Cognitive abilities and inflation expectations

by Francesco D‘Acunto, Daniel Hoang, Maritta Paloviita and Michael Weber

No. 126 | JANUARY 2019

WORKING PAPER SERIES IN ECONOMICS

KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft econpapers.wiwi.kit.edu

Impressum

Karlsruher Institut für Technologie (KIT)

Fakultät für Wirtschaftswissenschaften

Institut für Volkswirtschaftslehre (ECON)

Kaiserstraße 12

76131 Karlsruhe

KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft

Working Paper Series in Economics

No. 126, January 2019

ISSN 2190-9806

econpapers.wiwi.kit.edu

Cognitive Abilities and Inflation Expectations

By Francesco D’Acunto, Daniel Hoang, Maritta Paloviita, and Michael Weber∗

Over the last few years, interest has re-vived among economists in understandinghow households form and update their ex-pectations, as well as the determinants ofthe cross-sectional variation in economicexpectations across households (for a dis-cussion, see Gennaioli and Shleifer (2018);D’Acunto, Prokopczuk and Weber (2019);D’Acunto et al. (2018c)). Inflation expec-tations have been at the center stage of thisstrand of research, because of their rele-vance as a policy tool, especially when in-terest rates are low (Coibion et al. (2018)).

Forming and updating expectationsabout future macroeconomic variables suchas inflation requires the use of cognitiveresources at several stages, ranging fromcollecting information about the prevailinginflation rates to forecasting future poten-tial states of the world and their likelihood.Cognitive abilities thus appear as a naturalpotential determinant of the cross-sectionalvariation in inflation expectations.

Indeed, D’Acunto et al. (2018b) docu-ment in a representative population thatthe variation in cognitive abilities across in-dividuals is an important determinant of

∗ Carroll School of Management, Boston College,

Chestnut Hill, MA, USA. e-Mail: dacuntof@bc.edu.

Hoang: Department for Finance and Banking, Karl-sruhe Institute of Technology, Karlsruhe, B-W, Ger-

many. e-Mail: daniel.hoang@kit.edu. Palovi-ita: Bank of Finland, Helsinki, Finland. e-Mail:Maritta.Paloviita@bof.fi. Weber: Booth School of Busi-

ness, University of Chicago, Chicago, IL, USA and

NBER. This research was conducted with restricted ac-cess to data from the Finnish Armed Forces and Statis-

tics Finland. The views expressed here are those of theauthors and do not necessarily reflect the views of the

Bank of Finland, Finnish Armed Forces, or Statistics

Finland. We thank the project coordinator at Statis-tics Finland, Valtteri Valkonen, for his help with the

data and for very insightful comments. We gratefully

acknowledge financial support from the Deutsche Bun-desbank. Weber also gratefully acknowledges financial

support from the University of Chicago Booth School of

Business, the Fama Research Fund at the University ofChicago Booth School of Business, and the Fama-Miller

Center.

the variation in their macroeconomic ex-pectations. Individuals with mid-to-low IQlevels have absolute forecast errors for 12-month-ahead inflation rates 2.5 times aslarge as individuals with high IQ levels, andabout 4 times as large as the average re-alized inflation rate throughout the sam-ple period. D’Acunto et al. (2018b) furtherdocument mid-to-low IQ individuals areabout 4 times more likely to provide roundnumbers when asked about their inflationforecasts – which might reflect higher un-certainty in the forecasts – and are twice aslikely to report implausible values for theiraverage forecast.

In a related paper, D’Acunto et al.(2018a) further show cognitive abilities arean important friction to the effectivenessof economic policies that aim to affect theoverall economy through managing house-holds’ expectations (D’Acunto, Hoang andWeber (2016) discuss unconventional fiscalpolicy as one example). The facts that asubstantial part of a representative popu-lation of economic agents (i) barely formplausible inflation expectations and (ii) donot react to changes in their inflation expec-tations when making plans about consump-tion and saving choices drive this result.

A direct consequence of policies that aimto stimulate consumption expenditure viamanaging inflation expectations is that thepolicy measure might be less effective thana representative-agent model implies. Anindirect, unintended consequence of policiesthat aim to affect behavior via managingexpectations is, instead, that such policiesmight imply a redistribution of resourcesfrom the mid-to-low parts of the distribu-tion by cognitive ability to the higher endof the distribution.

I. IQ: Subcomponents

The research described so far has treatedcognitive abilities as a catch-all concept.

1

2

Cognition though is a complex human char-acteristic and includes several facets thatare not necessarily highly correlated. Forinstance, individuals who have strong quan-titative cognitive skills need not have alsostrong verbal or visuospatial skills.

At the same time, one might expect thatall these three facets of cognition are im-portant in the determination of expecta-tions regarding future economic variables.Quantitative skills are crucial for individu-als to map changes in price levels into infla-tion rates. Visuospatial skills are importantbecause they allow individuals to abstractfrom their personal situation and form plau-sible scenarios about future general infla-tion and other macroeconomic variables aswell as attach plausible probabilities tothese future states of the world. Ver-bal skills might matter because individualsneed to obtain information about current,past, and potentially future states of theworld through sources such as newspaperarticles, television, or family and friends.

Understanding which types of cognitiveabilities might matter to explain the ob-served differences in expectations is still awidely open question. Answering this ques-tion is important to assess the viability ofpotential interventions that aim to increasethe effectiveness of economic policies.

Because short-term interventions are un-likely to affect the cognitive abilities of eco-nomic agents, policymakers need to under-stand the specific skills agents lack, in or-der to design policies that could effectivelysubstitute for such skills. For instance,if agents cannot process the informationabout economic variables from newspaperarticles or official central-bank statements,policymakers might invest in communica-tion strategies that aim to make such con-cepts easy to grasp for the broader popu-lation (Coibion, Gorodnichenko and Weber(2019)). Alternatively, information mightbe provided to economic agents in a vividfashion that does not require high cognitionor knowledge about basic economic con-cepts but relies on agents’ comparison oftheir conditions with those of their peers(D’Acunto, Rossi and Weber (2019)). If,instead, information treatments were inef-

fective or required substantial investmentin financial education, cheap advice mech-anisms such as robo-advising tools indi-viduals could easily access through theirphones might improve the quality of low-IQ households’ expectations and choice(D’Acunto, Prabhala and Rossi (forthcom-ing) and D’Acunto et al. (2019)).

II. Data and Results

In this paper, we propose a first step tounderstand which types of cognitive abili-ties might matter more or less to explainthe heterogeneity in inflation expectationsacross individuals.

We build on the empirical setting ofD’Acunto et al. (2018b) and D’Acunto et al.(2018a), who merge administrative dataon cognitive ability test scores from theFinnish Armed Forces (FAF) for a repre-sentative sample of men with unique infor-mation on their inflation expectations fromStatistics Finland.1 Grinblatt, Keloharjuand Linnainmaa (2011) and Grinblatt et al.(2015) use the data on cognitive abilities infinance research and discuss in detail the in-centives test takers have to put effect intoanswering the questions.

For this paper, we exploit the fact thatthe 120 questions from the FAF test cog-nitive abilities across three types, namely,quantitative, verbal, and visuospatial abili-ties. FAF aggregate the scores within eachcategory and standardizes them within co-horts of test takers so that the IQ rank-ings follow a stanine distribution. Stanine(STAndard NINE) is a method of scalingtest scores on a nine-point standard scalewith a mean of five and a standard devi-ation of two. The respondents with thelowest 4% of test scores are at least 1.75standard deviations from the mean and areassigned a standardized IQ score of 1, andthe 4% with the highest test scores are as-signed a standardized score of 9. We thusobserve three scores per individual between1 and 9 based on the relative performancein each part of the test.

1D’Acunto et al. (2018b) and D’Acunto et al. (2018a)contain detailed descriptions of these data.

3

The verbal part of the test asks individu-als to compare words and pairs of words andfind synonyms and antonyms. In the arith-metic part, individuals have to solve simplearithmetic operations, solve verbal prob-lems, or compare pairs of numbers. Thevisuospatial part is similar to the Raven’sprogressive matrices test and requires indi-viduals to identify a missing item that com-pletes a pattern.

One might be concerned that the threescores we obtain are highly correlatedwithin individuals and that the differentportions of the test are ultimately unableto capture different sources of variation incognitive abilities across individuals. Infact, we find the point estimates for thepairwise correlations of the three scores inthe individual-level data range from 0.56 to0.66. These statistics suggest that even if,indeed, a positive correlation exists betweeneach pair of scores, we still detect enoughvariation across the scores of each individ-ual to make our test meaningful.

We merge this individual-level cognitive-ability information with the micro-data un-derlying the Consumer Survey of StatisticsFinland, in which respondents report theirnumerical 12-month-ahead forecast for in-flation on top of several other macro and in-dividual economic expectations and a largeset of demographic characteristics. Everymonth, the survey asks a representative re-peated cross section of approximately 1,500Finnish individuals questions about generaland personal economic conditions, infla-tion expectations, and willingness to spendon consumption goods. Statistics Finlandalso collects additional information throughsupplementary questions about households’plans to save and borrow. The sample pe-riod is from January 2001 to March 2015.We use the numerical inflation forecasts atthe individual level to construct our mainoutcome of interest – the absolute forecasterror for 12-month-ahead inflation. We de-fine this variable as the absolute value ofthe difference between the 12-month-aheadinflation forecast of individual i in a givenmonth and the ex-post realized inflationrate observed in Finland 12 months later.We then compute the average absolute fore-

cast error across individuals in the same binby arithmetic, verbal, and visuospatial cog-nitive abilities.2

Figure 1 plots the average absolute fore-cast error by bin of cognitive abilities or-dered from the lowest level of cognitive abil-ities (IQ=1) to the highest level of cognitiveabilities (IQ=9). Panel A refers to arith-metic abilities, panel B to verbal abilities,and panel C to visuospatial abilities.

Note we only have one absolute forecasterror per individual. Hence, to the extentwe observe different patterns in average ab-solute forecast errors across different sub-categories of IQ, we conclude that individu-als perform differently in different subpartsof the IQ test.

The three panels of Figure 1 beget afew comments. First of all, across allthree types of cognitive abilities, we detecta monotonically decreasing association be-tween IQ levels and average absolute fore-cast errors. This fact suggests none of thetypes of cognitive abilities behaves differ-ently from the aggregate score for overallcognitive abilities D’Acunto et al. (2018b)discuss in earlier research.

The second notable feature of Figure 1 isthe three scatter plots have different slopes,especially for the levels of cognitive abili-ties up to the median bins (IQ=5). Specifi-cally, the curve is steepest at lower values byarithmetic IQ followed by verbal IQ and isflattest at lower values by visuospatial IQ.One way to interpret this fact is that thevariation in visuospatial cognitive abilitiesacross individuals is less relevant in explain-ing the cross section of inflation forecastsrelative to the variation in arithmetic andverbal IQ.

A third relevant result – not depicted inFigure 1 – relates to statistical inference.We can reject the null hypothesis that anyof the forecast errors in any bin are equalto zero, as well as the null hypothesis that

2Because we compute the absolute value, we treatdeviations from the realized inflation rate in either direc-

tion identically. Our results are qualitatively unchanged

if we repeat the whole analysis computing the averageforecast error within each bin and hence allowing for

positive and negative forecast errors across individualswithin the same IQ bin to wash away.

4

Figure 1. Mean Absolute Forecast Error for Inflation by Subcomponenents of IQ

Panel A. IQ Arithmetic Panel B. IQ Verbal

23

45

Mea

n Ab

solu

te F

orec

ast E

rror

0 2 4 6 8 10Normalized IQ Arithmetic

23

45

Mea

n Ab

solu

te F

orec

ast E

rror

0 2 4 6 8 10Normalized IQ Verbal

Panel C. IQ Visuospatial

23

45

Mea

n Ab

solu

te F

orec

ast E

rror

0 2 8 104 6 Normalized IQ Visuospatial

This figure plots the average absolute forecast error for inflation across IQ levels of Finnish men. Forecast error

is the difference between the numerical forecast for 12-month-ahead inflation and ex-post realized inflation. IQ

is the standardized test score from the Finnish Armed Forces that obtains integer values between 1 and 9.

Panel A reports results for arithmetic IQ, panel B reports results for verbal IQ, and panel C reports results for

visuospatial IQ. The sample period is from January 2001 to March 2015.

the averages across any adjacent bins acrosseach of the three sorting schemes are equalfor most bins. This result suggests thateven though the relevance of different typesof cognitive abilities in explaining the crosssection of forecast errors for inflation mightbe higher or lower, higher scores in anyof the three components of IQ are system-atically associated with lower forecast er-rors on average, and hence each componentmight matter.

Based on our data, we conclude thatarithmetic, verbal, and visuospatial cogni-tive abilities are all likely to be relevant inexplaining the cross section of inflation ex-pectations in a representative population ofmen, even though arithmetic cognitive abil-ities seem to be most relevant, especially forthose scoring lowest.

III. Concluding Remarks

Large heterogeneity exists in how indi-viduals form, update, and act upon theirexpectations, which previous research re-lates to individual-level cognitive abilities.We show that all three subcomponents ofcognitive abilities we observe – arithmetic,verbal, and visuospatial – matter for theforecast accuracy for inflation, with the ef-fect of arithmetic cognitive abilities beingstrongest.

The results could suggest policymak-ers should design targeted communicationstrategies for different subpopulations. Forinstance, short and vivid messages suchtweets might be more successful in reachingthe lowest part of the population by cog-nitive abilities, whereas more detailed and

COGNITIVE ABILITIES AND INFLATION EXPECTATIONS 5

technical reports could still be relevant toshape the expectations and choices of high-IQ individuals.

Moreover, policymakers might proposesimple and salient policy measures to en-sure large parts of the targeted populationreact to the policy and to alleviate con-cerns about unintended consequences, suchas a redistribution from lower parts of theIQ distribution to higher parts (D’Acunto,Hoang and Weber (2018)).

REFERENCES

Coibion, Olivier, Yuriy Gorod-nichenko, Saten Kumar, andMathieu Pedemonte. 2018. “Infla-tion Expectations as a Policy Tool?”National Bureau of Economic Research.

Coibion, Olivier, Yuryi Gorod-nichenko, and Michael Weber. 2019.“Monetary Policy Communication andInflation Expectations.” Working Paper.

D’Acunto, F, Alberto Rossi, andMichael Weber. 2019. “CrowdsourcingFinancial Information to Change Spend-ing Behavior.” Working Paper.

D’Acunto, F, N Prabhala, andAlberto Rossi. forthcoming. “ThePromises and Pitfalls of Robo-Advising.”Review of Financial Studies.

D’Acunto, Francesco, Daniel Hoang,and Michael Weber. 2016. “The effectof unconventional fiscal policy on con-sumption expenditure.” National Bureauof Economic Research.

D’Acunto, Francesco, Daniel Hoang,and Michael Weber. 2018. “Unconven-tional Fiscal Policy.” AEA Papers andProceedings, 108: 519–23.

D’Acunto, Francesco, Daniel Hoang,Maritta Paloviita, and Michael We-ber. 2018a. “Human Frictions to theTransmission of Economic Policy.” Work-ing Paper.

D’Acunto, Francesco, Daniel Hoang,Maritta Paloviita, and Michael We-ber. 2018b. “IQ, Expectations, andChoice.” Working Paper.

D’Acunto, Francesco, MarcelProkopczuk, and Michael We-ber. 2019. “Historical Antisemitism,Ethnic Specialization, and FinancialDevelopment.” Review of EconomicStudies.

D’Acunto, F, Thomas Rauter,Christop Scheuch, and MichaelWeber. 2019. “Mobile Overdraft Fa-cilities and Consumption Behavior:Evidence from FinTech.” WorkingPaper.

D’Acunto, F, U Malmendier, J Os-pina, and M Weber. 2018c. “SalientPrice Changes, Inflation Expectations,and Household Behavior.” Working Pa-per.

Gennaioli, N, and A Shleifer. 2018. “ACrisis of Beliefs – Investor Psychologyand Financial Fragility.” Princeton Uni-versity Press.

Grinblatt, Mark, Matti Keloharju,and Juhani Linnainmaa. 2011. “IQand Stock Market Participation.” TheJournal of Finance, 66(6): 2121–2164.

Grinblatt, Mark, Seppo Ikaheimo,Matti Keloharju, and SamuliKnupfer. 2015. “IQ and MutualFund Choice.” Management Science,62(4): 924–944.

No. 126

No. 125

No. 124

No. 123

No. 122

No. 121

No. 120

No. 119

No. 118

No. 117

No. 116

Francesco D‘Acunto, Daniel Hoang, Maritta Paloviita and Michael Weber:

Cognitive abilities and inflation expectations, January 2019

Rebekka Buse, Melanie Schienle and Jörg Urban: Effectiveness of policy

and regulation in European sovereign credit risk markets - A network

analysis, January 2019

Chong Liang and Melanie Schienle: Determination of vector error correc-

tion models in high dimensions, January 2019

Rebekka Buse and Melanie Schienle: Measuring connectedness of euro

area sovereign risk, January 2019

Carsten Bormann and Melanie Schienle: Detecting structural differences

in tail dependence of financial time series, January 2019

Christian Conrad and Melanie Schienle: Testing for an omitted multiplica-

tive long-term component in GARCH models, January 2019

Marta Serra-Garcia and Nora Szech: The (in)elasticity of moral ignorance,

December 2018

Thomas Mariotti, Nikolaus Schweizer, Nora Szech and Jonas von Wangen-

heim: Information nudges and self-control, November 2018

Andranik S. Tangian: Methodological notes on composite indicators for

monitoring working conditions, October 2018

Andranik S. Tangian: Testing the improved third vote during the 2018

election of the Karlsruhe Institute of Technology student parliament,

September 2018

Yuri Golubev and Mher Safarian: On robust stopping times for detecting

changes in distribution, May 2018

recent issues

Working Paper Series in Economics

The responsibility for the contents of the working papers rests with the author, not the Institute. Since working papers

are of a preliminary nature, it may be useful to contact the author of a particular working paper about results or ca-

veats before referring to, or quoting, a paper. Any comments on working papers should be sent directly to the author.