Overconfidence: in psychology and finance

1
Overconfidence in psychology and finance

• Doct st. J. Michailova

2
Introduction

• Traditional view Homo Economicus
• Reality investor behavior is difficult to
  reconcile with rationality of predictions of
  standard finance models (Glaser et al., 2003).
• Explanation overconfidence

3
Theoretical Insights into the Phenomenon of
Overconfidence

• Profession. Gender Differences. Overconfidence
  in Children. Cultural impact
• Overconfidence found in students, members of the
  armed forces, CIA analysts, entrepreneurs,
  clinical psychologists, bankers, executives,
  managers, lawyers, and civil engineers.
• No differences between men and women in
  calibration. Men are more overconfident in their
  performance estimates.
• Overconfidence is already present in children.
• Cross-cultural differences in calibration.
  Overconfidence in general knowledge is stronger
  among Asian than among Western subject groups.

4
Confidence

• Is strongly related to the notion of certainty,
  and strength of the feeling of certainty is
  embodied in the corresponding level of
  confidence.
• Zakay and Tsal (1993) feeling of confidence
  accompany any type of mental activity.
• Degree of confidence can vary from zero to
  absolute level.
• Can be expressed
• -gtverbally - very, totally, absolutely, quite
• -gtnumerically- 100, 9 to 10, 50/50 etc.
• Confidence internal to a subject, feeling of
  certainty about some state of reality.

5
Degree of Confidence as a Subjective Probability

• Decisions are based on beliefs concerning the
  likelihood of uncertain events
• These beliefs are expressed in numerical form as
  subjective probabilities Subjective probabilities
  - the probabilities that people generate in their
  own minds to express their uncertainty about the
  possibility of the occurrence of various events
  or outcomes (Bar-Hillel, 2001).
• Subjectivist point of view a probability is a
  degree of belief in a proposition.
• Expresses a purely internal state there is no
  right, correct or objective probability
  residing somewhere in reality against which
  ones degree of belief can be compared.

6
Degree of Confidence as a Subjective Probability

• Adequacy of subjective probabilities to objective
  reality can be measured via a question with the
  verifiable as correct or non-correct answer, and
  a measure of the degree of confidence in that
  judgment.
• If accuracy does not match the level of
  confidence, then a judgmental bias exists.

7
Overconfidence

• Overconfidence concerns the fact that people
  overestimate how much they actually know when
  they are p percent sure that they have answered a
  question correctly or predicted correctly, they
  are in fact right on average less than p percent
  of the time (Bar-Hillel, 2001).
• Overconfidence arises from not knowing the limits
  of ones knowledge (Conger and Wolstein, 2004).
• Two types of overconfidence - specific and
  generic.
• gtSpecific overconfidence - overestimation of
  the probability of a specific hypothesis
  or a designated outcome gtGeneric
  overconfidence - overestimation of the likelihood
  of the hypothesis, considered to be the
  most probable by the individual.
• In this manner one draws distinction between the
  estimation of the likelihood that ones answer is
  correct and the likelihood of judgmental answer
  (Pulford, 1996).

8
Overconfidence

• Local and global confidences (May, 1887, 1988)
• gtLocal confidence - average confidence across
  questions
• gtGlobal confidence - confidence in the total
  number, or percentage, of questions
  individuals have answered correctly.
• There are several psychological findings that are
  often summarized under the concept of
  overconfidence in financial literature
  miscalibration, the better than average effect,
  illusion of control, and unrealistic optimism
  (Glaser and Weber, 2004) .

9
Miscalibration

• In studies of calibration, participants answer a
  series of multiple-choice questions and stipulate
  their confidence of being correct for each
  answer.
• Calibration is tested by comparing the percentage
  of questions that a participant has answered
  correctly with the participants average
  confidence in the answers to these questions.
• Individuals are considered to be well calibrated
  in their confidence in the correctness of their
  knowledge, if the following condition is
  satisfied over the long run, for all answers
  assigned by them a given probability of being
  correct (e.g., 0.8), the actual proportion
  correct (80) equals the probability assigned
  (Koriat, Lichtenstein, and Fischhoff, 1980).
• Of those responses made with confidence P, about
  P should be correct (Adams, 1957).

10
Miscalibration

• Most of the people are not well-calibrated and
  demonstrate miscalibration.
• Miscalibration - a systematic deviation from
  perfect calibration. It is defined as an
  unwarranted belief in the correctness of ones
  answer
• Usually for all questions the proportion of
  correct answers is lower than the assigned
  probability.
• Studies that analyze assessments of uncertain
  quantities using the fractile method usually find
  that peoples probability distributions are too
  tight.
• People are often asked to state a 90 percent
  confidence interval for several uncertain
  quantities.
• Findings the percentage of surprises, i.e. the
  percentage of true values that fall outside the
  confidence interval, is higher than 10 percent
  (Glaser and Weber, 2004).

11
Better than average effect

• Based on the fact, that people tend to exaggerate
  their talents.
• Taylor and Brown (1988) document in their survey
  that people have unrealistically positive views
  of the self, i.e. they think about themselves as
  possessing above the average abilities, compared
  to other people e.g. with regard to skills or
  positive personal traits.
• 82 of a group of students rank themselves among
  the 30 percent of drivers with the highest
  driving safety (Svenson,1981).

12
Illusion of control

• Illusion of control is embodied in the
  exaggeration of the degree to which one can
  control her fate. Subjects prone to the illusion
  of control, tend to underestimate the role of
  chance in human affairs and to misperceive games
  of chance as games of skill (Kahneman and Riepe,
  1998).
• An expectancy of a personal success probability
  inappropriately higher than the objective
  probability would warrant (Langer, 1975)

13
Unrealistic optimism

• Closely related to the illusion of control bias
  is the phenomenon of unrealistic optimism about
  future life events.
• Mentally healthy people tend to exhibit
  psychological biases that encourage optimism,
  collectively known as positive illusions
  (Johnson et al., 2006).
• Most peoples beliefs are biased in the direction
  of optimism (Kahneman and Riepe, 1998). Optimists
  also underestimate the likelihood of bad outcomes
  over which they have no control. Most
  undergraduates, for example, believe that they
  are less likely than their roommates to develop
  cancer or to have a heart attack before the age
  of fifty (Kahneman and Riepe, 1998).
• The question whether all these concepts
  (miscalibration, better than average effect,
  illusion of control and unrealistic optimism) are
  related is mainly unexplored.

14
Factors influencing degree of overconfidence in
the experimental settings

• Keasey and Watson (1989) identified four factors
  that have impact on the accuracy-confidence
  relationship the complexity of the task, amount
  of feedback given, motivation level of the
  subjects, and skill of the subjects.
• Hard-Easy Effect
• The hard-easy effect occurs when the degree of
  overconfidence increases with the difficulty of
  the questions, where the difficulty is measured
  by the percentage of correct answers.
• People are overconfident with general-knowledge
  items of moderate or extreme difficulty. As tasks
  get easier, overconfidence is reduced
  (Lichtenstein et al., 1982).

15
Factors influencing degree of overconfidence in
the experimental settings

• Is more information always better?
• Useless information tends to increase subjects
  confidence. However this had no impact on the
  improvement on accuracy.
• Oskamp (1965) found that often people do not
  realize uselessness of the information given to
  them by the experimenter, and their confidence
  increases without justification, whereas,
  accuracy doesnt.
• Peterson and Pitz (1986) theorized that when
  several pieces of useful information are given
  they reduce overconfidence.
• Calibration and motivation
• The effect of increasing the stakes in judgment
  tasks has been shown to have mixed results.
  Fischhoff (1982) found no improvement in
  judgment, but other researchers have found
  changes in both directions (Kruglanski and
  Freund, 1983).

16
Factors influencing degree of overconfidence in
the experimental settings

• Experience, Feedback and Calibration
• Even experts, who by definition know a lot about
  a specialized topic, are often unable to express
  precisely how much they do not know (Russo,
  Schoemaker, 1992).
• Russo and Schoemaker (1992) claim that (job)
  relevant experience to the confidence-quiz,
  reduces overconfidence.
• What concerns their professional experience,
  experts better know what they dont know (Russo
  and Schoemaker (1992).
• However, in the study by Heath and Tversky
  (1991), greater overconfidence was found for
  tasks which subjects considered they had more
  expertise in

17
Factors influencing degree of overconfidence in
the experimental settings

• Experience, Feedback and Calibration
• It is one of the big mistakes to equate
  experience and learning. Experience is
  inevitable learning is not. The failure of
  individuals to learn from the experience is the
  cause of persisting, despite experience,
  overconfidence (Ruso et al, 1992).
• In order to learn individuals need to get
  feedback about the accuracy of their answers. On
  time received feedback can reduce bias towards
  overconfidence in professionals (Russo et.al,
  1992) .

18
Measurement of Overconfidence

• There are two types of calibration assessment
  techniques used in the psychological experiments
  
• 1) making probability judgments about
  discrete propositions, and
• 2) the calibration of probability density
  functions assessed for uncertain numerical
  quantities.
• Measurement Of Calibration In Discreet
  PropositionsTasks
• Procedure there is a questionnaire offering no
  alternatives, one alternative, two, three or more
  alternatives.
• Assessor either provides his own answer (no
  alternatives) or selects from the given ones.
• Then she states the probability that this choice
  is correct.

19
Calibration curve

• Derivation procedure 1st step is collection of
  probability assessments for events whose correct
  answer are known 2nd step is grouping of similar
  assessments 3rd step computation of ratio of
  correct items in each constructed category 4th
  step is plotting of mean response of each
  category against ratio of correct items.
• Calibration is measured by how close points are
  to the identity line. Distribution of all points
  on the identity line would correspond to perfect
  calibration.
• Identity line - where confidence exactly equals
  accuracy.
• Overconfidence the proportions correct are less
  than the assessed probabilities, so that the
  calibration curve falls below the identity line
• Underconfidence the proportions correct are
  greater than the assessed probabilities and the
  calibration curve lies above the identity line.

Example of a calibration curve with an identity
line (Source Pulford, 1996).
20
Calibration Score

• The Brier (1950) score, or the probability score
  (PS), is a measure of goodness of a set of
  probability assessments. The Brier score is
  defined as the average deviation between
  predicted event probabilities and their real
  outcomes.
• r is the assessed probability of item i,
• c is the associated outcome. c1 if event occurs,
  zero otherwise.
• N is a number of forecasts of subject
• The smaller is the Brier score, the higher
  accuracy in predictions it represents.
• The range of PSB is the closed interval 0, 2
• Briers score doesnt discriminate between
  overconfidence and underconfidence.

21
Calibration score

• Bias Score
• To discriminate between over- and underconfidence
  the bias score is used. Bias score is calculated
  as the difference between the mean confidence
  level across all questions and the proportion of
  correct answers
• Bias score average confidence average
  correct
• A positive bias score represents overconfidence,
  and a negative bias score represents
  underconfidence.
• A bias score of zero indicates an accurately
  calibrated person.

22
Calibration of the PDF (The
fractile method)

• Uncertainty about the value of a continuous
  quantity (e.g. what is the shortest distance from
  England to Australia) may be expressed as
  probability density function across the possible
  values of that quantity (Lichtenstein et al.,
  1982).
• Subjects usually do not have to draw the entire
  function. In this method, the assessor states
  values of the uncertain quantity that are
  associated with a small number of predetermined
  fractiles of the distribution.
• For the median or 0.5 fractile, for example the
  assessor states a value of the quantity such that
  the true value is equally likely to fall above or
  below the stated value.
• The 0.01 fractile is a value such that there is
  only 1 chance in 100 that the true value is
  smaller than the stated value.

23
Calibration of the PDF (The
fractile method)

• There are two calibration measures for continuous
  items.
• Interquartile index is the percentage of items
  for which the true value falls inside the
  interquartile range (i.e., between the 0.25 and
  0.75 fractiles). Interquartile index of a
  perfectly calibrated person is 0.5.
• Surprise index is the percentage of true values
  that fall outside the most extreme fractiles
  assessed. Large surprise index shows the
  inability of the assessor state confidence bounds
  wide enough to include as much as possible of the
  true values this indicates overconfidence.

24
Ways to Reduce Overconfidence and Improve
Calibration/ Debiasing

• Psychologists have tried to train subjects to
  avoid judgmental errors, or in other words tried
  to debias them.
• Training assessors by giving them feedback about
  their calibration has shown mixed results.
• Adams and Adams (1958) have found modest
  improvement after five training sessions .
  Fishfoff (1982) reports some successful training
  exercises, mostly using large amounts of
  well-structured feedback.

25
Ways to Reduce Overconfidence and Improve
Calibration/ Debiasing

• Lichtenstein et al., (1982) examined the example
  of calibration of weather forecasters. They
  conclude that continuance (they have been making
  probabilistic forecasts for years),
  repetitiveness of the task and the fact that the
  outcome feedback for weather forecasters is well
  defined and promptly received have high impact
  on accuracy of their predictions.
• Calibration might be improved for discrete
  propositions tasks, but not for the
  probabilities assessment by fractile method.
• Overconfidence might also be reduced by asking
  subjects, prior to the assessment of their
  confidence, to think of the reasons why their
  answers might be wrong.

26
Overconfidence in Finance

• Recently, interest in the consequences of
  traders overconfidence on financial decision
  making and the functioning of financial markets,
  has occurred in behavioral economics.

27
Theoretical Overconfidence Finance Literature

• Behavioral finance theories incorporate findings
  of psychology research into standard finance
  models, and model psychological bias of
  overconfidence.
• Overconfidence is usually modeled as
  overestimation of precision of ones information
  (Glaser et al, 2003). Overconfident investors
  underestimate the variance of their private
  signals, which contain not only information but
  also a random error. Stated equivalently
  overconfident investors underestimate the
  variance of the error term contained in their
  signal.
• This way of modeling overconfidence captures the
  idea that people overestimate the precision of
  their knowledge.
• Some models assume that the degree of
  overconfidence is a stable individual trait and
  thus constant over time. However, other models
  assume that overconfidence dynamically changes
  over time (Glaser et al., 2003).

28
Empirical and Experimental Tests of
Overconfidence Models

• Empirical studies, experiments, quasi-experiments
  are used to test validity of overconfidence
  models.
• There are two directions in testing the empirical
  validity of theoretical models model assumptions
  and model predictions.
• Model assumptions of investors overconfidence
  can be evaluated by experiments and
  questionnaires
• Model predictions of the link between market
  outcomes and investors overconfidence are tested
  by correlation of proxies (e.g. gender, high
  returns) or measures of overconfidence (bias
  score) and economic variables (e.g. trading
  volume).
• Its important to mention, that studies using
  proxies of overconfidence share the shortcoming
  that overconfidence is never directly observed.
• Experimental and quasi-experimental design allows
  more confident inference about cause and effect
  relations.

29
Theoretical Models of Overconfidence in Financial
Markets
Macro-level impact of overconfidence (Market-level) Author Micro-level Impact (investor level) Author
Excess trading volume DeBondt and Thaler, 1985 Shiller (2000), Overconfident investors trade more aggressively, i.e. higher trading volume. Odean (1998), Gervais and Odean (2001)
Excess price volatility Scheinkman J.A. and Xiong, W (2003), Benos 1998, Reduction of expected utility DeLong, Shleifer, Summers, and Waldman (1990), Odean (1997)
Overconfidence causes asset price bubbles Scheinkman J.A. and Xiong, W (2003) Undiversified portfolios/ Riskier portfolio choice Odean (1997), Odean (1998)
Icreases market depth Odean (1998), Kyle and Wang (1997), and Benos (1998) Increased tendency to herd Hirshleifer, Subrahmanyam and Titman (1994)
Underrecaction or overreaction of markets to new info Odean (1997) Active portfolio management Lakonishok, Shleifer and Vishny (1992)
Return predictability of financial assets Daniel et al. (1998, 2001) Scheinkman J.A. and Xiong, W (2003) Higher expected returns DeLong, Shleifer, Smmers, and Waldman (1990)

30
Empirical Findings and Experimental Results
Macro-level impact of Overconfidence Author Micro-level Impact Author
Trading (overconfidence generates trading) Odean (1997), Statman and Thorley (1997) Level of overconfidence vary with past market returns. Statman et al., 2004
High trading volume in the market. Statman, Thorley, and Vorkink (2004), Kim and Nofsinger (2003). Too much of trading activity. Men trade more than women. Barber and Odean, 2000 Barber and Odean, 2001.
Probability of Bubbles (Top rank belief variablehas a positive and significant effect on the probability of bubbles). Jörg Oechssler, Carsten Schmidt and Wendelin Schnedler (2007) There is relationship between degree of overconfidence, and degree of professionalism Maciejovsky et al., (2003) , Menkhoff et al. (2005) and Biais et al. (2004)
Underrecaction or overreaction of markets to new info Loughran et al. (1995) - overreaction Bernard et al. (1989) - underreaction. Unrealistically positive self-evaluation increases trading volume. Glaser and Weber (2004)
Higher degree of overconfidence reduces traders performance/welfare. Biais et al. (2004), Odean (1999), Barber and Odean (2002).
Mistakes in financial decision making Biais et al. (2004)

1 However when overconfidence was measured as
miscalibration Glaser and Weber (2004) do not
find relation between trading volume and
overconfidence. Same results are found in an
experiment by Biais et al. (2004).
31

• Thank You For Your Attention