To facilitate such coordination, a cross-refer-
ence of exercises in this book with relevant chapters in the leading econo-
metric theory textbooks is provided in the Instructor's Resource Guide, Sec-
ond, in most chapters theinitial exercises involve students in examining the
data before using them in further econometric analyses.I believe that expler-
alory data analysis is an important componentofgood econometric research.Third, the exercises have been tested and have served students well in a num-
ber of courses at MIT and elsewhere. But the practice of econometrics isdiverse, and instructors might wish to augmentorrevise the exercises to mecttheir particular needs and preferences. The text allows for such flexibility, andT expect considerable diversity in how this text is used.
Fourth, data for each of the exercises are provided on a floppy diskaccompanying this text. Classic data sets can be updated by consulting thecited data sources. Ideas and sources for additional data are discussed in theInsteuctor’s Resource Guide. Fifth, rather than presenting students with only
one data base and oneset of exercises per topic, in most chapters a numberof data series are provided; students (and instructors) are invited 1c chooseamong them and experiment. This selection permits a greater degree ofindi-vidualization andself-learning. Sixth, while the empirical exercises encourageexperimentation and are designed particularly with the personal computerinmind,they can of course be carried out on mainframe computers.
Finally, each application chapter contains exercises that correspond towolevels ofdifficulty, Most ofthe exercises initially focus on the use ofecon-ometric procedures, but the final few exercises often introduce proceduresthat might not yet have been covered in class. Students are therefore encour-aged to return to a particular topic later in a typical course. perhaps severaltimes,after more sophisticated econometric techniques have beenpresented
and learned.The emphasisin this text on empirical implementationandself-learning
containsatleast one significant potential liability. Since all the data requiredto perform the exercises are provided on a floppy disk students are notreallyrequired to “dig”for data or gain understanding ofhow data are actually gath-ered and compiled. That is unfortunate. Nonctheless, | believe that the strat-egy ofgiving students complete access to data is justifiable, sinceit is impor-tant for students to begin to understand quickly and at minimum cost thatthe process ofreplication is extremely importantin empirical research.
Supplements
Three manuals ase available to help readers implement the “hands-on”exer-
cises in each chapter of the text using well-known econometrics softwarepackages for microcomputers (as well as mainframes with PC-TSP andSHAZAM):
Greenberg, 4 Computer Handbook Using MicroTSPWhite/Bui, A Computer Handbook Using SHAZAMWhite/Hall, A Computer Handbook Using PC-TSP
The data disk bundled with the textbook is written in ASCII code andcan be used with these software packages or with other common statisticalsoftware.
aa
A detailed Instructor's Resource Guide is also available and includes the
following:
Solutions to many “hands-on” exercises
suggested course outlines | ;
ete for using the textbook with major econometric theory books
Sources for additional data sets
Acknowledgmentsinsi fa large num-
i book has benefited from the comments andinsightsof
weet‘people. particularly my students at MIT. Numerous colleagues and
friends have also provided constructive and challenging suggestions. | ove
special thanks to G. Campbell Watkins, Ken White, and Pauleenne
Ci h chapter with great care andoffered many constructivesi e s.
fe “Steve Lerman of Project Athena at MIT provided seed money for ls
project as part ofthe process of introducing personal computers and wort i
tions into the MIT curriculum. In addition, is a pleasure to thank the fol lo a
ing people whoin various ways contributed to one of more of the chapters:
Susanne Ackum, Dennis Aigner, William Alberts, Richard Anderson
G. Chris Archibald, David Atlas, Richard Blundell,Linda.Be«
" i Wang Chiang, Rol irinko,Capone,Jr., Ron Cartwright, Judy W { ie
iswil ji Linda Datcher-Loury, George Day,Chiswick, Klaus Conrad, Clint Cummins, : core Dye
in DiRay Fair, Hank Farber, Frant
Erwin Diewert, Ellen Dulberger, Robert Engle. pone
isBonFullerton, Mel Fuss, Davie
Fisher, M. Therese Flaherty, Peter Fortune, ullertor
Mi Zsi Griliches, GudmundurGarman, Robert J. Gordon, William Greene. 1SGar
ci , Morley Gunderson, Robert Halvorsen, Robert Hall, Ham,
DenHamermesh "Ray Hartman, Arnoldo Hax, BertilHotrod.oes
i ik He is W. Jansen, le Jorgenson,Houlihan, Hendrik Houthakker, Dennis W. ¢ chs
illingsw ; R. Klein, Jan Kmenta, RichaKemerer, Mark Killingsworth, Lawrence TO a
jal i ia Lane, Edward Lazear, Pierre xKopcke, Nalin Kulatilaka, Julia r ome,
Marsh, Ben McCallum,Robert E. Lucas, Jacques Mairesse, Terry E 4
i agli Mathew J. Morey, CatherineMcFadden, Claudio Migliore, Jacob Mincer, M ! vied
iAlice Nakamura, Masa
Morrison, Thomas Mroz, Stewart Myers, Mass
thy, Ronald Oaxaca, TaeNakamura, Charles Nelson, J. Randolph Norsworthy ma Oaad
istian Palda, Leslie Papke, Stephen Peck, John encavel,
Polly wiunnesPoterba, Harvey Rosen. Shelley Rosenstcin JulioRotemberk. i Fiona Scott-Morton,
Thomas Sargent, Richard Schmalensce,\
in Sil bn Solow, Robert Solow, BruceShowalter, Alvin Silk, V. Kerry Smith, Jol ue
i ’ Faylor, Lester Telser, Jactangle, Leslic Sundt, Paul Taubman, John 0 f
we Luc Valle, Joachim Wagner, G. C. Watkins, David Wood, Eric
Wruck.C.K, Woo, Bengt-Christer Ysander, and Amold Zeliner.Although each ofthese individuals has made direct contributions teve
chapters ofthis text, it is of course the case that the contents and style of the
book reflect in many ways the lasting impacts of those who have taught me
vii Preface
Most about the practice of econometric theory: Lau Christensen, ErwinDiewert, Art Goldberger, Zvi Griliches, Date Jorgenson, and Larry Lau. lamdeeply indebted to each of them.
The experience oftransforming an overdue manuscript into final bookform has been surprisingly enjoyable. This process is due in large part to thededication, thoroughness, and highest professional standards of the publish-ing team at Addison-Wesley. fn particular,it is a pleasure for me to thank myeditor, Barbara Rifkind, as well as Christine O’Brien, Kari Heen, KaziaNavas, Ann Kilbride, Jason Jordan, Mary Dyer, and Marshall Henrichs. Spe-cial thanks also to LynnSteele, my assistant at MIT.
This effort would not have been possible without the patient suppon ofmyfriend and wife, Cathy Morrison. The essential ideas for this text germi-nated while we were both on leave at Stanford University in 1985, andCathy’s support encouraged me to risk limiting attention on traditionalresearch, and instead devoting energies and efforts to this textbook.
While this effort has been hard andat times frustrating, you have stoodwith me. To you, Cathy, I dedicate this book. Thank you,
Cambridge, ALA E.R.B.
Cover:Astronomers analyze and predict the movementsof planets, stars and galaxies, com-bining theory with observation. Just so, the practicing econometrician endeavors topredict changes in economic variables aver timeand space, using economic theory andempirical observationstypically based on non-eaperimental data. Surprises, measure-menterrors. und the lack ofa completelysatisfactory theory often result in predictionerrors for both astronomers and cconometricians. This inability to predict with suthi-cient precision induces revisions to the underlying theory. In this way. theory andobservation work together in helping us belter understand our world.
Contents
Chapter}
COMPUTERS AND THE PRACTICE OF ECONOMETRICS
istori ives oe cr d EconometricsLL Historical Perspectives on Computers ant :
2 Background: Computer Hardware and Computer Software
3 Accessing Data from Diskettes for Use in Computer Software
Programs |
14 A Note onthe End-of-Chapter Exercises
(5. Hands On with an Exploratory Data Set
Exercise 1: Ensuring That Data Are Properly Entered and.
Transformed 3 _ .
Appendis At An OverviewofStatistics and Econometrics Software
fort 15 . .
Appendia B: Statistics and Econometrics Software for the IBM-PC
and Compatibles und for the Apple Macintosh: A
Partial Listing of Products and Vendors 7
Notes 18
Chapter 2
THE CAPITAL ASSET PRICING MODEL: AN APPLICATION OF
BIVARIATE REGRESSION ANALYSIS
1 Definitions and Basie Finance Concepts
Diversification and Portfolio Optimatity ,
Derivation of a Linear Relationship beiween Risk and Return .
FurtherInterpretation of the CAPM Risk-Retura Linear Relationship
Econometric Issues Involved in Imptcmenting the CAPM
2.6 Hands On with the CAPM
EXERCISES
2.2
Getting Started with the Data 43
Least Squares Estimates off 44
WhyGoldIs Special” 45 |
Enercise 4, Consequences of Runningthe Regression
Backward as |
Exercise 5: Using the CAPMto Construct Portfolios a7
Exercise 6: Assessing the Stability of # over Time and among
Companies 48
2i23732M41a
Contents
Exercise 7: Three Mile Island and the Conoco Takeover: EventSudies 48
Exercise 8: Is January Different? 50
Exercise 9: Comparing the Capital Asset and Arbitrage PricingModels 52
Exercise 10: What about Our Assumptions Concerning Disturbances?Did We Mess Up? 53
Notes $4References 56Further Readings 59
apter 3
COSTS, LEARNING CURVES, AND SCALE ECONOMIES:‘SIMPLE TO MULTIPLE REGRESSION MIS: FROM
3432
33
343.5
36
TheUnderlying Economic Theory of Cost and ProductionBrief Overview of Learning Curve Literature
Derivation of Cost Function Based on the Cobb-Douglas ProductionFunctionIntegrating the Learning Curve with the Cobb-Douglas Cost FunctionEconometric Issues
3.5.1 McasurementIssues 753.5.2 Omitted Variable Bias 763.5.3 Individual and Joint Hypothesis Tests 783.5.4 Brief Overview of Empirical Findings on Returns to Scale and
Learning Curves 79Hands On with Costs, Learning Curves, and Scale Economics
EXERCISESExercise 1: Estimating Parameters of a Learning Curve 85Exercise 2: Testing the Simple Learning Curve Specification 86Exercise 3; 8? in Simple and Multiple Regressions: A
Surprise 87Exercise 4: Replicating Nerlove’s Classic Results on Scale
Economies 87Exercise 5: Assessing Alternative Returns-to-Scale
| Specifications 88Exercise 6: Comparing Returns-to-Scale Estimates from 1955 with
. Updated 1970 Data 90
Exercise 7: Autocorrelation in the Learning Curve Model 92Exercise 8: Misspecification in the Nerlove Returns-to-Scale
. Modet 92Exercise 9: Testing for the Equality of Coefficientsin the 1955 and
. 1970 Returns-to-Scale Models 93Exercise 10: Forecasting Unit Costafter Further Learning
Occurs 94Notes 95References 98
Further Readings 100
60
266
687
78
8385
Contents
Chapter 4
THE MEASUREMENT OF QUALITY CHANGE: CONSTRUCTING AN
HEDONIC PRICE INDEX FOR COMPUTERS USING MULTIPLE
REGRESSION METHODS
4.1. Traditional Procedures for Incorporating Quality Changes into PriceIndexes
4.2. Analyzing the Price-Quality Relationship at a Given Point in Time4.3. Measuring Price-Quality Relationships over Time4.4 Applicationsofthe Hedonic Method to Price Tadexes for Computers
4.5. Econometric Issues in the Estimation of Hedonic Price Equations4.5.A fleteroskedasticity 1274.5.B Choice of Functional Form 1274.5.C Choice of Variables, Make Effects, and Effects of Omitted
Variables 1284.5.D Identification of Underlying Supply and Demand
Functions 1294.5.E FinalComments —131
4.6 Hands On with Hedonic Regressions, Price Indexes, and ComputersEXERCISES
Exercise f: Examining Waughs 1927 Asparagus Data 132Exercise 2: Exploring Relationships among R?, Coefficients of
Determination, and Correlation Coefficients 134Exercise 3: Constructing Alternative Price Indexes for Computers:
Based on Chow's Data 136Exercise 4: Price Indexes for Disk Drives: A Closer Look at the IBM
Study 138Exercise 5: Assessingthe Stability of the Hedonic Price Equation for
First- and Second-Gencration Computers 139Exercise 6: Using Time-Varying Hedonic Price Equationsto
Construct Chained Price Indexes for Computers 141Exercise 7; Exploring Alternative Functional Formsfor the Hedonic
Price Equation Using Box-Cox Procedures 142Notes 143References 146Further Readings 149
Chapter 5
ANALYZING DETERMINANTS OF WAGES AND MEASURING
WAGEDISCRIMINATION: DUMMY VARIABLES IN REGRESSION
MODELS
5.1 Considerations from Economic Theory: The Human Capital Model3.1.4 Adam Smith and Equalizing Differences 1525.1.B Schooling as tnvestment 154S.C On-the-Job Training as Investment 1555..D Screening as an Alternative to Human Capital Theory 157
xi
104
105Hd7
126
ih132
150
152
Contents
5.2 Issues in Econometric implementation of the Human Capital Model
5.2.A Measurement issues and Common Data Sources 1595.2.B Functional Formsfor Statistical Earnings Functions 1615.2.C Dummy Variables in Earnings Functions 1645.2.D Omitted Variables Bias: Abi 166
5.3, Selected Empirical Results on Determinants of Wages5.3.A Rates of Return to Education 1695.3.B Returns to On-the-Job Training and Experience #725.2.C The Productivity of Highly Paid Workers 175,5.3.D Union-Nonunion Relative Wages 176
5.4 Estimating the Wage Effects of Discrimination5.4.4 Issues in Defining and Estimating the Wage Effects of
Discrimination 180.
m by Race 1845.4.C Discrimination by Gender 188
5.5 Other Econometri¢ Issues in Modeling Determinants of Wages
5.6 Hands On with Estimating Determinants of Wages and EarningsEXERCISES
ExerciseExercise
: A Reviewof Essential Facts: Inspecting the Data 194Confirming Relationships among Alternative DummyVariable Specifications: Earnings and Returns toSchooling 196
Exercise
Married Males and Females 197
Exercise 4: Examining Experience-Earnings Profiles 199Exercise 5: Do Union Workers Earn Premium Wages? 201
Exercise 6: Measuring and interpreting Wage Discrimination byRace 203
Exercise 7; Measuring and Interpreting Wage Discrimination byGender 205
Exercise 8: Heteroskedasticityin the Statistical EarningsFunction 208
Notes 210
References 214
Further Readings 222
apter 6
EXPLAINING AND FORECASTING AGGREGATE INVESTMENTEXPENDITURES: DISTRIBUTED LAGS AND AUTOCORRELATION
6.t Investment and Capital Stock: Definitions and General Framework6.1.4 Definitions 2276.1.B A General Frameworkfor Representing Various
Models 2326.1.C Stochastic Specification 232
6.2 The Accelerator Model6.2.4 Theory 2336.2.B Dealing with Alternative Distributed Lag Forms 2346.2.C Empirical Implementations 237
; DummyInteraction Variables: The Earnings of Single and
188
167
179
19092
224
207
Contents
6.3 The Cash Flow Model64 The Neoclassical Model
64.8 Theory 2436.4.8 On Measuringthe User Cost of Capital 24564.C Toward Empirical Implementatiun 248
65 Tobin's g Modet6.5.A Theoty 2576.5.B Issues in Empirical Implementation 259
6.6 Time Scrivs/Autoregressive Models of Aggregate Investment6.6.A Measurement without Theory? 2636.6.B Empirical Implementation 264
6.7 Additional Econometric Specificationsand Issues6.7.A Simultaneity Issues 2666.7.8 Moving Average Errors 2676.7.C Nonstatic and Rational Expectations 2686.7.D Adjustment Costs 269
6.8 —Empirical ComparisonsofFive Investment Models69 Concluding Remarks6.10 Hands-OnEstimation and Forecasting ofInvestment EquationsEXERCISES
Chapter 7
Exercise 1: Examining the Data 279Exercise 2: Testing for Autocorrelation with Lagged Dependent
Variables 281Regression Estimationof the Time Series/‘Autoregressive Model 283
4: AlmonLagEstimationofthe Cash Flow Model 284Putty-Clay in the Neactassical Model ofInvestment —286
Exercise 6: Almon PDL Endpoint Restrictions in Tobin's ¢Model 287Box-Jenkins Identification, Estimation, and Forecastingofan investment Equation 289Estimation with Simultancity andAutocorrelation —-294.Levels and First Differences of the CES CapitalDemand Equation Autocorrelation —.292.
: A “Horse Race”Project Based on More RecentData 294
Notes 295References 298Further Readings 304
Exercise 3:
ExercitExercise
Exercise
Exercise 8:
THE DEMANDFORELECTRICITY: STRUCTURAL AND TIME SERIESAPPROACHES
FA} Essential Background Facts Concerning Electricity Demand7.2 Econometric ModelsofElectricity Demand with Equipment Stocks
Included Explicitly
xiii
239242
263
265
2702777279
306
308,
dv Contents
7.3 Econometric Models of Electricity Demand with EquipmentStocksIncluded Indirectly
7.4 Additional Econometric Issues.7.4.4 Effects of Omitting the Intramarginat Price as a
Regressor 32!7.4.8 Simultaneity Between Electricity Price and Quantity
=
3247.4.C On the Choice of Functional Form 326
8 q Brief Survey of Empirical Findings.6 Hands On with Modeling and i ‘citEXEROes 1B Forecasting Electricity Demand
Exercise : Examining Houthakker's U.K. Data on 42 Towns 337Exercise 2: Examining the Nelson-Peck NERC U.S. TimeSeries
. Data 338Exercise 3: Replicating Houthakker’s U.K. Study 339Exercise 4: Omitted Variable Bias: Intramarginal Electris
. Price=
343Exercise 5: The Fisher-Kaysen Specification with U.S. TimeSeries
| Data 344Exercise 6: Partial Adjustment Specifications with U.S. Time Series
| Data 346.Exercise 7; Forecasting Using Extrapolation and Smoothing
—
349.Exercise 8: Combined Structural, Time-Series Forecasting of
Electricity Demand 352Notes 354References 356Further Readings 359
tapter 8
CAUSALITY AND SIMULTANEITY BETWEEN ADVERTISINGAND SALES
Br
8.2
84
Considerations from Economic Theory8.1. Determinants of Advertising Outlay 3658.1.8 Determinants ofSales 371Issues in Econometric implementation8.2.4 MeasurementIssues 3738.2.B Simultancity, Identification, and the Hausman Specification
Test 3758.2.C Granger Causality 3808.2.D Consistent Market Share Specifications 3838.2.£ Distributed Lags and the Measurementof the Cumulative
Effects of Advertising 3858.2.F Temporal Aggregation and the Data Interval Bias 389DossAssregate Advertising Affect Aggregate Consumption? Empirical
alOverview ofSelected Classic and Recent Empirical Studies on Sales-Advertising Relationships with and without Rivals
85 Advertising and Public Policy: Banning Broadcast Ads for Cigarettes
315320
328335337
361
365
373
393
400
409
Contents
8.6 Other Current Empirical issues8.7 Hands-On EstimationofSales-Advertising Relationships.
EXERCISESExercise 1: Assessing Price and Quantity Endogencityin a Sales-
Advertising Model for Oranges 417Exercise 2: Estimating 90% Duration Levels for Ad
Annual and Monthly Data for the LydiMedicine Company 420
Exercise 3: Choosing Between Current and LingeringEffectsModels 422
Exercise 4: Avoiding Embarrassment with Consistent Market ShareSpecifications 424
Exercise 5: Using Granges's Method to Determine Causality BetweenAggregate Advertising and Aggregate Sales 426
Exercise 6: Estimating a Simultaneous Equations Modelof AggregateSales and Aggregate Advertising 429
Exercise 7; Evaluating the Effects of the Cigarette BroadeastBan 432
Exercise 8: Distinguishing the Sales Effects of Advertising Quality andAdvertising Quantity 436
Notes 437References 441Further Readings 447
ertising withE. Pinkham.
Chapter 9
MODELING THE INTERRELATED DEMANDS FOR FACTORS OFPRODUCTION:ESTIMATION ANDINFERENCEIN EQUATION
SYSTEMS
9.1 Historical Overview9.2. The Generalized Leontief Cost Function9.3 Statistical Inference and the MeasurementofFit in Equation Systems.9.4 The TranslogSpecification9.5 Autoregressive Stochastic Processes in Multivariate Equation Systems9.6 The Translog Function in an Interindustry General Equilibrium
Model9.7 Issues Addressed by Recent Research on Factor Demand Models9.8 Hands-On Estimation of Interrelated Demandsfor Factors of
ProductionEXERCISES
Exercise |: Inspecting the Berndt-Wood KLEMDatafor U.S.Manufacturing 487
Exercise 2: Single-Equation Estimation of Cobb-Douglas and CESForms 488
Exercise 3: Comparing Equation-by-Equation and [ZEF
Estimates 489Exercise 4: Special Issues in Estimating Singular Equatior
Systems 494
xy
aia4164i7
449450460465469
46
479ABL
486487
s xvixvi Contents
Contents x!
Exercise 5: Substitution Elasticities and Curvature Checks 492 Exercise 8: Maximum Likelihood Structural and Reduced FormExercise 6: Obtaining Statistical Inference in Equation Estimation of Klein's Model | 573
Systems 494 Exercise 9: Estimation ofa Structurally Recursive Model 575Exercise 7: Goodness of Fit in GL or Translog Equation Exercise 10: Replicating the Estimation ofTaylor's REH
Systems 496 Model 578Exercise 8: Estimating interrelated Factor Demand Models with Notes 579
Vector Autocorrelation 497 References 584Exercise 9: Obtaining Estimates of Multifactor Productivity Further Readings 592
Growth 497
Notes 498References 51 Chapter 11FurtherReadings 506 —— +:AND HOW MUCH WOMEN WORKFORPAY:
REPLICATIONS OF LIMITED DEPENDENT VARIABLE PROCEDURES 593
chapter 10 | ILL Definitions and Common Data Sources 596
PARAMETER ESTIMATION IN STRUCTURAL AND REDUCED FORM 11.2 ThsEsonomic Tacryof Labor Fores Paris 399SupplyEQUATIONS OF SMALL MACROECONOMETRIC MODELS 507 | TLZA Individual Labor Supply 60010.1. Measurement Issues: The Unemployment Rate 510 11.2.B Household Labor Supply 60710.2 Extraordinary Econometrics: The Original Phillips Curve 512 11.2. Effgcis on Labor Supply of Changes in Aggregate10.3 Theory and Observation: Should the Phillips Curve Be Stable? Is It? $17 ; Demand 609
10.3.4 Early Extensionsof Phillips's Findings 517 11.2.D The Allocation of Time: A More General Treatment 61010.3.8 Phelps, Friedman. and the Expectations-Augmented Phillips TE2E Implications of Costs of Employment 613
Curve 521 11.3 Econometric Issucs and Representative Empirical Findings 61410.3.C Parameter Stability When GovernmentPolicies 11.3.4 First-Generation Studies 614
Change 524 11.3.B Second-Generation Studies 617
(1.3.C OntheSensitivity of Resulis to Alternative Statistical and
Economic Assumptions: The Mroz Study 637
11.3.D Issues in Dynamic Labor Supply 645 .
11.4 Additionat Remarks on Econometric Analyses of Labor Supply a9
10.3.D_ inverting the Phillips Curve: The Lucas-RappingEquilibrium Model 527
10.3.E A Brief Overview of Related Research 53610.4 A Two-Equation Macroeconometric Model with Rational
Expectations 54) 11.5. Hands On with Limited Dependent Variable Techniques to Assess10.5 Parameter Estimates of Kicin’s Model [ Using Alternative the Labor Supply of Married Women 630
Simuttancous Equations Estimators 350 EXERCISES 6st10.6 Hands On with Parameter Estimation in Structeral and Reduced Exercise I: Inspecting Mroz’s 1975 Panel Study ofIncome
Form Equations of Small Macroeconometric Models 555 mics Data 651 .EXERCISES 557 Exercise ating the Hours Worked Equation Using
Exercise [: OLS, 25LS, and GLS Estimation of Kicin’s Procedurel 654Model1 $57 Exercise 3: Comparing OLS,Probit, und Logit Estimates of the
Exercise 2: The Two OLS Steps in 2SLS: A Lucas-Rapping Faux LaborForce Participation Decision 654Pas? S58 Exercise 4; Relating the Tobit and Conditional OLS
Exercise 3: REH-Consistent Estimation Using 2SES_—561 Estimates 657Exercise 4: Testing for Exogencity Using Hausman’s Specification ‘ Exercise $: Identifying Parameters in a Reduced Form
Test 565 L Estimation 658 .Exercise 5: Testing for Serial Corrclation in the Lucas-Rapping 4 Exercise 6: Implementing the Heckit Generalized Tobit
Model 567 i Estimator 661
Exercise 6: The Equivalence of Alternative Estimators in 3 Just : Exercise 7: Incorporating Income Taxes into a Modetof Labor
Identificd Model $70 Supply 663Exercise 7: Comparing 2SLS, ISLS, and I3SLS Estimatesin an Exercise 8: Specifying and Estimating an Extended Tobit
Overidentified Model 572 Model 664
xviii Contents
Notes 666.References 671Further Readings 679Author Index 681Subject Index 695
“Ufyou think ... that anything like a romanceis preparingyou, reader, you were never more mistaken. Do you anticipate
sentiment, and poetry, and reverie? Do you expect passion.
and stimulus, and melodrama? Calm your expectations,reduce them to a lowly standard. Something real, cool and
solid lies before you; something unromantic as Mondaymorning, when all who work wakewith the consciousness thatthey must rise and betake themselves thereto.”
CHARLOTTE BRONTEPrelude to Shirley
Chapter I
Computers and the
Practice of Econometrics
“[ think thereis a world market for about five computers. ”
THOMASJ. WATSON.' Chairmanof the Board—IBM. 1943
“Thereix no reasonfor anyindividual to have a computer in their
home.”
KENNETH OLSON.? President, Digitat Equipment Corporation, 1977
“One machine cando the work offifty ordinary men. No machine
can do the work ofoneextraordinary man.”
ELBERT (GREEN) HUBBARD?
Computers andthe Practice of Econometrics
| The purpose ofthis bookis to engage you in direct “hands-on” empir-ical applications of econometric tools, using classic and contemporary datasets. To becomeinvolved in such experiences, you will need to work with acomputer. Although this text is designed primarily with personal microcom-puters in mind (especially the IBM-PC andits compatibles or the Macintosh).all the exercises in this book can be performed using mainframe computers,
The goal efthis chapteris 10 help you get going with the implementationof econometrics. In Section 1.1 we begin with a briefhistorical overview ofcomputers in econometrics, andthen in Section 1.2 weintroduce you to somebasics ofcomputer equipment (hardware) and computer programs (software),focusing in particular on the most important software accompanying anycomputer,called its operating system.
With this as background, in Section 1.3 we acquaint you with proce-dures necessary for accessing the classic and contemporary data sets from thefloppy diskette accompanying this text. Although particular procedures fordoing this vary among computers, certain steps are almost universal, and wefocus attention on them. in Section 1.4, we briefly overview the structure ofthe exercises found at the end of each chapter. Finally, in Section 1.5 weinvolve youin a very simple hands-on application, to makesure you are readyand able to take on the applications and exercises of subsequent chapters.Two appendices accompanythis chapter. Jn Appendix A we overview a vari-ety of statistics and econometric software packages for the [BM-PC and theApple Macintosh. In Appendix B we provide you with references for obtain-ing further information and productreviews.
HISTORICAL PERSPECTIVES ON COMPUTERS ANDECONOMETRICS
The empirical implementation of econometric procedures is much simplerand easier today than it was even a decade ago, owing in large part to thedevelopment and diffusion of increasingly inexpensive and powerful com-puter hardware and software, Although recent developments appearparticu-larly dramatic, remarkable improvements in computational capabilities foreconometrics have been occurring for quite sometime.
For example, one might characterize thefirst generation of empiricaleconometrics as one in which computations were doneentirely by hand. Theadventof hand-crank andelectronic calculators improved matters consider-ably, but such second-generation developments still implied enormous com-putational times for typical problems in empirical econometrics. For exam-ple,at ‘Meetings in 1946 of members of the Cowles Commission staff (aninstitution associated with numerouspioneering developments in economet-Fics), the following exchange took place concerning the estimated timerequired to undertake econometric computations:
1d Historical Perspectives 3
In answer to a question by Girshick, Koopmans mentioned that onesupervisor and oneto two computers had worked two to three months on
an cight equation system,by handcalculating. Rubin estimated that a ten
equation system with fifty unknown parameters by hand computing with
an ordinary calculator requires seventy 24-hour computerdays."
with personal computers and modernstatistical software, such com-
ionstake less than 20 seconds.The history of computation in econometrics is not yet well chronicled.’
A few interesting facts are known, however. Oneis that in 1944, when Guy
H. Orcutt became an assistant professor of economics at the Massachusetts
Institute of Technology, he also began working for Professor George Wads-
worth of the MIT Mathematics Department to build and then operate a
“regression analyser,”the design of which was based on Orcutt’s Ph.D. thesis
in economicsat the University of Michigan,a thesis complete with electronic
circuit diagrams.° Although Orcutt’s prototype analog machine had only two-
digit accuracy and stored data for only 33 item timeseries on punched cards.
which were continuously read many times per second, one might charitably
think ofit as being the first portable computer. In particular, according to
Orcutt,
Todaputz
As events unfolded, the planned prototype machine was built and used
Successtully for a few months as World WarIL was beginning to wind
down. | took my regression analyser with me to Cleveland and demon-
strated it al the annual Christmas meetings. Since | was grossly over~
loaded, | was unable to take along the cathode ray tube whichserved as a
graphical display device for showing time series and scatter diagrams
derived from the three timeseries segmentsbeing read into the regression
analyser. But 1 wasable to borrow onefrom Case Universityin Cleveland.”
The regression analyser traveled even farther. During the 1945-1946
academic year, Orcutt was invited by Richard Stone to cometo the newly
established Department of Applied Economics at Cambridge University, with
the expressed hope that “I would bring the machine and put it to use in devel-
oping tests ofsignificance of apparentrelationships between autocorrelated
economic timeseries.”* it was there that Orcutt used his regression analyser
in working with autocorrelated timeseries generated by stochastic time series
models that were estimated by using data from Jan Tinbergen’s pioneering
macroeconometric mode!. The regression analyser played an essential role in
makingitfeasible to carry out Monte Carlo computationsfortesting the sig-
nificance ofapparentrelationships between economictimeseries having auto-
correlated characteristics.”Apparently, the first article published by an econometrician based on
results from a digital clectronic calculator having a stored memory program
is that by Hendrik S. Houthakker."” Houthakker’s research involved working
Computers and the Practice of Econometrics.
GUY ORCUTT
A Pioneer in Econometric Computation
om in a Detroitsuburb in 1917,Guy Orcutt spent
muchofhis spare time asa high schoo!student ex-perimenting with clectric-ity and chemistry, evenaccidentally creating an2xplosion that redeco-sated part of his bedroomceiling. After graduatingTom the University ofMichigan with majors inyhysics and mathematics, Orcuttpent the summer of 1939 at aQuaker work campin a severely de-wressed coal mining region of West/irginia. This experience kindled annterest in economics and led him tonroll as a graduate student in eco-comics at Michigan, even though head not taken a single undergraduateconomics course.Mentored by Arthur Smithies, as a
taduate student Orcutt was encour-ged to expand on his interests insing electronic devices for undertak-ag complex mathematical computa-ons. Thisled to Orcutt’s design of avegression analyser,” the theoryand
reuit diagrams of which are promi-
:ntly displayed in his Ph.D. dis-mation.Since the actual construction ofch a machine required consider-ile time, funds for materials, and‘lp from machinists (support thatts not available to Orcutt as a grad-te student), it was not until late‘45 that a prototype of Orcutt's
regression analyser wassuccessfully constructed.
This took place at theMassachusetts Institute ofTechnology, where Orcuttwas hired in 1944 toteach beginning —eco-nomics courses to engi-neering undergraduatesbut was simultaneouslyinvolved with ProfessorGeorge Wadsworth oftheMIT Mathematics De-
partment in an effort to improveweather forecasting by relating theweather at a given location to theweather somewhat carlicr at other lo-cations.
Orcutt’s regression analyser wassuccessfully demonstrated at the an-nual meetings of the American Eco-nomic Association in Cleveland in1945, andhis interest in autocorre-lated time series attracted the atten-tion of Richard Stone, who invitedOrcutt to Cambridge University be-ginningin thefall of 1946.
There, in the newly created De-
partment of Applied Economicsunder the direction of Stone, Orcuttrefurbished his regression analyser,availing himself offacilities at theCavendish Laboratory. He then usedthe regression analyser in MonteCarlo studies that focused on devel-opingtests ofsignificance of apparentrelationships between time serieshaving autocorrelated characteristicssimilar to actual time series beingused in econometric modeling of na-
tional economies. These efforts re-
sulted in two papers, one in the 1948
volume of the Journal ofthe Royal
Statistical Society, Series A, and the
other with S.F, James in the Decem-
ber 1948issue of Biometrika.
Together with a Cambridge grad-
uate student named Donald Coch-
rane, Orcutt also developed the now
widely used Cochrane-Orcutt estt-
mator for models with first-order au-
tocorrelated disturbances, which was
described in the March and Septem-
ber 1949 issues of the Journalof the
AmericanStatistical Association.
with the EDSAC—the Electronic Det:
University Mathematical Laboratoryat ¢
7 ofthis book you will have an opportunity to wo!
data set. and to reproduce his EDSACresults
1.1 Historical Perspectives
In additionto his pioneering work
‘on electronic computations and the
estimation of autocortelated time se-
ries models, Orcutt is widely known
for research on the measurement of
price elasticities in international
trade andfor the development of mi-
croanalytic models of socioeconomic
behavior. |
Guy Orcutt retired from Yale Uni-
versity in 1988 and nowlives with his
wife Geil (and bis personal com-
puter) in Grantham, New Hamp-
shire.
ous advantagesoverelectric hand-calculators:
tt takes about 7 minutes on the EDSACto compute all the 55 weighted
sums of squares and cross-products of 10 variables with 40 observations
jn addition to about 4 hours for punching and checking the numbertape
Photo of EDSAC—the Electronic Delay Storage Automatic Caicuator.
5
jay Storage Automatic Calculatorof the
at Cambridge University."' (In Chapter
rk with Houthakker’s classic
.) The EDSAC provided numer-
4.2
Computers and the Practice of Econometrics
and verifyingthe results bya sum-check. A human computerwith an clec-
tric desk machine would probably nced about 75 hours for this job. so that71 hours oflaborare replaced by 7 minutes of machine time."
Since the EDSAC, numerousother very important developments haveoccurred that have lowered enormously the computational costs of imple-
menting empirical econometrics. For example, major product innovationsinclude the installation of the first UNIVAC machineat the U.S. CensusBureau in 1951; the delivery of IBM's first electronic computer, the 701, toLos Alamosin 1953; the IBM debut of FORTRAN,thefirst high-level com-puter language,in 1957, the Digital Equipment Corporation introduction ofthe minicomputerin 1963; the unveiling of the [BM System/360in 1964, thefirst family of computers; the introduction of mass-produced pocket calcula-tors into the United States in 197); the initial offering of the AppteII personalcomputerin 1977; andthe entry of [BM into the personal computer marketin 198),
Although the history of these developments makes forfascinating read-
ing. for our purposesit is more useful here to focus on thefast of these devel-opments, namely, the personal computer and its uscfulness for econometricresearch.” To this we now turn ourattention.
BACKGROUND: COMPUTER HARDWARE ANDCOMPUTER SOFTWARE
At the heart of the personal or microcomputeris the microprocessor, a chipthat performs most of the computer's work. Quite frequently, microproces-sors are characterized in terms of the numberofdirs they are able to processat one time. The word “bit,” incidentally, is a contraction of the words“binary digit," each bit being either a zero or a one. A 32-bit microprocessorcan process 32 zeros and ones simultancously. Although carlier microcom-puter models often had 8- or 16-bit microprocessors, today models in the IBMPS/2 series contain a 32-bit Intel 80386 microprocessor. while the Macintosh1) utilizes a 32-bit Motorola 68020 unit.
Most personal computersoftware and statistical data comein the formof a floppydisk, made from a thin film of mylar piastic coated with ferrousoxide and placed inside a lubricated plastic protective jacket; commonsizesfor these diskettes are 5.25” and 3.5”. When blank floppy disks are purchasedfrom a store, they are typically usable in any personal computer, since the
millionsoftiny magnets in the ferrous oxide coating are arranged randomlyonthedisk, thatis, the disk is unformatted.
Thefirst operation users typically perform with a floppydisk is toformatit, which means magnetically inscribing pie-shaped sectors and concentrictracks onto the disk. Precisely how these sectors and tracksare laid out on thedisk varies depending on the specifications adopted by the various hardwareand software manufacturers. When a disk is formatted, each of the tiny mag-
1.2 Background
=
7
soaps . ees direee
nets in the ferrous oxide coating is “lined up” so that each points in the
i indicates “zero.” i tie“en "Goce the disk is formatted, data can be stored on it. Each magn
icS the
dipole in the ferrous oxide coating bas northand asPol "
i is pointing left, a “zero” or “off” is indicated, and Wh toes
gina . Feht sone” or “on”is implied. Eachbit is either “on” or “oft”A
aanne eitbits is called a byte. a byte is also often called 2 characterI.
, a number, or some other symbol. ; |
on Sh 2tondard code used to convert leners, numbers, and symbols into
‘oy the ASCH (American Standard Code for Information
i is called: c En
bitsaneyecode, Examples of ASCII code include the following:
ASCHCode0100 Cou!0100 00100100 001!0100 0100
6011 0000011 GOOT
001! GO10
011 GOI
0010 1010
0010 000!
Character
mwswrHe
eoDOW>
‘The standard ASCTI cade contains 256 distinct characters (2 raised to the
CASE rease alpha-
eighth power), including numbers, symbols, uppel andlowe aseamhae
betical elements, and some foreign language characters. dat owing
ided on the floppy disk accompanying this text have been codes lowing
ascll conventions and therefore should be readable by almost any x
ini inframe computer. | ; |
inpoydisks canstore a great deal of information. The reelsamour
of storage capacity of a disk depends on the number of sectors a!
inser i ang system when the disk was most recently for-
inserorne neeeecommon formatting procedures for an 1M or
matted. Onefrmachine was the 360K format, in which 512 bytes Petocho
aegenerated, there were nine sectors per track, and 40aesTaOtyor
f the twosides of the floppy disk. This yielded a total of 3 aedme
on 949. 120 bits); such disks are ‘commonlyreferred to as having K (kilo
Se) mapacity, Disks with 720K, 1.2 MB (for megabyte)and¢ AMBeanaciy
are now commonly wed froeaeHopsk ive manyPersonal computers usually fa) ‘ a d a
iti i disk drive, the latter essentially
ave an additional floppy drive and/or ahard kd on
7 i several floppydrives but permitting more con’ "
beingandprograms on the computer. Data, programs, and files con:
8 Computers and the Practice of Econometrics
tained in the floppy disk drives can usually be accessed by addressing theappropriatedrive; typically, A: and B: refer to the floppy drives, while C: (orDs)refers to the hard disk drive.
The operating system is the lowest level of sofiware on the personalcomputer. It interacts directly with the computer's microprocessorto performthe most basic and important housekeeping functions, such as formattingdisks, copying disks, saving and loading files, checking the status ofa disk,and keepinga directoryofthefiles on the disk. For the IBM-PC the operatingsystemis typically stored partly on a ROM (Read Only Memory) chip insidethe microcomputer and partly on the DOS(Disk Operating System) masterdiskette or on the hard disk. This means that certain commands—thosestored in ROM—canbe performed by the IBM-PC computer without evenloading the DOS diskette. We now overview some of the most importantfunctions performed by DOS on the [BM-PC andits compatibles.
With data stored in files on disks it is often convenient to Organize thefiles into separate groups. One function of DOSisto create distinct directoriesOr groupsoffiles, where thefiles within cach directory are kept separate fromthefiles in other directories. With DOS,each disk has a main,or root, direc-tory. Tothis root directory, one can add new subdirectories where files canbe keptin distinct groups.
Consider, for example, the data files that must be accessed to performthe empirical exercises at the end of each chapter in this text, The variousdata series for cach chapterare stored as distinctfiles in chapter-specifi sub-directories. For example, in the data diskette provided with this text, the rootdirectory contains the names of one documentation file. README,DOC,one data file, COLORTV. and ten subdirectories: CHAP2.DAT.CHAP3.DAT, CHAP4.DAT, ... , CHAPIL.DAT. Wi each of the tensubdirectories, you will find a numberofdata files containing data for hands-on exercises from the appropriate chapter.
Asan example,take the CHAP2.DATsubdirectory. It contains over 20data fites, each containing data onrates of return on securities for some par-ticular company. such as MOBIL, GERBER. DELTA, and TANDY.Simi-larly, the CHAP3.DATsubdirectory contains data files on production costsfor the manufacture ofpolyethylene (POLY)andtitanium dioxide (TI02),aswell as a rather strange data set named WEIRD,
One panticularly important function of DOSis to make backup capiesof importantfloppydisks. Right now is a very good time to make a backupcopy of yourdata disk. To dothis, use the XCOPY /S command.as noted inyour DOS manuat. (/mportant: Do not use the COPY command,since thedata diskette contains a numberofsubdirectories; the COPY commandwillusually not copy subdirectories. If your computer uses a version of DOSear-lier than 3.2, however, the XCOPY commandis not available; instead youcan employ the DISKCOPY command, even though it is somewhatcumbersome.)
Since subdirectories branch off ofdirectories, they are said to form a“tree”structure, Paths or patinamesare used to climb through the branches
1.2 Background 9
i i ing device that indicates,f the trees. More specifically, the pathnameis a routing
through a sequence ofaddresses, just where a file can be accessed. Foroan
ple, the data file GOLDin the subdirectory CHAP2.DAT ona floppy in is k
drive B can be called from any current working subdirectory by typing the
pathname
B:\CHAP2.DAT\GOLD
Notethat the backslashes (\) are used toseparate the names ofditectones and
files in the pathname.Also, when operating the computer, you wil It notice al
one directory is always the default:it
is
called your current working d rec o} i
To move between directories, type “cDv" (Change Directory) fol lowe
by the pathnameofthe directory to which you want to move °create a
directory, type the “MD”(MakeDirectory) command, followed by the mame
of the directory. For example, if you want to create a directory calle
BACKUPonthe disk deive B, type
B:\MD BACKUP
isk drives, type the letter name of the drive you want to access,
Taeaeion Forexample. if you are currentlyin the A drive and want
B drive, type “B:”. ;© Ootobe named Brusing a combination ofup to eight letter and/or
numbercharacters. (Somestatistical programs, however, restrict the number
of characters 10 six or fewer, Check your econometne software program doce
umentation for furtherdetails.) File names often have an extension,whiich
a period just afierthe filename,followed by a maximum of three ch raters
Extensions such as EXE and .COMindicate to the computer that e was
been rearranged (compiled) by software so that it can becee
ately. Often, however, users employan extensionto aid in quickly identifying
file’: Ss. .
afl Filescan be copied by using the COPYcommand. Forexample.“ you
wished to copya file named GOLDin thedirectory CHAP, wi ichis
currently on a floppy disk in the A drive, onto a file named GOLD in ac i rece
tory named BACKUPon a floppy disk in the B drive (assuming thai
BACKUPdirectory bas been created), you would type
COPY A:\CHAP2.DAT\GOLD B:\BACKUP\GGLD
i i ye ind 2 ide to renamea file. ThisAt times you might change your mind and deci ea file. s
can he done ery simply. For example, the GOLD data file mentioned in the
previous paragraph could be that for the gold market in London; to remind
yousself of that, you might want to change its name from GOLD to
GOLD.LONbytyping
RENAME B:\BACKUP\GOLD B:\BACKUP\GOLD.LON
‘A very common operation involves examining the contents ofa file
visually on the monitor screen. For example, you might have performed an
operation using an cconometric software program and might then wal
‘Computers and the Practice of Econometrics
examine the outputfile to determine whether you made any programmingmistakes. Anyfile stored in ASCII characters can bedisplayed on the monitorscreen by using the command
TYPE FILENAME .EXT
where thelocation ofthefilenameis identified by a pathname. If you want tohave the file contents shown onescreen al a time, add “/P” to the TYPEFILENAME.EXTcommand.Ifyou decide to have the material currentlydis-played on the monitorscreen printed, simultancouslyhit the keyboard keys(Shift) and (PrtSc). Finally, if you want to have the entire contents of anASCII file printed out (not just the portion displayed on your monitorscreen),type out the DOS command
TYPE FILENAME.EXT > PRN
Finally, it should be noted that to work with datafiles, to create sets ofcommandsforstatistical software programs, or to modify any ASCH] files, itis often most convenient to employ computer software that is specificallydesigned to allow youto edit files. For the IBM PC and its compatibles themost common text editor software is EDLIN (a component of DOS) and theBASIC Program Editor. More sophisticated and powerfultext editor softwareProgramsare typically called word processors. A wide variety of word proces-sor programsare available commercially, and some are cven in the publicdomain at virtually no cost. Reviews and comments on various word praces-sor software packages can be found in numerous magazines available at news-stands. You will find that having access to and being familiar with a wordprocessor software package will help you enormously in doing the exercisesin this textbook.
While this review of the IBM-PC and the DOS operating system isadmittedly verybrief, it shouid be sufficient to gel you started. Whenever youare in doubt about the appropriate choice of commands,refer to your DOSmanual;ifyou are working on a computerother than the IBM-PC orits com-patibles or on a mainframe computer, examineils documentation, check withthe mainframe computer opcrators, or see yourinstructor."
ACCESSING DATA FROM DISKETTES FORUSE INCOMPUTER SOFTWARE PROGRAMS
Aswas noted in the previoussection, the data necessary to do the empiricalexercises at the end of each chapter have been written onto a data diskette.These data are written using the almost universal ASCII free format proce-dures and thus can be read and edited or modified by virtually any word pro-cessor program. The data have been organized accordingto the structure pre-sented in Table 1.1,
Regression and statistical/forecasting software programs vary consider-
1.3 Accessing Data from Diskettes 11
Table 1.1 Structure of Data Diskette Accompanying the Practice of
Econometrics
NumberofFiles in Numbers ofCharacters
Root Directory Subdirectory Subdirectory in Subdirectory
Contains names of CHAP2.DAT 23 38,781
subdirectories, a CHAP3.DAT 6 14,774
documentationfile CHAP4.DAT 4 29,840
called README.DOC, CHAPS.DAT 3 59,798
and a data file called CHAP6.DAT 2 16,392
COLORTV. CHAP7.DAT 3 5,558
CHAP8.DAT 8 52,385
CHAP9.DAT 2 6.717
CHAPIW.DAT 4 10,397
CHAPI1.DAT 2 61,624
ably in the way in which theyinput or read data from diskettes. In most sta-tistical software programs the data are first described in terms of the numberof observations and the names ofthe variables. This part of the data inputistypically called the Aeader. The header provides information enabling thesoftware program to read the data correctly. For example, are the data orga-nized byvariable or by observation? The second portion ofdata input in moststatistical software programsconsists of the data itself—rows and columns ofnumbers, delimited by spaces or commas. Somesoftware programs requirethat the data be organized (formatted) in a particular manner, while a goodnumberofprogramsare now more flexible and permit “free format”entry ofdata. Thethird and final componentofa typical data inputis a word, symbol,or set of characters immediately following the last data entry, indicating that
the data series has terminated. Thisis often called a data terminator.You should check yourstatistical sofiware manual for complete infor-
mation on how the header. data, and data terminator or their equivalentsmust be specified in order that your input data can be read correctly. Notethat it will most likely be necessary for you to use a word processor programto modify (edit) the data files on the data diskette accompanying this text.
The numberof powerful statistics, econometrics, and forecasting sofi-ware programs that can be executed on the personal computeris growing veryrapidly, as are the features that these programs incorporate. Although this textcannotpossibly provide readers with up-to-datelists of alt available softwareand features—the computer environmentis changing too rapidly—two typesof important information can be given here.
First, in Appendix A to this chapter we provide somegeneral commentsconcerning several “extended families” of software programs. Software thatis currently eitherfree orrelatively inexpensive in the form ofspecial studentorinstructionaleditionsis highlighted.
La
Computers and the Practice of Econometrics
Second,since the available menuofstatistical software is changing rap-idly, in Appendix B to this chapter we furnish the addresses and phone num-bers of vendors of the most commonstatistics, econometrics, and forecastingsoftware for the [IBM-PC and compatibles and for the Macintosh. While thislist is not exhaustive,it is representative.
Youare strongly encouraged to contact vendors listed in Appendix B toobtain currentinformation on prices, hardware requirements, and features.When examining prices, you might examine not only the list price, but alsothe availability of quantity discounts, special educational prices, and sitelicense fees. In addition to standard statistical features, you might also com-pare the programsin termsoftheirability to display data graphically and their
amenability to importing or exporting data consistent with protocols of otherwidely used data base programs, such as Lotuse 1-2-3e or Symphony®,
Statistical software programsare frequently reviewed in computer mag-azines such as PC Magazine, PC Week, and info World (for the IBM-PC andcompatibles) and MacWorld, MacUser, and MacW’eek (for Apple Macintoshmachines). These magazines are typically available at most newsstands."® Fur-ther, The American Statistician, a quarterly publication of the American Sta-
tistical Association, regularly contains a special section devoted to summariz-ing and reviewing recentstatistical software for the personal computer.” Also,Springer-Verlag Press publishes a readable journal devoted to statistics andcomputing, called CHANCE."Finally, for a periodic survey and update ofavailable statistical software. see the Directory of Statistical MicrocommuerSofiware.”
A NOTE ON THE END-OF-CHAPTER EXERCISES
Theexercises at the end of each remaining chapter in this book have beencarefully designed to enable you to obtain hands-on experience in imple-
menting empirical research. One importantfeatureofthis textis that its chap-ters are ordered to correspond with ever increasingly advanced econometricprocedures—techniques typically learned in a sequential manner in mosteconometrics and forecasting courses. Therefore it is expected that you willwork through Chapter 2 before considering, say, Chapter 8.
However, the empirical topics are of considerable inhcrentinterest, and.much insight and understanding would be lost if the specific topics wereaddressed only once during a course. For that reason, while most ofthe exer-cises at the end ofeach chapter deal with a particularset ofeconometrictools,in almostall chapters the final two orso exercises use more advanced proce-dures, This means that you can undertake the vast majority ofexercises al theend of each chapter as youinitially proceed through a typical econometrics/forecasting course. However,thelast few exercises usually involve techniquesand procedures that are customarily learned later in such courses; you mightnot be able to do them until your instructor introduces the appropriateProcedures,
45
Exercises 13
You are mostlikely to gain a deeper appreciation of the empirical
aspects of eachspecific topicif, as your course of studies proceeds, you return
to the final, more difficult exercises of earlier chapters. In this way you can
examine issues of robustness, observing by yourself how results might be
affected by changes in assumptions andestimation methods. More generally,
returning ta the sametopicseveraltimes during a course enables youto expe-
rience the special satisfaction of observing your own maturation as you
employincreasingly sophisticated research tools and methods.
HANDSON WITH AN EXPLORATORY DATASET
The purpose ofthis section is to involve you in practicing some basic com-
puterskills to ensure that you are ready and able to perform the empirical
exercises found at the end of each remaining chapter in this book. Thefol-
lowing exercise involves entering data directly from the keyboard and then,
alternatively, accessing the same data from a data file on the data diskette
provided. It is assumed that you have access to a personal, mini, or main-
frame computer, that you have available and understand how to use a text
editor/word processor software program, and that vou also havestatistical
software at your disposal. .
Table 1.2 displays time series data for the unit price and cumulative
production ofcolortelevision sets in the United States over the 1964-1979
time period. This type ofdatais often used to analyze theeffects of learning
on production costs and prices, whereit is assumed that, as cumulative pro-
duction increases, learning occurs, resulting in lower unit costs and prices.
Thesubject of learning curves, incidentally,is discussed in considerable detail
in Chapter 3 ofthis text. As is seen in Table 1.2, the average price ofcolor
TVs has tended to decline as cumulative production has increased.
EXERCISES
Ensuring That Data Are Properly Entered andTransformed
EXERCISE 4:
The purpose ofthis exercise is to have you practice somebasic computerskillsto ensure that you are able to enter data into yourstatistical software program,both from the keyboard directly and from a data file provided on your data
diskette.
(a) Boot(start) up your computer, makesure that DOSorits equivalentisloaded, and then load into memory yourstatistical software program.
(b) Usingdirect keyboard entry, note the appropriate sample size (16 obser-vations) and then enter into your program historical values ofthe vari-
4 Computers and the Practice of Econometrics
Table 1.2 Prices and Cur imulative Pr iictaetnnes ToneProduction Of Color Tetevision Sets in the
ice Inder Cumulative ProductiYear(TVPRICE)
crveunans ™UMP!!S04 m7~
1969
3.1041968787
3.798
Ms 16333731968
662
i sn ire197] oui
62soss373
19722520
wn a50.218
i975ia
eto
aied
9191976
fc)
(d)
Why?(ce) Now check wheth ier you obtthen ye tain the sameresults when You use data fromiskette rather than typing it directly in from the keyboard. Read
ro
Appendix A 15
yoursiatistical software program manual and discover how data on diskfiles must be formatted in orderto be read in properly by yourstatisticatsoftware. (You might also want to reread Section 1.3 of this chapter.)Next, to ensure that you are able to enter data indirectly from your datadiskette, read in the data file COLORTYfoundin the root directory onyour data diskette, containing the variables YEAR, PRICETYV, and
CUMPRTV.(f) In parts (a) through (d), you used data that you entered directly via the
keyboard. Now,having entered data indirectly via datafiles on diskettes,repeat parts (c) and (d), noting that in the COLORTVdata file theTVPRICEvariable is called PRICETV and that TYCUMPRis namedCUMPRTV.Verify that the results obtained here are identical to thoseyou obtained originally with parts (c) and (d).
(g) Next, perform a simple linear regression of the logarithm of PRICETV.on a constant term and the logarithm ofTYCUMPR.Do theregressionresults seem plausible? Why or why not?
(h) Compute andplot the residuals from yourregression, and comment onthe timeseries pattera ofthe residuals.
(i) Having completed this practice exercise, you should now be able to pro-ceed with exercises from subsequent chapters of this book. Good luck!
APPENDIX A
An Overview ofStatistics and EconometricsSoftware for the PC
Although software vendors typically expend considerable efforts in differen-tiating their products from others, cerlain commonalities exist within groupsofsoftware. Oneofthefirst majorstatistical programs written for use by econ-ometricians on mainframe computers was TSP {Time Series Processor).It isa widely used program, and various incarnations and modificationsofit withdiffering characteristics are available for the PC. These include PC-TSP,MicroTSP®, ESP, and Soritec. MicroTSP® offers a special student edition ata relatively inexpensive price. A somewhat smaller version of Soritec, calledSoritec Sampler, is available free of charge from the Sorites Group.
Another software program originally written for econometricians work-ing on mainframe computers, but now designed for personal computers, isSHAZAM.Likethe various incarnations of TSP, SHAZAMincorporates awide rangeoffeatures that are typically employed by practicing econometri-cians. Personal computer versions of SHAZAMare available both for theiBM-PC and compatibles and for the Apple Macintosh.
A widely used microcomputer software program is RATS (RegressionAnalysis of Time Series). It is a very powerful package with particularstrengths in time series applications, although it incorporates a wide varietyofotherfeatures as well, Researchers who enjoy and are somewhatskilled atprogramming have tended to prefer the RATS software overthat in the TSP
16 Computers and the Practice of Econometsies
family: the latter tends to be more user-friendly and well-documented butisnotas efficient, although differences between them are diminishing with time.RATSisavailable for the 1BM-PC and compatibles and for the Macintosh. Arelatively inexpensive student version of RATS is also available. PDQ, SCASYSTEM, PC-GIVE, and DATA-FIT are examples of other software pro-grams with particular strengths in time series analysis, model specificationanalysis, and forecasting. The vendor of PDQ has special financial urrange-ments for classroom use of the program.
A large familyofstatistical software has principal roots in disciplinesother than economics, such as in statistics, or other social or physical sciences.While this software might not haveail the features currently used bystate-of-the-art practitioners of econometrics. the power and range of the featuresoffered are often considerable. In many cases this type of softwareis particu-larly designed to deal with very large data bases and produces excellentgraphic display summariesofdata. This family ofsofiware includes BMDP/PC, MINITAB, P-STAT, SAS/STAT, SPSS/PC+, STATVIEW 512+.STATA, STATGRAPHICS, STATPRO. and SYSTAT. SYSTAToffers adownsized version, called MYSTAT, free of charge for instructional use.BASSSTATis somewhatsimilar ta SAS but is smaller and considerablylessexpensive.
Another family of software was originally designed for estimation ofmodels in which the dependent variable is discrete, rather than continuous.This software group includes LIMDEP and SST;bothof these programs. aswell as a recent entrant from the LEMDEPfamily called ET, have expandedthesetofoffered features considerably beyond the originaldiscrete dependentvariable orientation and now include many common econometric pro-cedures.
A different set of software was originally designed to perform matrixalgebra calculations in a computationaily efficient manner and then wasexpanded to include a numberofthe more commoneconometric procedures.as simple sets of commands. Curtenily, the best-known such program isGAUSS. Another example is MATLAB,which is available for the [BM-PC,the Macintosh, and a numberofother microcomputers, minicomputers, andworkstations.
Finally, several large data base service companies have developed andmarketed software enabling users lo download data from mainframe com-puters and then to perform statistical operations an decentralized personalcomputers. Users are also-typically given the option of performing moresophisticated or exotic econometric procedures by using the large library ofprograms available on the mainframe computers. Two examples are theECONOMIST WORKSTATION from Data Resources/McGraw-Hill, andAREMOS/PC from Wharton Econometric Forecasting Associates, Inc.
This concludesourbrief overview ofthe rapidly growingsetofstatistics,econometrics, and forecasting sofiware programs that are available for thepersonal computer. It is worth repeating that prices, features, and hardware
Appendix B17
san ing a
requirements for these programs change rapidly and that pee mee
i isi ov should contact vendors ane 3major purchase decision yt is 10 and iete
i i i tion. The information contar ereviews to obtain current information. The c ined in Ay
dix B should be of considerable help in this process. Final tyesagewine
aainfes c" act your instructor or col eca mainframe computer, contac rin on Ope Ne
i jon ce i vailable statistical software. Manyolformation concerning availabl 5 fa "
grams mentioned aboveare also available for selected mainframe computers.
APPENDIX 8
i i the IBMPCistics and Econometrics Software for # vi
SoeCmpatibies and for the Apple Macintosh: A
PartialList of Products and Vendors
IBM-PC and Compatibles /
AREMOS/PC. Wharton Econometric Forecasting Associiies, 3
" Aphia, PA 19104, (215}-386-9000. | . -
aassetat.Bass institute Ine.. P.O. Box 349, Chapel Hill, NC 27514. (YE9EYI3-
T7096. .
BMDP/PC. BMDP§ stical Software Inc.,
‘Angeles, CA 90025. (213)479-7799DATA-FIT.Oxford Electronic Publishin
X2 6DP, U.K.. — .
ECONOMIST WORKSTATION.Data Resources/McGiraw-Hill, 24 Hartwell Ave-
" nue, Lexington, MA 02173. (617}-863-5 100.
ESP. Economic Software Package. 76 Bedford St..(617)-861-8852.
ET. William H. Green100 Trinity Place,
GAUSS. Aptech Systems Inc.. 26679, . .
{AMDEP.William H. Greene. Stern Graduate School of Business,
ity, inity Place. New York. NY 10006,
matheManWorks Inc., 20 N, Main St., Sherborn, MA 01770,017)-66 ,
MICRO. TSP. Quantitative Micro Soltware, 452) Campus Drive, Suite 33,
5. (714)-856-3368.MUNTEAB.Mi ay308) Enterprise Dr., State College. PA Voxal CHA2I
PC-GIVE, University of Oxford, Institute ‘of EconomiesanuStatistics. St. Crass Build-
ing, Manor Rd., Oxford OXI 3UL U.K. (0865}-249631. 15p326-1927
PC-TSP. TSP International, P.O. Box 61015, Palo Alto, CA 94306.( 32 27.
PDQ, Charles R. Nelson, 4921 NE 39thSt,Seattle, WA 810s, 249100
PSTAT. PSTAT Inc., P.O. Box AH, Princeton, NJ 08542. (60959240100,
RATS. VAR Econometrics, P.O, Box 18! 8, Evanston, IL 60204-1818. (3 12}-864-
8772. ;SAS/STAT.SASInstitute Ine.,
(919)-467-8000. vn ascites
SCA SYSTEM.Scientific Computing Associates,Suite 106, Liste,iL 60532. (312)-960-1698.
624 Science Center,
1440 Sepulveda Blvd, Suite 316, Los
zg. Oxford University Press, Walton Street,
Suite 33, Lexington, MA 02573,
Stern Graduate School of Business. NewYork Unit
y York, NY 10006.
ae 16250 196th Place SE. Kent, WA 98042. (206)-631-
NewYork Univer
P.O. Box 8000, SAS Circle. Cary. NC 27511-8000.
Lincoln Center, 4513 Lincoln Ave.,
18 Computers and the Practice of Econometrics
SHAZAM. Kenneth J, White, Deparment of ies, Unik . White, Econo: i itissonenubs, Vancouver, BC VET 1¥2 Canad, coop236-5062BetsEC. The Sorites Group Inc., P.O. Box 2939, 81 1 Formba, VA 22152, (703)-569-1400. * $138 Old Keene Mill Roud,/PC-+. SPSS Inc., 444 N. Michigan Ave., Chicago, IL.SST boty SS hi ‘
|
Ave, 80,
I
L
60611. (312)-329.3600.aseenRivers Research, 1510 Ontario Ave., Pasadena, CA 91103. (818}377-
STATA. Computing Resource Center, 10801 Nati; ional Bivd.. 3rd Firao20064800)STATAPC:in California, (213-470-4341.. STSC Inc.. 2115 E. Jefferson St... Rockvilrein Maryland, (301)984-5123, aieMD 20852 (800).592-‘ATPRO.Penton Software ing. 420 Lexington Avenue, Suite 2 vow10017. (800}-221-3414; in New York. (212)-878-9600.ws NW York. NYSYSTAT.Systat Inc, 1800 Sherman Ave., Evanston, IL 60201. (312}-864-5670,Apple Macintosh
Ir, Les Angeles,
MATLAB.MathWorks, Inc., 20 N Main SL., Shes‘ ey . .. Sherborn, MA 01770. (617)-653-14Raevee toternational P.O. Box 61015. Palo Alto, CA 94306. wipeiere ‘conometrics, P.O. Box 1818, Evanston. IL 60204-1818, (312)-864-SHAZAM. Kenneth J, White, De i iversil
K . > Department of Economics, Ui 7 itiColumbia, Vancouver, BC V6T 1¥2 Canada. (604)228-5062. of BatsnSTATVIEW 512+. Brain Pow .10) ot? +: Brain Power, inc. Suite 250, 24009 Ventura Blvd... Calabasas, CA
CHAPTER NOTES
|. From remarks attributed to Thomas J. W: ft; . Watson, quoted from Chris MorganDavia Langford, Facts and Fallacies, Exeter, England: Webb & Bower,108e2. From David H. Abt in an interview wit1H. ith Kenneth Olson. QuoteCerf and Victor Navasky, The Experts Speak: The Definitive4, Aitheritarive Misinformation, New York: Pantheon Books. 1984, p- 409a qimenican b ssman, writer,and Printer, lived from 1856 to 1915. Quote taken‘ m Martin H. Manser, The Chambers Book ofBusiness Quotations, Edinburgh;4 MER Chambers, Lid 1987, p, 42. . en|. Quote taken from Minutes of Discussion ofPa ~Multivaria
pers on “Muiltiva iCeExperimental Data and Related Problems of Matri: ComputationTe‘owles Commission, August 23-24, 1946, Ithaca, New York ’5 dore Ww. Anderson and Tjalling Koopmans,p.8. ors authored by Theo. For3beoverview, sec Lawrence R. Klein, “The History of Computation inoe hetrics,” remarks preparedat the Institute for Advanced Studyat Princetonniversity honoring John von Neumann, undated and unpublishedS Fora description ofthis machine, see Guy H. Orcutt, "A New Regression Ana-lyser,” Journal the Roy , ‘ict Ser,pp. S70, tal ofthe Roval Statistical Suciety, Vol, 111, Series A, Part}, 1948,
from ChristopherCompendiunof
47,
18,
~ From JAC.
. Incidentally, mathematical operations on the EBM-
Notes 19
. Guy H. Orcutt, “From Engineering to Microsimulation,” an autobiographicalreflection,to be published in 1990 in the Journal ofEconontic Behavior and Orga-nization, along with a collection of papers presented in connection with a confer-ence in honor of Orcutt sponsored by the Social Systems Research Institute andthe Department of Economicsat the University of Wisconsin. May 1988.
. Ibid.
. Results from these computations were published in Guy Orcutt, “A StudyoftheAutoregressive Nature ofthe TimeSeries Used for Tinbergen’s Model of the Eco-homie System of the United States, 1919-1932," Journal of the Royal Statistical
Society, Vol. 10, No. 1. Series B (Mcthodologicat), 1948, pp. 1-45; and Orcutt
and S.F. James, “Testing the Significance ofCorrelation between Time Series,Biometrika, Vol. 25, Parts IRt & IV, December 1948.
. Sce Hendrik $. Houthakker, “SomeCalculations on Electricity Consumptionin
Great Britain,” Journal ofthe Royal Statistical Society, Series A, No. 114, PartTH, F951, pp, 351-371,
- For a discussion of the EDSAC computer and its development, sce Maurice V.Wilkes, Memoirs ofa Computer Pioneer, Cambridge. MA: The MIT Press, 1985,especially Chapter 13, pp. 127-142.
. Brown, H.S. Houthakker, and S.J. Prais, “Electronic Computa
ionin Economic Statistics," Journal ofthe American Statistical Association, 48:263,September1953, p. 423.
. For a well-written historical overview of computers farge and small, see StanAugarten,Bit byBiFields, 1984; histoHisturyofEconometric
in HMustrated HistoryofComputers, New York: Ticknor &issues in econometrics are addressed by Roy J. Epstein, 4Amsterdam: North-Holland, 1987.
. the IBM-PC AT, or theIBM PS/2 can-be speeded up considerably by installing an Intel 8087, 80287, or80387 mathematical coprocessor, respectively. Some statistical regression pro-grams require such a mathematical coprocessor. Check your econometrics soft-ware documentation to determine whether your machine is properlyconfigured.
. Numerous booksare available to help you learn DOS. One ofthe bestis that byPeter Norton, A/S-DOS and PC-DOS User's Guide, Bowie, MD: Robert J. BradyCompanyfor Prentice-Hall, 1984. Also see Peter Norton, Programmer's Guide tothe IBMPC, Bellevue, WA: MicroSoft Press, 1985.Forfurther information, contact PC Magazine, One Park Avenue, New York, NY10016, PC Week, P.O. Box 5970, Cherry Hill, NJ 08034 (609-428-5000); fifoWorld, 1060 Marsh Road, Suite C-200, Menlo Park, CA 94025 (415-328-4602):MacWorld, PCW Communications, Inc., 501 Second St, San Francisco, CA94107; MacUser, P.O, Box 56986, Boulder, CO 80321-2461 (800-627-2247), andAfac Week, 525 BrannanSt, San Francisco, CA 94107 (415-882-7370).The AmericanStatistical Association, 1429 Duke St.. Alexandria, VA 22314.For further information, write to CHANCE,Springer-Verlag New York, Inc..Journal Fulfillment Services, 44 Hartz Way, Secaucus, NJ 07096-249t.
. Published by Marcel Dekker, Inc., 270 Madison Avenue, New York, NY 0016(212-696-9000).
top related