* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Lecture 04: Risk Preferences and Risk Preferences and Expected
Survey
Document related concepts
Transcript
Fin 501: Asset Pricing
Lecture 04: Risk Preferences and
Expected Utility Theory
• Prof. Markus K. Brunnermeier
Slide 04
04--1
Fin 501: Asset Pricing
O
Overview:
i
Ri k Preferences
Risk
P f
1. State
State--byby-state dominance
2. Stochastic dominance
3. vNM expected utility theory
a) Intuition
b) Axiomatic
A i
ti foundations
f d ti
[DD4]
[L4]
[DD3]
4. Risk aversion coefficients and portfolio choice [DD5,L4]
5 Prudence coefficient and precautionary savings [DD5]
5.
6. Mean
Mean--variance preferences
[L4.6]
Slide 04
04--2
Fin 501: Asset Pricing
St t -by
StateState
b -state
byt t Dominance
D i
- State-by-state
y
dominance Ä incomplete
p
rankingg
- « riskier »
Table 2.1 Asset Payoffs ($)
t=0
Cost at t=0
investment 1
i
investment
t
t2
investment 3
- 1000
- 1000
- 1000
t=1
Value at t=1
π1 = π2 = ½
s=1
s=2
1050
1200
500
1600
1600
1050
- investment 3 state by state dominates 1.
Slide 04
04--3
Fin 501: Asset Pricing
St t -by
StateState
b -state
byt t Dominance
D i
((ctd.)
td )
Table 2.2 State Contingent ROR (r )
Investment 1
Investment 2
Investment 3
State
St
t C
Contingent
ti
t ROR (r
( )
s=1
s=2
Er
σ
5%
20%
12.5%
7.5%
-50%
60%
5%
55%
5%
60%
32.5%
27.5%
- Investment 1 mean-variance dominates 2
- BUT investment 3 does not m-v dominate 1!
Slide 04
04--4
Fin 501: Asset Pricing
St t -by
StateState
b -state
byt t Dominance
D i
((ctd.)
td )
Table 2.3 State Contingent Rates of Return
investment 4
investment 5
State Contingent Rates of Return
s=1
s=2
3%
5%
3%
8%
π1 = π2 = ½
E[r4] = 4%; σ4 = 1%
E[r5] = 5.5%;
5 5%; σ5 = 2.5%
2 5%
- What is the trade-off between risk and expected return?
- Investment
I
4 hhas a hi
higher
h Sharpe
Sh
ratio
i (E[r]-r
(E[ ] f)/σ
)/ than
h iinvestment 5
for rf = 0.
Slide 04
04--5
Fin 501: Asset Pricing
O
Overview:
i
Ri k Preferences
Risk
P f
1. State
State--byby-state dominance
2. Stochastic dominance
3. vNM expected utility theory
a) Intuition
b) Axiomatic
A i
ti foundations
f d ti
c) Risk aversion coefficients
[DD4]
[L4]
[DD3]
[DD4,L4]
4 Risk aversion coefficients and portfolio choice [DD5,L4]
4.
[DD5 L4]
5. Prudence coefficient and precautionary savings [DD5]
6. Mean
Mean--variance preferences
[L4.6]
Slide 04
04--6
Fin 501: Asset Pricing
St h ti Dominance
Stochastic
D i
Still incomplete
p
ordering
g
“More complete” than state-by-state ordering
State-by-state dominance ⇒ stochastic dominance
Risk preference not needed for ranking!
independently of the specific trade-offs (between return, risk and other
characteristics of probability distributions) represented by an agent
agent'ss
utility function. (“risk-preference-free”)
Next Section:
Complete preference ordering and utility
representations
H
Homework:
k Provide
P id an example
l which
hi h can be
b ranked
k d
according to FSD , but not according to state dominance.
Slide 04
04--7
Table 3-1 Sample Investment Alternatives
States of nature
Payoffs
Proba Z 1
Proba Z 2
1
2
10
100
.4
.6
.4
.4
EZ 1 = 64, σ z 1 = 44
Fin 501: Asset Pricing
3
2000
0
.2
EZ 2 = 444, σ z 2 = 779
Pr obability
F1
10
1.0
0.9
F2
0.8
07
0.7
0.6
0.5
F 1 and F 2
0.4
0.3
0.2
01
0.1
Payoff
0
10
100
2000
Slide 04
04--8
Fin 501: Asset Pricing
Fi t Order
First
O d Stochastic
St h ti Dominance
D i
Definition 33.1
1 : Let FA(x) and FB(x) , respectively,
respectively
represent the cumulative distribution functions of two
random variables ((cash payoffs)
p y ff ) that,, without loss off
generality assume values in the interval [a,b]. We say
that FA(x) first order stochastically dominates (FSD)
FB(x) if and only if for all x ∈ [a,b]
FA(x) ≤ FB(x)
H
Homework:
k Provide
P id an example
l which
hi h can be
b ranked
k d
according to FSD , but not according to state dominance.
Slide 04
04--9
Fin 501: Asset Pricing
Fi t Order
First
O d Stochastic
St h ti Dominance
D i
1
0.9
0.8
FB
0.7
FA
0.6
0.5
0.4
0.3
0.2
0.1
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
X
Slide 04
04--10
Fin 501: Asset Pricing
Table 3-2 Two Independent Investments
Investment 3
Payoff
4
5
12
Investment 4
Prob.
0 25
0.25
0.50
0.25
Payoff
1
6
8
Prob.
0 33
0.33
0.33
0.33
1
0.9
0.8
0.7
0.6
investment 4
0.5
0.4
0.3
0.2
investment 3
0.1
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
Figure 3-6 Second Order Stochastic Dominance Illustrated
Slide 04
04--11
Fin 501: Asset Pricing
S
Second
dO
Order
d St
Stochastic
h ti Dominance
D i
Definition 3.2:
3 2: Let FA ( ~x ) , FB ( ~x ) , be two
cumulative probability distribution for
random payoffs in [a, b]. We say that FA ( ~x )
second order stochastically dominates
(SSD) FB ( ~x ) if and only if for any x :
x
∫
-∞
[ FB (t) - FA (t) ] dt ≥ 0
(with strict inequality for some meaningful
interval of values of t).
Slide 04
04--12
Fin 501: Asset Pricing
Mean Preserving Spread
xB = xA + z
((3.8))
where z is independent of xA and has zero mean
for normal distributions
fA (x)
f B (x)
μ = ∫ x fA (x)dx
d = ∫ x f B (x)dx
d
~
x , Payoff
Figure 3-7 Mean Preserving Spread
Slide 04
04--13
Fin 501: Asset Pricing
Mean Preserving Spread & SSD
Theorem 3.4 : Let FA(•) and FB(•) be two distribution
functions defined on the same state space with identical
means. Then the follow statements are equivalent :
x ) SSD FB ( ~
x)
FA ( ~
x ) is a mean p
preserving
ese ving spread
sp ead of FA ( ~x )
FB ( ~
in the sense of Equation (3.8) above.
Slide 04
04--14
Fin 501: Asset Pricing
O
Overview:
i
Ri k Preferences
Risk
P f
1. State
State--byby-state dominance
2. Stochastic dominance
3. vNM expected utility theory
a) Intuition
b) Axiomatic
A i
ti foundations
f d ti
[DD4]
[L4]
[DD3]
4. Risk aversion coefficients and portfolio choice [DD4,5,L4]
5 Prudence coefficient and precautionary savings [DD5]
5.
6. Mean
Mean--variance preferences
[L4.6]
Slide 04
04--15
Fin 501: Asset Pricing
AH
Hypothetical
th ti l Gamble
G bl
Suppose someone offers you this gamble:
"I have a fair coin here. I'll flip it, and if it's tail I pay
you $1 and the gamble is over. If it's head, I'll flip
again. If it's tail then, I pay you $2, if not I'll flip again.
With every round, I double the amount I will pay to
you if it's
it s tail
tail."
Sounds like a good deal. After all, you can't loose.
So here
here'ss the question:
How much are you willing to pay to take this
gamble?
g
Slide 04
04--16
Fin 501: Asset Pricing
P
Proposal
l 1:
1 Expected
E
t d Value
V l
With probability 1/2 you get $1.
With probability 1/4 you get $2.
$2
With probability 1/8 you get $4.
etc.
Ê
(12 )1 times
(12 )2 times
(12 )3 times
20
21
22
The expected payoff is the sum of these
payoffs weighted with their probabilities,
payoffs,
probabilities
∞
t
so
∞
1
⎛1⎞
∑
t =1
t −1
=
⎜ ⎟ ⋅ 2
2⎠
⎝{
123
probability
∑2
t =1
1
=∞
payoff
Slide 04
04--17
Fin 501: Asset Pricing
An Infinitely Valuable Gamble?
You should pay
everything you own and
more to purchase the right
to take this gamble!
Yet, in practice, no one is
prepared to pay such a
high price. Why?
E
Even though
th
h the
th expected
t d
payoff is infinite, the
distribution of payoffs is
not attractive…
probability
0.5
0.4
0.3
0.2
0.1
0
0
20
40
60 $
With 93% probability
we get $8 or less, with
99% probability
b bili we get
$64 or less.
Slide 04
04--18
Fin 501: Asset Pricing
What Sho
Should
ld We Do?
How can we decide in a rational fashion about such
gambles (or investments)?
Proposal 2: Bernoulli suggests that large gains should be
weighted less. He suggests to use the natural logarithm.
[Cremer - another great mathematician of the time - suggests the
square root
root.]]
∞
∑
t =1
t
expected utility
⎛1⎞
t −1
<∞
⎜ ⎟ ⋅ ln(2 ) = ln(2) =
of gamble
2⎠
⎝{
14243
probability
Ê
utility of payoff
Bernoulli would have paid at most eln(2) = $2 to
participate in this gamble.
Slide 04
04--19
Fin 501: Asset Pricing
Ri
Risk
Riskk-Aversion
A
i andd C
Concavity
it
u(x)
( )
The shape of the von
Neumann Morgenstern (NM)
utility function contains a lot
of information.
Consider a fifty-fifty lottery
with final wealth of x0 or x1
u(x1)
E{u(x)}
u(x0)
x0
E[x]
x1 x
Slide 04
04--20
Fin 501: Asset Pricing
Ri
Risk
Riskk-aversion
i andd concavity
it
u(x)
( )
Risk-aversion means that the
certainty equivalent is smaller
than the expected prize.
prize
Ê
We conclude that a
risk-averse vNM
utility function
must be concave.
u(x1)
u(E[x])
E[u(x)]
u(x0)
x0
u-1(E[u(x)])
E[x]
x1 x
Slide 04
04--21
Fin 501: Asset Pricing
J
Jensen’s
’ IInequality
lit
Theorem 3.1
3 1 (Jensen
(Jensen’ss Inequality):
Let g( ) be a concave function on the interval
[[a,b],
] and be a random variable such that
Prob (x ∈[a,b]) =1
Suppose the expectations E(x) and E[g(x)] exist;
then
E [g ( ~
x )] ≤ g [E ( ~
x )]
Furthermore, if g(•) is strictly concave, then the
inequality is strict.
Slide 04
04--22
Fin 501: Asset Pricing
R
Representation
t ti off Preferences
P f
A preference ordering is (i) complete, (ii) transitive, (iii)
continuous and [(iv) relatively stable] can be represented
by a utility function, i.e.
(c0,cc1,…,ccS) Â (c
(c’0,cc’1,…,cc’S)
⇔ U(c0,c1,…,cS) > U(c’0,c’1,…,c’S)
(preference ordering over lotteries –
(S+1)-dimensional
(S
)
space)
p )
Slide 04
04--23
Fin 501: Asset Pricing
P f
Preferences
over P
Prob.
b Di
Distributions
t ib ti
Consider c0 fixed, c1 is a random variable
Preference ordering over probability distributions
Let
P bbe a sett off probability
b bilit distributions
di t ib ti
with
ith a finite
fi it
support over a set X,
 a (strict) preference ordering over P
P, and
Define % by p % q if q ¨ p
Slide 04
04--24
Fin 501: Asset Pricing
S states of the world
Set of all possible lotteries
Space with S dimensions
C
Can we simplify
i lif the
th utility
tilit representation
t ti off
preferences over lotteries?
Space
S
with
ith one dimension
di
i – income
i
We need to assume further axioms
Slide 04
04--25
Fin 501: Asset Pricing
E
Expected
t d Utility
Utilit Th
Theory
A binary relation that satisfies the following
three axioms if and only
y if there exists a function
u(•) such that
p  q ⇔ ∑ u(c) p(c) > ∑ u(c) q(c)
i.e. preferences correspond to expected utility.
Slide 04
04--26
Fin 501: Asset Pricing
vNM
NM Expected
E
t d Utility
Utilit Theory
Th
Axiom 1 ((Completeness
p
and Transitivity):
y)
Agents have preference relation over P (repeated)
Axiom 2 (Substitution/Independence)
For all lotteries p,q,r ∈ P and α ∈ (0,1],
p < q iff α p + ((1-α)) r < α q + ((1-α)) r ((see next slide))
Axiom 3 (Archimedian/Continuity)
For all lotteries p,q,r
p q r ∈ P,
P if p  q  r,
r then there
exists a α , β ∈ (0,1) such that
α p + (1( α) r  q  β p + ((1 - β) r..
Problem: p you get $100 for sure, q you get $ 10 for sure, r you are killed
Slide 04
04--27
Fin 501: Asset Pricing
I d
Independence
d
Axiom
A i
Independence of irrelevant alternatives:
π
p<q
p
π
q
<
⇔
r
r
Slide 04
04--28
Fin 501: Asset Pricing
Allais Paradox –
Violation of Independence Axiom
10%
10’
≺
0
9%
15’
0
Slide 04
04--29
Fin 501: Asset Pricing
Allais Paradox –
Violation of Independence Axiom
10%
10’
≺
9%
0
100%
15’
0
10’
90%
15’
Â
0
0
Slide 04
04--30
Fin 501: Asset Pricing
Allais Paradox –
Violation of Independence Axiom
10%
10’
9%
≺
0
100%
0
10’
90%
Â
10%
15’
15’
10%
0
0
0
0
Slide 04
04--31
Fin 501: Asset Pricing
vNM
NM EU Th
Theorem
A binary relation that satisfies the axioms 1-3 if
and only
y if there exists a function u(•)
( ) such that
p  q ⇔ ∑ u(c)
( ) p(
p(c)) > ∑ u(c)
( ) q(
q(c))
i.e. p
preferences correspond
p
to expected
p
utility.
y
Slide 04
04--32
Fin 501: Asset Pricing
E
Expected
t d Utility
Utilit Th
Theory
U(Y)
U( Y0 + Z 2 )
~
U( Y0 + E( Z))
~
EU( Y0 + Z)
U( Y0 + Z1 )
~
CE( Z)
Y0
Y0 + Z1
Π
~
~
CE ( Y0 + Z) Y0 + E(Z)
Y0 + Z 2
Y
Slide 04
04--33
Fin 501: Asset Pricing
Expected Utility & Stochastic Dominance
Theorem 3. 2 : Let FA ( ~x ) , FB ( ~x ) , be two cumulative
x ∈ [a , b].
probability
b bilit distribution
di t ib ti for
f random
d payoffs
ff ~
Then FA ( ~x ) FSD FB ( ~x ) if and only if
for all non decreasing utility functions U(
U(•)).
E A U(~
x) ≥ E B U(~
x)
Slide 04
04--34
Fin 501: Asset Pricing
Expected Utility & Stochastic Dominance
Theorem 3. 3 : Let FA ( ~x ) , FB ( ~x ) , be two cumulative
probabilityy distribution
p
x defined on [a , b] .
for random payoffs ~
Then, FA ( ~x ) SSD FB ( ~x ) if and only if E A U(~x) ≥ E B U(~x)
f all
for
ll non decreasing
d
andd concave U.
Slide 04
04--35
Fin 501: Asset Pricing
Digression:
S bj ti EU Theory
Subjective
Th
Derive perceived probability from preferences!
Set S of prizes/consequences
Set Z of states
Set of functions f(s) ∈ Z
Z, called acts (consumption plans)
Seven SAVAGE Axioms
Goes
G
bbeyond
d scope off thi
this course.
Slide 04
04--36
Fin 501: Asset Pricing
Digression:
Ell b
Ellsberg
Paradox
P d
10 balls in an urn
Lottery 1: win $100 if you draw a red ball
L tt
Lottery
2:
2 win
i $100 if you ddraw a bl
blue bball
ll
Uncertainty: Probability distribution is not known
Risk:
Probability distribution is known
(5 balls are red, 5 balls are blue)
Individuals are “uncertainty/ambiguity averse”
(non-additive probability approach)
Slide 04
04--37
Fin 501: Asset Pricing
Digression: Prospect
P
t Th
Theory
Value function ((over ggains and losses))
Overweight low probability events
Experimental evidence
Slide 04
04--38
Fin 501: Asset Pricing
I diff
Indifference
curves
x2
Any point in
this plane is
a particular
lottery.
Where is the
set of riskffree
lotteries?
45°
x1
If x1=x
x2,
then the
lottery
contains no
risk.Slide 04
04--39
Fin 501: Asset Pricing
I diff
Indifference
curves
x2
π
z
45°
z
x1
Where is the
set of lotteries
with expected
prize E[L]=z?
It's a
straight line,
and the
slope is
given by the
relative
probabilities
of the two
states.
Slide 04
04--40
Fin 501: Asset Pricing
Suppose the
agent is risk
averse. Where
is the set of
lotteries which
are indifferent
to (z,z)?
That's not
right! Note
th t there
that
th
are risky
lotteries
with smaller
expected
x1 prize and
which are
Slide 04
04--41
preferred.
I diff
Indifference
curves
???
x2
π
z
45°
z
Fin 501: Asset Pricing
I diff
Indifference
curves
x2
π
z
So the
indifference
curve must be
tangent to the
iso-expectedprize line.
This is a direct
p
of
implication
risk-aversion
alone.
45°
z
x1
Slide 04
04--42
Fin 501: Asset Pricing
I diff
Indifference
curves
x2
π
But riskrisk
aversion does
not imply
convexity.
This
indifference
d ff
curve is also
compatible
p
with riskaversion.
z
45°
z
x1
Slide 04
04--43
Fin 501: Asset Pricing
I diff
Indifference
curves
x2
∇ V(z,z)
π
z
The tangency
implies that
the gradient
of V at the
point (z,z) is
collinear to π.
Formally,
∇ V(z,z) = λπ,
for some λ>0.
45°
z
x1
Slide 04
04--44
Fin 501: Asset Pricing
C i
Certainty
Equivalent
E i l andd Risk
Ri k Premium
P
i
(3.6)
(3.7)
~
~
EU(Y + Z ) = U(Y + CE(Y, Z ))
~
~
= U(Y +E Z - Π(Y, Z ))
Slide 04
04--45
Fin 501: Asset Pricing
C i
Certainty
Equivalent
E i l andd Risk
Ri k Premium
P
i
U(Y)
U( Y0 + Z 2 )
~
U( Y0 + E( Z))
~
EU( Y0 + Z)
U( Y0 + Z1 )
~
CE( Z)
Y0
Y0 + Z1
Π
~
~
CE ( Y0 + Z) Y0 + E(Z)
Y0 + Z 2
Y
Figure
g
3-3 Certainty
y Equivalent
q
and Risk Premium: An Illustration
Slide 04
04--46
Fin 501: Asset Pricing
O
Overview:
i
Ri k Preferences
Risk
P f
1. State
State--byby-state dominance
2. Stochastic dominance
3. vNM expected utility theory
a) Intuition
b) Axiomatic
A i
ti foundations
f d ti
[DD4]
[L4]
[DD3]
4. Risk aversion coefficients and portfolio choice [DD4,5,L4]
5 Prudence coefficient and precautionary savings [DD5]
5.
6. Mean
Mean--variance preferences
[L4.6]
Slide 04
04--47
Fin 501: Asset Pricing
M
Measuring
i Ri
Riskk aversion
i
U(W)
tangent lines
li
U(Y+h)
U[0.5(Y+h)+0.5(Y-h)]
0.5U(Y+h)+0.5U(Y-h)
U(Y-h)
Y-h
Y
Y+h
W
Figure 3-1 A Strictly Concave Utility Function
Slide 04
04--48
Fin 501: Asset Pricing
A
ArrowArrow
-Pratt
P tt measures off risk
i k aversion
i
and their interpretations
absolute risk aversion = - U" ((Y)) ≡ R A (Y)
U' (Y)
relative risk aversion
Y U" (Y)
=≡ R R (Y)
U' ((Y))
risk tolerance
=
Slide 04
04--49
Fin 501: Asset Pricing
Ab l t risk
Absolute
i k aversion
i coefficient
ffi i t
π
Y+h
1−π
Y-h
Y
Slide 04
04--50
Fin 501: Asset Pricing
R l ti risk
Relative
i k aversion
i coefficient
ffi i t
π
Y(1+θ)
1−π
Y(1-θ)
Y
Homework: Derive this result.
Slide 04
04--51
Fin 501: Asset Pricing
CARA and
d CRRA
CRRA--utility
tilit functions
f ti
Constant Absolute RA utility function
Constant
C
R
Relative
l i RA utility
ili function
f
i
Slide 04
04--52
Fin 501: Asset Pricing
Investor ’s
’ Levell off Relative
l i Risk
i k Aversion
A
i
1− γ
( Y + CE)
1- γ
Y=0
Y=100,000
( Y + 50,000 )1 − γ 12 ( Y + 100,000 )1 − γ
=
+
1- γ
1- γ
1
2
γ=0
γ=1
γ=2
γ=5
γ = 10
γ = 20
γ = 30
CE = 75,000 (risk neutrality)
CE = 70,711
CE = 66,246
66 246
CE = 58,566
CE = 53,991
CE = 51,858
CE = 51,209
γ=5
CE = 66,530
Slide 04
04--53
Fin 501: Asset Pricing
Ri k aversion
Risk
i andd Portfolio
P tf li Allocation
All ti
No
N savings
i
ddecision
i i (consumption
(
i occurs only
l at t=1)
1)
Asset structure
One risk free bond with net return rf
One risky asset with random net return r (a =quantity of risky assets)
Slide 04
04--54
Fin 501: Asset Pricing
• Theorem 4.1: Assume U'( ) > 0,, and U"( ) < 0 and let â
denote the solution to above problem. Then
â > 0 if and only if E~r > rf
â = 0 if and only if E~r = r
f
â < 0 if and only if E~r < rf .
• Define W(a ) = E{U (Y0 (1 + rf ) + a (~r − rf ))}. The FOC can
then be written W′(a ) = E[U′(Y0 (1 + rf ) + a (~r − rf ))(~r − rf )] = 0 .
By risk aversion (U''<0)
(U''<0), W′′(a ) = E[U′′(Y0 (1 + rf ) + a (~r − rf ))(~r − rf )2 ]
< 0, that is, W'(a) is everywhere decreasing. It follows that
â will be positive if and only if W′(0) = U′(Y0 (1 + rf ))E(~r − rf ) > 0
(since then a will have to be increased from its value of 0 to
achieve equality in the FOC). Since U' is always strictly
positive this implies aâ > 0 if and only if E(~r − rf ) > 0 .
positive,
W’(a)
W
(a)
The other assertion follows similarly.
a
0
Slide 04
04--55
Fin 501: Asset Pricing
P tf li as wealth
Portfolio
lth changes
h
• Theorem 4.4 (Arrow, 1971): Let
solution to max-problem above; then:
be the
(i)
(ii)
(iii) .
Slide 04
04--56
Fin 501: Asset Pricing
P tf li as wealth
Portfolio
lth changes
h
• Theorem 4.5 (Arrow 1971): If, for all wealth levels Y,
(i)
(ii)
( )
(iii)
where
η=
da/a (elasticity)
dY /Y
Slide 04
04--57
Fin 501: Asset Pricing
L utility
Log
tilit & Portfolio
P tf li All
Allocation
ti
U(Y) = ln Y.
2 states
states, where r2 > rf > r1
Constant fraction of wealth is invested in risky asset!
Slide 04
04--58
Fin 501: Asset Pricing
Portfolio of risky assets as wealth changes
Now -- many risky assets
Theorem 4.6 (Cass and Stiglitz,1970). Let the vector
⎡ â1 ( Y0 ) ⎤
p
y invested in the J risky
y assets if
⎢ . ⎥ denote the amount optimally
⎢
⎥
⎢ . ⎥
⎡ â1 ( Y0 ) ⎤ ⎡ a1 ⎤
⎢
⎥
â
(
Y
)
⎣ J 0 ⎦
⎢ . ⎥ ⎢.⎥
⎥ = ⎢ ⎥f ( Y0 )
the wealth level is Y0.. Then ⎢
⎢ . ⎥ ⎢.⎥
⎢
⎥ ⎢ ⎥
â
(
Y
)
J
0
⎣
⎦ ⎣a J ⎦
if andd only
l if either
ith
Δ
(i) U ' ( Y0 ) = ( θY0 + κ )
(ii) U ' ( Y ) = ξe − νY0 .
or
0
In words, it is sufficient to offer a mutual fund.
Slide 04
04--59
Fin 501: Asset Pricing
LRT/HARA--utility
LRT/HARA
tilit functions
f ti
Linear Risk Tolerance/hyperbolic absolute risk aversion
Special Cases
B=0, A>0 CARA
B ≠ 0, ≠1 Generalized Power
B=1 Log utility
u(c) =ln (A+Bc)
B=-1 Quadratic Utility
u(c)=-(A-c)2
B ≠ 1 A=0
A 0 CRRA Utilit
Utility function
f ti
Slide 04
04--60
Fin 501: Asset Pricing
O
Overview:
i
Ri k Preferences
Risk
P f
1. State
State--byby-state dominance
2. Stochastic dominance
3. vNM expected utility theory
a) Intuition
b) Axiomatic
A i
ti foundations
f d ti
[DD4]
[L4]
[DD3]
4. Risk aversion coefficients and portfolio choice [DD4,5,L4]
5 Prudence coefficient and precautionary savings [DD5]
5.
6. Mean
Mean--variance preferences
[L4.6]
Slide 04
04--61
Fin 501: Asset Pricing
I t d i Savings
Introducing
S i
• Introduce savings decision: Consumption at tt=00 and tt=11
• Asset structure 1:
– risk free bond Rf
– NO risky asset with random return
– Increase Rf:
– Substitution effect: shift consumption from t=0 to t=1
⇒ save more
– Income
I
effect:
ff t agentt is
i “effectively
“ ff ti l richer”
i h ” andd wants
t to
t
consume some of the additional richness at t=0
⇒ save less
– For log-utility (γ=1) both effects cancel each other
Slide 04
04--62
Fin 501: Asset Pricing
P d
Prudence
and
dP
Pre-cautionary
Preti
Savings
S i
• IIntroduce
t d
savings
i
decision
d ii
Consumption at t=0 and t=1
• Asset structure 2:
– NO risk free bond
– One risky asset with random gross return R
Slide 04
04--63
Fin 501: Asset Pricing
P d
Prudence
and
dS
Savings
i
B
Behavior
h i
Risk aversion is about the willingness to insure …
… but not about its comparative statics.
How
H ddoes th
the behavior
b h i off an agentt change
h
when
h
we marginally increase his exposure to risk?
An old hypothesis (going back at least to
J.M.Keynes) is that people should save more now
when they face greater uncertainty in the future.
The idea is called precautionary saving and has
intuitive appeal.
Slide 04
04--64
Fin 501: Asset Pricing
P d
Prudence
and
dP
Pre-cautionary
Prei
Savings
S i
Does not directlyy follow from risk aversion alone.
Involves the third derivative of the utility function.
ba ((1990)
990) de
defines
es abso
absolute
ute p
prudence
ude ce as
Kimball
P(w) := –u'''(w)/u''(w).
p
Precautionaryy savingg if anyy onlyy if theyy are prudent.
This finding is important when one does comparative
statics of interest rates.
Prudence seems uncontroversial, because it is weaker
than DARA.
Slide 04
04--65
Fin 501: Asset Pricing
Pre--cautionary Saving (extra material)
Pre
(+)
((-))
in s
Is saving s increasing/decreasing in risk of R?
Is RHS increasing/decreasing is riskiness of R?
Is U’() convex/concave?
Depends on third derivative of U()!
N.B: For U(c)=ln c, U’(sR)R=1/s does not depend on R.
Slide 04
04--66
Fin 501: Asset Pricing
P -cautionary
PrePre
ti
Saving
S i (extra material)
2 effects: Tomorrow consumption is more volatile
• consume more today, since it’s not risky
• save more for precautionary reasons
~
~
R
R
Theorem 4.7 (Rothschild and Stiglitz,1971) : Let A , B
be
~ two
~ return distributions with identical means such that
RB = RA + e, (where e is white noise) and let s and s be,
A
B
respectively, the savings out of Y0 corresponding to the
~
~
return distributions R
and
.B
R
A
If R ' R ( Y ) ≤ 0 and RR(Y) > 1, then sA < sB ;
If R ' R ( Y ) ≥ 0 and RR(Y) < 1, then sA > sB
Slide 04
04--67
Fin 501: Asset Pricing
Prudence & PrePre-cautionary Saving
P(c) =
− U ' ' ' (c)
U ' ' (c)
− cU ' ' ' (c)
P(c)c =
U ' ' (c)
~ ~
Th
Theorem 4.8
4 8 : Let
L tR
, be
b two
t return
t
distributions
di t ib ti
suchh
A RB
~ SSD ~ , and let s and s be, respectively, the
that R
RB
A
B
A
savings out of Y0 corresponding to the return distributions
~
~
R Aand R B. Then,
s A ≥ s B iff cP(c) ≤ 2, and conversely,
iff cP(c) > 2
sA < sB
Slide 04
04--68
Fin 501: Asset Pricing
O
Overview:
i
Ri k Preferences
Risk
P f
1. State
State--byby-state dominance
2. Stochastic dominance
3. vNM expected utility theory
a) Intuition
b) Axiomatic
A i
ti foundations
f d ti
[DD4]
[L4]
[DD3]
4. Risk aversion coefficients and portfolio choice [DD4,5,L4]
5 Prudence coefficient and precautionary savings [DD5]
5.
6. Mean
Mean--variance preferences
[L4.6]
Slide 04
04--69
Fin 501: Asset Pricing
M -variance
MeanMean
i
Preferences
P f
Early researchers in finance, such as Markowitz
and Sharpe, used just the mean and the variance
of the return rate of an asset to describe it.
Mean-variance characterization is often easier
than using an vNM utility function
But is it compatible with vNM theory?
The answer is yes … approximately … under
some conditions.
Slide 04
04--70
Fin 501: Asset Pricing
M -Variance:
MeanMean
V i
quadratic
d ti utility
tilit
Suppose utility is quadratic, u(y) = ay–by2.
Expected utility is then
2
E [u ( y )] = aE [ y ] − bE [ y ]
= aE
E [ y ] − b ( E [ y ] 2 + var[[ y ]).
])
Thus, expected utility is a function of the
Thus
mean, E[y], and the variance, var[y], only.
Slide 04
04--71
Fin 501: Asset Pricing
M -Variance:
MeanMean
V i
joint
j i t normals
l
Suppose all lotteries in the domain have normally
distributed prized. (independence is not needed).
This
Thi requires
i an infinite
i fi i state space.
Any linear combination of normals is also normal.
The normal distribution is completely described by its
first two moments.
Hence, expected utility can be expressed as a function of
just these two numbers as well.
Slide 04
04--72
Fin 501: Asset Pricing
MeanMean
e -V
Variance:
ce:
linear distribution classes
Generalization of joint nomarls.
Consider a class of distributions F1, …, Fn with the
f ll i property:
following
for all i there exists (m,s) such that Fi(x) = F1(a+bx) for all x.
Thi
This iis called
ll d a li
linear di
distribution
t ib ti class.
l
It means that any Fi can be transformed into an Fj by an
appropriate shift (a) and stretch (b).
(b)
Let yi be a random variable drawn from Fi. Let μi = E{yi}
and σi2 = E{(yi–μi)2} denote the mean and the var of yi.
Slide 04
04--73
Fin 501: Asset Pricing
MeanMean
e -V
Variance:
ce:
linear distribution classes
Define then the random variable x = (yi–μi)/σi. We
denote the distribution of x with F.
Note that the mean of x is 0 and the variance is 1, and F
is p
part of the same linear distribution class.
Moreover, the distribution of x is independent of which
i we start with.
Ê
We want to evaluate the expected utility of yi ,
+∞
∫− ∞ v(z)dFi (z).
Slide 04
04--74
Fin 501: Asset Pricing
MeanMean
e -V
Variance:
ce:
linear distribution classes
But yi = μi + σi x, thus
+∞
+∞
∫− ∞ v(z)dFi (z) = ∫− ∞ v(μi + σi z)dF (z)
=: u(μ i , σi ).
The expected utility of all random variables
d
drawn
ffrom the
h same linear
l
distribution
d
b
class can be expressed as functions of the
mean and the standard deviation only.
only
Slide 04
04--75
Fin 501: Asset Pricing
M -Variance:
MeanMean
V i
small
ll risks
ik
Justification for mean-variance for the case of small risks.
use a second order (local) Taylor approximation of vNM
U(c).
( ) is concave,, second order Taylor
y approximation
pp
is a
If U(c)
quadratic function with a negative coefficient on the
quadratic term.
Expectation
E
i off a quadratic
d i utility
ili function
f
i can be
b
evaluated with the mean and variance.
Slide 04
04--76
Fin 501: Asset Pricing
Mean--Variance: small risks
Mean
Let f : R Æ R be a smooth function.
function The Taylor
approximation is
( x − x0 )1
( x − x0 )2
+ f ' ' ( x0 )
+
f ( x ) ≈ f ( x0 ) + f ' ( x0 )
1!
2!
( x − x0 )3
+L
f ' ' ' ( x0 )
3!
Use Taylor approximation for E[u(x)].
Slide 04
04--77
Fin 501: Asset Pricing
Mean--Variance: small risks
Mean
Since E[u(w+x)] = u(cCE), this simplifies to
var( x)
w − cCE ≈ RA( w)
.
2
Ê
w – cCE is the risk premium.
Ê
We see here
W
h
that
th t the
th risk
i k premium
i
is
i
approximately a linear function of the
variance of the additive risk
risk, with the
slope of the effect equal to half the
coefficient of absolute risk.
Slide 04
04--78
Fin 501: Asset Pricing
Mean--Variance: small risks
Mean
The same exercise can be done with a multiplicative risk.
Let y = gw, where g is a positive random variable with
unit mean.
Doing
D i the
th same steps
t
as before
b f
leads
l d to
t
var( g )
1 − κ ≈ RR ( w)
,
2
where κ is the certainty equivalent growth rate,
u( w) = E[u(gw)].
u(κw)
E[u(gw)]
The coefficient of relative risk aversion is
relevant for multiplicative risk
risk, absolute risk
aversion for additive risk.
Slide 04
04--79
12:30 Lecture
Risk Aversion
Ê
Fin 501: Asset Pricing
E t material
Extra
t i l follows!
f ll
!
Slide 04
04--80
Fin 501: Asset Pricing
Joint savingsaving-portfolio problem
• Consumption at t=0 and t=1. (savings decision)
• Asset structure
– One risk free bond with net return rf
– One riskyy asset ((a = quantity
q
y of riskyy assets))
max U ( Y0 − s ) + δEU ( s(1 + rf ) + a ( ~r − rf ))
{ a ,s }
FOC:
s:
a:
((4.7))
U’(ct)
= δ E[U’(ct+1)(1+rf)]
E[U’(c
E[U
(ct+1)(r-r
)(r rf)] = 0
Slide 04
04--81
Fin 501: Asset Pricing
for CRRA utility functions
s:
a:
( Y0 − s) − γ ( −1) + δE ([s(1 + rf ) + a ( ~r − rf )]− γ (1 + rf ) ) = 0
[
]
E (s(1 + rf ) + a ( ~r − rf )) − γ ( ~r − rf ) = 0
Where s is total saving and a is amount invested in risky asset.
asset
Slide 04
04--82
Fin 501: Asset Pricing
MultiM lti-period
Multi
i d Setting
S tti
Canonical framework (exponential discounting)
U(c) = E[ ∑ βt u(ct)]
prefers earlier uncertainty resolution if it affect action
indifferent, if it does not affect action
Time-inconsistent (hyperbolic discounting)
Special case: β−δ formulation
U(c) = E[u(c0) + β ∑ δt u(ct)]
Preference for the timing of uncertainty resolution
recursive utility formulation (Kreps-Porteus
(Kreps Porteus 1978)
Slide 04
04--83
Fin 501: Asset Pricing
MultiM lti-period
Multi
i d Portfolio
P tf li Choice
Ch i
Theorem 4.10 ((Merton, 1971):
) Consider the above canonical
multi-period consumption-saving-portfolio allocation problem.
Suppose U() displays CRRA, rf is constant and {r} is i.i.d.
Then a/st is time invariant.
invariant
Slide 04
04--84
Fin 501: Asset Pricing
Preference
e e e ce for
o thee
timing of uncertainty resolution
Digression:
g ess o :
π
0
$100
$150
$100
$ 25
$150
$100
Kreps-Porteus
π
$ 25
Early (late) resolution if W(P1,…) is convex (concave)
Marginal rate of temporal substitution risk aversion
Slide 04
04--85