R.&U. GENERAL THEORY 10.10.

2024

1. EXPECTED UTILITY MODEL

Risky Environment. A risky environment is described by:

• Enumeration of all possible outcomes:
o 𝑪 is the set of all possible outcomes that a<ect the well-being of the decision
maker;
o 𝐶 is assumed to be finite.
• Characterization of the risky environment:
o Probability vector of the possible outcomes;
o 𝒑𝑺 is the probability of outcomes 𝑺 occurring, with ∑#"$% 𝑝"

Lottery and set of lotteries. A lottery 𝑳 is a vector (𝒑𝟏 , … , 𝒑𝑺 ) and the set of all lotteries on 𝑪 is
𝑳 = -(𝒑𝟏 , … , 𝒑𝑺 ) ∈ [𝟎, 𝟏]𝑺 3 ∑𝒊 𝒑𝒊 = 𝟏4.
A compound lottery is a lottery whose outcomes are lotteries.

Rational Preference and Expected Utility. A rational preference relation over the set of lotteries
𝐿 can be written:

Discrete random variable Continuous random variable

𝒘
7𝒂 ≽ 𝒘
7𝒃 ⇔ 𝑬𝒖(𝒘 7𝒂 ) ≥ 𝑬𝒖(𝒘7𝒃 ) 𝒘
7𝒂 ≽ 𝒘
7𝒃 ⇔ 𝑬𝒖(𝒘 7𝒂 ) ≥ 𝑬𝒖(𝒘
7𝒃 )
# #
⇔ @ 𝑢(𝑤)𝑑𝐹+ (𝑤)𝑑𝑤 ≥ @ 𝑢(𝑤)𝑑𝐹, (𝑤)𝑑𝑤
⇔ = 𝑝*+ 𝑢(𝑤* ) ≥ = 𝑝*, 𝑢(𝑤* )
*$% *$%

2. RISK AVERSION

This part focuses on changes in risk aversion, i.e. in utility functions.

2.1. Characterization

Risk Averse Agent. It is an agent who dislikes all zero-mean risks (“pure risks”) at all wealth
levels.
• An agent is RA if 𝑬𝒖(𝒛D) ≤ 𝒖F𝑬(𝒛D)G = 𝑢(𝑤) where 𝑧̃ = 𝑤 + 𝑥D with 𝑥D a pure risk (i.e. 𝐸(𝑥D) =
0.
𝑬𝒖(𝒛D) is the expected utility of the wealth. This is the utility the agent would get from
facing an uncertain outcome, like a gamble.
𝒖F𝑬(𝒛D)G is the utility of the expected wealth. This is the utility the agent gets from
having the certain amount of wealth equal to the expected value of their uncertain final
wealth.
• An agent is risk averse if replacing an uncertain final wealth by its expected value makes
him better off.

Risk Averse and Utility Function. An agent with von Neumann–Morgenstern (VNM) utility
function 𝑢 is risk averse if 𝒖 is concave (i.e. ∀𝑥, 𝑢′′(𝑥) ≤ 0).
• An agent is risk neutral if his vNM utility function is linear.
• An agent is risk lover if his vNM utility function is convex (i.e. ∀𝑥, 𝑢′′(𝑥) ≥ 0).

1
 R.&U. GENERAL THEORY 10.10.2024

Risk Premium. A risk premium is the maximum amount of money 𝝅 that one is ready to pay to
escape a pure risk (i.e. a zero-mean risk).

S) = 𝒖(𝒘𝟎 − 𝝅) where 𝐸(𝑥D) = 0

𝜋 is such that 𝑬𝒖(𝒘𝟎 + 𝒙

For a risk averse agent, 𝜋 is positive

𝜋 should be > to the expected loss/actuarial value pxL

Certainty Equivalent. The certainty equivalent is the sure amount of money that makes the
agent indiXerent between playing the lottery or not.

Considering a lottery whose net gain is described by 𝑦D = 𝜇 + 𝑥D, we have:

S) = 𝒖(𝒘𝟎 + 𝑪𝒆 )
𝑬𝒖(𝒘𝟎 + 𝒚

𝑪𝒆 = 𝝁 − 𝝅

Arrow Pratt Approximation of the Premium. In case of a small risk 𝑦D, i.e. 𝑦D = 𝜇 + 𝑥D where 𝑥D is a
pure risk and 𝜇 ≪ 𝑤/ :
𝟏 𝟐
𝝅= 𝝈 𝑨(𝒘𝟎 + 𝝁)
𝟐 𝑿
𝒖!! (𝒘) %
𝑨(𝒘) = − 𝒖! (𝒘) is the coeXicient of absolute risk aversion (of dimension € )

The risk premium for a small pure risk is approximatively proportional to its variance

788(9)
For a risk averse agent, 𝐴 > 0. Because: 𝐴(𝑤) = 78(9)
with 𝑢8 (𝑤) > 0 and 𝑢88 (𝑤) < 0

CoeXicient of Relative Risk Aversion. The coe<icient of relative risk aversion is:

𝒖′′(𝒘)
𝑹(𝒘) = −𝒘 = 𝒘𝑨(𝒘)
𝒖(𝒘)

R is independent of the monetary unit contrary to A

2.2. Comparison in risk aversion

Comparison in risk aversion. An agent 𝑢 is more risk averse than an agent 𝑣 if 𝒖 dislikes all
lotteries that 𝒗 dislikes, independently of their initial wealth.

The risk premium of any risk is larger for agent 𝑢 than for agent 𝑣

Properties of risk aversion comparison. An agent 𝑢 is more risk averse than an agent 𝑣 when:
7!! (9) ; !! (9)
• ∀𝑤, 𝑨𝒖 (𝒘) ≥ 𝑨𝒗 (𝒘) ⇔ − 7! (9) ≥ − ; ! (9) .
OR
• The function 𝒖 is a concave transformation of function 𝒗: ∃𝜙(. ) 𝑤𝑖𝑡ℎ 𝜙 8 > 0 𝑎𝑛𝑑
𝜙 88 ≤ 0 𝑠𝑢𝑐ℎ 𝑡ℎ𝑎𝑡 ∀𝑤, 𝑢(𝑤) = 𝜙F𝑣(𝑤)G.

2
 R.&U. GENERAL THEORY 10.10.2024

Decreasing absolute risk aversion (DARA). Preferences exhibit a decreasing absolute risk
aversion (DARA) if the risk premium associated to any risk is a decreasing function of wealth:

𝑑𝜋(𝑤/ , 𝑢, 𝑥D)
≤ 0 𝑓𝑜𝑟 𝑎𝑛𝑦 𝑤/ 𝑎𝑛𝑑 𝑥D
𝑑𝑤/

Wealthier people are generally less willing to pay for the elimination of fixed risks

Decreasing risk premium property. The risk premium is a decreasing function of wealth when
7!! (9)
the absolute risk aversion 𝑨(𝒘) is decreasing with wealth, i.e. − 7! (9) is decreasing.

2.3. Some utility functions

Harmonic Absolute Risk Aversion (HARA) utility function. HARA utility functions are linear with
wealth:

𝑤 %<= 𝑤
𝑢(𝑤) = 𝜉 r𝜂 + u , 𝑤𝑖𝑡ℎ 𝜂 + > 0
𝛾 𝛾

HARA utility function is the inverse of 𝐴(𝑤), i.e. the prudence 𝑇(𝑤)
𝑢8 (𝑤) 1
𝑇(𝑤) = − 88 =
𝑢 (𝑤) 𝐴(𝑤)

Constant Relative Risk Aversion (CRRA) utility function. CRRA utility functions are HARA utility
functions with 𝜼 = 𝟎 and 𝑹(𝒘) = 𝜸 = 𝒄𝒐𝒏𝒔𝒕𝒂𝒏𝒕:

𝑤 %<=
, 𝑖𝑓 𝛾 ≠ 1
𝑢(𝑤) = €1 − 𝛾
ln(𝑤), 𝑖𝑓 𝛾 = 1

In case of CRRA utility function, the ratio of the risk premium to the risky profit
remains constant:
If we scale the profit from 𝑥D to 𝛼𝑥D (𝛼 > 1), the risk premium should also scale
proportionally with the magnitude of the profit:
𝑹𝒊𝒔𝒌 𝑷𝒓𝒆𝒎𝒊𝒖𝒎 𝒇𝒐𝒓 𝜶𝒙 S = 𝜶 × 𝑹𝒊𝒔𝒌 𝑷𝒓𝒆𝒎𝒊𝒖𝒎 𝒇𝒐𝒓 𝒙
S

Constant Absolute Risk Aversion (CARA) utility function. CARA utility functions exhibit
increasing relative risk aversion and 𝐴(𝑤) = 𝐴 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡:

𝑒 <>9
𝑢(𝑤) = −
𝐴
Quadratic utility function. Quadratic utility functions are HARA utility functions with 𝜸 = −𝟏
and they exhibit increasing absolute risk aversion:

𝑤?
𝑢(𝑤) = 𝑤 − 𝜌
2

3
 R.&U. GENERAL THEORY 10.10.2024

3. CHANGE IN RISK

This part focuses on changes in risk that have an unambiguous eXect on the risk premium
independently of the utility function.

Some restrictions on the utility functions are imposed:

• Risk averse agents (𝑢88 < 0);
• Usually, prudent agents (𝑢888 ≥ 0).
More precisely, as:
𝜋(𝑤/ , 𝑢, 𝑥
7)
% ≥ 𝜋(𝑤/ , 𝑢, 𝑥
7)
? ⟺ 𝐸𝑢(𝑤/ + 𝑥
7)
% ≤ 𝐸𝑢(𝑤/ + 𝑥
7)
?

The objective is to determine conditions under which this inequality is true for any
utility function
A

Mean-Preserving Changes in Risks. There are two approaches to study an increase in risk that
preserve the expected outcome:

• Adding noise: From 𝑤 S + 𝜀̃ with 𝐸(𝜀̃) = 0

S to 𝑤
• Simple Mean Preserving Spread (SMPS):

𝑤
7% is a SMPS of 𝑤
7? if:

There exists an interval 𝐼 such that

∀𝒘 ∈ 𝑰, 𝒇𝟏 (𝒘) ≤ 𝒇𝟐 (𝒘) and ∀𝒘 ∉ 𝑰, 𝒇𝟏 (𝒘) ≥ 𝒇𝟐 (𝒘)
7𝟏 ) = 𝑬(𝒘
𝑬(𝒘 7𝟐 ) AND
OR
𝑭𝟏 (𝒘) − 𝑭𝟐 (𝒘) is positive and then negative as 𝑤 increases

3.1. Second order stochastic dominance

Second order stochastic dominance (SSD) is the condition under which every agent
prefers one risk to the other in the case where the utility function is increasing and
concave

Second Order Stochastic Dominance. 𝑤

7? dominates 𝑤
7% in the sense of second order stochastic
7? 𝑆𝑆𝐷 𝑤
dominance (𝑤 7)
% i<:

@
𝐸(𝑤
7) 7? )
% = 𝐸(𝑤 AND ∀𝑎 ∈ [𝑎, 𝑏], 𝑆(𝜃) = @ F𝐹% (𝑠) − 𝐹? (𝑠)G 𝑑𝑠 ≥ 0
+

,
𝐸(𝑤
7)
A = 𝑏 − @ 𝐹" (𝑠) 𝑑𝑠
+

SSD Theorem (Rothschild and Stiglitz, 1970). An increase in risk from 𝒘

7𝟐 to 𝒘 7𝟏 ) =
7𝟏 with 𝑬(𝒘
7𝟐 ) can be defined by:
𝑬(𝒘

• The means are the same and risk-averse agents unanimously prefer 𝒘
7𝟐 to 𝒘
7𝟏 :
∀𝑢 𝑐𝑜𝑛𝑐𝑎𝑣𝑒, 𝐸𝑢(𝑤7% ) ≤ 𝐸𝑢(𝑤
7? )
OR
• 𝒘 7𝟏 : ∀𝑎 ∈ [𝑎, 𝑏], 𝑆(𝜃) ≥ 0
7𝟐 𝑺𝑺𝑫 𝒘

4
 R.&U. GENERAL THEORY 10.10.2024

OR
• 𝒘7𝟏 is obtained from 𝒘
7𝟐 by a sequence of simple mean preserving spreads (SMPS)
OR
• 𝒘 7𝟐 by adding a white noise 𝝐D to it, with ∀𝑤, 𝐸(𝜀̃|𝑤
7𝟏 is obtained from 𝒘 7? = 𝑤) = 0

Risk and variance. A decrease (increase) in risk implies less (more) variance. But, less (more)
variance does NOT imply a decrease (increase) in risk.

3.2. First order stochastic dominance

First order stochastic dominance (FSD) is the condition under which every agent prefers
one risk to the other in the case where the utility function is only increasing

First Order Stochastic Dominance. 𝑤

7? dominates 𝑤
7% in the sense of first order stochastic
dominance (𝑤7? 𝐹𝑆𝐷 𝑤
7)
% i<:

∀𝒘, 𝑭𝟏 (𝒘) ≥ 𝑭𝟐 (𝒘)

FSD properties. 𝑤
7? FSD 𝑤
7% when:

• All agents with a non-decreasing utility function prefer 𝒘

7𝟐 to 𝒘7𝟏 :
∀𝑢 𝑛𝑜𝑛𝑑𝑒𝑐𝑟𝑒𝑎𝑠𝑖𝑛𝑔, 𝐸𝑢(𝑤 7)% ≤ 𝐸𝑢(𝑤 7? )
OR
• 𝒘7𝟏 is obtained from 𝒘 7𝟐 by adding a nonnegative noise 𝝐D to the possible outcomes of
𝒘
7𝟐 : 𝑤
7% = 𝑤
7% + 𝜀̃, where 𝜖̃ ≥ 0 with probability 1

3.3. Main limits of FSD and SSD

• Incomplete orderings (just a subset of random variables can be ordered).

• Not able to characterize the level of \riskiness" of a single random variable.
• Need for an index of risk that would parallel the Arrow Pratt's index of risk aversion.
• A good index should both increase with more dispersion (SSD), and a lower location (FSD).