The Beginning

The price of an asset is determined by:

1) Investor Risk Preferences

2) Distribution of risky asset Payments
3) Distribution of Information
4) Rationality

The first two are essential to classic finance theory and the later two are important to
behavioral finance

Failure to account for item 1) is at the heart of the famous St. Petersburg Paradox.

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition

1
 St. Petersburg Paradox
A coin is flipped as long as heads is the outcome. At the first tails outcome the game
is stopped. The payoff is 2n-1 where n is the number of heads tossed. So….

E[ payoff ] = 12 (1) + 14 (2) + 81 (4) + ...

= 12 (1 + 1 + 1 + 1 + ...)
=

But would anyone actually pay infinity for this gamble… no.

Why?

Risk Aversion ≡ Diminishing Marginal Utility of Wealth

To solve this Formally, we need to build up some machinery

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition

2
 Axioms of Utility
A lottery P includes probabilities { p1 , p2 ,..., pn } and outcomes {x1 , x2 ,..., xn }

For preferences: we use P>P* to mean P is preferred to P*. P≥P* to mean P is

preferred or there is indifference between P and P*. P~P* means there is indifference
between P and P*

1) completeness: either P>P*, or P*>P, or P~P

2) transitivity: if P**≥P* and P*≥P then P**≥P
3) continuity: If P**≥P*≥P then there exists λϵ[0,1] such that P*~[λP**+(1-λ)P]
4) Independence: If λP*+(1-λ)P**> λP+(1-λ)P** for every λϵ(0,1) then P*>P
If every combination of milk and cereal is preferred to every combination of beer and
cereal then milk is preferred to beer—see be careful—this doesn’t happen with
monetary outcomes though
5) Dominance:
Given P1= λ1P*+(1-λ1)P** and P2= λ2P*+(1-λ2)P**
If P*>P** then P1> P2 iff λ1>λ2

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 3

 Utility Under Uncertainty

We need to build up some machinery to show that the “von-Neumann Morgenstern” is

the correct form of utility to use with uncertainty

Define a “primitive” or “elementary” security as one that pays off only if one of the
possible outcomes to a gamble occurs (if the payoffs are all 1 unit it is called a “pure”
security too). So…

ei =  x1 ,..., xi ,..., xn  with probabilities 0,...,1,..., 0

Assume: Utility of en  Utility of en −1  ...

which does not necessarily mean: xn  xn −1  ...

Because some states may represent a greater need for a payout than others.

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 4

 Developing vN-M Utility
Step 1. By continuity (Ax 3) Let’s say you have a personal valuation function, U(w), that
values every possible outcome on a relative basis, ei. By the continuity axiom:

U i  [0,1]  ei = U i en + (1 − U i )e1

Note: Here, U is from 0 to 1 but in reality this restriction is not necessary. The lower
and upper limit may be adjusted. One can think of Ui as a “scaled” utility…which can
be unscaled later. That will be explained shortly.

Your valuation of any one primitive security is a linear combination of the

value of the best and worst outcomes.
Now by the dominance axiom we only need to look at Ui to know if one
outcome is preferred to another. That is ei>ej if and only if (iff), Ui>Uj.

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 5

 Developing vN-M Utility
Now, we consider an extension.
Let us treat two lotteries P as compound lotteries across the elementary
securities. i.e.

 
P ~ p1e1 + ... + pn en ~  pi ei = en  piU i + e1 1 −  piU i 
i i  i 

Now, consider two lotteries P* and P**

 
P* ~ p1*e1 + ... + pn*en ~  pi*ei = en  pi*U i + e1 1 −  pi*U i 
i i  i 
 
P ** ~ p1**e1 + ... + pn**en ~  pi**ei = en  pi**U i + e1 1 −  pi**U i 
i i  i 
By the dominance axiom we can compare the gambles by looking only at the first summation
of the right hand side expressions. So, we can summarize by saying:
n n n n
P* =  pi*ei P ** =  pi**ei iff  pi*U i   pi**U i
i =1 i =1 i =1 i =1

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 6

 Developing vN-M Utility
Now, we can define the vonNeumann-Morgenstern Utility function:
V ( p ) = p 'U

Also, known as expected utility. This is the theoretically correct measure to compare things
given our utility axioms. vN-M utility is not ordinal it is cardinal. The vN-M utility
maintains ordering for linear transformations but not non-linear. It is not necessary that U
be between 0,1, because, rescaling the actual U to [0,1] is a linear transformation (you are
adding the most negative value to the U’s dividing by the range). So the original U gives an
equally valid utility.

This is half of what we need to solve the Saint Petersburg Paradox. The other half is Risk
Aversion. As we will see risk aversion is the assumption that U is a concave function of
wealth.

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 7

 Nature of Risk Aversion
U(w) We assume that U’(w) is always greater than 0.
U(w*) Risk Aversion is the assumption that U”(w) is always
E[U(w)]
less than 0. That is, U(w) is concave.

A concave function is also know as a decreasing

marginal utility function. This means that from any
π
fixed point on the curve a gamble of mean zero will
w* w*+ϵ lower utility. That is, if you have a gamble which will
w*-ϵ raise your wealth by ϵ or lower it by ϵ the expected
cew
utility will be lower than without a gamble. This is
caused by the fact that a negative second derivative
means that the utility gained by + ϵ is not as great as
the utility lost by - ϵ

π is known as the Arrow-Pratt risk premium (i.e. w*-cew), so π is the value that sets
cew such that U(w*- π)=E[U(w)]

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 8

 Deriving the Risk Premium
U ( w * − )  U ( w*) −  U '( w*) − 12  2 U "( w*)
 U ( w*) −  U '( w*)
And ,
E[U ( w*)]  E[U ( w*) + U '( w*)dw + 12 U " dw2 ]
 U ( w*) + U '( w*) E[dw] + 12 U "( w*) E[dw2 ]
 U ( w*) + 0 + 12 U "( w) w2
because E[dw] = 0 and  w2 = E[dw2 ] − E[dw]2

Let E[U ( w*)] = U ( w * − ) →

 = − 12  w2 U "( w*) U '( w*)

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 9

 Absolute and Relative Risk Aversion
Only the derivatives of the utility function matter in the previous equation for
differentiating people. Therefore, we give this ratio a name (as well as one related
measure):

-U”(w)/U’(w)=R(w): Absolute risk aversion

-w [U”(w)/U’(w)]=Rr(w): Relative risk aversion

The larger R(w), the bigger the required risk premium. The larger Rr(w) is the higher the
risk premium as a percentage of wealth must be.

ALSO:

 − U "( w) = −  (ln U '( w))

U '( w) w→ You can recover the ordinal
−  R( w)dw = ln(U '( w)) + c1 → utility function knowing only risk
aversion.
−  R ( w ) dw
 = U ( w) + c2  U ( w)
c1
e e

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 10

 Examples
Power Utility:
1 w ,   1,   0 This is an example of a CRRA or
U(w)=  Constant Relative Risk Aversion Utility function.
ln( w),  = 0 (by l'hopital's rule) We will use it a great deal, because it carries the
U '( w) = w −1 intuition that people consider equal % returns
U ''( w) = (  − 1) w − 2 and standard deviations the same, regardless of
R( w) =
(1 −  ) Rr ( w) = (1 −  )
nominal dollar value
W

Hara Utility:

( )( )

U(w)= 1 −  w +
 1− 
Hyperbolic absolute risk aversion (HARA) utility
with   1,   0, 1−w +   0,  = 1if  = −,   0 if   1
is very versatile.
( )
 −1
U '( w) =   w + Constant Absolute Risk Aversion (CARA) if
1− 
γ=-∞ and β=1;
( )
 −2
U "( w) = −  w 2
+ CRRA if γ<1 and β=0;
1− 
Quadratic if γ=2
( )
−1
R( w) = w +
1−  

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 11

 Resolution of St. Petersburg Paradox

Let us pick a risk averse utility function and apply v-NM utility to the game:
Let’s pick U(w)=SQRT(w)

i
    
 1  1
V ( p) =  piU i ( wi ) = = 2 = 2 = 
i −1 − i + i −21 − i2
1i
2 2  − 1 −  1.707
i =1 i =1 i =1 i =2 i =0  2 2
V 2  2.914

So a payment of 2.914 for sure yields the same utility as the gamble.

People have suggested making the payout structure such that the payout
overwhelms the risk aversion. The resolution there is nobody believes the payout
can actually occur. One can promise anything, but will the payout actually get
made. This lowers many of the probabilities to effectively zero. We will largely
ignore this last problem.

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 12

 A First Investment Model
Given: Know this trick with the budget constraint very well:
w(A)=w=(W0 − A)(1 + rf ) + A(1 + r ) Investing part of wealth in risk free asset and the rest in
the risky asset is equivalent to investing it all in the risk
= W0 (1 + rf ) + A(r − rf )
free asset and the same risky part in an asset that pays
the Excess return over the risk free rate.

Maximization Problem: The objective function has the budget constraint folded
into it. This is because you are maximizing expected utility
MaxA E[U ( w)]
of wealth but that is a function of A. The budget constraint
is not separated with a Lagrange multiplier. That would
cause unnecessary complication.

First Order Condition: The first order condition is arrived at by the chain rule.
E[U'(w)(r-r f )] = 0  U ( w)  w( A)
=0
w A

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 13

 Results from the First Order Condition
If E[r]=rf then A=0, why?
Look at F.O.C. =E[U’(w)]E[(r-rf)]+cov(U’(w),(r-rf))=0
A=0 is a solution because E[w]=W0(1+rf) regardless of what A is (so cov=0, E[(r-rf)]=0).
It is the only solution because every other solution of A has some uncertainty and a risk
averse person will not accept uncertainty without additional expected return.

If E[r]>rf then A>0, why?

Assume E[r]>rf but A=0 Now, cov=0 but E[(r-rf)]≠0. So F.O.C can’t be satisfied. A<0 Also
not possible because then portfolio has expected value less than the risk free rate and
no risk averse person would hold that.

So, this is a truly interesting result. Any risk averse person will hold at least some of a
risky asset that has Expected return greater than the risk free rate. It doesn’t matter how
high the risk aversion or how little above risk free the risky asset is the person will hold
something of it.

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 14

 Tool #1: The Envelope Theorem

Let’s say we want to know exactly how much the optimized utility will change if we
change a parameter.

Define: V (r , rf , w0 ) = U ( w( A*))
where A* is the optimized value of A
 V ( P1 ,...) =
 obj ( P1 ,...; c1 *...)
According to the envelope theorem:  P1  P1

That is the derivative of the derived utility. So here the change in derived utility for a given
change in rf is:  U ( w)  w
rf = E U '( w)(W0 − A) rf
 w  rf
Why,
V V  A * V  R f The FOC ensures all the derivative with respect to
= + choice variables are zero, and only one parameter is
 Rf  A* Rf  Rf  Rf
0 1 changing so the result holds

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 15

 Tool #2: Comparative Statics
Let’s say we want to know exactly how much a choice variable will change if we
change a parameter. Let’s Say we want to know dA*/dW0

Taking a total derivative of the F.O.C. yields:

FOC : E[U '( w)(r − rf )] = 0  w = W0 (1 + rf ) + A(r − rf ) 
→
E[U "( w)(r − rf )(1 + rf )]dW0 + E[U "( w)(r − rf ) 2 ]dA* = 0
E[U "( w)(r − rf )(1 + rf )]
dA * =−
dW0 E[U "( w)(r − rf ) 2 ]

More Generally:
Given V (c1 ,..., cn : p1 ,..., p m )
If we wish to find dci /dp j
V11 ... V1n   dc1 / dp j   Vc1 p j  The desired value can be found with
      Cramer’s rule analytically or by Gaussian
    = −  
Vn1 Vnn   dcn / dp j  Vc n p j 
  elimination numerically.

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 16

 “Income” vs. “Substitution” Effect
E[U "( w)(r − rf )(1 + rf )]
Let us look at the quantity: dA * =−
dW0 E[U "( w)(r − rf ) 2 ]

Intuitively, Should this be positive or negative? There are two competing effects,
which I call income and substitution. These are not exactly like the effects the
Economists mean when they use the terms:

The Income Effect says that raising current wealth should make you feel richer, and
therefore, you can risk more and still be well off.

The Substitution Effect says that raising current wealth makes you’re marginal utility
of extra wealth less so you substitute in more risk free investments.

Needless to say, which dominates depends on the nature of risk aversion of the
individual.

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 17

 dA*/dW0 as a Function of dARA/dw
E[U "( w)(r − rf )(1 + rf )] This has a positive denominator. What about the numerator?
dA * = We can use concavity of U(w) to answer this question in
dW0 − E[U "( w)(r − rf ) 2 ]
various contexts.
Let rh be a realization of r>rf , let rl be a realization of r<rf, and let wh and wl be the
corresponding wealths.
If dARA/dw<0 then R( wh )  R(W0 (1 + rf )) →
U "( wh ) Multiplying by negative reverses the
− h 
 −U '( wh )(r h − rf )    −U '( wh )(r h − rf )  R(W0 (1 + rf )) →
U '( w )
inequality
U "( wh )(r h − rf )   −U '( wh )(r h − rf )  R(W0 (1 + rf ))
Next ,
R( wl )  R(W0 (1 + rf ))
U "( wl ) Multiplying by positive doesn’t change
− l 
 −U '( wl )(r l − rf )    −U '( wl )(r l − rf )  R(W0 (1 + rf )) →
U '( w ) the inequality
U "( wl )(r l − rf )   −U '( wl )(r l − rf )  R(W0 (1 + rf ))
→ Since the inequality is all the same way
E[U "( w)(r − rf )]  − E[U '( w)(r − rf )]R(W0 (1 + rf )) = 0 the expectation is too. Zero is because of
Numerator >0 FOC

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 18

 dA*/dW0 as a Function of dARA/dw
The arguments can be made parallel for dARA/dw equal 0 and less than 0.

This leaves us with the following table:

dARA dA *
 0→ 0
dw dW0
dARA dA *
= 0→ =0
dw dW0
dARA dA *
 0→ 0
dw dW0

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 19

 dA*/dW0 as a Function of dRRA/dw
W0
Income Elasticity of A
Define  = dA
dW0 A
Add 1-A/A on the right
 = 1+ 
(dA dW0 )W0 − A
A
W0 (1 + rf ) E U "( w)(r − rf ) + AE U "( w)(r − rf ) 2   Substitute in the comparative
 = 1+ statics solution for dA/dW0
− E U "( w)(r − rf ) 2 
E U "( w)(r − rf ) w
 = 1+ This denominator is always positive again
− AE U "( w)(r − rf ) 2 

If dRRA/dw<0
w h R( wh )  W0 (1 + rf ) R(W0 (1 + rf )) This is almost the exact argument from a
w U "( w )(r − rf )  −U '( w)(r − rf )W0 (1 + rf ) R(W0 (1 + rf ))
h h h h
couple slides earlier, except now we have
and, shown η-1>0 if dRRA/dw<0.
w lU "( wl )(r l − rf )  −U '( w)(r l − rf )W0 (1 + rf ) R(W0 (1 + rf )) →
E[wU "( w. )(r − rf )]  − E[U '( w)(r − rf )]W0 (1 + rf ) R(W0 (1 + rf )) = 0

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 20

 dA*/dW0 as a Function of dRRA/dw
η-1>0 means that dA*/ dW0>A/W0 That is, if the relative risk aversion is decreasing in
wealth the percentage of wealth invested increases. Similar arguments lead us to the
following table:

dRRA dA * A *
 0→ 
dw dW0 W0
dRRA dA * A *
= 0→ =
dw dW0 W0
dRRA dA * A
 0→ 
dw dW0 W0

Note: it is possible to have increasing ARA but decreasing RRA, and other such similarities.

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 21

Arrow, K. J. Essays in the Theory of Risk Bearing (1971) Noth-Holland, Amsterdam

Bernoulli, D. “Exposition of A New Theory on the Measurement of Risk” (1954)

Econometrica 22, 23-36

Pennacchi, G. Theory of Asset Pricing (2007), Pearson, London, ch. 1

Pratt, J.W. “Risk-Aversion in the Small and in the Large” (1964) Econometrica (32),
122-136

Von Neumann, J. and Morgenstern, 0. Theory of Games and Economic Behavior

(1944) Princeton University Press, Princeton

Lecture 1--Risk, Risk Aversion, Basic Techniques and Intuition 22