Part II

Risk and Time Preferences

1
Chapter 7

Risk Preference

Until now we have discussed a consumer’s problem where if the consumer pays the
price he would get the commodity for sure. However, in real life we face many
scenarios where there is uncertainty in the outcome. In this section we will discuss
how the consumer makes choices in face of such uncertainty.
Consider the following situation, a consumer are considering whether to buy
a fire insurance or not for his new home. To buy the insurance he needs to pay
premium every year however he only gets back money in terms of compensation
from the insurance company only if there is a fire. Thus the payoff from buying an
insurance is uncertain.
In most asset markets there are uncertainty about the realization of the payoff
from the asset. Under this uncertainty how would an economic agent makes an
optimal choice. To explore this, we need to first define a preference relation over
uncertain objects and then derive a utility function that represents the preference
over uncertain objects.

7.1 Preference over Lotteries

To define a preference relation over uncertain objects, we start with a set of prizes
Z with contains N prizes, where N is finite. For most of our discussion the set of
prizes will have only monetary prizes. However, the set of prizes need not be only
in terms of money, e.g., health outcomes can define set of possible prizes. We can
have a set of prizes comprising two health states, one where the DM is healthy and
the other one where he needs medical attention.
A lottery p over the set of prizes Z = {z1 , z2 , . . . zN } is defined as follows:
N
X
p = (p1 , p2 , . . . pN ) [Link] ≥ 0, pn = 1,
n=1

where pn denotes the probability of winning the prize zn . The set of lotteries X
over the set of prizes Z is defined as,
( N
)
X
X = p ≡ (p1 , . . . pN ) ∈ RN
+| pn = 1 .
n=1

A lottery over Z can be represented in several ways. Suppose Z = {0, 1}. Consider
the following two lotteries, one, when a fair coin is tossed if H is obtained prize 0 is

3
4 CHAPTER 7. RISK PREFERENCE

earned and if T is obtained prize 1 is earned. So the probability of earning 0 and 1

both are 1/2. The second lottery is as follows, a fair dice is rolled, if an odd number
is realized then we toss a fair coin and H earns 0, but if an even number is realized
then again a fair coin is tossed and T earns 0. For this lottery also probability of
prize 0 and 1, both are half. The tree below illustrates,

Figure 7.1: Compound Lottery

The second lottery is called a compound lottery, since it is a lottery over lotteries.
Any simple lottery in X can be represented by many possible compound lotteries.
Instead of defining preferences over prizes in Z we will define preferences over the
set of lotteries X. However, our approach is consequentialist here, which means any
compound lottery will be treated same as the corresponding simple lottery over X.
So when defining preferences the above two lotteries are indifferent to each other.
Let us consider a preference relation defined over set of lotteries X, defined over
the set of prizes Z, that satisfies completeness and transitivity. To obtain a well-
defined utility function that represents this preference relation We add a technical
assumption as follows:

Continuity axiom (A1) : Given a prize set Z and any three lotteries p, q, r ∈
X such that p  q  r there exists λ ∈ (0, 1) such that the compound lottery
λp + (1 − λ)r is indifferent to q,
λp + (1 − λ)r ∼ q.
A complete and transitive preference relation that also satisfies continuity has a
utility function representation.
However, the structure of the set of lotteries implies some structure on the
possible preference. For example, consider the set of prizes Z = {z1 , z2 } such that
the consumer (1, 0)  (0, 1), that is the consumer prefers the first prize over the
second. In this case we want a lottery that put higher probability on better prize
should be preferred over a lottery that puts lower weight on the better prize, i.e.,
(.75, .25)  (.25, .75). To ensure this we assume the property of the preference
relation,

Substitution axiom (A2) Given p, q, r ∈ X if p  q then for any given λ ∈ (0, 1]

λp + (1 − λ)r  λq + (1 − λ)r
7.1. PREFERENCE OVER LOTTERIES 5

. If p ∼ q then for any given λ ∈ [0, 1]

λp + (1 − λ)r ∼ λq + (1 − λ)r.

It is straightforward to see why the substitution axiom leads to the desired

property. Consider any two lotteries over Z = {z1 , z2 }, namely, (.75, .25) and
(.25, .75). We want to show that under A2, if (1, 0)  (0, 1) then (.75, .25) 
(.25, .75). Note that we can rewrite the two lotteries as follows:

(0.75, 0.25) = 0.5(0.5, 0.5) + 0.5(1, 0);

(0.25, 0.75) = 0.5(0.5, 0.5) + 0.5(0, 1).

Thus by A2, since (1, 0)  (0, 1) we get (.75, .25)  (.25, .75).
Math Aside
If general, for any Z = {z1 , z2 } if (1, 0)  (0, 1) then for any (p, 1 − p) and (q, 1 − q)
such that p > q, we can show (p, 1 − p)  (q, 1 − q). Let us rewrite the lotteries as
follows:

(p, 1 − p) = β(αp + (1 − α)q, 1 − αp − (1 − α)q) + (1 − β)(1, 0);

(q, 1 − q) = β(αp + (1 − α)q, 1 − αp − (1 − α)q) + (1 − β)(0, 1);

Since the simple lottery and the compound lottery generate the same consequence
we can get the value of α and β such that

1 − p = β(1 − αp − (1 − α)q)
q = β(αp + (1 − α)q)

Or,
q
β= .
αp + (1 − α)q

Plugging back we get,

q
1−p= (1 − αp − (1 − α)q)
αp + (1 − α)q
(1 − p)(αp + (1 − α)q) = q(1 − αp − (1 − α)q)
q
αp + (1 − α)q =
1−p+q
q
α(p − q) = −q
1−p+q
q(p − q)
α(p − q) =
1−p+q
q
α= ; β = (1 − p + q)
1−p+q

By substitution axiom, since (1, 0)  (0, 1) we get (p, 1 − p)  (q, 1 − q).

6 CHAPTER 7. RISK PREFERENCE

7.2 Expected Utility

As already mentioned with only completeness, transitivity and continuity any pref-
erence defined over X can be represented by a utility function but the substitution
axiom gives a rather useful structure on the set of possible utility functions which
makes the analysis of optimal choice easy.

Expected Utility Theorem A complete and transitive binary preference re-

lation satisfies continuity and substitution if and only if there exists a function
u : Z → R such that for any two lotteries p, q ∈ X
N
X N
X
p%q if and only if pn u(zn ) ≥ qn u(zn ).
n=1 n=1

Note that the expression, n = 1N pn u(zn ) denote the mathematical expression

P
for the expectation over the utility value associated with the set of prizes given
lottery p. Thus the statement of the theorem implies a lottery p is preferred to q if
the expected utility from p is higher than that of q.
Note that, we compare the expected utility of a lottery and not the expected
value of a lottery. This was first noted by famous mathematician Daniel Bernoulli
in context of a betting game called St Petersberg Paradox. In that game a player
earns 2k amount for any k consecutive H until the first T appears. Even though the
expected value of the lottery is infinite no player is willing to pay above a meager
amount (average 25) for playing the game. This is because the expected utility can
still be bounded and determines the decision of the consumer.
The u(.) function is called the vNM utility function after von Neumann and
Morgenstein who provided the first formulation of the expected utility theorem.

Math Aside
Here we provide a simple proof of the expected utility theorem. Consider three
distinct prizes z1 , z2 and z3 such that z1  z2  z3 . WLOG let u(z1 ) = 1 and
u(z3 ) = 0. Find a constant ū ∈ (0, 1) such that ūz1 + (1 − ū)z3 ∼ z2 and let
u(z2 ) = ū. Such a ū exists by continuity.
Now consider any two arbitrary lotteries (p1 , p2 , p3 ) and (q1 , q2 , q3 ) both of which
can be rewritten as

(p1 , p2 , p3 ) = (p1 + p2 ū, 0, (1 − ū)p2 + p3 )

(q1 , q2 , q3 ) = (q1 + q2 ū, 0, q2 (1 − ū) + q3 )

By substitution axiom

(p1 , p2 , p3 ) % (q1 , q2 , q3 ) ⇔ p1 + p2 ū ≥ q1 + q2 ū.

This is equivalent to
3
X 3
X
EU (p) = pi zi ≥ qi zi = EU (q).
i=1 i=1
7.3. MONETARY PRIZES AND CERTAINTY EQUIVALENCE 7

7.2.1 *Critique of EU Theorem

The EU theorem is the workhorse model for analyzing behavior under uncertainty,
the application of which extends to macro, finance and various other sub-disciplines.
But it is also one of the most critiqued theory in economics. Part of behavioral
economics came as a response to understanding when and how the EU theorem
fails. The substitution axiom seems to often fail in the data.
One of the problematic pattern for EU emerges when the probability of winning
is close to zero. Consider the following example from Kahneman and Tversky
(1979). Between the two sets of lotteries as follows:
A1 = [($6000; 0.45), ($0; 0.55)]; A2 = [($3000; 0.9), ($0; 0.1)]
B1 = [($6000; 0.001), ($0; 0.999)]; B2 = [($3000; 0.002), ($0; 0.998)]
KT(1979) found that A2(86%) and B1 (73% ) are popular winners. According to
the EU theorem,
A2  A1 ⇔ 0.45u(6000) + 0.55u(0) < 0.9u(3000) + 0.1u(0)
u(6000) + u(0) < 2u(3000)
and
B1  B2 ⇔ 0.001u(6000) + 0.999u(0) > 0.002u(3000) + 0.998u(0)
u(6000) + u(0) > 2u(3000),
which is a clear contradiction. KT(1979) mentioned is as possibility effect where
agents behave differently for a very high or very low probability event. It is straight-
forward to check that the substitution axiom fails here.
Another example, also borrowed from KT (1979) is as follows: between the two
sets of lotteries
C1 = [($2500; 0.33), ($2300; 0.66), ($0; 0.01)]; C2 = [($2300; 1.0)]
D1 = [($2500; 0.33), ($0; 0.67)]; D2 = [($2300; 0.34), ($0; 0.66)]
Using EU theorem would lead to the conclusion that if C2  C1 then D2  D1.
But KT found popular winners to be C2(82%) and D1(83%), which violates the
prediction of the EU theorem. In this case also, the substitution axiom fails. KT
coined this behavior as certainty effect. They claimed agents behave different in
case of a certain event, over and above what is predicted by standard EU analysis.
The failure of EU theorem hints at considering prizes not necessarily in mone-
tary units. Fear and anxiety may play a significant and so does cognition. Current
behavioral research tries to incorporate psychological prizes and imperfect compre-
hension to explain the data.

7.3 Monetary Prizes and Certainty Equivalence

For the rest of our analysis we will only focus on monetary prizes. Let us assume
Z ∈ R+ contains finitely many monetary prizes. We can write a lottery over
monetary prizes as [(p1 , c1 ), (p2 , c2 )] where cn ≥ 0. Since the prize only considers
monetary prizes, we will assume that u(z) is strictly monotone, that is the consumers
strictly prefers more money than less. This allows us to represent the EU of a lottery
in a simple diagram. For example, consider the following three lotteries,
8 CHAPTER 7. RISK PREFERENCE

• L1 : $100 for sure.

• L2 : $200 with probability 0.5, $50 with probability 0.5.

• L3 : $200 with probability 0.25, $150 with probability 0.5, $50 with probability
0.25.

In a x-y plane, where the horizontal axis denotes the amount of monetary prize
and the vertical axis denote the expected utility for any lottery we can represent
the three lotteries using chords of the vNM utility function. The diagram below
illustrates,

Figure 7.2: Black: EU (L1 ), Red: EU (L2 ), Blue: EU (L3 )

To represent the EU of any sure prize, e.g., L1 , we find the point on utility
function (black dashed line) that corresponds to the sure prize ($ 100). For a
lottery between two prizes, e.g., L2 the EU lies on the chord (red dashed line). The
chord joining u($50) and u($200) denotes all possible convex combination between
the two values. Since the expected utility of L2 is the midpoint of these two values,
we can find the midpoint of the chord joining to find the EU of lottery L2 . For a
lottery between three prizes, e.g., L3 the EU lies on the cord joining third point
with the other point on the chord between first and second point (blue dashed line)
by a similar logic. For any lottery defined over finite set of prizes we can obtain the
EU by repeatedly joining the chords.
For monetary prizes we want to investigate the risk preference of different eco-
nomic agents. A relevant concept in that regard is that of certainty equivalence(CE).
Suppose a consumer is offered with a lottery and a sure prize worth of z. Certainty
equivalence is the value of z such that if he gets z with certainty he would be
indifferent between z and the lottery. The diagram below illustrates,
7.4. RISK PREFERENCE 9

Figure 7.3: Certainty Equivalence

For any lottery once we find the EU, in the diagram we need to find a value of
z that generates the same height of the utility function. This value will be called
the certainty equivalent of the lottery.

7.4 Risk Preference

In reality different consumers have different appetite for risk taking and we can
represent them by comparing a lottery with its’ certainty equivalent. For any lottery
over monetary prizes, say, p = [(p1 , c1 ), . . . , (pN , cN )] where cn ∈ R+ we can define
the expected value (in dollar terms) as follows:
X
E(p) = pn cn

We define three types of risk preferences. The expected utility function u :

R+ −→ R is risk averse if

CE(p) ≤ E(p) for all p ∈ X,

and strictly risk averse if the inequality is strict for all gambles with at least two
prizes. The expected utility function u : R+ −→ R is risk loving if

CE(p) ≥ E(p) for all p ∈ X,

and strictly risk loving if the inequality is strict for all gambles with at least two
prizes. Finally, the expected utility function u : R+ −→ R is risk neutral if

CE(p) = E(p) for all p ∈ X.

For the risk averse consumer, since the equivalent sure outcome, i.e., the cer-
tainty equivalent is better than the lottery the vNM utility function is concave. This
is the case because for a concave function if we join two points on the curve the line
10 CHAPTER 7. RISK PREFERENCE

would be below the curve, this implies the average (any convex combination) of any
two points of the curve, i.e., u(x) would be lower than the height of the curve for
the corresponding average values of x.

Figure 7.4: Risk averse

In the above diagram suppose cL and cH are the two monetary prizes. Any prize
in between these two points would generate a higher utility than the corresponding
convex combination of u(xL ) and uxH . Since the average of u(xL ) and uxH denote
the expected utility and average of cL and cH denote the expected value of the
lottery, for a risk averse agent the utility of the expected value (sure outcome of
getting expected value) is preferred over the lottery (expected utility).

Figure 7.5: Risk lover

Just the opposite logic is true for a risk-loving agent. The line joining any two
extreme points would be above the curve since the expected value would be smaller
than expected utility. Combining the two cases together we see that a risk-averse
agent would want to get a lottery that reduces the distance between cL and cH (this
is known as consumption smoothing across states) and the risk loving agent would
do the opposite. (Think whether a risk-loving agent would ever buy an insurance?)
7.4. RISK PREFERENCE 11

For a risk neutral consumer both the lottery and the sure outcome of getting
the expected value is same, hence the vNM utility function would be linear. The
diagram below illustrates the behavior of the EU function for all three types of the
consumers.

Figure 7.6: left: risk averse; middle: Risk lover, right: risk neutral

In most economic context, the standard risk preference is given by a strictly

increasing, concave utility function u : R+ −→ R. But economist have found often
the same consumer has different risk preferences for different scenarios.
Friedman and Savage argues that consumers behave like a risk-averse agent for
lower wealth levels and become more risk loving for higher wealth levels. This, they
argued, explains many financial decisions made by consumer than the standard
assumption that the consumer is always risk-averse for any level of wealth.
Moreover, Kahneman and Tversky showed that the risk-preference is also not
stable for smaller prizes. This is best explained in the following example. Suppose
a consumer, Johnny has a risk preference that can be represented by a EU function.
If Johnny refuses a bet that offer a $11 gain with 50% chance and a $10 loss with
50% chance then he will reject any lottery of $100 loss with 50 % chance and any
positive amount (up to infinity) gain with 50% chance. This shows that the risk
preference reversal holds even for a smaller change in wealth level.
12 CHAPTER 7. RISK PREFERENCE
Chapter 8

Financial Markets

In this chapter, we will explore the optimal behavior of consumers in several financial
markets. For the rest of chapter, we will consider EU preferences defined over R2+ .
The two dimensions represent two states of the world. For example, state 1 can
denote a state where the DM stays healthy and state 2 can denote the state where
the DM falls sick and needs to incur medical expenditure.
Instead of defining lotteries over health outcomes, we will consider the monetary
prizes in each state of the world and define lotteries over the monetary prize. Let
c1 and c2 respectively denote the contingent consumption levels in monetary terms
in state 1 and state 2. We assume that the preference satisfies CT, A1, and A2.
Then according to the EU theorem, we can write the utility function as

EU (p) = p1 u(c1 ) + p2 u(c2 ).

Note that the utility function is additively separable in the two states, i.e., the
utility is one state is independent of the other state. If we assume the probability
of state 1 is π (which implies probability of state 2 is (1 − π)) then we can rewrite
the EU function as
πu(c1 ) + (1 − π)u(c2 ).
For a standard monotone, concave EU function we can draw the ICs as follows:

Figure 8.1: EU preferences for concave u(c)

13
14 CHAPTER 8. FINANCIAL MARKETS

The underlying assumption of risk aversion generates the standard ICs that
increase in the NE direction. For risk-neutral DM the ICs will be given by a
straight line and for the risk-loving consumer it will not have the standard shape.
Given the Ic we can find the MRS between the two states as follows,
πu0 (c1 ) + (1 − π)u0 (c2 (c1 ))c02 (c1 ) = 0
−πu0 (c1 )
=⇒ c02 (c1 ) = .
(1 − π)u0 (c2 (c1 ))
One important scenario is where the contingent consumption in both states is iden-
tical. This will be given by the 45o -line. Note that, on the 45o line the slope of the
IC is determined only by the ratio of probabilities,
−π
c1 = c2 (c1 ) =⇒ c02 = .
(1 − π)

8.1 Insurance
We can now explore how consumers make decisions in the insurance market. Con-
sider a consumer who has a wealth of W . In state 1, with probability π < 1 he faces
a health risk which would require a treatment of cost L and will drive his wealth
down to W − L. In state 2, there is no health risk, hence his wealth would be W .
Insurance is a type of financial asset that is called ”Arrow security”. An arrow
security provides a promised sum in a given state of the world. For example, suppose
the consumer purchases Y units of insurance at a premium price of γ < 1. This
means upfront he pays γY , in state 1 the DM gets Y back and nothing in state
2. Thus purchasing insurance increases the consumption in state 1 by (1 − γ)Y
amount and decreases the consumption in state 2 by γY .
An insurance is actuarially fair if the expected payout of the insurance is equal
to the upfront payment for the insurance. To illustrate, in the example, with prob-
ability π a payout of Y dollars is made, i.e., the expected payout for the insurance
is πY and the upfront payment is γY . If π = γ then the insurance is actuarially
fair. If the expected payout is lower than the upfront payment, i.e., π < γ then the
insurance is actuarially unfair.
Availability of insurance gives rise to a set of feasible contingent consumption
levels as follows:

Figure 8.2: regions of insurance on the budget line

8.2. PORTFOLIO ALLOCATION 15

The red solid line denotes the situation where the consumption in both states are
equal, i.e., full insurance is available. Any point to the right of the red line denotes
that the consumer gets more in state 1 than in state 2, which would happen under
over-insurance. To the left of the red line, we have the region of under-insurance.
However, to the left of the endowment point (W − L, L) the consumer consumes
even more in state 2, which can happen if he sells insurance and obtains additional
resources in state 2 only but loses some resources in state 1.
Under the assumption of monotonic and concave EU function, we can exclude
the region of selling insurance and over-insurance. We want to find the optimal
amount of insurance the consumer will purchase. To do so we will maximize the
expected utility of the consumer as follows:
max πu(W − C + x − γx) + (1 − π)u(W − γx)
x
FOC: π(1 − γ)u (W − C + x − γx) = (1 − π)γu0 (W − γx)
0

πu0 (W − C + x − γx) γ
=
(1 − π)u0 (W − γx) (1 − γ)
Solving for x from the FOC would give us the optimal amount of insurance.
π
Note that on the 45o the MRS (given in the LHS) would become 1−π . Thus for
a fair insurance where, π = γ we get,
π γ
= .
(1 − π) (1 − γ)
This implies if the insurance is fair, then the consumer will buy full insurance. If
γ > π, then from the FOC we get that tangency will occur at a point to the left
of 45o line, where the slope of the IC is higher. Thus for an actuarially unfair
insurance, the consumer will optimally choose to under-insure.

8.2 Portfolio Allocation

Another key application of risk preference is the portfolio allocation problem. Sup-
pose the agent has a wealth of $W , which he can invest in two possible assets. For
illustration purposes, let us assume, that one asset is risky and the other asset is
risk-free. Examples of risky assets include stocks, and examples of risk-free assets
include government bonds and term deposits.
Suppose the return from risk-free asset is given by r%, irrespective of the state
of the world. However, the risky asset gives different returns in different states. Let
us consider state 1 to be the good state where the return on the risky asset is y%.
In state 2, the bad state, the return is given by −z% where, y, z ≥ 0, and at least
one strict inequality.
Suppose the DM puts x proportion of his wealth W in the risky asset and the rest
in the risk-free asset. Then the total wealth in state 1 is given by xW (1 + y) + (1 −
x)W (1 + r), and the total wealth in state 2 is given by xW (1 − z) + (1 − x)W (1 + r).
If the probability of state 1 is π, The expected utility is given by,
EU (x) = πu(xW (1 + y) + (1 − x)W (1 + r)) + (1 − π)u(xW (1 − z) + (1 − x)W (1 + r))
Note that, since the vNM utility function is not necessarily linear, we cannot write
the riskless asset separately while calculating the expected utility.
16 CHAPTER 8. FINANCIAL MARKETS

To solve for the optimal portfolio allocation problem we find the FOC for the
EU with respect to x ∈ [0, 1], which is given by,
πu0 (xW (1 + y) + (1 − x)W (1 + r))(W (1 + y) − W (1 + r))
+(1 − π)u0 (xW (1 − z) + (1 − x)W (1 + r))(W (1 − z) − W (1 + r)) = 0.
We can solve x ∈ [0, 1] from the above FOC to find the optimal share of wealth
W that should be invested in the risky asset. Since x ∈ [0, 1], the optimal solution
can be a corner solution, i.e., x = 0(1) if the risky asset is really bad (or good)
compared to the risk-free asset.

8.3 Mutual Insurance: idiosyncratic risk

The mutual risk-sharing arrangement is fundamental to many financial arrange-
ments in the world. A few examples would be health and life insurance, annu-
ities, sharing labor income within a household, and more specifically for developing
economies the microfinance groups.
We can analyze the mutual insurance behavior using EU theory with risk-averse
agents in an endowment economy. Consider an economy with two agents, A and
B. Instead of assuming two commodities, we will consider two possible states of
the world. The Edgeworth box denotes the endowments of the two agents in the
two states.
Idiosyncratic risk refers to a condition where the total endowment in the econ-
omy is constant but one of the two agents gets a lower endowment in one of the
two state. We can show this as a point in a square EB. The EB will be square be-
cause the total endowment in both the states are the same. Given any endowment
and preferences of the two agents we can solve for the price and optimal choice of
insurance in the exchange economy. If the terms of trade are such that both the
consumers consume same amount in each state then we obtain full insurance in the
economy. The diagram below illustrates,

Figure 8.3: EB: full insurance

The point (3/2, 1/2) denotes the endowment in the square EB. In state 1, con-
sumer B suffers a loss and gets 1/2 units of contingent consumption and in state
8.4. ASSET PRICES: AGGREGATE RISK 17

2, consumer A suffers loss. Under monotonic and concave EU, they trade such
that the equilibrium allocation on the 45o . As discussed in the earlier section, this
implies both the consumers get full insurance by exchange.
Note that, in reality most often we do not observe full insurance being traded.
This hints at several possibilities, including information and incentive problems.
As a simple consider a situation where one of the two agents can hide part of his
endowment from the other agent. This will change the price and the equilibrium
allocation and end up providing only partial insurance, or even worse no insurance
at all. However, discussion of these scenarios is beyond the scope of the discussion
here.

8.4 Asset prices: aggregate risk

In contrast, aggregate risk refers to a situation where the total endowment of econ-
omy changes between state. For example, consider the following rectangular EB,

Figure 8.4: EB: aggregate risk

Using similar analysis as before we can solve for the equilibrium price of risk
and consumption levels in different state.
Note that, given there is an aggregate it is not possible to fully insure both the
two agents, since in one of the two states, there are fewer resources to begin with.
Also, referring to our discussion from part 2, we can show that if the consumption
in both states are equally preferred, the price of the contingent consumption in
the good aggregate (macro state) would be lower. This is because of the relative
abundance argument.
Along with the asymmetric endowment, the preferences also affect the price of
risk. For example, if the agent are highly risk averse, then they will be willing to
give up a significant amount of good state endowment to get a small amount of
contingent consumption in the bad state. This leads to a flatter price line as shown
below,
18 CHAPTER 8. FINANCIAL MARKETS

Figure 8.5: Left: high risk aversion, Right:low risk aversion

However, the relative abundance argument fails to explain why the equity prices
(that pays out in good states only) are much higher than the risk-free rates. In US,
the treasury bills represent risk-free asset and the difference between the equity
prices and the treasury bills (known as the equity premium) measures the price of
risk. The micro-level measurement of risk aversion points to a much lower level
of equity premium compared to what’s observed in the data. This phenomenon is
called the equity premium puzzle. There has been significant research in economics
and finance that tries to explain the puzzle, but is out of scope of our discussion.
Chapter 9

Intertemporal Choice

In this section we will consider a consumer’s lifetime choice problem over multiple
periods. To decide the optimal choice over life cycle the DM needs to trade off
consumption today against consumption tomorrow. When considering the future
the consumer may decide to borrow or save for the future. We are interested in
understanding how market forces and consumer’s own preference affect his savings
decision.

9.1 Time Preference

Suppose the life cycle of a DM is defined over T periods. The time preference
of the DM will then be defined over the set of consumption vectors (c1 , c2 . . . cT )
over T period. This vector of consumption is called consumption stream or lifetime
consumption.
The standard assumption is that the utility function over the consumption
stream is separable across time. This assumption implies consumption today does
not generate utility tomorrow. Note that, if the DM forms habit then this assump-
tion may not hold true. However, in this chapter we will consider utility functions
separable across time.
Another important feature of time preference is that the DM prefers consuming
in the present vs consuming in the future, i.e., future consumption is discounted
relative to the present. Let β ∈ (0, 1) denote the constant discount factor, i.e., $ 1
tomorrow is values as $β today. The lower the value of β the higher the discounting
there is.
One crucial assumption is that β does not change over time. For example,
consider two periods, one-year from today and one-year and one-day from today,
call them t and t + 1 resp. Then because β is constant, getting $ 1 in t + 1 is as
valuable as getting $ 1 in t as well. This form of discounting is called exponential
discounting in the literature and used ubiquitously in economics. It assumes that
consumer does not consider any particular period in present or future specially. As
we will see later in this chapter, relaxing this assumption will lead to interesting
behavior by the consumer.
With these assumption we can write the utility function for any T periods as

19
20 CHAPTER 9. INTERTEMPORAL CHOICE

follows,

T
X
u(c1 , c2 , . . . , cT ) = β t−1 v(ct ).
t=1

where v : R+ −→ R is a strictly increasing concave function. The function v(x) is

called the period utility function since it denotes the utility from consumption in
each period. A special case is when T = 2. The utility function for the consumption
over two periods would be given by,

u(c1 , c2 ) = v(c1 ) + βv(c2 )

Note that, the utility function is defined from the perspective of period 1. Once she
reached period 2 she consumes whatever is left over.
The form of the utility function over consumption stream looks similar to the EU
function discussed in the last chapter. If we treat each time period as a particular
state of the world then the life cycle utility function takes the form of a risk-averse
EU agent. Thus the concavity of the period utility function can be interpreted as
the desire for the DM to equalize consumption over time periods, just as a risk
averse agent would want to insure himself in the bad state of the world.

9.2 Intertemporal Budget Constraint

For the rest of the analysis let us assume T = 2, i.e, the DM lives for only two
periods (you can think of working/studying and retired) t and t + 1, in other words
today or tomorrow. The consumer earns mt today and mt+1 tomorrow. He can
always consume what he earns in a given period in that period but to borrow or
save there needs to be a credit market. Instead of writing the budget constraint
separately for each period we want to write the budget constraint for both periods
together.
Let us first consider the most extreme case where there is no credit market.
Hence the DM can not borrow but he can still leave some of his resources from
period t to consume in period t + 1. Then in period t the maximum amount she can
consume is mt , her income but in period t+1 she can consume up to mt +mt+1 . The
savings rate is dollar per dollar since no interest rate in paid. Thus the constraint
in this case is

ct ≤ mt ;
ct+1 = (mt − ct) + mt+1

where mt − ct denotes the amount of savings in period t which the DM can consume
in period t + 1. This generates the budget constraint in figure 9.1.
9.2. INTERTEMPORAL BUDGET CONSTRAINT 21

Figure 9.1: Intertemporal budget constraint: no credit market

Now, let us assume that the consumer can save at a rate r but still cannot
borrow. This is a fairly common assumption In this case, period t consumption is
still at a maximum of mt but in period t + 1 she can consume (1 + r)mt + mt+1 .
This is called the future value of income, it is as if the price of future consumption
is at 1 and that of today’s consumption is at 1 + [Link] budget constraint in this
case will be given by,
ct ≤ m t ;
ct+1 = (1 + r)(mt − ct ) + mt+1
Consuming today is more expensive since the consumer foregoes the rate of interest
by consuming today. The diagram below illustrate the budget constraint.

Figure 9.2: Intertemporal budget constraint: no borrowing

Finally, let us consider a perfect capital market where the consumer can save
or borrow at an interest rate r. Thus her maximum consumption in period t will
22 CHAPTER 9. INTERTEMPORAL CHOICE

be mt + m1+r
t+1
and maximum consumption at t + 1 will be (1 + r)mt + mt+1 . The
expression mt + m1+r
t+1
is called the present value of income since it assumes the price
of period t good to be 1 and future income are worth less since the consumer cannot
earn interest rate on future income. Thus the constraint in this case is

ct+1 = (1 + r)(mt − ct ) + mt+1 .

Rearranging we can write the budget constraint as

ct+1 mt+1
ct + = mt + .
1+r 1+r

Multiplying by (1 + r) on both sides we can also get

(1 + r)ct + ct+1 = (1 + r)mt + mt+1 .

Both equations represent the same constraint one in terms of present value and the
other in terms of future value. The diagram below illustrate,

Figure 9.3: Intertemporal budget constraint: perfect credit market

In reality often consumer’s face different rates for borrowing vs saving. For
example, suppose the consumer can save at rate r but needs to borrow at a rate r0
where r0 > r. Then the budget constraint would be given by,
(
(1 + r)(mt − ct ) + mt+1 for ct ≤ mt
ct+1 =
(1 + r0 )(mt − ct ) + mt+1 for ct > mt .

The diagram below illustrates,

9.3. INTERTEMPORAL CHOICE PROBLEM 23

Figure 9.4: borrow at r0 , save at r; r < r0

This type of credit market is called imperfect credit market and is most common
in the world. However, for the sake of simplicity we will continue assuming perfect
capital market throughout the chapter.

9.3 Intertemporal choice problem

The consumer’s problem in two periods is thus given by,

max v(ct ) + βv(ct+1 )

ct ,ct+1
ct+1 mt+1
[Link] + = mt + .
1+r 1+r

Note that the price ratio in this case is (1 + r). The optimization condition is same
as before,

M RS = (1 + r)
v 0 (ct )
=1+r
βv 0 (ct+1 )

v 0 (ct ) = β(1 + r)v 0 (ct+1 ) (EE)

Equation EE is called the Euler equation. Solving Euler equation and intertemporal
budget constraint together we get the optimal consumption c∗t , c∗t+1 in both periods.
If c∗t > mt , we call the consumer a net borrower and if c∗t < mt then she is a net
saver.

9.4 Changing Interest rate

Changing interest rate in this model is similar to a price change analysis from last
chapter. However, the effect of change in r is different depending on whether the
consumer is a borrower or a saver. Consider the following two diagrams where r
increases. In the left panel the consumer is a borrower and in the right she is saver.
24 CHAPTER 9. INTERTEMPORAL CHOICE

Figure 9.5: Left: Borrower, Right: Saver

One important observation is that the change in r affects both the two intercepts.
The only point that remains unaffected is (mt , mt+1 ), i.e., the endowment point. So
the budget line swivels along the endowment point in a way that the x-intercept is
smaller and the y-intercept is larger.
Increase in interest rate r has an interesting impact on the lifetime wealth of a
DM based on their asset position. If a consumer is a borrower then an increase in r
decrease his lifetime wealth, since he can borrow less against his future income. On
the other hand, an increase in r would increase the lifetime wealth of a saver. Thus
with an increase in r since the budget line becomes steeper both types of consumers
reduce ct and increase ct+1 , but as the borrower moves to a lower level of utility the
saver will move to a higher level of utility.
In an economy, the interest rate is often influenced by the central bank as a
monetary policy tool. Monetary policies are important tool for adjusting the impact
of business cycle. However, the analysis in this section highlights the challenge
of such a policy. If poorer people in the economy are more likely to borrower
in an economy then monetary policy can amplify the inequality in the economy.
Modern macroeconomic models considers this issue carefully while providing policy
recommendation.
The following diagram shows the Slutsky decomposition. As before the black
line is the old budget line and red line is the new budget line. The blue line that
goes through the optimal consumption bundle at old prices but is parallel to the
new budget line denotes the hypothetical budget line that keeps the real income
same.
The movement from A to C is the substitution effect, since only price ratio
changes. The movement from C to B is called the Wealth effect since a change in
r changes the level of wealth for a consumer. The movement from A to B is the
Price effect.
The SE is negative as always, as r increases the consumer moves away from
ct to ct+1 . But since she is a borrower the WE is such that she cuts down the
consumption of both ct and ct+1 . In this diagram the borrower remains a borrower
since the WE is stronger than the SE for ct+1 . But if the SE were stronger, she
would become net saver and her utility may increase.
9.5. *SELF-CONTROL 25

Figure 9.6: Slutsky decomposition: Borrower remains borrower

The following diagram shows the Slutsky decomposition where the borrower
becomes a saver following an increase in r,

Figure 9.7: Slutsky decomposition: r increase, borrower becomes saver

In general, following an increase in r a saver remains saver but a borrower can

become a saver as well. Whereas following a decrease in r a borrower remains
borrower but a saver can become a saver.

9.5 *Self-control
Many predictions from exponential discounting that we discussed in the earlier
section has given rise to several interesting experimental and theoretical questions.
For example, consider the following problem faced by Bob. As a New Year resolution
Bob has bought a one-year gym membership in January. It is almost April and he
26 CHAPTER 9. INTERTEMPORAL CHOICE

has has been to the gym only a couple of times. If Bob were a rational consumer
with exponential discounting then he would have never bought the membership and
not follow through his plan.
Many of us have been in Bob’s shoes one time or the other. Many experiments
in economics have found the consumers do not stick to the plan they make at the
beginning of the play. It seems the present is often treated differently compared to
the future periods.
Let us construct a simple example to illustrate the problem. Consider a rational
consumer with exponential discounting who chooses for three periods. The utility
function is given by

u(c1 , c2 , c3 ) = ln c1 + β ln c2 + β ln c3

For simplicity let us assume that r = 0 and the consumer only earns in period t = 1.
So m1 = 1330, m2 = m3 = 0. The consumer’s problem is thus given by,

max ln c1 + β ln c2 + β ln c3
c1 ,c2 ,c3

s.c1 + c2 + c3 = 1330

Given the Cobb-Douglas utility function the optimal consumption plan will be

1330 1330β 1330β 2

c∗1 = ; c∗2 = ; c∗1 = ;
1 + β + β2 1 + β + β2 1 + β + β2

If β = 32 then the optimal consumption plan is c∗1 = 630, c∗2 = 420, c∗3 = 280.
Now consider the same consumer in period t = 2. In period t = 1 she has
consumed 630, so she has 700 left. Her optimization problem from period t = 2
perspective is

max ln c2 + β ln c3
c2 ,c3

s.t. c2 + c3 = 700

Using the same analysis again we the optimal consumption plan from period t = 2
perspective is
700 700β
c∗2 = ; c∗3 =
1+β 1+β

Again for β = 32 we get c2 = 420, c3 = 280. This is same as the planned consumption
in period t = 1. This means if the consumer were to forget her plan and choose
again then she will choose the same amount again. This property is called dynamic
consistency.
In the example earlier with Bob, there was a failure of dynamic consistency since
Bob did not stick to his plan. To model this we consider another type of discounting
where the present is treated differently. The utility function is as follows:

u(c1 , c2 , c3 ) = ln c1 + δβ ln c2 + δβ ln c3

where δ ∈ [0, 1]. The extra δ term appear everywhere except the first period, i.e.,
today. Note that, if δ = 1 then we are back to the previous example. So we will
9.5. *SELF-CONTROL 27

consider cases where δ < 1. This type of discounting is called hyperbolic discounting
and is used extensively in behavioral economics, psychology and neuroeconomics.
The agent’s problem is now given by

max ln c1 + δβ ln c2 + δβ ln c3
c1 ,c2 ,c3

s.c1 + c2 + c3 = 380

and the consumption plan will become,

1330 1330δβ 1330δβ 2

c∗1 = ; c∗2 = ; c∗1 = ;
1 + δβ + δβ 2 1 + δβ + δβ 2 1 + δβ + δβ 2

As before we assume β = 32 , also δ = 12 . Then the optimal consumption plan is

c∗1 = 855, c∗2 = 285, c∗3 = 190.
Now consider the same consumer in period t = 2, she has 475 left but her utility
function has changed since from period t = 2’s perspective the new “present” is
t = 2. So her problem is now,

max ln c2 + δβ ln c3
c2 ,c3

s.t. c2 + c3 = 475.

The optimal consumption plan now is given by

475 475δβ
c∗2 = ; c∗3 =
1 + δβ 1 + δβ

Given β = 32 and δ = 12 the new consumption plan is c∗2 = 356.25 and c∗3 = 118.75.
So relative to the plan made in period t = 1 the consumer will over-consume in
period t = 2 and under-consume in period t = 3.
28 CHAPTER 9. INTERTEMPORAL CHOICE
Chapter 10

Precautionary Savings

Life cycle choices in reality is made under uncertainty. DM often does not not his
future income for sure and needs to make a savings decision based on the uncertain
future income. This relates to many real-life decisions regarding retirement and
life-time savings. In this chapter we will use insights developed in the previous
chapters to formulate a model of choice over time under uncertainty.
In the simplest model consider a consumer who lives for two-periods, young and
old. For simplicity, we assume that the discount rate is β = 1, i.e., the consumer
does not discount the future and the rate of interest is r = 0, i.e., $ 1 today generates
$ 1 tomorrow as well.

10.1 Uncertain Future Income

Consider the consumer has an utility function
U (c1 , c2 ) = v(c1 ) + v(c2 )
. In the current period he earns Y1 = Y . But period 2 income is subject to
uncertainty and is given by
(
Y − L w.p. p
Y2 =
Y w.p. 1 − p.

where L ∈ (0, Y ) denotes a loss that can occur in the future period.
We want to investigate how the uncertainty in future income affects the optimal
choice of the DM. Given the utility function and the uncertain stream of future
income, the consumer’s problem would be,
max u(c1 , c2 ) = v(c1 ) + v(c2 )
c1 ,c2
(
2Y − L − c1 w.p. p
where c2 =
2Y − c1 w.p. 1 − p.
The constraint reflects the fact that the consumer can not borrow against uncertain
part of his future income. We can rewrite the decision problem in terms only c1 as
follows:
max U (c1 ) ≡ v(c1 ) + pL v(2Y − L − c1 ) + (1 − pL )v(2Y − c1 )
c1 ∈[0,2Y −L]

29
30 CHAPTER 10. PRECAUTIONARY SAVINGS

where pL v(2Y − L − c1 ) + (1 − pL )v(2Y − c1 ) denote the expected utility from

consumption in period 2. The first order condition is given by,

dE(v(c2 ))
v 0 (c1 ) =
dc1
dpv(2Y − L − c1 ) d(1 − p)v(2Y − c1 )
= +
dc1 dc1
0 0
= pv (2Y − L − c1 ) + (1 − p)v (2Y − c1 )

The interpretation of the optimal choice condition is that the consumer wants to
equalize the expected marginal utility from two periods.

10.2 Precautionary Saving

We want to compare the above optimal choice with the choice without any uncer-
tainty. Since β = 1 and r = 0 and v(x) is strictly increasing and concave we know
without any uncertainty the DM would choose to consumer the same amount in
both periods. Thus he should choose to consume 0.5(Y + pL (Y − L) + (1 − pL )Y )
in both periods if he were choosing same expenditure in both periods.
However, as can be seen in the optimal choice problem presented in the earlier
section, the consumer rather choose to equate expected marginal utilities between
two periods. For a concave v(c), the marginal utility is decreasing in c. Thus
with uncertainty the expected marginal utility from period 2 increases significantly
with uncertainty. To equate the marginal utilities the consumer thus needs to
reduce consumption in period 1 as well. This type of savings behavior is called
precautionary savings since agents save in reaction to the uncertainty in future
income. The example below illustrates the argument.
Let us assume Y = 100, L = 100, pL = 0.5 and v(c1 ) = ln c1 . If the consumer
equalizes the expected spending then he will consume (100 + .5 ∗ 100)/2 = 75 in
both periods, i.e., in the bad state of the second period he will have 25 to consume
and in the good state 125. The diagram below illustrates,

Figure 10.1: Same Expected Spending

10.2. PRECAUTIONARY SAVING 31

In the above diagram the expected spending is equalized between two periods.
As shown, in this case the expected utility in period 2 is higher than that of period
1.
However, the optimization condition implies (where Y is in units of hundreds),
1 1 1
= +
c1 2(1 − c1 ) 2(2 − c1 )
2(1 − c1 )(2 − c1 ) = c1 (3 − 2c1 )
4 − 6c1 + 2c21 = 3c1 − 2c21
4c21 − 9c1 + 4 = 0
√
9 − 17
c1 = ≈ 0.61 < 0.75.
8
In the following diagram the consumer is equating the expected marginal utility
across both periods,

Figure 10.2: Same Expected Utility

This gives the optimal consumption path.