Oxford - Essential Flashcards
[Reversed]
Prospect Theory:
S shaped value function, non-linear weighting function - “Possibility” and “Certainty” effects Prospect theory has two phases: Editing and evaluating phases of the decision process
An alternative theory of choice where value is assigned to gains and losses rather than to final assets.
Probabilities are replaced by decision weights. The value function is normally concave for gains, commonly convex for losses, and is generally steeper for losses than for gains.
Decision weights < corresponding probabilities, except in the range of low probabilities. Overweighting of low probabilities may contribute to the attractiveness of both insurance and gambling.
In addition, people generally discard components that are shared by all prospects under consideration. This tendency, called the isolation effect, leads to inconsistent preferences when the same choice is presented in different forms.
Pay more to eliminate the last bullet in Russian roulette. Certainty effect.
Kahneman and Tversky (1979) (Emtca)
[Reversed]
Empirical Evidence in Games:
- decide how to share endowment, avergae 10-25% but modes at 0 and 50%
- DG + accept / reject –> offer average 40-50%, offers <20% rejected around half the time, higher offers less likely to be rejected
- £y –> trustee gets 3x transfer, now how much to give back –> Keep all? No trust and reciprocity? –> No, investing 50% of endowment and trustees averge return slightly less than investment.
- Wage offer accepted / rejected, then if accept choose an effort level. GTh. would predict minimal wage, minimal effort –> In practice, generous wage offers rewarded with generous effort
Empirical Evidence in Games:
- Dictator Game
- Ultimatum Game
- Trust game / Investment game
- Gift exchange game
Benoit and Dubra (2011)
“Better than average” effect can be conssitent with Bayes’ rue when applied to correct beliefs:
- Plausible to have majority of people believing with Prob > 0.5 their skills lie above median
- Beliefs are P distributions not absolute, do not indicate a certainty of such a relative rank
Mobius et al. (2014)
Biased about own ability: track responses about an IQ quiz, overweighting positive feedback and are conservative, updating too little in response to both positive and negative feedback.
Less pronouned effect in a placebo experiment with no ego motives.
- IQ quiz: own performance vs. robot
- Driving ability
- CEO performance
Results
- Underinference: too conservative in updating (robust)
- Asymmetry: update more about good news (not so robust)
Speaks to various other ideas
- Confirmation bias: overweighting information confirming prior views
- Overconfidence
- Forecasting overconfidence, eg. in precision
Fundamental necessity to relax use of Bayes’ rule.
Fehr, Kirchsteiger and Riedl (1993) [QJE]
5 firms bidding for workers, each paid a wage to exert effort. Gift exchange game.
- Standard theory: pay a wage to satisfy IR, same effort for all wages
- Real data shows positive link between wage and effort
- Pattern is stable over time, not linked to experience
- Could be explained by inequity aversion: low wages –> worker behind –> low effort, vice. versa for high wages
- Could also be reciprocity
[Reversed]
Second mover behaviour in two trust games –> Evidence of Positive reciprocity
Would expect same behaviour in games, but we see much higher proportions of continue under voluntary game.
McCabe, Rigdon and Smith (2003)
DellaVigna and Linos (2020) (Emtca)
Comparison of nudge interventions, 126 RCTs + 26 RCTs analysed in US, over 23 million participants + 500k participants.
8.7% point effect on take up shows a large impact of nudges. Smaller average effect, 1.4% point but still 8% relative.
No publication bias, just combined work of two nudge units.
- No financial incentives, pure nudges
- No default treatments
Academic Journals (few 0 effect, positive but wide CI) vs. Nudge Units (many 0, small CI)
- Sample size –> Statistical power, determines how small of an effect can be detected
- Publication bias –> p-hacking
- AJ vs NU have different goals
Kahneman and Tversky (1973)
Personal profiles and job likelihoods.
- Subjects ignored base rate, predicted ratio only 1.2 despite should equal 5.4 given base rates
Ellsberg (1961)
Subjective expected utility, ambiguity aversion in 3 colour ball experiment.
- f1 = (100,0,0) vs. f2 = (0,100,0)
- f3 = (100,0,100) vs. f4 = (0, 100,100)
f1 > f2, f4 > f3 violation of consistent beliefs.
Cannot use EU here as we do not know the probabilities. SEU is required here!
[Reversed]
Anticipation utility: usually discount the future, but in fact some people pay to postpone things.
Couple also with memory utility, eg. weddings. Link to Benabou and Tirole (2016)
- Affective side: perceptions are a direct source of utility and disutility
- Functional side: beliefs may help us to achieve outcomes, and hence have instrumental value - internal vs. external goals.
Loewenstein (1987)
[Reversed]
Compare subjective likelihood of future desirable or adverse attempts in population eg. drinking problem, owning a home, being burgled
- Optimism: more likely to epxerience positve outcomes
- Overconfidence: Less likely to experience negative outcomes
Weinstein (1980)
[Reversed]
Logit models as an alternative theory, widely used in experimental work.
A: $1 0.9, $60 0.1
B: $5
- More risk averse, we expect go for more option B.
- However, increases after a while…why?
Logit is a cardinal model, cardinal differences matter for choice proabablities. Logit is only good if utilities have cardinal patterns.
CE representation: non-linear transformation gives very different cardinality scales.
Apesteguia and Ballester (2018)
[Reversed]
Interdependent preferences and motives: our care for others depends upon our beliefs about how nice or reciprocal they are.
Levine (1998)
Tversky and Kahneman (1974); Benjamin, Rabin and Raymond (2016)
Sample size neglect: non belief in LLN
- Smaller hospitals or larger hospitals more likely to have 60% boys when average overall is 50%?
- Most people say about the same –> but deviations more likely in small samples
“Non belief” in LLN.
[Reversed]
- Motivated ignorance - most confident individuals and information loving, least confident information averse.*
- Can we avoid information to choose our beliefs?
Ganguly and Tasoff (2016) - information avoidance in testing for herpes.
Oster et al (2013) - genetic risk of developing Huntington disease, <10% people decide to get tested.
- People who got tested and were positive substantially changed behaviour.
- Not tested behave as if they were low risk, overconfidence
Burks et al (2013)
DellaVigna, List and Malmendier (2012) (QJE)
Altruism vs. Societal Pressure: Charity Donations
Door-to-door fundraising drive, flyers giving recipients option to seek or avoid.
Flyers reduce share of households opening door by 10-25%, 30% if “Do Not Disturb” option offered.
Door to door campaigns on average lower the utility of potential donors. Welfare losses by flyer campaign, stronger impact for popular charity.
Linked also to “warm glow” - people like the act of giving and not giving itself per se. Image concerns.
Two stage game model: reference points, social costs, altrusim levels, solve via BI
- Low altruism –> Opt out of opening door
- No social pressure cost if opt out
[Reversed]
Homo economicus solves very complex problem: well defined preferences, unbiased beliefs, able to make fully optimal choices
Thaler (2016)
[Reversed]
Ambiguity and ambiguity aversion. Link to Subjective expected utility giving only prizes and no probabilities. “Acts”.
eg. Urn 1: 50 red balls, 50 black balls
Urn 2: 100 balls red or black
People prefer to bet on Urn 1. SEU, indifferences suggest the belief formed is 1/2 in both worlds but prefer to choose when known.
Machina and Siniscalchi (2014)
[Reversed]
Maxmin preferences: beliefs are not fixed, depend upon contemplated act f and simple way of dealing with ambiguity.
For a given act only one belif is considered, it is the one that damages the act the most. Pessimistic, think the experimentalist put all the balls against me!
Can be used to explain the choices over acts seen in Ellsberg paradowx problem.
Gilboa and Schmeidler (1989)
Fox and Clemen (2005)
Predict likelihood of US business schools being ranked #1
- Wharton, all else: 60% Wharton
- Chicago, Harvard, Kelogg, Stanford, Wharton, Other: 30% Wharton
Isolation impact? Sub-additivity?
Augenblick and Rabin (2019)
Beliefs: People dont like to be too wrong in their predictions
Forecast own effort level in greek letter transcription task, reminded of own effort before completion.
- Bonus payment didnt dramtically increase beliefs, small increase.
- People try to stay within a +/- 5 of their prior prediction!
- More -5 than +5, present bias
Gul (1991)
Disappointment aversion: add to model with an additional parameter capturing sensititvity to elation and disappointment
- Ex-post elation: outcome > CE
- Disappointment: outcome < CE
[Reversed]
Nudge: choice achitectrue to alter people’s behavriousl without forbidding options or significantly changing incentives.
Light touch behavioural interventions via simplification, personalisation or reminers etc.
Applications: tax compliance, beahvriour change, recycling, purchasing decisions
Thaler and Sunstein (2008)
[Reversed]
Comparison of nudge interventions, 126 RCTs + 26 RCTs analysed in US, over 23 million participants + 500k participants.
8.7% point effect on take up shows a lrge impact of nudges. Smaller average effect, 1.4% point but still 8% relative.
No publication bias, just combined work of two nudge units.
- No financial incentives, pure nudges
- No default treatments
Academic Journals (few 0 effect, positive but wide CI) vs. Nudge Units (many 0, small CI)
- Sample size –> Statistical power, determines how small of an effect can be detected
- Publication bias –> p-hacking
- AJ vs NU have different goals
DellaVigna and Linos (2020) (Emtca)
Gilboa and Schmeidler (1989)
Maxmin preferences: beliefs are not fixed, depend upon contemplated act f and simple way of dealing with ambiguity.
For a given act only one belif is considered, it is the one that damages the act the most. Pessimistic, think the experimentalist put all the balls against me!
Can be used to explain the choices over acts seen in Ellsberg paradowx problem.
Rabin and Vayanos (2010)
GF and HHF in single framework
- Belif in negative correlaiton –> GF
- Put P > 0 on process have HH
Agent therefore expects two things
- High frequency -ive autocorrelation “short streaks”
- +ive autocorreltion over long streaks
[Reversed]
Lottery choice experiment measuring risk aversion over a wide range of payoffs. Few $ to several 100 $. Hypothetical vs real incentives.
- Hypothetical ==> More erratic, no change when payoffs are scaled
Real Prizes:
- Small prizes ==> Low levels of risk aversion
- Larger prizes ==> Sharp increase in risk aversion
Power-Expo utility exhibiting IRRA and DARA
Highlight dangers of assuming risk neutrality, subjects also underestimate extent to which they will avoid risk.
Holt and Laury (2002) (AER)
Kahneman and Tversky (1979) (Emtca)
Prospect Theory:
S shaped value function, non-linear weighting function - “Possibility” and “Certainty” effects Prospect theory has two phases: Editing and evaluating phases of the decision process
An alternative theory of choice where value is assigned to gains and losses rather than to final assets.
Probabilities are replaced by decision weights. The value function is normally concave for gains, commonly convex for losses, and is generally steeper for losses than for gains.
Decision weights < corresponding probabilities, except in the range of low probabilities. Overweighting of low probabilities may contribute to the attractiveness of both insurance and gambling.
In addition, people generally discard components that are shared by all prospects under consideration. This tendency, called the isolation effect, leads to inconsistent preferences when the same choice is presented in different forms.
Pay more to eliminate the last bullet in Russian roulette. Certainty effect.
Eil and Rao (2011)
Experimental setting to analyse updating process relative to Bayesian process.
- IQ or beauty, rank 1-10 both ego relevant
- Report prior –> Signal: “you ranked above/below person X” –> Update belief
- Control task for guessing an integer
- Elicit WTP for learning true rank
[Reversed]
Subjective expected utility, ambiguity aversion in 3 colour ball experiment.
f1 = (100,0,0) vs. f2 = (0,100,0)
f3 = (100,0,100) vs. f4 = (0, 100,100)
f1 > f2, f4 > f3 violation of consistent beliefs.
Cannot use EU here as we do not know the probabilities. SEU is required here!
Ellsberg (1961)
[Reversed]
Rank dependent utility, order the lottery outcomes and then assign weights to each iteratively.
- Satisfied monotonicity
- Tries to capture data in a normative way, but does not fully explain data
Quiggin (1982)
[Reversed]
Beliefs: People dont like to be too wrong in their predictions
Forecast own effort level in greek letter transcription task, reminded of own effort before completion.
- Bonus payment didnt dramtically increase beliefs, small increase.
- People try to stay within a +/- 5 of their prior prediction!
- More -5 than +5, present bias
Augenblick and Rabin (2019)
[Reversed]
Altruism vs. Societal Pressure: Charity Donations
Door-to-door fundraising drive, flyers giving recipients option to seek or avoid.
Flyers reduce share of households opening door by 10-25%, 30% if “Do Not Disturb” option offered.
Door to door campaigns on average lower the utility of potential donors.
DellaVigna, List and Malmendier (2012) (QJE)
Benabou and Tirole (2006)
Charitable giving: wanting to appear as prosocial, dilemma vs liking $$. Bayesian games, trying to learn types.
- What do people infer about me…?
- I care about what others think - “image capital”
Read, Lowenstein and Kalyanaraman (1999)
High vs Low brow movies: now pick low, in future pick sophisticated
[Reversed]
Theory of Fairness: Difference aversion in payoffs.
Utility is a function of: my own monetary reward, minus two weighted objects –> sum/avg. of differences where I am behind, sum/avg. of differences where I am ahead.
Matches up with behaviour in:
- Ultimatum games
- Public goods games
- Other market games, sometimes with responders
Similar to Bolton and Ockenfels (2000)
Economic environment determisn whether the fair or dominant types determine competitive equilibrium.
Kinked diagram: Ui(xj | xi) on y, xj on x. Point on 45 degree line then two downward sloping lines, steeper when xj > xi.
Assume this guilt / envy does not over pwoer our own self interest. Can explain behaviours in the ultimatum game.
Fehr and Schmidt (1999) (QJE)
[Reversed]
Games testing social preferences to explain departures from self interest. 2x2 game payoff combinations for you and me. Dictator games, response games, also 3 person games of these types.
Individuals care less about reducing differences as in existing models (Fehr and Schmidt, 1999), more about sacrifices which boost welfare overall - especially for low payoff players.
Evidence for reciprocity. Stark contrast in proportion of outcomes over (400,400) vs (750,400) when seocnd player is given choice rather than first picking (750, 0).
Utility function is a weighted sum of A and B payoffs with parameters on weights for..
- Who is ahead / behind?
- Has the other player misbehaved?
Competitive preferences: caring directly for others but wanting do as well as possible as them
Charness and Rabin (2002) (QJE)
[Reversed]
Biased about own ability: track responses about an IQ quiz, overweighting positive feedback and are conservative, updating too little in response to both positive and negative feedback.
Less pronouned effect in a placebo experiment with no ego motives.
- IQ quiz: own performance vs. robot
- Driving ability
- CEO performance
Results
- Underinference: too conservative in updating (robust)
- Asymmetry: update more about good news (not so robust)
Speaks to various other ideas
- Confirmation bias: overweighting information confirming prior views
- Overconfidence
- Forecasting overconfidence, eg. in precision
Fundamental necessity to relax use of Bayes’ rule.
Mobius et al. (2014)
Samuelson and Zeckhauser (1988)
Masatlioglu and Ok (2005)
Status quo bias: eg. never show a prospective customer more than 3 options for a purchase
Reference points, only choose things strictly better than reference point, which can be damaging missing out on better bundles.
Falk, Fehr and Fischbacher (2008)
Take away some tokens from others or give more, studying role of intentions.
- Random device vs real person choosing –> sam expected distribution
- Steeper reaction under real person intentions
- Not flat under dice, so still some care for distribution
Positive and negative reciprocity!
[Reversed]
3 relaxations of standard model to accomodate human behaiours
- Non standard preference: unstable discounting, social preferences, risk preferences vs. reference points
- Non standard beliefs: misperception, overconfidence, law of small numbers, projection bias, confirmation bias
- Non standard decision making: limitted attention, framing, salience, heuristics
DellaVigna (2009)
Read and van Leeuwen (1998)
Snack choice: chocolate or apple
Following week wants chocolate, one week in advance want apple.
Cognitive dissonance: we want to be consistent in our decisions
Hungry vs Satiated: Projection bias, hard to put ourselves in the shoes of future self, we just like to think about the here and now
Attanasi and Nagel (2008) (Essays on BGT)
Battigalli (2007) - formulation proposed
Example of trust game with belief dependent motivations. Persistent lab deviations from maximising expected payoff –> Explain via psychological utilities.
Psychological games exacerbate multiplicity, making equilibrium coordination harder. Two stage trust game, continuation like the centipede game.
Guilt Aversion: inflicting harm on others induced guilt –> Incorporate this into the “steal” choice in the trust game discounting the second player benefit from being selfish. (Battigalli and Dufwenberg 2007) Guilt aversion gives opposite relation between sharing and second order beliefs.
Positive and Negative reciprocity: what goes around comes around. (Rabin, 1993) Consider player kindness, perceieved kindness –> Second order beliefs. B sharing decreases in second order beliefs, ie. his expectaion of how often A expects him to share.
Usually players learn to play NE via repeated interaction as they come to hold correct beliefs. In psychological eq*, utilities depend upon hierarchical beliefs implying they need to learn others’ beliefs.
[Reversed]
Foundation for behvaioural public economics focus on policy problems.
Can sometimes replace revealed preference with normative principles.
Applications: saving, addiction and public goods
Bernheim and Rangel (2005)
[Reversed]
Test a menu of dictator games: participants equalising payoffs interpreted as having social preferences not difference aversion.
Andreoni and Miller (2002)
[Reversed]
Independence assumptin of EU implies betweeness.
- Property may not be satisfied if we have compund risk aversion, Ie. we think more risky to combine two already risky lotteries.
- Independence implies common consequence principle –> indifference curves must be parallel
- More risk averse –> Steeper slope of IC in the triangles, more concave if not linear etc.
- More risk averse –> Lower CE
Bekel (1986)
[Reversed]
Base rate Neglect
- Correct answer 1.96%, but many smart people answer 95%
- Forgetting base rate
Casscells, Schoenberger and Graboys (1978)
[Reversed]
7 Good Properties of Economic Models
- Parsimony
- Tractability
- Conceptual insightfulness
- Generalizability
- Falsifiability
- Empirical consistency
- Predictive precision
Gabaix and Laibson (2008)
[Reversed]
“Better than average” effect can be conssitent with Bayes’ rue when applied to correct beliefs:
- Plausible to have majority of people believing with Prob > 0.5 their skills lie above median
- Beliefs are P distributions not absolute, do not indicate a certainty of such a relative rank
Benoit and Dubra (2011)
Quattrone and Tversky (1984)
Participants asked to submerge their hand into ice water, exercise on a bike, and told about chances of an undesirable event.
- Risk of heart disease beliefs varied with exercise / not
- Depending upon what you were told, you keep hand in water longer / shorter
- Deny purposefully trying to make favourable diagnosis
Loewenstein (1987)
Anticipation utility: usually discount the future, but in fact some people pay to postpone things.
Couple also with memory utility, eg. weddings. Link to Benabou and Tirole (2016)
- Affective side: perceptions are a direct source of utility and disutility
- Functional side: beliefs may help us to achieve outcomes, and hence have instrumental value - internal vs. external goals.
Quiggin (1982)
Rank dependent utility, order the lottery outcomes and then assign weights to each iteratively.
- Satisfied monotonicity
- Tries to capture data in a normative way, but does not fully explain data
[Reversed]
Methods to estimate present bias
Find X today that makes indifferent to $10 in one week, find Y in one week than makes you indifferent to $10 in two weeks.
Assume linear utility –> Back out parameters.
Issues:
- Static vs. dynamic preference reversals: time invariance required, chaing the time at which a question is asked does not matter
- Money vs. consumption: saving and interest considerations for $
- Linear vs. concave utility: small vs. large amounts –> introduce bias in amounts
- Uncertainty: shocks to future MU
Thaler (1981)
Toussaert (2018), Fedyk (2018)
We are not good at forecasting about ourselves, good at spotting flaws in others. People who demand commitment dont succumb to temptation. - “self-control” types.
- We think others are like us, more than they actually are
- Ask about others to remove some ego elements, perhaps can be a valid instrument
- Empirical estimates about others are closer to present bias, when asked about self, usually close to 1 –> “Time consistent”
Andreoni (1990)
Warm glow - donation becomes party a private good, “impure altruism”. Taxes cannot replace own contribution, people care about how public goods come about not just their existence.
O’Donoghue and Rabin (1999) (AER)
Self-control problems–modeled as time-inconsistent, present-biased preferences–in a model where a person must do an activity exactly once. Beta-delta preferences.
- Immediate costs or immediate rewards?
- Are people sophisticated or naive about future self-control problems?
Naive people procrastinate immediate-cost activities and preproperate–do too soon–immediate-reward activities. Beta < Estimate = 1
Sophistication mitigates procrastination but exacerbates preproperation. Beta = Estimate
Moreover, with immediate costs, a small present bias can severely harm only naive people, whereas with immediate rewards it can severely harm only sophisticated people.
Applications: savings, addiction, healthcare, pensions
Must perform activity exactly once within a finite time T, {Delay, Complete}. “Perception perfect strategy”: optimal in each t given current preferences and future behaviour perception.
Investment goods (delayed benefit, immediate cost): V=V, C = {3,5,8,13}, beta = 0.5, delta = 1
Time Consistent: {C,C,C,C}
Naifs: {D,D,D,C}
Sophisticates: {D,C,D,C} –> Compares 2 with 4, knowing he procrastinates at 3!
Leisure goods (immediate benefit, delayed cost): C=0, V = {3,5,8,13}, beta = 0.5, delta = 1 –> Watch a movie or Not…? Films getting better
Time Consistent: {N,N,N,Y}
Naifs: {N,N,Y,Y}: 8 > 0.5 * 13
Sophisticates: {Y,Y,Y,Y} –> 8 > 0.5* 13, 5 > 0.5* 8…full unravelling
[Reversed]
Belief in the las of small numbers: draw with replacement, but form beleifs as if without replacement.
- Can generate Gamblers Fallacy
- Can NOT generate Hot Hand effects
Rabin (2002)
Mayraz (2011)
Shown a chart of wheat prices and asked bakers vs. farmers
- Over optimising
- Wishful thinking
Gul and Pesendorfer (2002)
Random EU Theory: set of EU, with with given Prob of occuring. Realised EU determines choice vis standard maximisation. Distribution F over set of all EU.
eg. Choices under the influence of alcohol vs sober.
Diagram: pairs of lotteries A C with P = 0.8. Model CAN explain this!
- B > A ==> Red: IC flatter than red line (vice versa)
- A > B ==> Steeper IC than red, carries over in Blue
- In every mood I choose A, I should also choose C: I should shoose C more often than A. Vairation in EU and different realisations, so we can compare only when we have probabilities of things being chosen.
Gul and Pesendorfer (2001);
Fudenberg and Levine (2006,11,12)
People care about the process of choice and decision making.
- Menu dependent preferences
- Lower utility from taking sald if i have to resist a burger
- Utility = Commitment Utility - Self Control Cost
- Dual self models
- Planner (self at t) restrictions choice set for doer (self at t+1)
Rabin (2002)
Belief in the las of small numbers: draw with replacement, but form beleifs as if without replacement.
- Can generate Gamblers Fallacy
- Can NOT generate Hot Hand effects
[Reversed]
- fMRI machine - hardwired for optimism.*
- Asymmetry in good and bad news, brain diminished activity after bad news
Sharot et. al (2011)
Bernheim and Rangel (2009) (QJE)
Generalising standard choice theoretic welfare economics to nonstandard behavioural models.
Unambiguous choice relation: X is always chosen over Y when available.
Nests the standard choice framework when aximos satisfied. Small departures catered for.
Theory that delivers bounds on welfare based purely on choice data.
- Standard model, agents choose an option x from a choice set X –> Policy ID optimal x
- Behavioural models, agents choose from “generalized choice sets” G = (X, d) (henceforth, GCS)
- d is an “ancillary condition”: something that affects choice behaviour but (by assumption) does not affect experienced utility. I Examples: colour of paper, salience, framing, default option
[Reversed]
Exponential discounting to cater for time preferences, its time consistent and very appealing.
£100 today or £105 tomorrow vs £100 in 30 days vs £105 in 31 days?
Only the delay between two periods should matter, not the timing of the period.
Emprically invalid! Many studys show systematic preference reversals, humans and animals, different parts of the brain.
Samuelson (1937)
Suetens et al (2015)
Lottery players: 9.2% of people react to recent lottery draws
- Less likely to bet on it - GF
- If they see it many times, bet on it - HHF
- Correlation between the two driving forces of above: LSN and Overinference
Brunnermeier and Parker (2005)
Increasing cost of self-deception could reduce dynamic bias.
- Payment for correct recall
- Encoding vs. Retrieval stages
- People dont erase info from memory, just suppressing due to damage to ego
[Reversed]
Experimental setting to analyse updating process relative to Bayesian process.
- IQ or beauty, rank 1-10 both ego relevant
- Report prior –> Signal: “you ranked above/below person X” –> Update belief
- Control task for guessing an integer
- Elicit WTP for learning true rank
Eil and Rao (2011)
[Reversed]
Self-control problems–modeled as time-inconsistent, present-biased preferences–in a model where a person must do an activity exactly once. Beta-delta preferences.
- Immediate costs or immediate rewards?
- Are people sophisticated or naive about future self-control problems?
Naive people procrastinate immediate-cost activities and preproperate–do too soon–immediate-reward activities. Beta < Estimate = 1
Sophistication mitigates procrastination but exacerbates preproperation. Beta = Estimate
Moreover, with immediate costs, a small present bias can severely harm only naive people, whereas with immediate rewards it can severely harm only sophisticated people.
Applications: savings, addiction, healthcare, pensions
Must perform activity exactly once within a finite time T, {Delay, Complete}. “Perception perfect strategy”: optimal in each t given current preferences and future behaviour perception.
Investment goods (delayed benefit, immediate cost): V=V, C = {3,5,8,13}, beta = 0.5, delta = 1
Time Consistent: {C,C,C,C}
Naifs: {D,D,D,C}
Sophisticates: {D,C,D,C} –> Compares 2 with 4, knowing he procrastinates at 3!
Leisure goods (immediate benefit, delayed cost): C=0, V = {3,5,8,13}, beta = 0.5, delta = 1 –> Watch a movie or Not…? Films getting better
Time Consistent: {N,N,N,Y}
Naifs: {N,N,Y,Y}: 8 > 0.5 * 13
Sophisticates: {Y,Y,Y,Y} –> 8 > 0.5* 13, 5 > 0.5* 8…full unravelling
O’Donoghue and Rabin (1999) (AER)
Apesteguia and Ballester (2018)
Logit models as an alternative theory, widely used in experimental work.
A: $1 0.9, $60 0.1
B: $5
- More risk averse, we expect go for more option B.
- However, increases after a while…why?
Logit is a cardinal model, cardinal differences matter for choice proabablities. Logit is only good if utilities have cardinal patterns.
CE representation: non-linear transformation gives very different cardinality scales.
Dana et al. (2006)
Social pressure and image concerns:
- Opt out experiments, dictator game vs get $9
- Selfish: (10,0) > (9,0)
- Social preferences (9,1) > (9,0)
- 1/3 of subjects opt out at (9,0)
[Reversed]
Warm glow - donation becomes party a private good, “impure altruism”. Taxes cannot replace own contribution, people care about how public goods come about not just their existence.
Andreoni (1990)
McCabe, Rigdon and Smith (2003)
Second mover behaviour in two trust games –> Evidence of Positive reciprocity
Would expect same behaviour in games, but we see much higher proportions of continue under voluntary game.
Kube, Marechal and Puppe (2012) [AER]
Gift giving: positive reciprocity
- Small effect of 20% pay icnrease
- Real gift has high effect!
- Persistent effects on effort over time
Why is this true?
- Not due to perceived value, price tag makes no impact
- More kindness perceived, hence response is higher effort
- Even stronger impact with Origami money!
Koszegi and Rabin (2002) (QJE)
Model of reference-dependent preferences and loss aversion where “gain-loss utility” is derived from standard “consumption utility” and the reference point is determined endogenously by the economic environment.
Reference point = rational expectations held in the recent past about outcomes in a personal equilibrium
Gain-loss utility influences behavior when there is uncertainty.
WTP increasing in the expected probability of purchase and in the expected prices conditional on purchase.
In within-day labor-supply decisions, a worker is less likely to continue work if income earned thus far is unexpectedly high, but more likely to show up as well as continue work if expected income is high.
Inaction has a specific feature! “Status quo bias” - perhaps complexity of menu encourages no action?
Allows expectations, satisfaction, memory, frames, emotions etc.
[Reversed]
Investigate choices over consumption (real effort) in a longitudinal experiment. Subjects asked to complete boring tasks, changing interest rates gives expected downward slope.
We pair this effort study with a companion monetary discounting study.
- Very limited time inconsistency in monetary choices.
- Considerably more present bias in effort.
Repetitive transcription task and also a tetris game completion, ran over the course of 7 weeks.
Augenblick, Niederle and Sprenger (2015) (QJE)
Holt and Laury (2002) (AER)
Lottery choice experiment measuring risk aversion over a wide range of payoffs. Few $ to several 100 $. Hypothetical vs real incentives.
- Hypothetical ==> More erratic, no change when payoffs are scaled
Real Prizes:
- Small prizes ==> Low levels of risk aversion
- Larger prizes ==> Sharp increase in risk aversion
Power-Expo utility exhibiting IRRA and DARA
Highlight dangers of assuming risk neutrality, subjects also underestimate extent to which they will avoid risk.
[Reversed]
Charitable giving: wanting to appear as prosocial, dilemma vs liking $$. Bayesian games, trying to learn types.
- What do people infer about me…?
- I care about what others think - “image capital”
Benabou and Tirole (2006)
[Reversed]
Disappointment aversion: add to model with an additional parameter capturing sensititvity to elation and disappointment
- Ex-post elation: outcome > CE
- Disappointment: outcome < CE
Gul (1991)
Grether (1980); Benjamin (2018)
Inference neglect: Reduced form model, applying powers to Bayes’ Rule:
- Bias in likelihoods (c) –> Under / over inference on info contained in signals
- Bias in priors (d) –> Base rate neglect / over-use of prior information
Results:
- Underinference far more common than overinference
- Underinference is more pronounced for larger samples
- Underinfer even when only one signal
- Proportion based inferences
Explanations?
- Biased sampling distribution beliefs: use some beliefs instead of probabilities
- Conservatism bias: c < 1 given structural interpretation
- Extrem belief aversion: dislike beliefs close to certainty or impossibility
[Reversed]
Personal profiles and job likelihoods.
- Subjects ignored base rate, predicted ratio only 1.2 despite should equal 5.4 given base rates
Kahneman and Tversky (1973)
Gabaix and Laibson (2008)
7 Good Properties of Economic Models
- Parsimony
- Tractability
- Conceptual insightfulness
- Generalizability
- Falsifiability
- Empirical consistency
- Predictive precision
[Reversed]
Allais Paradox: 11% and 89% split across different things violate independence.
Allais (1953)
[Reversed]
Gift giving: positive reciprocity
- Small effect of 20% pay icnrease
- Real gift has high effect!
- Persistent effects on effort over time
Why is this true?
- Not due to perceived value, price tag makes no impact
- More kindness perceived, hence response is higher effort
- Even stronger impact with Origami money!
Kube, Marechal and Puppe (2012) [AER]
Fehr and Schmidt (1999) (QJE)
Theory of Fairness: Difference aversion in payoffs.
Utility is a function of: my own monetary reward, minus two weighted objects –> sum/avg. of differences where I am behind, sum/avg. of differences where I am ahead.
Matches up with behaviour in:
- Ultimatum games
- Public goods games
- Other market games, sometimes with responders
Similar to Bolton and Ockenfels (2000)
Economic environment determisn whether the fair or dominant types determine competitive equilibrium.
Kinked diagram: Ui(xj | xi) on y, xj on x. Point on 45 degree line then two downward sloping lines, steeper when xj > xi.
Assume this guilt / envy does not over pwoer our own self interest. Can explain behaviours in the ultimatum game, but some experiments feature sacrificing earnings when behind which is not compatible.
[Reversed]
Take away some tokens from others or give more, studying role of intentions.
- Random device vs real person choosing –> sam expected distribution
- Steeper reaction under real person intentions
- Not flat under dice, so still some care for distribution
Positive and negative reciprocity!
Falk, Fehr and Fischbacher (2008)
[Reversed]
Dynamic motivated beliefs in response to feedback .
- Positive feedback has persistent effect
- Negative feedback has only short run effects
Directly after feedback, subjects respond immediately in the correct direction, but asymmetries come in after one month. Negative accuracy is recalled with less accuracy than positive –> Higher % of “I don’t recall”
Less profound when not related to ego aspects like the IQ test.
Zimmermann (2020) (AER)
Enke and Zimmermann (2019) (Review of Econ. Studies)
Correlation Neglect: if you suffer, then you do not fully account for some signal incorporates another –> overshooting patterns, upward bias in beliefs
Experimental evidence that many people neglect the double-counting problem in the updating process in signals.
- Need to estimate continuous state, paid for accuracy
- Get 4 signals, either correlated or uncorrelated signals
- See s1 and then average of each pair s1 and another
Rational DM would extract independent information from each pair of signals. If you suffer from correlation neglect, do not fully incorporate correlation.
- Overshooting / undershooting –> Naivete shows heterogeneity, bi-modal in correlated case
- Beliefs are too sensitive to the ubiquitous “telling and re-telling of stories” and exhibit excessive swings.
Most participants have the computational skills to develop rational beliefs, many approach the problem in a wrong way when the environment is moderately complex.
- Conceptual vs. computational hurdles –> Conceptual is most important element
Thus, experimentally nudging people’s focus towards the correlation and the underlying independent signals has large effects on beliefs.
[Reversed]
Example of trust game with belief dependent motivations. Persistent lab deviations from maximising expected payoff –> Explain via psychological utilities.
Psychological games exacerbate multiplicity, making equilibrium coordination harder. Two stage trust game, continuation like the centipede game.
Guilt Aversion: inflicting harm on others induced guilt –> Incorporate this into the “steal” choice in the trust game discounting the second player benefit from being selfish. (Battigalli and Dufwenberg 2007)
Positive and Negative reciprocity: what goes around comes around. (Rabin, 1993)
Usually players learn to play NE via repeated interaction as they come to hold correct beliefs. In psychological eq*, utilities depend upon hierarchical beliefs implying they need to learn others’ beliefs.
Attanasi and Nagel (2008) (Essays on BGT)
Battigalli (2007) - formulation proposed
[Reversed]
Inference neglect: Reduced form model, applying powers to Bayes’ Rule:
- Bias in likelihoods (c) –> Under / over inference on info contained in signals
- Bias in priors (d) –> Base rate neglect / over-use of prior information
Results:
- Underinference far more common than overinference
- Underinference is more pronounced for larger samples
- Underinfer even when only one signal
- Proportion based inferences
Explanations?
- Biased sampling distribution beliefs: use some beliefs instead of probabilities
- Conservatism bias: c < 1 given structural interpretation
- Extrem belief aversion: dislike beliefs close to certainty or impossibility
Grether (1980); Benjamin (2018)
DellaVigna and Malmendier (2004) (QJE)
Analyze profit-maximizing contract design of firms if consumers have time-inconsistent preferences and are partially naive about it.
Two good types: 3 period model, two part tariff proposed involving lump sum fee and per usage price.
- goods with immediate costs and delayed benefits (investment goods) such as health club attendance,
- goods with immediate benefits and delayed costs (leisure goods) such as credit card-financed consumption.
We establish three features of contract design with naive time-inconsistent consumers. - Firms price investment goods below marginal cost.
- Firms price leisure goods above marginal cost.
- All types, firms introduce switching costs and charge back-loaded fees.
Targets consumer misperception of future consumption and underestimation of the renewal probability. Naifs offered a rebate, that they get less often than anticipated! Rebates helps sophisticates overcome present bias. Lower lump sum fee operates similarly, eg. credit cards no initial fee but high rates, casinos, cell phones
Applications: credit card, gambling, health club, life insurance, mail order, mobile phone, and vacation time-sharing industries.
Orhun (2018)
Reciprocity game: vary the amounts of one payoff to elicit reasons for choice - either fear of punishment
Result: people are nicer when there is no fear of punishment. More likely to choose “r”, rewarding the other person when there is no strategic incentive for them to have been kind.
Also measure second order beliefs, in line with behaviour.
Bekel (1986)
Independence assumptin of EU implies betweeness.
- Property may not be satisfied if we have compund risk aversion, Ie. we think more risky to combine two already risky lotteries.
- Independence implies common consequence principle –> indifference curves must be parallel
- More risk averse –> Steeper slope of IC in the triangles, more concave if not linear etc.
- More risk averse –> Lower CE
Charness and Rabin (2002) (QJE)
Games testing social preferences to explain departures from self interest. 2x2 game payoff combinations for you and me. Dictator games, response games, also 3 person games of these types.
Individuals care less about reducing differences as in existing models (Fehr and Schmidt, 1999), more about sacrifices which boost welfare overall - especially for low payoff players.
Evidence for reciprocity. Stark contrast in proportion of outcomes over (400,400) vs (750,400) when seocnd player is given choice rather than first picking (750, 0).
Utility function is a weighted sum of A and B payoffs with parameters on weights for..
- Who is ahead / behind?
- Has the other player misbehaved?
Competitive preferences: caring directly for others but wanting do as well as possible as them
Zimmermann (2020) (AER)
Dynamic motivated beliefs in response to feedback .
- Positive feedback has persistent effect
- Negative feedback has only short run effects
Directly after feedback, subjects respond immediately in the correct direction, but asymmetries come in after one month. Negative accuracy is recalled with less accuracy than positive –> Higher % of “I don’t recall”
Less profound when not related to ego aspects like the IQ test.
Similar findings of “forgetting bad feedback” in Benabou and Tirole (2002, 2004)
- “Recall bias”
Casscells, Schoenberger and Graboys (1978)
Base rate Neglect
- Correct answer 1.96%, but many smart people answer 95%
- Forgetting base rate
[Reversed]
Reciprocity game: vary the amounts of one payoff to elicit reasons for choice - either fear of punishment
Result: people are nicer when there is no fear of punishment. More likely to choose “r”, rewarding the other person when there is no strategic incentive for them to have been kind.
Also measure second order beliefs, in line with behaviour.
Orhun (2018)
[Reversed]
Analyze profit-maximizing contract design of firms if consumers have time-inconsistent preferences and are partially naive about it.
Two good types: 3 period model, two part tariff proposed involving lump sum fee and per usage price.
- goods with immediate costs and delayed benefits (investment goods) such as health club attendance,
- goods with immediate benefits and delayed costs (leisure goods) such as credit card-financed consumption.
We establish three features of contract design with naive time-inconsistent consumers. - Firms price investment goods below marginal cost.
- Firms price leisure goods above marginal cost.
- All types, firms introduce switching costs and charge back-loaded fees.
Targets consumer misperception of future consumption and underestimation of the renewal probability. Naifs offered a rebate, that they get less often than anticipated! Rebates helps sophisticates overcome present bias. Lower lump sum fee operates similarly, eg. credit cards no initial fee but high rates, casinos, cell phones
Applications: credit card, gambling, health club, life insurance, mail order, mobile phone, and vacation time-sharing industries.
DellaVigna and Malmendier (2004) (QJE)
[Reversed]
Correlation Neglect: if you suffer, then you do not fully account for some signal incorporates another –> overshooting patterns, upward bias in beliefs
Experimental evidence that many people neglect the double-counting problem in the updating process in signals.
- Need to estimate continuous state, paid for accuracy
- Get 4 signals, either correlated or uncorrelated signals
- See s1 and then average of each pair s1 and another
Rational DM would extract independent information from each pair of signals. If you suffer from correlation neglect, do not fully incorporate correlation.
- Overshooting / undershooting –> Naivete shows heterogeneity, bi-modal in correlated case
- Beliefs are too sensitive to the ubiquitous “telling and re-telling of stories” and exhibit excessive swings.
Most participants have the computational skills to develop rational beliefs, many approach the problem in a wrong way when the environment is moderately complex.
- Conceptual vs. computational hurdles –> Conceptual is most important element
Thus, experimentally nudging people’s focus towards the correlation and the underlying independent signals has large effects on beliefs.
Enke and Zimmermann (2019) (Review of Econ. Studies)
[Reversed]
Random EU Theory: set of EU, with with given Prob of occuring. Realised EU determines choice vis standard maximisation. Distribution F over set of all EU.
eg. Choices under the influence of alcohol vs sober.
Diagram: pairs of lotteries A C with P = 0.8. Model CAN explain this!
- B > A ==> Red: IC flatter than red line (vice versa)
- A > B ==> Steeper IC than red, carries over in Blue
- In every mood I choose A, I should also choose C: I should shoose C more often than A. Vairation in EU and different realisations, so we can compare only when we have probabilities of things being chosen.
Gul and Pesendorfer (2002)
[Reversed]
Shown a chart of wheat prices and asked bakers vs. farmers
- Over optimising
- Wishful thinking
Mayraz (2011)
Levine (1998)
Interdependent preferences and motives: our care for others depends upon our beliefs about how nice or reciprocal they are.
[Reversed]
Status quo bias: eg. never show a prospective customer more than 3 options for a purchase
Reference points, only choose things strictly better than reference point, which can be damaging missing out on better bundles.
Samuelson and Zeckhauser (1988)
Masatlioglu and Ok (2005)
Allais (1953)
Allais Paradox: 11% and 89% split across different things violate independence.
Augenblick, Niederle and Sprenger (2015) (QJE)
Investigate choices over consumption (real effort) in a longitudinal experiment. Subjects asked to complete boring tasks, changing interest rates gives expected downward slope.
We pair this effort study with a companion monetary discounting study.
- Very limited time inconsistency in monetary choices.
- Considerably more present bias in effort.
Repetitive transcription task and also a tetris game completion, ran over the course of 7 weeks.
Masatlioglu and Raymond (2015)
Chetty, Looney and Kroft (2009)
Masatlioglu and Raymond (2015)
Chetty, Looney and Kroft (2009)
[Reversed]
Generalising standard choice theoretic welfare economics to nonstandard behavioural models.
Unambiguous choice relation: X is always chosen over Y when available.
Nests the standard choice framework when aximos satisfied. Small departures catered for.
Theory that delivers bounds on welfare based purely on choice data.
- Standard model, agents choose an option x from a choice set X –> Policy ID optimal x
- Behavioural models, agents choose from “generalized choice sets” G = (X, d) (henceforth, GCS)
- d is an “ancillary condition”: something that affects choice behaviour but (by assumption) does not affect experienced utility. I Examples: colour of paper, salience, framing, default option
Bernheim and Rangel (2009) (QJE)
Levitt and List (2007)
- Contribution - The real world is less conducive to pro-social behaviour, e.g. Finance: high stakes, anonymity, disregard of future behaviour, factors include population representativeness, scrutiny degree, decision process emphasis, limits on action space
- Summary
Laboratory behaviour is influenced not only by monetary factors but also by these 5 other aspects
- Morals and Ethics
- Nature and Extent of scrutiny (of actions chosen) from others à exaggerates importance of pro-social behaviour (List, 2006), Image of eyes watching subjects can increase generosity (Haley and Fessler 2005)
- Decision context à Ross and Ward (1996), “Community” or “wall Street” language can swing proportions of prisoner’s dilemma cheaters
- Self-selection of individuals making decisions à volunteers typically higher educated,
- Game stakes – high stakes induce more acceptance of unfair offers due to more focus on monetary wealth
[Reversed]
- Contribution - The real world is less conducive to pro-social behaviour, e.g. Finance: high stakes, anonymity, disregard of future behaviour, factors include population representativeness, scrutiny degree, decision process emphasis, limits on action space
- Summary
Laboratory behaviour is influenced not only by monetary factors but also by these 5 other aspects
- Morals and Ethics
- Nature and Extent of scrutiny (of actions chosen) from others à exaggerates importance of pro-social behaviour (List, 2006), Image of eyes watching subjects can increase generosity (Haley and Fessler 2005)
- Decision context à Ross and Ward (1996), “Community” or “wall Street” language can swing proportions of prisoner’s dilemma cheaters
- Self-selection of individuals making decisions à volunteers typically higher educated,
- Game stakes – high stakes induce more acceptance of unfair offers due to more focus on monetary wealth
Levitt and List (2007)
Gul and Pesendorfer (2001)
“Temptation and Self Control”: two period model, ex-ante inferior option is tempting at period 2. Different rankings based on:
- Commitment –> stricly prefer a restricted subset
- Temptation –>
- Cost of self control –> Resist temptation and chooses higher ex-ante utility option
Key additional of Set Betweeness axiom,
A > B ==> A > A U B > B
[Reversed]
“Temptation and Self Control”: two period model, ex-ante inferior option is tempting at period 2. Different rankings based on:
- Commitment –> stricly prefer a restricted subset
- Temptation –>
- Cost of self control –> Resist temptation and chooses higher ex-ante utility option
Key additional of Set Betweeness axiom,
A > B ==> A > A U B > B
Gul and Pesendorfer (2001)
