Cognitive Economics

1
Cognitive Economics

• Miles Kimball
• University of Michigan
• Presentation at Osaka University

2
Definition of Cognitive Economics

• The Economics of What is in Peoples Minds

3
Named by Analogy to Cognitive Psychology

• Cognitive Psychology the area of psychology
  that examines internal mental processes such as
  problem solving, memory and language.
• Cognitive Psychology was a departure from
  Behaviorism--the idea that only outward behavior
  was a legitimate subject of study.

4
How is Cognitive Economics Different from
Behavioral or Psychological Economics?

• Cognitive economics is narrower.
• Much of cognitive economics is inspired by the
  internal dynamic of economics rather than by
  psychology.
• Cognitive economics is a field of study, not a
  school of thought.

5
Areas of Economics by Distinctive Data Type

• Standard Economics (including Mindless
  Psychological Economics a la Gul and
  Pesendorfer) actual market choices only.
• Experimental Economics choices in artificial
  situations but with real stakes.
• Neuroeconomics FMRI, saccades, skin conductance,
  
• Bioeconomics genes, hormones
• Cognitive Economics mental contents (based on
  tests and self-reports) and hypothetical choices.

6
Four Themes of Cognitive Economics

• New Types of Data
• Heterogeneity
• Finite and Scarce Cognition
• Welfare Economics Revisited

7
1. Innovative Survey Data

• fluid intelligence data
• crystallized intelligence data
• happiness data
• survey measures of expectations
• survey measures of preferences

8
2. Individual Heterogeneity

• heterogeneous expectations
• heterogeneous preferences
• heterogeneous emotional reactions
• heterogeneous views on how the world works (folk
  theories)

9
3. Finite and Scarce Cognition

• Finite cognitionthe reality that people are not
  infinitely intelligent.
• Scarce cognitionsome decisions required by our
  modern environmentat work and in private
  livescan require more intelligence for
  full-scale optimization than an individual has

10
4. Welfare Economics Revisited

• Scarce cognition means that people sometimes make
  mistakes. Thus, one can no longer use naïve
  revealed preference for welfare economics.
• Kimball and Willis, in Utility and Happiness,
  argue that happiness data is not a magical
  touchstone for diagnosing mistakes.
• Then, what does count as evidence of mistakes?
• Internal inconsistencies, such as lack of
  transitivity? But which choice then deserves
  respect?
• Regret?
• Modification of choices after experience?
• Differences in choices between those with high
  cognitive ability and those with low cognitive
  ability?
• e.g., Dan Benjamin and Jesse Shapiro show that
  low IQ students had more low-stakes risk aversion
  and short-horizon impatience

11
Some Research Questions in Cognitive Economics

• Seek to make innovations in economic theory and
  measurement to address
• What are peoples limitations in knowledge,
  memory, reasoning, calculation?
• What is the role of emotion, social context,
  conscious vs. unconscious judgments and
  decisions?
• What is the role of health as determinant,
  outcome and context for economic activity,
  decisions and well being?
• What is connection between economic welfare and
  measures of well being?
• Etc.

12
New Types of DataMeasurement of Cognition in
the HRS

• HRS has included cognitive measures from the
  outset, but mostly focused on memory in order to
  trace cognitive decline.
• Re-engineering HRS cognitive measures
• Led by Jack McArdle, a cognitive psychologist and
  HRS co-PI, we have begun a project to
  re-engineer our cognitive measures in order to
  improve our understanding of the determinants of
  decision-making about retirement, savings and
  health and their implications for the well-being
  of older Americans

13
New Types of Data Measurement of Cognition in
the HRS (cont.)

• Separate HRS-Cognition Study
• Begins with a separate sample of 1200 persons age
  50 who will receive about three hours of
  cognitive testing of their fluid and crystallized
  intelligence plus parts of the HRS questionnaire
  on demographics, health and cognition
• Followed a month later by administration of an
  internet or mail survey of questions designed by
  economists on financial literacy, ability to
  compound-discount, hypothetical decisions about
  portfolio choice, long term care
• Finally, telephone follow-up with HRS cognition
  items and subjective probability questions
• Analysis of data will guide re-engineering of
  cognitive items for HRS-2010

14
New Types of Data Survey Measures of Expectations

• What is the mapping between probability beliefs
  in peoples minds and the decisions they make?

(Robert Willis, Charles Manski, Mike Hurd, Jeff
Dominitz, Adeline Delavande)
15
Direct Measurement of Subjective Probability
Beliefs in HRS
Probability questions use a format pioneered by
Tom Juster and Chuck Manski
(Manski, 2004)
HRS Survival Probability Question Using a
number from 0 to 100, what do you think are the
chances that you will live to be at least target
age X? X 80 for persons 50 to 70 and
increases to 85, 90, 95, 100 for each five year
increase in age

16
Two Key Findings From Previous Research on HRS
Probability Questions

• 1. On average, probabilities make sense
• Survival probabilities conform to life tables and
  are
• predictive of actual mortality
• (Hurd and McGarry 1995,
  2002 Sloan, et. al., 2001 )
• Bequest probabilities behave sensibly
• (Smith 1999), Perry (2006)
• Retirement incentives can be analyzed using
  expectational data
• (Chan and Stevens, 2003)
• People can predict nursing home entry
• (Finkelstein and McGarry,
  2006)
• Early Social Security Claiming Depends on
  Survival Probability
• (Delevande, Perry and Willis,
  2006) , (Coile, et. al., 2002)
• 2. Individual probabilities are very noisy with
  heaping on focal values of "0", "50-50" and "100
  
• (Hurd, McFadden and Gan, 1998)

17
10 Year Mortality Rate vs. Subjective Survival
Probability to Age 75
Odds Ratio of Death by t10
Subjective Survival Probability at Time t
Source Mortality Computations from HRS-2002 by
David Weir
18
10 Year Mortality Rate vs. Subjective Survival
Probability to Age 75
Strongest relationship between subjective and
objective risks for people with low subjective
survival beliefs
Odds Ratio of Death by t10
Subjective Survival Probability at Time t
Source Mortality Computations from HRS-2002 by
David Weir
19
. Histograms of Responses to Probability
Questions in the HRS
A. General Events Social Security less
generous Double digit inflation B. Events
with Personal Information Survival to 75
Income increase faster than inflation C.
Events with Personal Control Leave
inheritance Work at age 62
20
Are Benefits of Greater Individual Choice
Influenced by Quality of Probabilistic Thinking?

• Trend of increasing scope for individual choice
  in public and private policy, especially as it
  affects those planning for retirement or already
  retired
• Private sector shift from defined benefit to
  defined contribution pension plans
• Proposals for individual accounts in Social
  Security
• Choice of when/whether to annuitize
• Choice of medical insurance plans and providers
  by employers and by Medicare, new Medicare
  Prescription Drug program
• Economists generally view increased choice as a
  good thing, but
• General public wonders whether people will make
  wise use of choice
• Decisions faced by older individuals balancing
  risks and benefits of alternative financial and
  health care choices are genuinely difficult

21
Quality of Probabilistic Thinking and Uncertainty
Aversion

• Lillard and Willis (2001) began to look at the
  pattern of responses to probability questions as
  indicators of the degree to which they indicate
  peoples capacity to think clearly about
  subjective probability beliefs
• We explored the idea that focal answers of 0,
  50 and 100 were perhaps indicators of less
  coherent or well-formed beliefs than non-focal
  (or exact) answers.

22
Index of Focal Responses

• We treated the probability questions like a
  psychological battery and constructed an
  empirical propensity to give focal answers of
  0, 50 or 100

We found that people who had a lower
propensity to give focal answers tended to have
higher wealth, had riskier portfolios, and
achieved higher rates of return, controlling for
conventional economic and demographic variables
23
Uncertainty Aversion

• We hypothesized that people who give more focal
  answers are more uncertain about the true value
  of probabilities
• If the uncertainty is about a repeated risk, such
  as the return to a stock portfolio held over
  time, we show that people who have more imprecise
  probability beliefs (i.e. are more uncertain
  about the true probability) will behave more
  risk aversely

24
Some Further Results on Subjective Probabilities

• There is optimism factor common across all
  probability questions which is correlated with
  stock-holding and associated with being healthy,
  wealthy and wise
• Kezdi and Willis (2003)
• HRS has added direct questions on stock returns
• stockholding is related to probability beliefs
• Kezdi and Willis (2003) and Dominitz and Manski
  (2006)
• most people do not believe that stocks have
  positive returns, despite the equity premium that
  economists know about
• Persons who provide more precise probability
  answers also exhibit less risk aversion on
  subjective risk aversion questions in the HRS,
  and they save a higher fraction of their full
  wealth.
• Sahm (2007), Pounder (2007)
• In 2006, HRS added questions to those who answer
  50 to see whether they mean equally probable
  or just uncertain. 75 indicate they are
  uncertain.

25
New Types of Data Survey Measures of
Preferences Based on Hypothetical Choices

• Examples
• Labor Supply Elasticities,
• Altruism,
• Social Rivalry,
• Risk Aversion,
• Elasticity of Intertemporal Substitution

26
Does Risk Tolerance Change?

• Claudia Sahm
• University of Michigan?Board of
• Governors

27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)
36
(No Transcript)
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
(No Transcript)
41
(No Transcript)
42
(No Transcript)
43
(No Transcript)
44
(No Transcript)
45
(No Transcript)
46
(No Transcript)
47
(No Transcript)
48
(No Transcript)
49
(No Transcript)
50
Measuring Time Preference and the Elasticity of
Intertemporal Substitution

• Miles S. Kimball, Claudia R. Sahm and Matthew D.
  Shapiro

September 6, 2006 Internet Project Meeting
51
Behavioral Model

• c is consumption,
• r is the real interest rate,
• s is the elasticity of intertemporal
  substitution, and
• ? is the subjective discount rate

52
Research Design
Estimate Parameters s, ?
53
Implementation

• Vary Interest Rate
• Vary cost of current consumption
• Vary length of time periods
• Measure Consumption Choice
• Choose among small set of paths
• Actively form a desired path
• Infer Preferences
• Summary statistics of responses
• Statistical model with response error

54
Previous Survey Measures

• HRS 1992 Module K, N 198
• Analyzed by Barsky, Kimball, Juster, and Shapiro
  (QJE 1997)
• HRS 1999 Mailout, N 1,210
• Similar content to part of Internet Survey

Questions explicitly vary the cost of current
consumption and offer a discrete choice over a
small set of consumption paths
55
MS Internet SurveyWave 2 (Fall 2004)
Use graphics on Internet to test other measures

• Version 1, N 350
• Vary cost of consumption
• Choose from set of pairs
• Version 2, N 155
• Vary cost of consumption
• Move bars to create pair
• Version 3, N 183
• Vary length of period
• Move bars to create pair

56
Series Introduction - Version 1 -

• Series includes four questions with varying
  interest rates

57
Introduction 0 Interest Rate

• Sequence r 0, 4.6, 9.2, 13.8 is random
• Introduction repeated for each interest rate

58
Patterns 0 Interest Rate

• Asked to choose two patterns
• Above screen (1 of 6) is identical to HRS Mail
  Out

59
Expansion Screen

• Follow-up if first choice on boundary (A or E)

• New feature on Internet

60
Randomize Pair C

• Choice C positive, zero, negative growth rate
• 3 values to the parameter
• New feature on Internet
• Top screen on mail out

61
Randomize Left-to-Right

• Growth rates increase or decrease left-to-right
• New feature on Internet
• Top screen on mail out

62
Randomize Shifts with Interest Rate

• Example with r 9.2
• Choice of (2750, 3900) moves from E to C to A
• 3 values to the parameter
• New feature on Internet
• Middle screen on mail out

63
Summary of Innovations in Internet Question
Series

• 18 different screen groups
• 6 different sequences of interest rates
• 11 discrete choices per question

Purpose of Innovations

• Encourage active choices
• Increase informative responses
• Isolate framing effects

64
Response Statistics

• Internet lower completion rate
• Internet fewer second choices
• Internet fewer non-informative responses

65
Consumption Growth at 0 Interest Rate

• Constant consumption is modal choice

66
Change in Consumption Growth as Interest Rate to
13.8 from 0

• Interest rates change consumption more on Internet

67
Change in Consumption Growth as Interest Rate
Increases - Internet

• Decrease in growth is a sign of survey response
  error

68
Estimates of Parameters

• Responses reveal low time preference and IES
• Median and modal values in both surveys equal 0

69
Effect of Changes in Choice Set
70
Estimates by Screen Group
71
More Graphical Questions- Version 2 -

• Move bars to select a consumption path

72
More Graphical Questions- Version 3 -

• Vary length of current and future periods

73
Extensions / Renewal

• Measure complementary parameters
• Diminishing marginal utility
• Labor supply elasticities
• Retirement elasticity

74
Four Themes of Cognitive Economics

1. New Types of Data
2. Heterogeneity
3. Finite and Scarce Cognition
4. Welfare Economics Revisited

75
Bounded Rationality vs. Scarce Cognition

• Same meaning, but bounded rationality seems a
  misnomer, since it is rational to recognize ones
  own cognitive limitations.
• Two obstacles have prevented Bounded
  Rationality from becoming part of the mainstream
• theoretical difficulties stemming from the
  importance of constrained optimization as a
  theoretical tool in economics
• paucity of data
• Scarce Cognition is meant to label a data-rich
  research agenda, using new theoretical tools.

76
The Reality of Finite Cognition

• Computers beat us at chess
• People dont get perfect scores on tests, even
  after they have studied the material
• For hundreds of years, we had no proof of
  Fermats last theorem

77
The Reality of Scarce Cognition

• Many people
• spend time and money learning math
• pay others with higher wage rates to do their
  taxes
• pay others to read law books for them
• pay for financial advice

78
Modeling Scarce Cognition is Hard The Infinite
Regress Problem (Conlisk)

• It is natural for economists to assume a cost of
  computation, just like any other costso why not
  more such models?
• Answer figuring out how hard to think about a
  problem is always a strictly harder problem than
  the original problem
• Need the solution to the original problem to
  calculate the benefit
• Need to know how to solve the problem to know how
  many computational steps it needs

79
Dodging the Infinite Regress Problem by Breaking
Taboos

• Ignoring computational costs at the outer level.
  (Maybe OK if the original problem is a repeated
  choice.)
• Using limited information transmission capacity
  as a metaphor for limited intelligence. (A
  thick skull.)
• Subhuman intelligence
• --agent-based modeling
• --rules of thumb (adaptive expectations, consume
  income, statistical models)
• Modeling folk theories ignorant of the maintained
  hypotheses

80
Modeling Unawareness Requires a Subjective State
Space Distinct from the True State Space

• (Dekel, Lipman and Rustichini)
• economic actor subjective state space
• analyst state space maintained as true

81
Two Levels of Theory

• Folk theory economic actors theory modeled in
  the subjective state space
• May look like an accounting framework in the
  sense of Herrnstein and Prelec in The Matching
  Law
• Metatheory the analysts theory which includes a
  description of the relevant folk theories.
• Preferences
• Technology
• Available Strategies
• Active Information Structure
• Folk Theories

82
Desirable Properties for a Model of a Folk Theory

• Accuracy in describing how people actually view
  the world
• Providing a clear prediction for how people will
  behave in various circumstances
• Representing clearly how people are confused and
  what they do understand.
• NOT REQUIRED deep logical consistency

83
An Example of Folk Physics

• Many people believe that if they swing a stone
  around on a string and let it go, then the stone
  will curve sideways in the direction they were
  swinging it around.
• Other than going up and down in the vertical
  direction, it actually goes straight once
  released.

84
An Example of Folk Finance

• It Misses These Ideas
• Link of diversification to the
• variance/covariance matrix
• 2. Diversification makes it safe
• enough to hold a lot of the
• risky asset
• 3. Role of human capital
• 4. Consumption as the ultimate
• objective

• This Folk Theory Models
• Three Ideas
• Mean return is good
• Risk is bad
• Diversification is good

I
85
(No Transcript)
86
(No Transcript)
87
(No Transcript)
88
(No Transcript)
89
(No Transcript)
90
(No Transcript)
91
(No Transcript)
92
(No Transcript)
93
Household Finance and Welfare Economics The
Possibility of Strong Normative Statements

• Fungibility money is money
• At the end of the day, only the total value of
  the portfolio matters, not the separate value of
  its constituent parts.
• Fungibility is a legal term OK to pay back a
  different piece of currency as long as the value
  is the same. (Not like a diamond ring)
• Very basic principle in economics fungibility of
  money is assumed in standard treatments of
  revealed preference
• Noneconomists do not always understand
  fungibility mental accounts

94
High and Low Savers? Circumstances, Patience, and
Cognition

• Laurie Pounder

95
Differences in Consumption (Saving) Rates Across
Households

• Circumstances?
• Such as income shocks differences in pensions
  (income replacement rate in retirement) income
  profiles etc.
• Or Types of Savers?
• By inherent characteristics such as
  preferences or ability/cognition

96
Getting Past Circumstances

• Circumstances create cross-household variation
  when measuring rates C/Income or C/NetWorth
• Difficult to isolate role of preferences or other
  inherent differences across households

97
The Right Rate Consumption and Full Wealth

• Lifecycle/PIH theory since Modigliani says
  consumption should depend on all current and
  future resources (including financial and human
  wealth.)
• Like a stock value of permanent income from today
  forward
• I call this PV of all resources
• Modigliani full wealth M

98
Data Allows Comparison Testing

• A credible estimate of M for older households,
  together with consumption, available for the
  first time in the HRS
• Compare observed propensity to consume C/M to
  neoclassical model of optimal consumption rates
• Use survey estimates of models factors that vary
  across households to indicate which factors play
  largest role in explaining observed variation in
  C/M

99
Neoclassical Model (Merton 1971)

• Mortality and rate of return the only sources of
  uncertainty

Subject to
Estimate mortality with Gompertz function ?age
?1 e(?2age)
100
Average Propensity to Consume
Infinite Horizon (no mortality)

• Implications
• C is proportional in M
• C/M depends only on preferences, stochastic
  return characteristics, and mortality.
• C/M does not depend directly on M, income
  profile, or outcome of past income shocks

101
Findings

• Survey estimates of model factors matter in
  expected direction
• Heterogeneity in observed C/M! Rich save more
  (lower propensity to consume)
• Inherent characteristics or types important in
  explaining C/M
• Within neoclassical/rational model must have
  heterogeneous time preference to explain C/M
• Looking outside standard model cognition
  planning matter to C/M

102
Model Factors
Dependent Variable ln(C/M) (1) (2)

Log Predicted C/M with variation by mortality only 0.013 (0.002)
Log Predicted C/M with variation by mortality and expected risky returns 0.017 (0.002)

N1842 R20.023 R20.027
103
Adding Survey Measures of Model Factors
Dependent Variable ln(C/M) with additional demographics with additional demographics with additional demographics

Subjective Life Expectancy Ratio -0.137 (0.044) -0.080 (0.043)
Probability of Bequest gt10k (Continuous) -0.002 (0.001)
Probability of Bequest gt100k (Continuous) -0.003 (0.001)
Risk Aversion Survey Measure -0.022 (0.007)

Model Prediction 1.33 (0.363) 1.04 (0.359) 1.45 (0.454)
Constant 3.69 -3.55 -5.77
R20.231 R20.287 R20.226
N1190 N1190 N894
104
Rich Save MoreC/M Varies by Income or Wealth
Level
105
Beyond the Neoclassical ModelAbilities
Cognition Planning

• Bounded cognition
• Propensity to plan
• Expectations formation
• (Lusardi, LillardWillis, CaplinLeahy)

106
Measures in HRS

• HRS asks questions on basic cognition (recall,
  counting, subtraction) plus planning horizon and
  subjective expectations
• Lillard Willis focal point answers
  precision of expectations formation related to
  financial decisions
• Measures matter such that lower cognition, less
  precision, and shorter planning horizons all
  imply higher propensities to consume

107
Cognition/Planning Predict C/M
Dependent Variable Residual of ln(C/M) after full regression

Long Financial Planning Horizon -0.070 (0.028)
Fraction of Precise Answers -0.088 (0.053)
High Word Recall -0.062 (0.030)
Counting Backwards -0.124 (0.048)
Hardest Subtraction Problem -0.049 (0.030)

N1645 R20.027
 
108
Further Evidence Loosely Related Preference
Covariates
Dependent Variable Residual of ln(C/M) after full regression

Ever Smoked 0.046 (0.028)
Reports would Spend all of hypothetical income increase 0.091 (0.053)
Reports would Save all of hypothetical income increase -0.055 (0.031)

N1645 R20.01
Dependent Variable Residual of ln(C/M) after full regression Dependent Variable Residual of ln(C/M) after full regression Dependent Variable Residual of ln(C/M) after full regression Dependent Variable Residual of ln(C/M) after full regression Dependent Variable Residual of ln(C/M) after full regression Dependent Variable Residual of ln(C/M) after full regression
Personality Questions
Seldom apprehensive about future 0.028 (0.022) 0.048 (0.023)
Strive for excellence -0.097 (0.024) -0.109 (0.029)
Clear set of goals and work toward them -0.026 (0.023) -0.006 (0.025)
Work hard to accomplish goals -0.063 (0.034) -0.011 (0.039)
N235 N235 N235 N235 N235 N235