Introduction to Stochastic
Dominance
Ekta Selarka
Portfolio dominance
• The approach to the maximization of expected utility discussed
so far is based on the assumption that the preferences of the
decision makers are known, easily obtained or quantified.
• In a number of cases one may be confronted with the necessity
of making a prediction about a decision maker’s preferences
between risky prospects with limited or no knowledge of the
underlying utility function.
• Under these conditions the decision-theoretic approach is of
limited value.
• SD is introduced to help solve this problem. It helps to resolve
risky choices while making the weakest possible assumptions.
The most general form of SD makes no assumptions about the
form of the probability distribution.
Stochastic dominance
• The concept of stochastic dominance is
designed to capture the technical properties
of statistical distributions for lotteries that
enable broad rankings of those lotteries (with
only limited information about the utility
function of a particular consumer).
• Practically speaking, it is a way of comparing
different lotteries or distributions of
outcomes.
State-by-State Dominance
State-by-state dominance => incomplete ranking
investment 3 state by state dominates 1.
Advantages: Stochastic dominance
1. Stochastic dominance can be defined independently of the
specific trade-offs (between return, risk and other characteristics
of probability distributions) represented by an agent's utility
function. (“risk-preference-free”).
2. Less “demanding” than state-by-state dominance
Stochastic Dominance - Notions
Let L1 be a lottery with cumulative distribution F(x) and L2 be a
lottery with cumulative distribution G(x). One approach to
comparing these lotteries (and thus examining stochastic
dominance) is to ask the following two questions:
1) When can we say that everyone will prefer L1 to L2?
2) When can we say that anyone who is risk averse will prefer L1
to L2?
[Link]-order stochastic dominance: when a
lottery L1 dominates L2 in the sense of first-
order stochastic dominance, the decision
maker prefers L1 to L2 regardless of what u
is, as long as it is weakly increasing.
2. Second-order stochastic dominance: when
a lottery L1 dominates L2 in the sense of
second-order stochastic dominance, the
decision maker prefers L1 to L2 as long as he is
risk averse and u is weakly increasing.
Stochastic dominance – first order
Let FA(x) and FB(x),respectively, represent the cumulative
distribution functions of two random variables (cash payoffs) that,
without loss of 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) for all X
FA(Xi) < FB(Xi) for some Xi
• A FSD B if the CDF of B always lies to the left of the CDF of A (receives
greater wealth from it in every state of nature)
Stochastic dominance – first order
Example1
• Let X ~ U(0,1) and Y ~ U(0,2). Then
Fx(x) = x,0 ≤ x ≤ 1
1, x ≥ 1
x / 2,0 ≤ x ≤ 2
Fy(x) =
1, x ≥ 2
So Fx(x) ≥ Fy(x) and Fx(1)=1 > Fy(1)=1/2 and we
have Y ≥fsd X.
Example2
• Let X ~ exp(1) and Y ~exp(2)
x
• Then Fx(x) = 1- e
2x
• Fy(x) = 1-e
• Therefore X ≥fsd Y.
Example 3
0 if x < 100
Reject Bet: F1 ( x) ≡ 1 if 100 ≤ x
0 if x < 90
Accept Bet: F2 ( x) = 1 / 2 if 90 ≤ x < 110
1 if 110 ≤ x
Empirical Implementation of SD
• Previous illustration depends on continuous probability
distributions. We need to come up with ways of using SD
when the distributions are discrete. We follow the following
4-step procedure:
• Step 1: Take the outcomes for all probability distributions
and arrange them in order.
• Step 2: Write the frequencies of observations against each of
the x levels for each distribution. Some of these frequencies
will usually be zero if for example an x level is observed in
one distribution and not in the other.
• Step 3: Form the CDF starting at the first value of x.
• Step 4: Do the comparisons.
Example 1
Example 2
Observations:
1. A FSD B, guarantees that the expected utility of wealth
provided by A will be greater than that offered by B for
all increasing utility function.
E[U (W )] = ∫ U (W ) f (W ) dW
FSD & Expected Utility
f(W) f(W)
fB(W) fB(W)
fA(W) fA(W)
fi(W)
W W
F(W) U(W)
FB(W) FA(W) U (A)
0.5
U (B)
W W
A FSD B FSD & E U
Intuition of FSD
• If we can write L1 = L2+ “something good”,
then everyone should prefer L1 to L2 for the
right definition of “something good”.
Intuition of Second order stochastic
dominance
• We want to write L2=L1+”risk”, and then find the
appropriate definition of risk so that every risk averse
person will prefer L1 to L2.
• Here we assume weak preference and weak
concavity, so u’’(x) ≤ 0
• To isolate the effect of risk, we want the two
distributions to have the same mean.
• We can specify L2=L1+L3, where L3 represents “risk”
and has a conditional mean of 0 for each value of L1.
75
1/2
+25
50
1/2 1/2 -25
50
1/2
25
1/2 1/2 -50
-50
L1 L1 + L3 = L2
Mean preserving spread
• Definition: A mean-preserving spread is a
lottery with mean 0 and some variation,
meaning that it is not a degenerate lottery
with 0 as the only possible outcome.
Measuring Risk: Mean-Preserving
Spread
Consider a distribution FA(x) and randomize each
outcome x further so that the final payoff is
xB = xA + z, where z has distribution function Hx(z)
with mean zero. We have ∫ xdH x (z ) = 0
Let the resulting distribution over xB be GB(x). The
mean of G is the same as the mean of F, since the
mean of Hx(z) is zero.
Definition: A distribution G is a mean preserving
spread of F if it can be obtained in the above
manner.
Mean Preserving Spread
xB= xA+ z
where z is independent of xA and has zero mean
In our example
• We will use MPS lottery for “risk”:
• L2 = L1+L3, where L3 is a MPS for each possible value in L1.
• Description:
– L3 is the 50-50 lottery between +25 and -25 if L1=50, and L3=0 for
certain if L1=-50.
– L2 can be written as: {+75,+25,-50;1/4,1/4,1/2).
• If u(x) is concave:
– U(50) ≥ ½ u(75) + ½ u(25) [Jensen’s inequality]
• That is, every risk averse person prefers 50 for sure to 50
plus the mean-preserving spread of L3.
Let’s try again
• Suppose L1 and L2 each have expected value 50, and that L2 is riskier
than L1 (why?)
20 0
1/2 1/2
1/2 80 1/2 100
• Let’s add an additional mean-preserving spread to L1 to create L2.
How do we create L3 that transforms L1 into L2 even when L2 is
clearly riskier than L1?
Hint
• Add a mean-preserving spread to +20 to
create outcomes 0 and 100. This will require a
lottery with outcomes -20 and 80. What will
be the probabilities to give expected value of
0?
L2 = L1 + Risk
$0
4/5
-$20
1/2 $80
$20 1/5
$100
1/2 $80
For all concave u(x)
L1 = L 2
Equivalence of Mean-Preserving
Spread and SOSD
Two statements are equivalent:
• If G is a m.p.s. of F, then F SOSD G.
• If and only if G is a m.p.s. of F, then
U ( F ) = ∫ u ( x)dF ( x) ≥ ∫ u ( x)dG ( x) = U (G )
for any concave function u(x)
DEFINITION
Conditions For Converting L1 to L2
with MPS’s
1) SOSD cannot hold if
F(x) = G(x) and
E(L1|L1≤ x) < E(L2|L2 ≤ x) for any x;
2) Statement 1 is equivalent to saying that F(z)
cannot SOSD G(z) if, for any x,
x x
∫0 F ( z )dz > ∫0 G( z )dz
Thus, to determine whether F(z) SOSD’s G(z), it is
only necessary to check that:
x x
−∞
∫ F ( z )dz ≤ ∫ G( z )dz for all x
−∞
SSD & Expected Utility
fA(W) f(W)
f(W) fA(W)
fB(W)
fB(W) fi(W)
W ∆W1 ∆W2 W
F(W) U(W)
∆U2
FA(W)
1
FB(W) ∆U1
U (A)
0.5
U (B)
W W
A SSD B SSD & EU
Combining FSD and SSD
• It is possible that every risk averse consumer
prefers L1 to L2 and yet we cannot use SSD
directly.
– This occurs when L1 has a higher mean than L2, so
it is not possible to apply the definition of SSD
which assumes E(L1) = E(L2)
• However, we can compare the two lotteries if
we use both FSD and SSD
Example
• L1={$60;1}
• L2={$0,$100;1/2}
E(L1)=$60
E(L2)=$50
L2 is riskier (why?)
Using stochastic dominance
• We want to show
– The lottery L3, which is $50 for certain, SSD’s L2
– L1 FSD’s L3
• Since 60>50, by transitivity, L1 is better than L2
for any risk averse consumer.
Answer: L1 FSD L3, so every consumer prefers L1 to
L3, and L3 SSD’s L2, so every risk averse consumer
prefers L3 to L2.
Thus, for any risk averse consumer EU(L1)≥EU(L3)
≥EU(L2)
HW 1
Classify each of the following pairs of lotteries into one of the
following three categories.
Category 1: Every consumer with (weakly)
increasing utility function for money
would prefer lottery 1 to lottery 2
Category 2: Every risk averse consumer with
(weakly) increasing utility function for
money would prefer lottery 1 to lottery
2, but some consumers who are not risk
averse would prefer lottery 2.
Category 3: Some risk averse consumers would
prefer lottery 1 to lottery 2, while other
risk averse consumers would prefer
lottery 2 to lottery 1
Pairs
Pair1: Lottery 1 has equal probabilities of the
outcomes 100, 40, and 10. Lottery 2 has a 0.3
probability of the outcome 80 and 0.7
probability of the outcome 0.
Pair2: Lottery 1 has equal probabilities of the
outcomes 75 and 25. Lottery 2 has an 0.25
probability of the outcome 100, an 0.5
probability of the outcome 50, and an 0.25
probability of the outcome 0.
Pair3: Lottery 1 gives 50 with certainty. Lottery
2 has equal probabilities of the outcomes 60 and
20.
Pair4: Lottery 1 has equal probabilities of 70 and
30. Lottery 2 gives 80 with probability 0.25 and
40 with probability 0.75.
HW 2
Probability A B
.1 0 -.5
.2 .5 -.25
.4 1 1.5
.2 2 3
.1 3 4
You have estimated the following probabilities for Earnings
per Share of companies A and B. Compare A and B, using
the SSD criterion.
