01 Expected Utility Theory

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 27

BE 510 Business Economics 1

1. Expected utility theory

Prof. Dr. Henrik Orzen

Office hour: Tuesdays, 16:00-17:00 (by appointment)


Room 4.01 (Department of Economics, L7, 3-5)
E-Mail: henrik.orzen@uni-mannheim.de
BE 510 Business Economics 1 • Bayes Nash Equilibrium

Decision-making under risk


What does ‘risk’ mean?
 Risky world: Future holds different possibilities but only one outcome will become a reality.
 My decisions may affect the probabilities for the different outcomes.
 In a risky world I know
 how likely each outcome is and what exactly each outcome implies, and
 how actions affect the probabilities of outcomes.
 In an uncertain world I do not have (full) knowledge about the probabilities and/or exact implications of the
outcomes.
BE 510 Business Economics 1 • Bayes Nash Equilibrium

Decision-making under risk


Simplest application: Monetary gambles
 Good outcomes yield higher monetary payoffs, bad outcomes yield lower monetary payoffs.
 Once taken for granted that rationality requires expected payoff maximization.
 Example:
 Gamble A: €25 with probability 0.4; €0 otherwise. 𝐸𝑉(𝐴) = €10
 Gamble B: €9 with certainty. 𝐸𝑉(𝐵) = €9
 Gamble C: €100 with probability 0.05; €0 otherwise. 𝐸𝑉(𝐶) = €5
BE 510 Business Economics 1 • Bayes Nash Equilibrium

Decision-making under risk


A coin game
 A fair coin is tossed repeatedly until “Heads” comes up.
 If “Heads” appears…
…the first time, you receive: €2 H 1/2
…the second time (after 1 “Tails”), you receive: €4 TH 1/4
…the third time (after 2 “Tails”), you receive: €8 TTH 1/8
…the fourth time (after 3 “Tails”), you receive: €16 TTTH 1/16
…the fifth time (after 4 “Tails”), you receive: €32 TTTTH 1/32
…and so on.
 If you play this game many times, how much money can you expect to earn on average?
 How much would you be willing to pay (as an entry fee) to be allowed to play this game?
BE 510 Business Economics 1 • Bayes Nash Equilibrium

Decision-making under risk


A coin game
 How much would you be willing to pay (as an entry fee) to be allowed to play this game?
 Probability of winning up to €16:
1 1 1 1 15
Pr 𝑋 ≤ 16 = + + + = = 0.9375
2 4 8 16 16
 Probability of winning more than €64:
1 1 1 1 1 1 1
Pr 𝑋 > 64 = 1 − + + + + + = = 0.0156
2 4 8 16 32 64 64
BE 510 Business Economics 1 • Bayes Nash Equilibrium

Resolving the St. Petersburg Paradox


Diminishing marginal utility
 Utility function in general:
 Mathematical tool to represent preferences. Utility
 Assigns a number value to each possible outcome. (Utils) 𝑢(𝑥)
 Different values imply a strict preference relationship.
 Utility increase from one extra monetary unit is
the smaller the more money a person has:

Utils
𝑢′ 𝑥 > 0 and 𝑢′′ 𝑥 < 0.
 Consequence for the coin game:
∞ 𝑖
1
𝐸𝑉 = 2𝑖 = ∞
2
𝑖=1
∞ 𝑖
1
𝐸𝑈 = 𝑢 2𝑖 = Some finite value 0 1 2 3 4 5 6 7 8 9 10
2 Amount Monetary units (𝑥)
𝑖=1
BE 510 Business Economics 1 • Preferences, utility and optimization • 22

The DMU principle implies risk aversion


A choice problem Utility (Utils)
 Option 1: Get €100 with probability
𝑢(100)
0.5 (else get nothing).
 Option 2: Get €50.
 An expected utility maximizer
characterized by DMU must pick
option 2.
𝐸𝑈 = 0.5𝑢 100 + 0.5𝑢(0)

𝑢(0)
0 10 20 30 40 50 60 70 80 90 100 𝑥 (€)
BE 510 Business Economics 1 • Preferences, utility and optimization • 22

The DMU principle implies risk aversion


A choice problem Utility (Utils)
 Option 1: Get €100 with probability 𝑢(𝑥)
𝑢(100)
0.5 (else get nothing).
 Option 2: Get €50.
 An expected utility maximizer
characterized by DMU must pick
option 2.
𝐸𝑈

𝑢(0)
0 10 20 30 40 50 60 70 80 90 100 𝑥 (€)
BE 510 Business Economics 1 • Preferences, utility and optimization • 22

The DMU principle implies risk aversion


A choice problem Utility (Utils)
 Option 1: Get €100 with probability 𝑢(𝑥)
𝑢(100)
0.5 (else get nothing).
 Option 2: Get €50.
 An expected utility maximizer 𝑢(50)
characterized by DMU must pick
option 2.
𝐸𝑈

𝑢(0)
0 10 20 30 40 50 60 70 80 90 100 𝑥 (€)
BE 510 Business Economics 1 • Preferences, utility and optimization • 22

The DMU principle implies risk aversion


A choice problem Utility (Utils)
 Option 1: Get €100 with probability 𝑢(𝑥)
𝑢(100)
0.75 (else get nothing).
𝑢(75)
 Option 2: Get €75.
 An expected utility maximizer 𝐸𝑈 = 0.75𝑢 100 + 0.25𝑢(0)
characterized by DMU must pick
option 2.

𝑢(0)
0 10 20 30 40 50 60 70 80 90 100 𝑥 (€)
BE 510 Business Economics 1 • Preferences, utility and optimization • 22

The DMU principle implies risk aversion


A choice problem Utility (Utils)
 Option 1: Get €100 with probability 𝑢(𝑥)
𝑢(100)
0.25 (else get nothing).
 Option 2: Get €25.
 An expected utility maximizer
characterized by DMU must pick
option 2.
𝑢(25)

𝐸𝑈 = 0.25𝑢 100 + 0.75𝑢(0)

𝑢(0)
0 10 20 30 40 50 60 70 80 90 100 𝑥 (€)
BE 510 Business Economics 1 • Preferences, utility and optimization • 22

The DMU principle implies risk aversion


A choice problem Utility (Utils)
 Option 1: Get €100 with probability 𝑢(𝑥)
𝑢(100)
0.25 (else get nothing).
 Option 2: Get €25.
 An expected utility maximizer
characterized by DMU must pick
option 2.

𝑢(0)
0 10 20 30 40 50 60 70 80 90 100 𝑥 (€)
BE 510 Business Economics 1 • Preferences, utility and optimization • 22

The DMU principle implies risk aversion


A choice problem Utility (Utils)
 Option 1: Get gamble 𝑢(𝑥)
𝑢(𝑥2 )
𝑔 = 𝑝 𝑥2 ⊕ 1 − 𝑝 𝑥1 .
 Option 2: Get monetary amount
𝑤 = 𝑝𝑥2 + 1 − 𝑝 𝑥1 .
 An expected utility maximizer
characterized by DMU must pick
option 2.

𝑢(𝑥1 )
𝑥1 𝑥2 𝑥 (€)
BE 510 Business Economics 1 • Preferences, utility and optimization • 22

Certainty equivalent and risk premium


A choice problem Utility (Utils)
 Option 1: Get gamble 𝑢(𝑥)
𝑢(𝑥2 )
𝑔 = 𝑝 𝑥2 ⊕ 1 − 𝑝 𝑥1 .
 Option 2: Get monetary amount 𝑢(𝑤)
𝑤 = 𝑝𝑥2 + 1 − 𝑝 𝑥1 .
 Certainty equivalent of 𝑔: 𝑢(𝑔)
Get amount 𝑐 such that
𝑢 𝑐 = 𝑝𝑢 𝑥2 + 1 − 𝑝 𝑢 𝑥1 = 𝑢(𝑔).

Risk
premium
=𝑤−𝑐
𝑢(𝑥1 )
𝑥1 𝑐 𝑤 𝑥2 𝑥 (€)
BE 510 Business Economics 1 • Bayes Nash Equilibrium

EUT can accommodate different risk attitudes


Three basic categories
 Risk aversion
 A person is risk averse in the context of a particular gamble if 𝐸𝑈(𝑔) < 𝑢 𝐸𝑉(𝑔) .
 Certainty equivalent of a gamble is less than 𝐸𝑉(𝑔). Risk premium is a positive number.
 Concave utility function.
 Risk neutrality
 A person is risk neutral in the context of a particular gamble if 𝐸𝑈 𝑔 = 𝑢 𝐸𝑉(𝑔) .
 Certainty equivalent of a gamble is equal to 𝐸𝑉(𝑔). Risk premium is zero.
 Linear utility function.
 Risk proneness
 A person is risk prone in the context of a particular gamble if 𝐸𝑈 𝑔 > 𝑢 𝐸𝑉(𝑔) .
 Certainty equivalent of a gamble is greater than 𝐸𝑉(𝑔). Risk premium is a negative number.
 Convex utility function.
BE 510 Business Economics 1 • Bayes Nash Equilibrium

EUT can accommodate different risk attitudes


Being an expected value maximizer
 Example from before:
 Gamble A: €25 with probability 0.4; €0 otherwise. 𝐸𝑉(𝐴) = €10
 Gamble B: €9 with certainty. 𝐸𝑉(𝐵) = €9
 Gamble C: €100 with probability 0.05; €0 otherwise. 𝐸𝑉(𝐶) = €5
EUT can accommodate different risk attitudes
Allowing heterogeneity in risk attitudes
 Person 1: 𝑢 €0 = 10 utils; 𝑢 €9 = 13 utils; 𝑢 €25 = 15 utils; 𝑢 €100 = 20 utils
 𝑢 A = 0.4×15 utils + 0.6×10 utils = 𝟏𝟐 utils
 𝑢 B = 𝑢 €9 = 𝟏𝟑 utils
 𝑢 C = 0.05×20 utils + 0.95×10 utils = 𝟏0.5 utils
 Person 2: 𝑢 €0 = 0 utils; 𝑢 €9 = 81 utils; 𝑢 €25 = 625 utils; 𝑢 €100 = 10,000 utils
 𝑢 A = 0.4×625 utils + 0.6×0 utils = 𝟐𝟓𝟎 utils
 𝑢 B = 81 utils = 𝟖𝟏 utils
 𝑢 C = 0.05×10,000 utils + 0.95×0 utils = 𝟓𝟎0 utils
Can actual choice behavior be inconsistent with EUT?
How EU maximizers evaluate gambles: Von Neumann/Morgenstern utility
 Consider a choice between a safe outcome (𝑥2 ) and a gamble (50-50 chance of either 𝑥1 or 𝑥3 ).
 An expected utility maximizer evaluates these options as follows:
 𝑢(Safe) = 𝑢(𝑥2 ). Von Neumann/Morgenstern
 𝑢 Gamble = 𝑢 0.5 𝑥1 ⊕ 0.5 𝑥3 = 0.5𝑢(𝑥1 ) + 0.5𝑢(𝑥3 ). expected utility function

 In EUT a number of assumptions are made implicitly.


 For example: Rationality.
 Or: More of something is preferred to less of the same thing (ceteris paribus).
 Most of these assumptions are very conventional in economics.
 However, it does not follow from the conventional assumptions that
people must evaluate gambles in the von Neumann/Morgenstern way.
 Thus, the von Neumann/Morgenstern expected utility function requires an
additional assumption about human choice behavior.
Oskar Morgenstern John von Neumann
 This extra assumption is called the independence axiom.
1902 – 1977 1903 – 1957
The independence axiom
Definition
If 𝑔1 and 𝑔2 are two gambles such that 𝑔1 ≿ 𝑔2 , then for any 𝛼 ∈ [0,1] and any gamble 𝑔3 :

𝛼 𝑔1 ⊕ 1 − 𝛼 𝑔3 ≿ 𝛼 𝑔2 ⊕ 1 − 𝛼 𝑔3

In words, if you construct two compound gambles by (1) combining in some way gamble 𝑔1
with gamble 𝑔3 and by (2) combining 𝑔2 with 𝑔3 in the same way, then the preference
ordering of the two compound gambles is identical to the preference ordering of 𝑔1 and 𝑔2 .

Thus, the preference ordering is independent of the characteristics of the third gamble.
The independence axiom
An example
Situation 1 Situation 2
Option A: Throw a coin. Heads: You get the nuts.
Option A: A packet of nuts.
Tails: You get a glass of beer.

Option B: Throw a coin. Heads: You get the yoghurt.


Option B: A pot of strawberry yoghurt.
Tails: You get a glass of beer.

 Suppose you prefer A to B in Situation 1.


 Then, says the independence axiom, you should also prefer A to B in Situation 2.
 Key argument:
 The beer on the one hand and the nuts/yoghurt on the other hand are mutually exclusive events.
 Therefore, your preference ranking between nuts and yoghurt should be independent of the presence or absence of the beer.
 Likewise, the probability (here: 50-50) and the nature of the common outcome (beer) should not matter.
The common consequence effect
The idea
 Originally described by Maurice Allais in 1953, and also known as the “Allais Paradox”.
 Choice problem 1:
 Alternative A: 1.00[€1,000,000]
 Alternative B: 0.85 €1,000,000 ⊕ 0.05 €0 ⊕ 0.10[€5,000,000]
 Choice problem 2:
 Alternative C: 0.85 €0 ⊕ 0.15[€1,000,000]
 Alternative D: 0.90 €0 ⊕ 0.10[€5,000,000]
 A reasonable response: Choose A in choice problem 1, D in problem 2. Maurice Allais
1911 – 2010
 But this violates EUT.
 Problem 1: 𝑢(1m) > 0.85𝑢(1m) + 0.05𝑢(0) + 0.10𝑢(5m) ⇔ 0.15𝑢(1m) > 0.05𝑢(0) + 0.10𝑢(5m)
 Problem 2: 0.85𝑢(0) + 0.15𝑢(1m) < 0.90𝑢(0) + 0.10𝑢(5m) ⇔ 0.15𝑢(1m) < 0.05𝑢(0) + 0.10𝑢(5m)
The common consequence effect
Experimental evidence
 For example Starmer and Sugden (1991), AER.
 Choice problem 1:
 Alternative A: 1.00[£7] D
 Alternative B: 0.75 £7 ⊕ 0.05 £0 ⊕ 0.20 £10
 Choice problem 2: D
 Alternative C: 0.75[£0] ⊕ 0.25[£7]
 Alternative D: 0.80[£0] ⊕ 0.20[£10] C
C

A Choices B Choices
Consequences of EUT violations in the lab
Normative interpretation of EUT as a way out?
 EUT provides sensible guidance to decision-making under risk.
 Violations are essentially mistakes.
 If this was properly explained to subjects they would correct their behavior.

Accepting the findings and moving forward


 Used to be a minority view. Now more or less mainstream.
 Development of alternative theories of decision-making under risk.
 Dialog between theoretical development and empirical/experimental tests.
Prospect theory
Modifying EUT in several ways
 Two foundation papers:
 D. Kahneman and A. Tversky (1979). Prospect Theory: An Analysis of Decision under
Risk. Econometrica, 47(2), 263-292.
 A. Tversky and D. Kahneman (1992). Advances in Prospect Theory: Cumulative
Representation of Uncertainty. Journal of Risk and Uncertainty, 5, 297-323.
 Plus: Further work on this theory by other authors.
 In PT decision makers evaluate outcomes against a reference point.
Daniel Kahneman Amos Tversky
Outcomes are perceived as gains or as losses relative to that. 1934 – 1937 – 1996
 Decision makers are assumed to be loss averse.
Losses loom larger than gains, i.e. incurring a loss reduces the DM’s utility more than an equally sized gain lifts it.
 Also incorporates a switchover in risk attitudes at the reference point:
Risk aversion in the domain of gains and risk proneness in the domain of losses (reflection effect).
Prospect theory
Value function
 Based on these ideas the utility function (referred to as value function in PT) looks like this:
Psychological value

𝜋 𝑝

Monetary value minus 1.0


reference value .9
.8
.7
.6
.5
.4
 Finally, PT also assumes a transformation of probabilities: .3
Decision makers have a distorted perception of probabilities: .2
 There is overweighting of small probabilities .1
.0
 There is underweighting of moderate/high probabilities.
.0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1.0 𝑝
Prospect theory
Back to the common consequence effect
 Recall the two choice problems:
 Choice problem 1: A = 1.00[€1m] versus B = 0.85 €1m ⊕ 0.05 €0 ⊕ 0.10 [€5m]
 Choice problem 2: C = 0.85 €0 ⊕ 0.15[€1m] versus D = 0.90 €0 ⊕ 0.10[€5m]
 Utility function (“value function”): Probability weighting function:
𝑥 0.4 if 𝑥 ≥ 0 𝑝0.6
𝑢 𝑥 = 0.4 𝜋 𝑝 = 1
−2.5 −𝑥 if 𝑥 < 0
𝑝0.6 + 1−𝑝 0.6 0.6

 Reference point: Receive €1m.


 Problem 1: 𝜋 1 𝑢 ±0 > 𝜋 0.85 𝑢(±0) + 𝜋 0.05 𝑢(−1m) + 𝜋 0.10 𝑢(+4m)
 Problem 2: 𝜋 0.85 𝑢 −1m + 𝜋 0.15 𝑢 ±0 < 𝜋 0.90 𝑢 −1m + 𝜋 0.10 𝑢(+4m)
Summary
EUT: Decision making under risk Violations of EUT
 What is “rational” depends on a person’s risk attitude.  E.g. the common consequence effect.
 Expected value maximization is a special case.  Not random but systematic.

Risk aversion/neutrality/proneness Prospect theory


 Concave/linear/convex utility function  Changes EUT in fundamental ways.
 𝑢(𝐸𝑉(Gamble)) ⋛ 𝑢(Gamble)  Decision makers evaluate outcomes as gains or
 Certainty equivalent ⋛ 𝐸𝑉(Gamble) losses relative to a reference point.
 Risk premium is pos./zero/neg.  Assumes loss aversion and a change of risk
attitudes at the reference point.
Independence axiom  Decision makers are modelled as perceiving
 The assumption that characterizes EUT specifically. probabilities in a distorted way.
 Leads to von Neumann-Morgenstern EU function.  Is compatible with some behavior that is not
compatible with EUT.

You might also like