Developments in 2019-2020 Starmer C.

Non-Expected Utility Theory (1-4.1)

Developments in Non-Expected Utility Theory: The Hunt for a

Descriptive Theory of Choice under Risk (1-4.1)

1. Descriptive Limitations of Expected Utility Theory

With hindsight, it seems that violations of EUT fall under two broad headings: those which have possible
explanations in terms of some "conventional" theory of preferences and those which apparently do not.
There is now a large body of evidence indicating that actual choice behavior may systematically violate the
independence axiom. Journal Example; L1Paper = 50, L2Electro=30: L2> L1, then L3 Paper/Electro=60:
L1<L2<L3. Independence Axiom should be L1<L3<L2

1.1.Consequence Effects
Allais Paradox (Allais, 1953):
Allais expected that people faced with these choices might opt for s1 in the first problem, lured by the
certainty of becoming a millionaire, and select r2 in the second choice where the odds of winning seem very
similar, but the prizes very different. Evidence quickly emerged that many people did respond to these
problems as Allais had predicted.
Example: First, imagine choosing between the two prospects: s1 = ($1M,1) or r1 = ($5M,0.1; $1M, 0.89;
0, 0.01). Then, second problem, between s2 = ($1M,0.11; 0,0.89) or r2 = ($5M,0.1; 0,0.9).
Choice between pair of prospect with the following form:
s = (y,p; c,1-p) and r = (q,p; c,l-p), where q=(x,γ; 0,1-γ) and 0< γ< 1
The payoffs c, x and y are nonnegative (usually monetary) consequences such that x > y. Both prospects s*
and r* give outcome c with probability (1 – p): this is the "Common Consequence" and it is an obvious
implication of the independence axiom of EUT that choices between s* and r* should be independent of
the value of c.
Individuals choose s* when c = y, and r* when c = 0.

1.2.Common Ratio Effects

Example: Choice between $3000 for sure, or entering a gamble with an 80 percent chance of getting $4000
(otherwise nothing). What would you choose? Now think about what you would do if you had to choose
either a 25 percent chance of gaining $3000 or a 20 percent chance of gaining $4000. A good deal of
evidence suggests that many people would opt for the certainty of $3000 in the first choice and opt for the
20 percent chance of $4000 in the second.
This phenome-non is observed in choices among pairs of problems with the following form:
s = (y, p; 0, 1-p) and r = (x, γp; 0, 1 - γp), where x > y
Assume that the ratio of "winning" probabilities (γ) is constant, then for pairs of prospects of this structure,
EUT implies that preferences should not depend on the value of p, yet numerous studies reveal a tendency
for individuals to switch their choice from s** to r** as p falls.
Developments in 2019-2020 Starmer C.
Non-Expected Utility Theory (1-4.1)

1.3.Conventional Theories:
The general spirit of the approach is to seek "well behaved" theories of preference consistent with observed
violations of independence.
Assumptions:
Procedure Invariance: Preferences over prospects are independent of the method used to elicit them.
Description Invariance: Preferences over prospects are purely a function of the probability distributions of
consequences implied by prospects and do not depend on how those given distributions are described.

Failure of Procedure Invariance - Preference Reversal:

Example: The first task , a choice between two prospects: one prospect ($-bet) offers a small chance of
winning a "good" prize; the other (P-bet) offers a larger chance of winning a smaller prize. The second
task, assignment of monetary values (usually minimum selling prices denoted M($) and M(P)) to the two
prospects.
Preference Reversal Phenomenon: Individuals choose the P-bet (i.e., reveal P > $) while placing a higher
value on the $-bet (i.e., M($) > M(P)).
Choice and valuation tasks may invoke different mental processes which in turn generate different orderings
of a given pair of prospects. Consequently, the rankings observed in choice and valuation tasks cannot be
explained with reference to a single preference ordering.
Interpretation explains Preference Reversal as a failure of transitivity:
(M($) ~ $; M(P) ~ P) ≠ (P > $ ~ M($) > M(P) ~ P)

Failures of Description Invariance:

There is also widespread evidence that very minor changes in the presentation or "framing" of prospects
can have dramatic impacts upon the choices of decision makers.
Example: Read p. 339 example with US Asian disease
The two pairs of options are stochastically equivalent. The only difference is that the group I description
presents the information in terms of lives saved while the information presented to group II is in terms of
lives lost. (72% prefer Option A over B / 22% prefer C over D).
Observations:
1. Framing effects have a bearing on issues of genuine economic relevance
2. Cannot be reduced, as other theories, to a single preference function V (.) defined over individual
prospects.
Developments in 2019-2020 Starmer C.
Non-Expected Utility Theory (1-4.1)

2. Non-Expected Utility Theories

2.1 The Conventional Strategy

Consider the class of prospects defined over three outcomes x1, x2, x3 such that x1 < x2 < x3. Since any such
prospects can be described as a vector of probabilities (p1, 1-p1-p3, p3) we can also locate them, graphically,
in two-dimensional probability space. The addition of the Independence Axiom of EUT restricts the set
of indifference curves to being upward sloping (left to right), linear and parallel.
In EUT, the slope of the indifference curves reflects attitude to risk and may vary between individuals:
The more risk averse the individual, the steeper the slope of their indifference curves.
On the Left Figure: Person 1 has indifference
curves with the slope of the dashed line (hence
s to r); Person 2 has indifference curves with
the slope of the solid line (hence s to r').
Person 2 can be seen to be the more risk
averse in the sense that, as we move
northwest along the hypotenuse, relative to
person 1, we must give her a higher chance of
winning the best outcome in the riskier
prospect in order to generate indifference with
the safe prospect s.

1. Allais Paradox and Consequence Effect: Indifference curves could have a steeper slope than the lines
connecting prospects, in which case s1 > r1 and s2 > r2. People often violate EUT, revealing s1 > r1 in the
left-hand problem, r2 > s2 in the right-hand problem. Relative to the predictions of EUT, in choosing r2 over
s2 these people are being more risk seeking than they should be, given their choice of s1 over r1.
2. Common Ratio Effect (Right Figure), the pair of prospects {s1**, r1**}, near the left edge of the triangle,
correspond with the Common Ratio problems where p = 1. As p falls, we generate pairs of prospects like
{s2**, r2**}, located on parallel lines further to the right in the triangle. Assuming EUT preferences, an
individual must either prefer the "safer option" in both choices or the "riskier option" in both choices, yet
many people choose s1** over r1** and r2** over s2**. Relative to the predictions of EUT, there is an
"inconsistency" in the risk attitudes revealed across their choices.
The Predictions of EUT: Choices between prospects located in the bottom right-hand corner appear more
risk prone than should be expected given preferences revealed for choices located leftwards and/or upwards
in the triangle.
The Violations of EUT: Indifference curves determine preferences over pairs of prospects located near the
right-hand corner of a given triangle, say {s2**, r2**} will need to be relatively flat (reflecting more risk-prone
behavior), compared with indifference curves determining choices over pairs of prospects, like {s1**, r1**},
near to the left- hand edge of the triangle.
Developments in 2019-2020 Starmer C.
Non-Expected Utility Theory (1-4.1)

2.2 The “Fanning-out” Hypothesis

Machina proposed an analytical extension of EUT with a hypothesis on the shape of non-expected utility
indifference curves,
where V(q) = Σ U(xi). pi, the utility values U(xi) = d.V(q)/d.pi are the probability derivatives of V(.)

Given the existence of phenomena like the Common Ratio and Common Consequence Effects, Machina
hypothesized that the local utility functions U(.;q) become more concave as we move from (first order)
stochastically dominated to stochastically dominating distribution.
Hypothesis II: Hypothesis that implies a tendency for agents to become
more risk averse as the prospects they face get better; in the context
of the triangle, it means that indifference curves become steeper, or "fan
out" as we move northwest.
Since indifference curves are relatively steeply sloped in the
neighborhood of prospect m, m lies on a higher indifference curve than
q or r. Flatter indifference curves in the bottom right-hand corner of the
triangle are such that t lies on a higher indifference curve than s. A
common consequence effect (e.g. m > q and t > s) and a common ratio
effect (e.g. m > r and t > s).

2.2.1 Weighted Utility Theory:

V(q) = [Σpi . g(xi) . u(xi)] / [Σpi . g(xi)]
where u(.) and g(.) are two different functions assigning non-zero
weights to all consequences. The model incorporates EUT as the
special case in which the weights assigned by g(.) are identical for
every consequence.
Weakened form of the independence axiom constrains indifference
curves to be linear without requiring them to be parallel. The location
of this point, which could lie inside or outside of the triangle boundary,
depends upon the specifications of the functions u(.) and g(.).

2.2.2 Theory of Disappointment:

V(q) = Σi pi [u(xi) + D(u(xi)-U]
where u(xi) is interpreted as a measure of "basic" utility (that is, the utility of xi, considered in isolation
from the. other consequences of q) and U is a measure of the "prior expectation" of the utility from the
prospect.
The model assumes that if the outcome of a prospect is worse than expected (i.e., if u(xi) < U) a sense of
“disappointment” will be generated. On the other hand, an outcome better than expected will stimulate
"elation." With D(.) = 0, the model reduces to EUT. This additional function, however, is intended to
capture a particular intuition about human psychology: that people dislike disappointment and so act to
avoid it. More specifically, this is captured by assuming that agents are "Disappointment Averse" (D(h) is
concave for h < 0) and "Elation Prone" (D(.) is convex for h > 0).
Developments in 2019-2020 Starmer C.
Non-Expected Utility Theory (1-4.1)

2.2.3 Axiom of Betweenness:

if q > r, then q > (q,p; r,(1-p)) > r for all p < 1
Betweenness implies that any probability mixture of two lotteries will be ranked between them in terms of
preference and, given continuity, an individual will be indifferent to randomization among equally valued
prospects.
Left: Quasi-Convex Preference. No Randomization

Right: Quasi-concave Preference. Randomization

With linear indifference curves, the individual

indifferent between q and r is also indifferent to
randomization between them. Once betweenness
is relaxed, this indifference to randomization no
longer holds and two important cases can be
distinguished: Quasi-Convex Preferences and
Quasi-Concave Preferences.

Quasi-Convex Preference: if for every q ≠ r, V(q,p; r,(1 - p)) < max[V(q) ,V(r)] for all p. When
preferences are quasi-convex, indifference curves are concave and consequently the individual will be
averse to randomization among equally valued prospects.
Quasi-Concave Preference: if for every q ≠ r, V(q,p; r,(1 - p)) < max[V(q) ,V(r)] for all p. When
preferences are quasi-concave, indifference curves are convex and consequently the individual will prefer
to randomize among equally valued prospects.

2.2.4 Axiom that do not impose betweenness:

Quadratic Utility Theory:
if q ~ r then (q, p; r, (1 -p)) ~ (q, (1 -p); r, p).
In this model, indifference curves may switch from concave to convex (or vice versa) as we move across
the triangle. This theory relies on a weakened form of betweenness called Mixture Symmetry.
Each of the few last fanning hypothesis operates by assigning subjective weights/utilities to consequences.
The value assigned to any given prospect is then determined by some function that combines these utilities
with objective probabilities.

2.3 Theories with Decisions Weights

Another variant of the conventional strategy involves the use of probability transformation functions which
convert objective probabilities into subjective decision weights. An important feature of these models is
that, excepting special cases, betweenness does not hold.

2.3.1 Decision Weight Form

Individuals tend to underestimate data from relative frequent causes, and to overestimate others data due to
relatively infrequent causes. These effects might be revealing misperception of objective probabilities or a
tendency for individuals to subjectively weight objective probabilities.
V(q) = Σi wi . u(xi)
Developments in 2019-2020 Starmer C.
Non-Expected Utility Theory (1-4.1)

Subjective Expected Value: Consequences are treated in the way that probabilities are handled in the
standard theory and enter "raw" with u(xi) =xi for all i.
In the version presented by Handa, the decision weight attached to each outcome is determined by a
probability weighting function π (pi) which transforms the individual probabilities of each consequence
directly into weights. As in most theories that incorporate probability weights, π (.) is assumed to be in-
creasing with π (1) = 1 and π (O) = 0, and I will retain these assumptions from now on.

2.3.2 Simple Decision Weighted Utility

Individuals are assumed to maximize the function: V(q) =Σ π (pi) . u(xi)
Both this and subjective expected value transform the probabilities of individual consequences directly into
weights (i.e., wi = π (pi)), have the property that V(q) will not generally satisfy Monotonicity.
Example: π (.) is convex, then apart from the extremes of the probability scale, π (p) + π (l -p) < 1 and
there will be some ε > 0 such that gambles of the form (x, p; x + ε, 1-p) will be rejected in favor of (x,1)
even though they stochastically dominate the sure option.
A similar argument applies for any departure from linearity, and the only way to ensure general
monotonicity in this type of theory is to set decision weights equal to objective probabilities (i.e., Wi = π
(pi) = pi for all i) in which case the theory reduces to EUT.

2.3.3 Rank-Dependent Expected Utility Theory

The conventional route have proposed decision weighting models with more sophisticated probability
transformations designed to ensure monotonicity of V(.).
In the rank-dependent expected utility theory, the weight attached to any consequence of a prospect depends
not only on the true probability of that consequence but also on its ranking relative to the other outcomes
of the prospect.
With consequences indexed as before such that xi is worst and xn best, we can state rank dependent expected
utility theory as the hypothesis that agents maximize the decision weighted form with weights for i = 1,...,
n - 1 given by:
Wi= π (pi+…+pn) - π (pi+1+… +pn)
and Wi = π (pi) for i = n

In this model there is a meaningful distinction between Decision Weights (W) and Probability Weights (π).
Notice that π (pi +…+ pn) is a subjective weight attached to the probability of getting a consequence of xi
or better, and π (pi+1 +…+ pn) is a weight attached to the probability of getting a consequence better than
xi, so, π (.) is a transformation on cumulative probabilities. Ensure that V(.) is Monotonic.
A less appealing feature of the model is that a small change in the value of some outcome of a prospect can
have a dramatic effect on its decision weight if the change affects the rank order of the consequence. A
change in the value of an outcome, no matter how large the change, can have no effect on the decision
weight if it does not alter its rank.
 Developments in 2019-2020 Starmer C.
Non-Expected Utility Theory (1-4.1)

The predictions of the rank-dependent model depend crucially on the form of π (.). If π (.) is convex, this
generates a set of concave indifference curves (implying aversion to randomization) which are parallel at
the hypotenuse but fan out as we move left to right across the triangle and fan in (i.e., become less steep)
as we move vertically upwards.
Curvature of π(.) in the rank-dependent model has been interpreted as reflecting "optimism" and/or
"pessimism" with respect to probabilities. Hence the weight attached to the lower ranking consequence, x1,
will be higher than the weight attached to the larger consequence. Pessimism = Risk Aversion
The early evidence of EUT violation can be explained either by
assuming a simple convex π (.)
One possibility is the function which has π (p) = p for a unique value
of p = p*; it is concave below p* and convex above it, hence "low"
probabilities (below p*) are over weighted.
The essential difference between EUT and rank-dependent expected
utility is that the latter theory relies on a weakened form of
independence called Co-monotonic Independence.
Co-monotonic Independence: It asserts that preferences between
prospects will be unaffected by substitution of Common
Consequences so long as these substitutions have no effect on the rank
order of the outcomes in either prospect.

Powered by TCPDF (www.tcpdf.org)

 Psychological Aspects 2019-2020 Paul J. H. Schoemaker

Psychological Aspects

1. Decomposed Fashion
Each alternative being assigned a separate level of utility. In the first type, alternatives are compared against
a preset standard and eliminated if they fail to measure up.
The model is conjunctive, if it is required that all attributes meet certain minimum standards.
The model is disjunctive, if satisfaction of at least one dimensional standard is sufficient.
For example, Melvin W. Reder (1947) found that investment projects are often eliminated from
consideration because the probability of ruin exceeds some critical level (i.e., a disjunctive rejection rule).
In the second type, no preset standards are used; instead the alternatives are compared directly (in a
decomposed manner). One such approach is elimination-by-aspects, a type of lexicographic model.
Another lexicographic model was proposed by Payne and Braunstein (1971) in their study of pairs of
duplex gambles. Subjects first compared probabilities. If the difference was sufficiently large a choice
would be made on this basis alone. If not, dollar dimensions would be considered. (P-bet and $-bet)

2. Task Complexity
When alternatives and dimensions are numerous, subjects tend to use conjunctive and lexicographic models
as initial screening rules. With fewer attributes, holistic evaluations are more probable. However, with just
two alternatives and many different attributes the additive difference model is predominant.
Interesting evidence comes from studies on decision time and task complexity by Charles Kiesler.
For example, Charles Kiesler (1966) examined how much time children took in choosing one piece of candy
from four alternatives. Surprisingly, the decision times were shorter when all four pieces were about equally
attractive than when two were attractive and two unattractive!
This finding suggests a greater motivation to be optimal with simple than with complex choices.

3. Isolation
Isolation proved some violations of the first-order stochastic dominance. For instance, a decision
concerning a particular insurance policy would, in theory, depend on all other risks one faces. However,
people may not approach decisions in this comprehensive way, and simply treat problems in isolation.
Another example they discuss concerns the treatment of a $20 loss. Say that in one scenario you purchase
a $20 theater ticket, which you then lose while waiting in the lobby. Would you buy a new ticket? In the
other scenario you are about to purchase this same theater ticket, but upon opening your wallet you
discover that $20 is missing. Would you still buy the ticket?
Because of the way people partition decision contexts, the $20 loss seems less relevant to the second than
to the first choice. Similar isolation effects are demonstrated in prospect theory (Kahneman and Tversky,
1979) concerning windfall profits, and in Shoemaker (1980) with insurance.
 Psychological Aspects 2019-2020 Paul J. H. Schoemaker

4. Reference Points and Aspiration Levels

Although EU theory suggests that alternatives are evaluated with respect to their effects on final wealth
levels, it is cognitively easier to assess options in terms of gains and losses relative to some reference point.
This target point will often be the status quo, but might also be a target return level, or an aspired future
wealth position. Importantly, risk-taking attitudes are likely to be quite different above versus below this
reference point.
For instance, in prospect theory v(x) is assumed to be convex for losses and concave for gains. In risk-
return models, risk is often related to the probability and consequences of failing to meet a target return
level.
An experimental test of aspiration level effects was offered by Payne who examined binary choices of
students and managers between three-outcome gambles whose expected values were equal. In each pair,
the outcomes of the second gamble all fell within the range of the first. By adding a constant amount to all
outcomes, the pairs were translated through the origin to different degrees.
Such translations commonly induced changes in preferences, especially if they were such that in one
gamble all outcomes were positive (or negative) while in the other gamble there were mixed outcomes.
Another example is about the $5 money experience p. 550 about percent savings, that instead of looking
at final asset positions, people's reference dimension is percent savings. The price variance increased
markedly with price level, but was fairly constant when expressed as a percentage of price.
Although general behavior could be rationalized through income effects and transaction costs, it probably
involves a psychological difference between opportunity costs and actual out-of-pocket costs.

5. The Psychology of Probability Judgements

5.1 Subjective probabilities relate non-linearly to objectives ones
Typically, the subjective probability curves overweigh low probabilities and underweigh high one.
For example, Menahem Yaari (1965) explained the acceptance of actuarially unfair gambles as resulting
from optimism regarding low-probability events, as opposed to convexities in the utility function.
Slovic found that subjective probabilities tend to be higher as outcomes become more desirable.
Furthermore, subjective "probabilities" often violate mathematical properties of probability. In such cases
we refer to them as decision weights w(p).

5.2 People are generally overconfident

Many biases in subjective probability judgments stem from the type of heuristics (i.e., simplifying rules)
people use to estimate relative likelihoods.
For instance, a doctor may judge the likelihood of a patient having disease A versus B solely on the basis
of the similarity of the symptoms to textbook stereotypes of these diseases.
This so-called representativeness heuristic, however, ignores possible differences in the a priori
probabilities of these diseases.
 Psychological Aspects 2019-2020 Paul J. H. Schoemaker

Another common heuristic is estimation on the basis of availability. In judging the chances of dying from
a car accident versus lung cancer, people may base their estimates solely on the frequencies with which
they hear of them. Due to unrepresentative news coverage in favor of car accidents, these estimates will be
systematically biased.
Finally, people suffer from a serious hindsight bias resulting in suboptimal learning. Events that did (not)
happen appear in retrospect more (less) likely than they did before the outcome was known. Fischhoff
(1975) explains this phenomenon as due to reconstructive memory.

5.3 Bayesian Interference

Two somewhat opposite phenomena are encountered depending on the context.
One is conservatism (Edwards, 1968) according to which new information is underweighted in the revision
of opinions. It reflects an anchoring onto the old information with insufficient adjustment in the direction
of the new information.
The other phenomenon involves ignorance of stated prior probabilities, similar to the earlier example of
disease A vs. B. One psychological explanation is that people's heuristic for assessing the relevance of
information is based on its causal relationship.
Example: This would explain why most people consider the probability of a daughter having blue eyes,
given that her mother has blue eyes, to be higher than the probability of a mother having blue eyes, given
that her daughter has blue eyes. However, assuming that successive generations have the same incidence
of blue eyes, these probabilities are the same.

Powered by TCPDF (www.tcpdf.org)

Doing Now or Later 2019-2020 O’DONOGHUE

Doing It Now or Later

1. Present-Biased Preference
Let ut be a person’s instantaneous utility in period t. A person in period t cares not only about her present
instantaneous utility, but also about her future instantaneous utilities. We let Ut (ut , ut+1 , ... , uT ) represent
a person’s intertemporal preferences from the perspective of period t, where Ut is continuous and increasing
in all components.
The standard simple model employed by economists is exponential discounting: For all t, Ut (ut , ut+1 , ... ,
uT ) ≡ ∑ δt ut , where δ C (0, 1] is a ‘‘discount factor.’’ .Exponential discounting parsimoniously captures
the fact that people are impatient.
Time-Consistent: A person’s relative preference for well-being at an earlier date over a later date is the
same no matter when she is asked.
Present-Biased Preferences: When considering trade-offs between two future moments, present-biased
preferences give stronger relative weight to the earlier moment as it gets closer.

Definition 1: (ꞵ, δ) -preferences are preferences that can be represented by:

In this model, δ represents long-run, time consistent

discounting. The parameter ꞵ, on the other hand,
represents a ‘‘bias for the present’’—how you favor now
versus later. If ꞵ=1, then (ꞵ, δ) -preferences are simply
exponential discounting. But ꞵ<1 implies present-biased
preferences: The person gives more relative weight to
period ŧ in period ŧ than she did in any period prior to
period ŧ.

A sophisticated person know exactly what her future selves’ preferences will be.
A naive person believe her future selves’ preferences will be identical to her current self ’s, not realizing
that as she gets closer to executing decisions her tastes will have changed. If you were naive, you would
never worry that your tomorrow self might choose an option you do not like today.

2. Doing It Once
Suppose there is an activity that a person must perform exactly once, and there are T < ∞ periods in which
she can do it. Let v ≡ (υ1 , υ2 , ... , υT ) be the reward schedule, and let c ≡ (c1 , c2 , ... , cT ) be the cost
schedule, where υT ≥ 0 and ct ≥ 0 for each t ∈ {1, 2, ... , T}.
In each period t ≤ T - 1, the person must choose either to do it or to wait. If she does the activity in period
t, she receives reward υT but incurs cost ct , and makes no further choices. If she waits, she then will face
the same choice in period t + 1. Importantly, if the person waits she cannot commit in period t to when later
she will do it. If the person waits until period T, she must do it then.
Doing Now or Later 2019-2020 O’DONOGHUE

The reward schedule υ and the cost schedule c represent rewards and costs as a function of when the person
does the activity. However, the person does not necessarily receive the rewards and costs immediately upon
completion of the activity.
Some activities, such as writing a paper or mowing the lawn, are unpleasant to perform, but create future
benefits. We refer to activities where the cost is incurred immediately while the reward is delayed as
activities having immediate costs. Other activities, such as seeing a movie or taking a vacation, are
pleasurable to perform, but may create future costs. We refer to activities where the reward is received
immediately while the cost is delayed as activities having immediate rewards.

A person’s intertemporal utility: The activity in period ŧ ≥ t, which we denote by U’ (ŧ).

We refer to people with standard exponential, time-consistent preferences (i.e., ꞵ=1) as TCs. We then focus
on two types of people with present-biased preferences (i.e., ꞵ<1), representing the two extremes. We call
people with sophisticated perceptions sophisticates, and people with naive perceptions naïfs. Sophisticates
and nails have identical preferences (throughout we assume they have the same b), and therefore differ only
in their perceptions of future preferences.
A person’s behavior can be fully described by a strategy s ≡ (s1 , s2 , ... , sT ), where st ∈ {Y, N} specifies
for period t ∈ {1, 2, ... , T} whether or not to do the activity in period t given she has not yet done it. The
strategy s specifies doing it in period t if st = Y, and waiting if st = N. In addition to specifying when the
person will actually complete the activity, a strategy also specifies what the person ‘‘would’’ do in periods
after she has already done it; e.g., if st = Y, we still specify st, for all t* > t. This feature will prove useful
in our analysis. Since the person must do it in period T if she has not yet done it, without loss of generality
we require sT = Y.
A perception-perfect strategy: A strategy that in all periods (even those after the activity is performed ) a
person chooses the optimal action given her current preferences and her perceptions of future behavior.
Reflecting the fact that TCs do not have a self-control
problem, Definition 2 says that in any period, TCs will
complete the activity if and only if it is the optimal period
of those remaining given her current preferences.
Naifs have present-biased preferences (since b < 1 ), but
naifs believe that they are time-consistent. As a result,
the decision process for naifs is identical to that for TCs.
Definition 3 says that in any period, naifs will complete
the activity if and only if it is the optimal period of those
remaining given her current preferences.
Doing Now or Later 2019-2020 O’DONOGHUE

Sophisticates also have present-biased preferences and

a self-control problem. But unlike naifs, sophisticates
know they will have self-control problems in the future,
and therefore correctly predict future behavior.
Definition 4 says that in period t, sophisticates calculate
when their future selves will complete the activity if they wait now, and then do the activity now if and only
if given their current preferences doing it now is preferred to waiting for their future selves to do it.
Note that in Definitions 2, 3, and 4, we have assumed that people do it when indifferent, which implies that
there is a unique perception-perfect strategy for each type. In addition, this assumption implies that a
perception-perfect strategy must be a pure strategy. For generic values of υ, c, and ꞵ, nobody will ever be
indifferent, so these assumptions are irrelevant.

3. Behavior

3.1. Immediate Cost and Reward

We compare the behavior of TCs, naifs, and sophisticates who have identical long-run preferences.
Comparing naifs or sophisticates to TCs reflects how people with present-biased preferences behave
relative to how they would like to behave from a long run perspective; and comparing sophisticates to naifs
reflects the implications of sophistication about self-control problems.

Immediate Cost - Importance to read the Example 1, page 109.

TCs: They always do the activity in the period t that maximizes υt - ct.
Naifs: They incorrectly predict that they will not procrastinate in the future, and consequently underestimate
the cost of procrastinating now.
Sophisticates: They have a tendency to procrastinate (they do not write the report right away, which their
long-run selves prefer ), perfect foresight can mitigate this problem because sophisticates will do it now
when they (correctly) foresee costly procrastination in the future.

Immediate Reward - Importance to read the Example 2, page 110.

TCs: They wait and see the Depp movie since it yields the highest reward.
Naifs: For activities with immediate rewards, the self-control problem leads naifs to do the activity too
soon.
Sophisticates: They have even worse self-control problems in this situation. Knowing about future self-
control problems can lead you to give in to them today, because you realize you will give in to them
tomorrow.
Doing Now or Later 2019-2020 O’DONOGHUE

3.2. Present-Bias Effect & Sophistication Effect

The Present-bias effect: When costs are immediate people with present-biased preferences tend to
procrastinate (wait when they should do it) while when rewards are immediate they tend to preproperate
(do it when they should wait).
For immediate costs, they wait in periods where they should do it because they want to avoid the immediate
cost. For immediate rewards, they do it in periods where they should wait because they want the immediate
reward now.
Proposition 1 captures that naifs are influenced solely by the present-bias effect. For immediate costs naifs
always procrastinate, and for immediate rewards naifs always preproperate.

PROPOSITION 1: (1) If costs are immediate, then ŧn ≥ ŧtc

(2) If rewards are immediate, then ŧn ≤ ŧtc

Proposition 1 is as simple as it seems: Naifs believe they will behave like TCs in the future but are more
impatient now. Hence, the qualitative behavior of naifs relative to TCs intuitively and solely reflects the
present-bias effect.

The Sophistication Effect: Sophisticates are fully aware of any self-control problems they might have in
the future, and this awareness can influence behavior now. The sophistication effect is captured in
comparisons of sophisticates to naifs.
In our one-activity model, the sophistication effect is straightforward: Because sophisticates are (correctly)
pessimistic that they will behave themselves in the future, they are more inclined than naifs to do it now,
irrespective of whether it is costs, rewards, or both that are immediate.

PROPOSITION 2: For all cases, ŧn ≤ ŧtc

For immediate costs, that sophisticates do it before naifs reflects that sophistication helps mitigate the
tendency to procrastinate. For immediate rewards, that sophisticates do it before naifs reflects that
sophistication can exacerbate the tendency to preproperate.

Immediate Cost (and Reward) - Importance to read the Example 3, page 112.
TCs: They write the paper on Saturday because the marginal benefit of a better paper outweighs the
marginal cost of giving up college football for pro basketball
Sophisticates: Since the example involves immediate costs, the present-bias effect suggests that
sophisticates should procrastinate. However, they behave exactly opposite from what present-biased
preferences would suggest. Indeed, when rewards are immediate, sophisticates always preproperate
because the sophistication effect exacerbates the self-control problem.
Doing Now or Later 2019-2020 O’DONOGHUE

4. Welfare
Welfare comparisons for people with time inconsistent preferences are in principle problematic; The very
premise of the model is that a person’s preferences at different times disagree, so that a change in behavior
may make some selves better off while making other selves worse off.
Since present-biased preferences are often meant to capture self-control problems, where people pursue
immediate gratification on a day-to-day basis, we feel the natural perspective in most situations is the
‘‘long-run perspective’’.
To formalize the long-run perspective, we suppose there is a (fictitious ) period 0 where the person has no
decision to make and weights all future periods equally. We can then denote a person’s long-run utility
from doing it in period ŧ by U0 (ŧ) ≡ υt - ct.

When costs are immediate, sophistication is good because it mitigates the tendency to procrastinate.
Indeed, it is straightforward to show that when costs are immediate, sophisticates always do at least as well
as naifs [i.e., U0 (ŧs) ≥ U0 (ŧn) ]. Intuitively, since sophisticates never procrastinate in a period where naifs
do it, the only way their utilities can differ is when sophisticates preempt costly procrastination.
When rewards are immediate, on the other hand, implied that sophistication is bad because it exacerbates
the tendency to preproperate. More severe preproperation will often lead to lower long-run utility.
Since sophisticates, naifs, and TCs have identical long-run utility, we can measure the welfare loss from
self-control problems by the deviation from TC long-run utility [i.e., U0 (ŧtc) - U0 (ŧs) and U0 (ŧtc ) - U0 (ŧn)].

Immediate costs, where the self-control problem leads you to procrastinate.

Naifs: They can compound self-control problems by making repeated decisions to procrastinate, each time
believing they will do it next period. With each decision to procrastinate, they incur a small welfare loss,
but the total welfare loss is the sum of these increments. No matter how small the individual welfare losses,
naifs can suffer severe welfare losses if they procrastinate enough times. (Proposition 3)

Sophisticates: They know exactly when they will do it if they wait, so delaying from period ŧtc to period ŧs
is a single decision to procrastinate. Hence, for sophisticates small self-control problems cannot cause
severe welfare losses. (Proposition 3)
 Doing Now or Later 2019-2020 O’DONOGHUE

Immediate Reward, where the self-control problem leads you to preproperate.

Naifs: They believe that if they wait they will do it when TCs do it, so doing it in period ŧn as opposed to
waiting until period ŧtc is a single decision to preproperate for naifs. Hence, for naifs small self-control
problems cannot cause severe welfare losses. (Proposition 4)
Sophisticates: They can compound self-control problems because of an unwinding: In the end,
sophisticates will preproperate; because they realize this, near the end they will preproperate; realizing this
they preproperate a little sooner, etc. For each step of this unwinding, the welfare loss may be small, but
the total welfare loss is the sum of multiple steps. (Proposition 4)
As with naifs and immediate costs, no matter how small the individual welfare losses, sophisticates can
suffer severe welfare losses if the unraveling occurs over enough periods.

5. Smoking Guns

Commitment Devices: External device that limit future choice sets, because the use of such devices
provides smoking guns that prove time consistency wrong.
Dominance: First property that claims that one strategy
dominates another if it yields in every period an
instantaneous utility at least as large as the instantaneous
utility from the other strategy, and strictly larger for
some periods.
In our model, one strategy is dominated by another if and only if the first strategy implies doing it at a cost
with no reward while the second strategy implies doing it for a reward with no cost.
Time-consistent Behavior: The second property that a
person with time-consistent behavior will never violate
is independence of irrelevant alternatives, eliminating
an option from the choice set that is not chosen should
not change the person’s choice from the remaining
options.
A time-consistent person will never violate dominance nor independence of irrelevant alternatives. These
results hold for any time-consistent preferences, including time-consistent preferences that discount
differently from period to period, and even time-consistent preferences that are not additively separable.
Proposition 5 establishes that these results do not hold for people with present-biased preferences.
Sophisticates violate dominance when they choose a dominated
early time to do an activity because they (correctly ) worry that
their future selves will not choose the dominating later time.
Naifs can violate dominance because of incorrect perceptions
about future behavior.

Powered by TCPDF (www.tcpdf.org)

Behavioral Economics 2019-2020 Fehr & Schmidt

A Theory of Fairness, Competition and Cooperation

1. A simple model of Inequity Aversion

1.1 Introduction

Inequity averse individual: Individual that dislikes outcomes that are perceived as inequitable. This
definition raises, of course, the difficult question of how individuals measure or perceive the fairness of
outcomes.
Fairness judgments are inevitably based on a kind of neutral reference outcome.
The reference outcome that is used to evaluate a given situation is itself the product of complicated social
comparison processes.
One key insight of this literature is that relative material payoffs affect people's well-being and behavior.
There is, moreover, direct empirical evidence for the importance of relative payoffs. Agell and Lundborg
show that relative payoff considerations constitute an important constraint for the internal wage structure
of firms.
Subjects exhibit a strong and robust aversion against disadvantageous inequality: for a given own income
xi, subjects rank outcomes in which a comparison person earns more than xi substantially lower than an
outcome with equal material payoffs.

1.2 Utility Function of a Reference Group

The subjects enter the laboratory as equals, they do not know anything about each other, and they are
allocated to different roles in the experiment at random. Thus, it is natural to assume that the reference
group is simply the set of subjects playing against each other and that the reference point, i.e., the equitable
outcome, is given by the egalitarian outcome.
First, in addition to purely selfish subjects, there are subjects who dislike inequitable outcomes. They
experience inequity if they are worse off in material terms than the other players in the experiment, and
they also feel inequity if they are better off. Second, however, we assume that, in general, subjects suffer
more from inequity that is to their material disadvantage than from inequity that is to their material
advantage. Formally, consider a set of n players indexed by i C (1, . .. , n), and let x = x1, . . . , Xn denote
the vector of monetary payoffs.

Where we assumed that ꞵi ≤ αi and 0 ≤ ꞵi ≤ 1. In the two-player case simplifies to:

The second term measures the utility loss from disadvantageous inequality, while the third term measures
the loss from advantageous inequality.
Behavioral Economics 2019-2020 Fehr & Schmidt

This figures illustrates the utility of player i as

a function of xj for a given income xi.
Given his own monetary payoff xi, player i's
utility function obtains a maximum at xj = xi.
The utility loss from disadvantageous
inequality (xj > xi) is larger than the utility loss
if player i is better off than player j(xj <xi).

To evaluate the implications of this utility function, let us start with the two-player case. For simplicity, we
assume that the utility function is linear in inequality aversion as well as in xi. This implies that the marginal
rate of substitution between monetary income and inequality is constant.
Note that ꞵi ≤ αi essentially means that a subject is loss averse in social comparisons: negative deviations
from the reference outcome count more than positive deviations.
ꞵi ≥ 0 means that we rule out the existence of subjects who like to be better off than others. We impose this
assumption here, although we believe that there are subjects with ꞵi < 0. The reason is that in the context
of the experiments we consider individuals with ꞵi < 0 have virtually no impact on equilibrium behavior.
ꞵi < 1, suppose that player i has a higher monetary payoff than player j. In this case ꞵi = 0.5 implies that
player i is just indifferent between keeping one dollar to himself and giving this dollar to player j. If ꞵi = 1,
then player i is prepared to throw away one dollar in order to reduce his advantage relative to player j
which seems very implausible.
Example: To see this, suppose that player i has a lower monetary payoff than player j. In this case player i
is prepared to give up one dollar of his own monetary payoff if this reduces the payoff of his opponent by
(1 + αi)/αi dollars. For example, if αi = 4, then player i is willing to give up one dollar if this reduces the
payoff of his opponent by 1.25 dollars.
If there are n > 2 players, player i compares his income with all other n - 1 players. In this case the
disutility from inequality has been normalized by dividing the second and third term by n - 1. This
normalization is necessary to make sure that the relative impact of inequality aversion on player i's total
payoff is indepen- dent of the number of players.

2. Ultimatum and Market Games

2.1. The Ultimatum Game
For definition of “Ultimatum Game”, see previous tutorial – Too lazy to rewrite everything
There is a unique subgame perfect equilibrium in which the proposer offers s = 0, which is accepted by the
responder. If there is a smallest money unit ε, then there exists a second subgame perfect equilibrium in
which the responder accepts any s C [ε,1] and rejects, s = 0 while the proposer offers ε.
Behavioral Economics 2019-2020 Fehr & Schmidt

PROPOSITION 1
Proposer Preference: α1, ꞵ1 Responder Preference: α2, ꞵ2

It is a dominant strategy for the responder to accept any offer s ≥ 0.5, to reject s if
and to accept s > s'(α2).
If the proposer knows the preferences of the responder,

If the proposer does not know the preferences of the responder but believes that α2 is distributed
according to the cumulative distribution function F(α2).

The optimal offer of the proposer,

If ꞵ1 > 0.5, his utility is strictly increasing in s for all s ≤ 0.5. This is the case where the proposer prefers to
share his resources rather than to maximize his own monetary payoff, so he will offer s = 0.5.
If ꞵ1 = 0.5, he is just indifferent between giving one dollar to the responder and keeping it to himself; i.e.,
he is indifferent between all offers
If ꞵ1 < 0.5, the proposer would like to increase his monetary payoff at the expense of the responder.
However, he is constrained by the responder's acceptance threshold. If the proposer is perfectly informed
about the responder's preferences, he will simply offer s'(α2). If the proposer is imperfectly informed about
the responder's type, then the probability of acceptance is F(s/(l - 2s)).
The probability of acceptance, F(s/(l - 2s)), is increasing in s for s'<s'(α)<0.5. Note also that the
acceptance threshold s'(α2)= α2/(l + 2α2) is nonlinear and has some intuitively appealing properties. It is
increasing and strictly concave in α2, and it converges to 0.5 if α2 -> ∞. Furthermore, relatively small values
of α2 already yield relatively large thresholds.

2.2. Market Game with Proposer Competition

For definition of “Market Game”, see previous tutorial – Too lazy to rewrite everything
There is a unique subgame perfect equilibrium outcome in which at least two proposers make an offer of
one, and the responder reaps all gains from trade.
Behavioral Economics 2019-2020 Fehr & Schmidt

PROPOSITION 2
Suppose that the utility functions of the players are given by the utility function of any player i. For any
parameters (αi, ꞵi), i C (1, ..., n), there is a unique subgame perfect equilibrium outcome in which at least
two proposers offer s = 1 which is accepted by the responder.
The responder must accept any s ≥ 0.5. Suppose that he rejects a "low" offer s < 0.5. This cannot happen
on the equilibrium path either since in this case proposer i can improve his payoff by offering si = 0.5 which
is accepted with probability 1 and gives him a strictly higher payoff. Hence, on the equilibrium path s must
be accepted.
Consider now any equilibrium candidate with s < 1. If there is one player i offering s i < s, then this player
should have offered slightly more than s. There will be inequality anyway, but by winning the competition,
player i can increase his own monetary payoff, and he can turn the inequality to his advantage.
A similar argument applies if all players offer si = s < 1. By slightly increasing his offer, player i can increase
the probability of winning the competition from 1/(n - 1) to 1. Again, this increases his expected monetary
payoff, and it turns the inequality toward the other proposers to his advantage. Therefore, s < 1 cannot be
part of a subgame perfect equilibrium.
Hence, the only equilibrium candidate is that at least two sellers offer s = 1. This is a subgame perfect
equilibrium since all sellers receive a payoff of 0, and no player can change this outcome by changing his
action.

3. Cooperation Games
We start with the following public good game. There are n - 2 players who decide simultaneously on their
contribution levels gi C [O,y], i C {1, ... , n}, to the public good. Each player has an endowment of y. The
monetary payoff of player i is given by

where a denotes the constant marginal return to the public good Since a < 1, a marginal
investment into G causes a monetary loss of (1 - a); i.e., the dominant strategy of a com- pletely selfish
player is to choose gi = 0. Thus, the standard model predicts gi = 0. However, since a > 1/n, the aggregate
monetary payoff is maximized if each player chooses gi =y.
Consider now a slightly different public good game that consists of two stages. At stage 1 the game is
identical to the previous game. At stage 2 each player i is informed about the contribution vector (g1, . . .,
gn) and can simultaneously impose a punishment on the other players; i.e., player i chooses a punishment
vector pi = (Pi1, Pin), where pij ≥ 0 denotes the punishment player i imposes on player j. The cost of this
punishment to player i is given by Player i, however, may also be punished by
the other players, which generates an income loss to i of Thus, the monetary payoff of
player i is given by

Since punishments are costly, players' dominant strategy at stage 2 is to not punish. Therefore, if selfishness
and rationality are common knowledge, each player knows that the second stage is completely irrelevant.
Behavioral Economics 2019-2020 Fehr & Schmidt

Without Punishment:
In the final period of n-person cooperation games (n > 3) without punishment the vast majority of subjects
play the equilibrium strategy of complete free riding. If we average over all studies, 73 percent of all
subjects choose gi = 0 in the final period. It is also worth mentioning that in addition to those subjects who
play exactly the equilibrium strategy there are very often a nonnegligible fraction of subjects who play
"close" to the equilibrium.
With Punishment:
However, if we turn to the public good game with punishment, there emerges a radically different picture
although the standard model predicts the same outcome as in the one-stage game. Note that the same
subjects generated the distribution in the game without and in the game with punishment. A strikingly large
fraction of roughly 80 percent cooperates fully in the game with punishment. Researchers report that the
vast majority of punishments are imposed by cooperators on the defectors and that lower contribution levels
are associated with higher received punishments. Thus, defectors do not gain from free riding because
they are being punished.

PROPOSITION 4
(Part a) By spending one dollar on
the public good, he earns a dollars
in monetary terms. In addition, he
may get a non-pecuniary benefit of
at most ꞵi dollars from reducing
inequality. Therefore, since αi + ꞵi
< 1 for this player, it is a dominant
strategy for him to contribute
nothing.

(Part b) of the proposition says that if the fraction of subjects, for whom g i = 0 is a dominant strategy, is
sufficiently high, there is a unique equilibrium in which nobody contributes. The reason is that if there are
only a few players with αi + ꞵi > 1, they would suffer too much from the disadvantageous inequality caused
by the free riders. The proof of the proposition shows that if a potential contributor knows that the number
of free riders, k, is larger than a(n - 1)/2, then he will not contribute either.
(Part c) shows that if there are sufficiently many players with αi + ꞵi > 1, they can sustain cooperation
among themselves even if the other players do not contribute. However, this requires that the contributors
are not too upset about the disadvantageous inequality toward the free rider. goes up. To put it differently,
the greater the aversion against being the sucker, the more difficult it is to sustain cooperation in the one-
stage game. We will see below that the opposite holds true in the two-stage game.
Fairness and Retaliation 2019-2020 Fehr & Gächter
The Economics of Reciprocity

Fairness and Retaliation

The Economics of Reciprocity

Reciprocity: In response to friendly actions, people are frequently much nicer and much more cooperative
than predicted by the self-interest model; conversely, in response to hostile actions they are frequently much
more nasty and even brutal.
We term the cooperative reciprocal tendencies "positive reciprocity" while the retaliatory aspects are
called "negative reciprocity."
When people face strong material incentives to free ride, the self-interest model predicts no cooperation at
all. However, if there are individual opportunities to punish others, then the reciprocal types vigorously
punish free riders even when the punishment is costly for the punisher.
Indeed, the power to enhance collective actions and to enforce social norm is probably one of the most
important consequences of reciprocity. Details of the institutional environment, like the presence of
incomplete contracts or of costly individual punishment opportunities, determine whether the reciprocal or
the selfish types are pivotal.

1. Positive and Negative Reciprocity: Some Evidence

Reciprocity is fundamentally different from cooperative or retaliatory behavior in repeated interactions.

Reciprocity is also fundamentally different from altruism. Altruism is a form of unconditional kindness;
that is, altruism given does not emerge as a response to altruism received. Again, reciprocity is an in-kind
response to beneficial or harmful acts.
Positive Reciprocity is deeply embedded in many social interactions. Psychological studies show, for
example, that smiling waitresses get tipped much more than less friendly one.
The normative power of reciprocity is also likely to have an important impact on social policy issues.
Social policies are much less likely to be endorsed by public opinion when they reward people independent
of whether and how much they contribute to society.

In the ultimatum bargaining experiment, responders do not behave in a self-interest maximizing manner.
In general, the motive indicated for the rejection of positive, yet "low," offers is that subjects view them as
unfair.
In a trust game (Money x and 3y), many Proposers send money and that many Responders give back some
money. There is frequently a positive correlation between y and the amount sent back at the individual as
well as at the aggregate level. Again, positive reciprocity does not appear to diminish even if the monetary
stake size is rather high.
The fraction of subjects who show a concern for fairness and behave reciprocally in one-shot situations is
relatively high.
Fairness and Retaliation 2019-2020 Fehr & Gächter
The Economics of Reciprocity

Male behavior in the ultimatum game is systematically linked to testosterone levels. Males who reject unfair
offers have higher testosterone levels than males who accept unfair offers. This is interesting because
testosterone levels are thought to be important mediators of male willingness to engage in aggressive
behavior.

Reciprocal Behavior: An emerging consensus that the propensity to punish harmful behavior is stronger
than the propensity to reward friendly behavior.
A fourth group of researchers, in contrast, views the reciprocal actions in laboratory experiments as a form
of boundedly rational behavior.
Reciprocity: Stable behavioral response by a nonnegligible fraction of the people that can be reliably
elicited under appropriate circumstances.

2. Public Goods

2.1. Reciprocity and Punishments

Public Good: Good that is both non-excludable and non-rivalrous in that individuals cannot be excluded
from use or could be enjoyed without paying for it, and where use by one individual does not reduce
availability to others or the goods can be effectively consumed simultaneously by more than one person.
For a group of self-interested agents, of course, public goods present the difficulty that since all agents will
want to be free riders on the efforts of others, no agent will contribute willingly to the public good. The
highest level of social welfare would be achieved if everyone contributed all of their assets to the public
good, but it is in the self-interest of each individual to free ride, regardless of what others contribute, and to
contribute nothing.

Positive reciprocity implies that subjects are willing to contribute something to the public good if others
are also willing to contribute, because a contribution to the public good represents a kind action, which
induces reciprocally motivated people to contribute.
Negative reciprocity has not played a role, because in the game as described there are no opportunities for
direct retaliation in response to observed free riding. However, negative reciprocity can play the role that if
subjects expect that others free ride, and if they interpret that as a hostile act, then they can "punish" others
by free riding.
The result is likely to be that self-interested types choose to free ride because they are self-interested, and
reciprocal types free ride because they observe others free riding. Although the motivation to free ride is
different for the reciprocal type, in the end the behavior of the selfish and the reciprocal type is
indistinguishable.
Fairness and Retaliation 2019-2020 Fehr & Gächter
The Economics of Reciprocity

Suppose that a group has the possibility to observe the contributions of others, and to punish those who do
not contribute. It is important that punishment be costly to the agent who imposes it.
1. If all subjects were purely self-interested, contribution decisions would be unaffected by the
punishment opportunity.
2. Negatively reciprocal subjects are willing to pay a price to act reciprocally and will use the costly
punishment opportunity to punish free riders. This will induce self-interested subjects to contribute
to avoid the punishment.
The public good game with direct punishment opportunities provides, therefore, an example where the
reciprocal types can induce the selfish types to make "cooperative" choices. Even a minority of reciprocal
subjects is capable of inducing a majority of selfish subjects to cooperate in these circumstances
2.2. Perfect Stranger and Partner Version
Perfect Stranger Version: Version that ensures that the actions in a particular period have no rewards in
future periods. All subjects knew the total number of periods in advance.
Partner Version: Version that ensures that there are possible strategic spillovers across periods so that
present actions can have future return.
This figure shows how much a subject is
punished for a given negative deviation
from the average contribution of other
members in the group. The punishment is
measured by the average percentage
reduction in the incomes of the punished
subject.
It turns out that the negative deviation from
others' average contributions to the public
good is a strong determinant of punishment.

The more a subject free rides relative to the others the more it gets punished. Moreover, this pattern is
almost the same in the two versions of the game: Free riders are punished irrespective of whether there are
future rewards for the punisher.
Next figure reflect the heavy punishment of free riders, in turn, has a large disciplining effect on subjects'
cooperation behavior. It compares the time paths of the average contributions of the two versions.
The first observation is that in the absence of a
punishment opportunity, average cooperation converges
to very low levels in the later periods.
When subjects are perfect strangers they can at least
stabilize contributions at relatively high levels. In the
Partner version they almost converge to the maximum
level of contributions, and this indicates the disciplining
force of the punishment opportunity.
Fairness and Retaliation 2019-2020 Fehr & Gächter
The Economics of Reciprocity

3. From Public Goods to Social Norms

The problem of public goods may seem a rather limited economic application, and it may seem farfetched
to link the experiment here to government spending on basic research and development or national defense.
We believe that the analytical structure of the public good problem is a good approximation to the question
of how social norms are established and maintained.
Social Norm: A behavioral regularity that is based on a socially shared belief of how one ought to behave
which triggers the enforcement of the prescribed behavior by informal social sanctions.
A social norm can be thought of as a sort of behavioral public good, in which everybody should make a
positive contribution, and also where individuals must be willing to enforce the social norm with informal
social sanctions, even at some immediate cost to themselves.

It has been observed that social norms influence work morale and behavior against "rate busters". Social
sanctions by peer members are probably a very important determinant of effort behavior in work relations
Social norms also constitute perhaps one of the most important elements of what recently has been termed
"social capital. There are powerful arguments that social norms also have a decisive impact on the
functioning of markets.

4. Reciprocity as a Contract Enforcement Device

Real-world contracts are often highly incomplete, which gives rise to strong incentives to shirk. Labor
contracts often take the form of a fixed wage contract without explicit performance incentives and in which
workers have a considerable degree of discretion over effort levels. In such a situation, a worker's general
job attitude, loyalty becomes important
Loyalty: An affirmative job attitude whereby gaps are filled, initiative is taken, and judgment is exercised
in an instrumental way.
The requirement of a generally cooperative job attitude renders reciprocal motivations potentially very
important in the labor process. If a substantial fraction of the work force is motivated by reciprocity
considerations, employers can affect the degree of "cooperativeness" of workers by varying the generosity
of the compensation package, even without offering explicit performance incentives.
Experience Fehr’s & all: Experimental employers could offer a wage contract that stipulated a binding
wage w and a desired effort level ê. If an experimental worker accepted this offer, the worker was free to
choose the actual effort level e between a minimum and a maximum level. Employee level e and effort cost
c(e) determine the profit π = 10e – w, which then indicate the monetary payoff u = w - c(e).
Observations:
1. Selfish workers have no incentives to provide effort above the minimum level e = 1.
2. In response to generous job offers, people are on average willing to put forward extra effort above
what is implied by purely pecuniary considerations.
Fairness and Retaliation 2019-2020 Fehr & Gächter
The Economics of Reciprocity

Alternative to Fehr’s Experience: Employers can also respond reciprocally by rewarding or punishing
workers after they observe actual effort choices.
Observations:
1. A selfish employer will never reward or punish since rewarding and punishing is costly.
2. Reciprocity substantially contributes to the enforcement of contracts and provides incentives for
the potential cheaters to behave cooperatively or at least to limit their degree of noncooperation.

5. Work Motivation and Performance Incentives

Another Alternative to Fehr’s Experience: Additional implementation of a treatment with explicit
performance incentives. Employers now have the possibility to stipulate a fine, to be paid by the worker to
the employer in case of verified shirking.
The offered rent is implied by the original contract
offer; it is defined as the wage minus the cost of
providing the desired effort level.
Due to the presence of many reciprocal workers, the
average effort level is strongly increasing in the
offered rent and rises far above the selfish level of
e = 1.
Except at the low rent levels, the average effort is
lower in the presence of the explicit incentives.
This result suggests that reciprocity-based effort elicitation and explicit performance incentives may indeed
be in conflict with each other. The average effort taken over all trades and the aggregate monetary surplus,
is lower in the incentive treatment than in the control treatment.
For the employers, their profits are higher because, in the incentive treatment, they rely much less on the
"carrot" of generous wage offers. Instead, they threaten the maximal fine in most cases. The savings in
wage costs more than offset the reductions in revenues that are caused by the lower effort in the incentive
treatment. Since the perceived fairness of the distribution of the gains from trade affects the effort behavior
of the reciprocal types, different distributions are associated with different levels of the aggregate gains.

6. Wage Rigidity, Rent-Sharing and Competition

Gift Exchange: Employers offer a gift of pay which is more than labor's opportunity cost and employees
offer a gift of more than minimal effort.
Competitive Double Auction: Most competitive environment where both experimental firms and workers
can make wage bids.
The presence of reciprocal types in the labor market gives rise to downward wage rigidity has been
demonstrated in a number of experiments, such in the competitive double auction.
Reciprocity causes wage rigidity in a double auction provides a strong piece of evidence in favor of the
importance of reciprocity in markets.
 Fairness and Retaliation 2019-2020 Fehr & Gächter
The Economics of Reciprocity

It exist two conditions in the context of a competitive double auction in a labor market:
Complete Labor Contract:
1. The experimenter enforced a given effort level.
2. Employers take full advantage of the low wage offers made by the workers and, as a consequence,
wages are close to the competitive level in this market.
3. The only method for a worker to punish a firm who offers a low wage is to reject such an offer.
4. Due to the presence of purely self-interested workers, reciprocal workers have no possibility to
punish firms, which induces them to accept low wage offer.
Incomplete Labor Contract:
1. Employees could choose an effort level between a minimum and maximum after the wage contract
was concluded, and the employers could neither stipulate nor enforce an effort level above the
minimum level.
2. Employers are very reluctant to accept workers' underbidding of prevailing wages.
3. A reciprocal worker can punish the firm by choosing a low effort level after the labor contract has
been concluded. Since firms anticipate this possibility, they have a reason to pay generous wages.
Left:
Complete
Right:
Incomplete

Powered by TCPDF (www.tcpdf.org)

Unraveling in Guessing Games 2019-2020 Nagel
An Experimental Study

Unraveling in Guessing Games:

An experimental study

1. Introduction:
Consider Experiment with choosing Number between (1;100) with p=2/3 and 4 consecutive rounds where
the winner is the person whose chosen number is closest to the mean times parameter p. (p is a
predetermined positive parameter of the game; p is common knowledge.)

Zero-Order Belief: Simplest case where a player selects a strategy at random without forming beliefs or
picks a number that is salient to him.
First-Order Beliefs: A somewhat more sophisticated player forms beliefs on the behavior of the other
players. He thinks that others select a number at random, and he chooses his best response to this belief.
Second-Order Beliefs: Player forms beliefs on the first-order beliefs of the others and maybe nth order
beliefs about the (n-1) th order beliefs of the others, but only up to a finite n, called the n- depth of reasoning.

For 0 ≤ p < 1, there exists only one Nash equilibrium: all players announce zero. The structure of the game
is favorable for investigating whether and how a player's mental process incorporates the behavior of the
other players in conscious reasoning.
For p = 1, one cannot distinguish between different steps of reasoning by actual subjects in an experiment.
In a two-person coordination games, players use Schelling Salience where they coordinate by either
applying depth of reasoning of order 1 or by picking a focal point. Note that in a two-player games with
p<1, Player that chose 0 always wins.
In a 3X3 games, subjects were using depths of reasoning of orders 1 or 2 or a Nash-equilibrium strategy.

2. Model of Boundedly Rational Behavior

2.1. Games Theoretic Solutions:

1. For 0 ≤ p < 1, there exists only one Nash equilibrium at which all players choose 0.
Assume that there is an equilibrium at which at least one player chooses a positive number with positive
probability. Let k be the highest number chosen with positive probability, and let m be one of the players
who chooses k with positive probability. Obviously, in this equilibrium p times the mean of the numbers
chosen is smaller than k. Therefore, player m can improve his chances of winning by replacing k by a
smaller number with the same probability. Therefore no equilibrium exists in which a positive number is
chosen with positive probability.
All announcing 0 is also the only strategy combination that survives the procedure of infinitely repeated
simultaneous elimination of weakly dominated strategies. (Note of 2-player games)
Unraveling in Guessing Games 2019-2020 Nagel
An Experimental Study

2. For p = 1 and more than two players, the game is a coordination game, and there are infinitely
many equilibrium points in which all players choose the same number.

3. For p > 1 and 2p < M (M is the number of players), all choosing 0 and all choosing 100 are the
only equilibrium points. Note that there are no dominated strategies, but only equilibrium points.
It is straightforward to show that all choosing 0 and all choosing 100 are equilibria: it does not pay to deviate
from 0 (100) if all other players choose 0 (100) and the number of players is sufficiently large. There is no
other symmetric equilibrium since with a unilateral small increase a player improves his payoff. Also, other
asymmetric equilibria or in mixed strategies cannot exist for analogous reasons, as in the case p < 1.

2.2. Nagel Experimental Model:

In the First-Period,
1. A player is strategic of degree 0 if he chooses the number 50. (Randomly choice by symmetric
distribution of number 0 and 100, by using Schelling Salience)
2. A player is strategic of degree n if he chooses the number 50pn (with a reference point r of 50 from
degree 0 player), which I will call iteration step n. A person whose behavior is described by n = 1 just
makes a naive best reply to random behavior. However, if he believes that the others also employ this
reasoning process, he will choose a number smaller than 50p, say 50p2, the best reply to all other players
using degree-1 behavior. A higher value of n indicates more strategic behavior paired with the belief that
the other players are also more strategic; the choice converges to the equilibrium play in the limit as n
increases.

For Period 2-4,

The reasoning process of period 1 can be modified by replacing the initial reference point r = 50 by a
reference point based on the information from the preceding period.
A natural candidate for such a reference point is the mean of the numbers named in the previous period.
With this initial reference point, iteration step 1, which is the product of p and the mean of the previous
period.
Anticipatory Learning: Learning of players in which an increase in iteration steps is expected of the other
players. Specifically, one can ask whether, with increasing experience, higher and higher iteration steps
will be observed.

2.3. Nagel Experimental Results:

In the First-Period,
Low percentage of player choose a number between 0 and 10 for p<1, and similar percentage for number
between 95 and 100 for player with p>1. Thus, the sessions with different parameters do not differ
significantly with respect to frequencies of equilibrium strategies and choices near the equilibrium
strategies.
 Unraveling in Guessing Games 2019-2020 Nagel
An Experimental Study

According to the reference point r, results suggest that many players do not choose numbers at random but
instead are influenced by the parameter p of the game. Another results revealed that when p was increased,
the mean of the chosen numbers was higher.

For Period 2-4,

With p<1, subject have a tendency to decrease their choices over times, as the median and the mean (apart
for p=1/2 for mean). With p>1, subject have a tendency to increase their choices and converge to 100 in
the 3rd or 4th period.

Adjustment Process: Process that refers to the behavioral process of individuals to adjust their answer
according to the previous period t-1.
Zero is never the best reply in the 1/2 and 2/3 sessions, given the actual strategies. Instead the best reply is
a moving target that approaches 0. Choices of subject in 1/2 sessions converge faster toward 0 than those
in 2/3 sessions. A rate of decrease of the means and medians from period 1 to period 4 within a session
would then help to understand the behavior difference between session of ½ and 2/3.
The rates of decrease of the session medians in the 1/2 sessions are higher than those in the 2/3 session.
There is no significant difference in the rates of decrease of the means. The median seems more informative
than the mean, since the mean may be strongly influenced by a single deviation to a high number. Thus,
we can conclude that the rate of decrease depends on the parameter p.

2.4. Adjustment Process Due To Individual Experience

Two possible experiences; Player gains a share or all the prize in the previous period. Or, player earns
nothing in the previous period.
There exist two adjustment factor:
Subject own adjustment factor at in period t (after facing payoff of 0) is defined as the relative deviation
from the mean in period t-1.
The "optimal" adjustment factor aopt,t in period t is defined as the optimal deviation from the mean of
period t - 1 that leads to p times the mean of period.

The idea of this simple learning-direction theory is that in an ex post reasoning process a player compares
his adjustment factor at, with the optimal adjustment factor aopt,t. In the next period he most likely adapts in
the direction of the optimal adjustment factor.
If his chosen number was above p-times the mean in the previous period then he should decrease
his rate.

If his number was below p times the mean, he should then increase his adjustment factor.

Powered by TCPDF (www.tcpdf.org)