When engaged in safe occupations, and living in healthy coun-

tries, men are much more apt to be frugal, than in unhealthy, or

hazardous occupations, and in climates pernicious to human life.
Sailors and soldiers are prodigals. In the West Indies, New Orleans,
the East Indies, the expenditure of the inhabitants is profuse. The
same people, coming to reside in the healthy parts of Europe, and
not getting into the vortex of extravagant fashion, live economically.
War and pestilence have always waste and luxury, among the other
evils that follow in their train.

1.5.1-Q2. Suppose that the pleasure that an agent gets from consumption
ct at time t depends on the amount ct 1 that he consumed at time t 1. In
particular, suppose that what he cares about is the di↵erence ct ct 1 (assume
that ct 1 = 0 when t = 0).
(i) Provide an intuitive interpretation of how this agent evaluates consump-
tion at any time.
(ii) Provide an intertemporal choice model for this agent.

1.5.1-Q3. Consider the philosophy of “living in the moment,” as ex-

pounded in the following words:

Living in the moment is all about living like there’s no tomorrow.

It takes practice, but in the end, you’ll lead a fuller life. This is your
life, now live!

(i) Model an agent who follows the above philosophy in its extreme.
(ii) Provide a behavioral implication of your model that reflects the philos-
ophy.
(iii) Suppose this agent lives for 3 periods (youth, middle age and old age),
and has a choice between the following streams at time 0 (beginning of youth):

A = ( 50, 1000, 1000), B = (1000, 0, 0) and C = ( 1, 100, 100).

Stream A denotes working miserably hard in youth, but maximizing the stan-
dards of living for the remainder of one’s life; stream B denotes enjoying youth
to the fullest without any investment in the future; and stream C denotes work-
ing moderately hard in youth, and attaining moderate living standards for the
remainder of one’s life. Formally determine which stream your agent would
choose.

1.5.1-Q4. Suppose that there are only 3 periods, and that the agent’s
preference %0 at period 0 respects the (static) Discounted Utility model where
0 < < 1, and u(c) = c.
(a) Does the agent’s preference %0 satisfy Order?

24
 (b) What is the agent’s choice in each of the following choice problems? (If
you think there is not enough information available to determine the choice,
write "not enough information")
(i) {(10, 20, 30), (30, 20, 10)}
(ii){(10, 20, 30), (20, 20, 20)}
(iii) {(30, 20, 10), (20, 20, 20)}

1.5.1-Q5. Suppose that an agent’s preference % over consumption streams

is represented by the following utility function:
1p
U (c0 , c1 ) = c0 + c1 .
2
(a) Does % satisfy Impatience?
(b) Does % satisfy Separability?
(c) Does % satisfy the following ‘Scale Independence’ property? Say that %
exhibits Scale Independence if for any c, c0 , t and t0 (where t < t0 ) and for any
k > 0,
[c, t] ⇠ [c0 , t0 ] =) [kc, t] ⇠ [kc0 , t0 ].

1.5.1-Q6. Suppose that an agent’s preference % over consumption streams

is represented by the following utility function:
✓ ◆c 0
1 1
U (c0 , c1 , c2 ) = c0 + c1 + c2 .
2 4

Consumption cannot be negative (that is, c0 , c1 , c2 0).

(a) Does % satisfy Separability?
(b) Does % satisfy Stationarity?

1.5.1-Q7. Consider the Dynamic DU model. Dynamic Consistency ex-

presses the idea that the period t + 1 self agrees with the period t self. What
behavior would express that the period t + 2 self agrees with the period t self?

1.5.2 Exercises on Addiction

1.5.2-Q1. In this problem, we consider a simple model of addiction. The
agent consumes bundles c = (x, d) where x denotes ‘normal’ consumption and
d denotes drug consumption. Take the DU model of the following form as a
benchmark:
T
X
t
DU (c0 , .., cT ) = u(ct ), where u(ct ) = xt (✓ + dt ).
t=0

Here, utility u(c) from a bundle c = (x, d) takes a particular multiplicative form
with parameters ✓ and . To keep matters simple, in what follows we suppose

25
that there are only two periods (today and tomorrow). We consider variants of
the benchmark model to highlight various aspects of addiction.
(a) Dependence refers to the need to take drugs in order to avoid the painful
e↵ects of withdrawal. Dependence can be captured as follows. Suppose

u(x0 , d0 ) = x0 (1 + d0 ) and u(x1 , d1 ) = x1 (1 2d0 + d1 ).

That is, ✓ is 1 in period 0, but 1 2d0 in period 1. Note that the higher the
value of d0 in period 0, the lower the value of ✓ in period 1.
Suppose that c0 = (1, 1) in period 0, that is, the agent uses 1 unit of normal
consumption, and takes 1 unit of drugs. He is then o↵ered the option of either
getting (1, 1) tomorrow, or (1 + x, 0). Note that the first option provides a
unit of drugs, whereas the second option o↵ers no drugs, but compensates with
greater normal consumption. Prove that no matter how large x is,

(c0 , (1 + x, 0)) (c0 , (1, 1)).

This is a behavioral expression of dependence. Intuitively, not having drugs is

so painful that he will give up very large rewards for one unit of it. Can you
also see how dependence is expressed in the utility representation?

(b) Tolerance refers to the need for more drugs in order to get the same
e↵ect. Suppose
1
u(x0 , d0 ) = x0 (1 + d0 ) and u(x1 , d1 ) = x1 (1 2d0 + d1 ),
1 + d0
1
so that in period 1, = 1+d 0
compared to the benchmark model.
Suppose that c0 = (1, 1) in period 0, with one unit of drug consumption as
before. The agent is then o↵ered the option of either getting (1, 2) tomorrow,
or (1 + x, 1). Note that the first option provides more drugs than in period 0,
whereas the second option o↵ers the same amount of drugs, but provides greater
normal consumption. Prove that no matter how large x is,

(c0 , (1 + x, 1)) (c0 , (1, 2)).

This is a behavioral expression of tolerance. Intuitively, having just the same

amount of drugs as before does not satisfy the craving, and indeed is so painful
that he will give up very large rewards for more drugs. Can you also see how
tolerance is expressed in the utility representation?

1.5.2-Q2. Studies in psychology show that drugs and discounting are

closely related. For instance:
– “on average, heroin addicts’ discount rates were twice those of controls”
(Kirby, Petry and Bickel (1999)).
– “heavy social drinkers ... and problem drinkers ... both showed greater
temporal discounting than light social drinkers” (Vuchinich and Simpson (1998)).

26
 – “a rapidly growing literature indicates that individuals with substance
dependencies might be generally more susceptible to behavioral problems that
may stem from diminished sensitivity to delayed outcomes ... drug-dependent
individuals are controlled to a greater extent by smaller more immediate conse-
quences at the expense of larger more delayed consequences” (Giordano, Bickel,
Loewenstein, Jacobs, Marsch and Badger (2002)).

Does the benchmark model of addiction in the previous problem capture

this? If not, provide the simplest extention of the model that does.

1.5.2-Q3. In his book, Addictive Thinking, psychiatrist A. Twersky writes:

For the addict, time may be measured in minutes or even seconds.

Certainly in the quest for the e↵ect of a chemical, the addict thinks
in terms of minutes....All the substances addicts use produce their
e↵ects within seconds or minutes...Addicts tell me they wouldn’t
buy a drug, regardless of its great e↵ect and of its low cost, if there
were a long delay in the onset of its action...Delay is not within the
addict’s frame of reference...The addict does think about the future,
but only in terms of moments, not years.

The author is just saying, in so many words, that addicts have very low
discount factors. To elucidate the relationship, define an agent’s time horizon
by the number of years that it takes for a unit of utility to lose (say) 90% of
its value. Thus, any given unit of future consumption is out of the agent’s time
horizon if he cannot visualize at least 10% of its utility value.
(i) Some macroeconomic estimates of discount factors give rise to a value of
approximately 0.97 (a unit of time being a year). Compute the time horizon of
an agent with such a discount factor.
1 th
(ii) What is the discount factor of an agent whose time horizon is 8760 of
a year (1 hour)?

1.5.3 Solutions for Intertemporal Choice Theory

1.5.1-A1. This passage says that when mortality risk is high, people tend
to be more impatient, and thus less frugal in their expenditures. That is, high
mortality risk is correlated with a smaller discount factor. Recall the discussion
in the lecture notes on how discounting may be viewed in terms of risk.

1.5.1-A2. This agent cares about how his consumption changes relative to
the previous period. This is captured by a model of preference over consumption
streams with the following representation
2 T
DU (c0 , .., cT ) = u(c0 ) + u(c1 c0 ) + u(c2 c1 ) + .... + u(cT cT (70)
1)
T
X
t
= u(c0 ) + u(ct ct 1 ). (71)
t=1

27
 1.5.1-A3. The extreme case can be captured by the DU model where
= 0. A peculiar behavioral implication (and there might be others) is that
the agent will never accept a reduction in current consumption, regardless of
how high a compensating increase there will be in future consumption: For any
streams (c0 , .., cT ) and (c00 , .., c0T ),

c0 > c00 =) (c0 , .., cT ) (c00 , .., c0T ).

The proof is left to you. Note that the above behavioral implication of the
model implies that the agent will choose B among streams A,B,C.

1.5.1-A4. (a) Yes. Since the preference has a utility representation, it

must satisfy Order.
(b)(i) (30,20,10). This is because
DU (10, 20, 30) = 10 + 20 + 30 2
= 10 + 20 + 10 2 + 20 2
< 10 + 20 + 10 2 + 20 since < 1
= 30 + 20 + 10 2
= DU (30, 20, 10).
That is, DU (30, 20, 10) > DU (10, 20, 30), and thus the agent chooses (30,20,10)
over (10,20,30). The intuition behind this result is that, because < 1, the agent
is impatient, and consequently prefers higher rewards earlier rather than later.
That is, he prefers to have 30 now, and 10 in period 3, rather than 10 now and
30 in period 3.
(ii) (20,20,20).
(iii) (30,20,10).

1.5.1-A5. p
(a)pNo. Observe that U [c, 0] = c and U [c, 1] = 12 c. It is indeed possible
1
that p2 c > c. Observe that
1
2 c>c
() 4c2 < c
() 4c2 c < 0
() c(4c 1) < 0. This is possible if c > 0 and 4c 1 < 0, that is, 0 < c < 14 .
Thus, Impatience is violated.
(b) Yes. Exploit the additive feature of the utility function, as in the proof
of Separability for the DU model in the lecture notes.
(c) No. Note that in this question, since there are only two periods, the con-
dition "t and t0 (where t < t0 )" required in the definition of Scale Independence
means that t = 0 and t0 = 1. So take any c, c0 and k > 0 and observe that
[c, 0] ⇠ [c0 , 1]
=) (c, 0) ⇠ (0, c0 )
0
=) U (c, 0) p= U (0, c )
1 0
=) c = 2 c
p
which in turn does NOT imply that kc = 12 kc0 , so you can’t proceed to
show that [kc, t] ⇠ [kc0 , t0 ]. A counter example is then readily constructed:

28
for instance, take c = 1 and c0 = 4 and k = 4. Then [1, 0] ⇠ [4, 1] (that is
[c, 0] ⇠ [c0 , 1]) but you’ll see that [4 ⇥ 1, 0] [4 ⇥ 4, 1] (that is, [kc, 0] [kc0 , 1]).

1.5.1-A6.
(a) No. Clearly the evaluation of two streams (c0 , c1 , c2 ) and (c0 , c01 , c02 )
depends on the value of c0 . For instance, inspect the utility function and note
that two consumption streams (c0 , c1 , c2 ) and (c0 , c01 , c02 ) may be non-indi↵erent
when c0 = 1, but they must be indi↵erent when c0 = 0.
(b) No. You can see that (0, c, 0) and (0, c0 , 0) must always be indi↵erent
because c0 = 0, but (c, 0, 0) and (c0 , 0, 0) may well be non-indi↵erent, thus
contradicting Stationarity.

1.5.4 Solutions for Addiction

1.5.2-A1.
(a) Determine that U (c0 , (1 + x, 0)) = 2 (1 + x) and U (c0 , (1, 1)) = 2.
(b) Determine that U (c0 , (1 + x, 1)) = 2 (1 + x) 21 and U (c0 , (1, 2)) = 2.

1.5.2-A2. We need to depend on how much drugs d0 the agent has

1
today. So should be a decreasing function of d0 . For instance, d0 = 1+d 0
.

1.5.2-A3. The length T ⇤ of the time horizon must satisfy the equation
T⇤
= 0.1.

1.6 Non-Standard Intertemporal Choice

1.6.1 Exercises on Beta-Delta model
1.6.1-Q1. Suppose that there are three periods, and preferences %0 , %1 , %2
in each period are respectively represented by the following utility functions:
2
U0 (c0 , c1 , c2 ) = u(c0 ) + u(c1 ) + u(c2 ),
I
(72)
U1 (c1 , c2 ) = u(c1 ) + u(c2 ), (73)
U2 (c2 ) = u(c2 ). (74)

Note that this is not the standard DU model, and neither is it the - model.
(Make sure you see why!).
(a) Can %0 exhibit preference reversals when and are strictly between 0
and 1?
(b) Are the preferences %0 and %1 dynamically consistent when and are
strictly between 0 and 1? (Use the definition in the lecture notes)
(c) Are the preferences %1 and %2 dynamically consistent when > 0?
(d) Does %0 satisfy Impatience
p if = 2 and = 1? As Later is better
(e) Suppose u(c) = c, = 1 and = 0.9. In period 0 the agent has the
choice between the three consumption plans: (2, 0, 0), (1, 2, 0), (1, 1, 1). Which
plan would he adopt?

29
 (f) Will the agent follow through with his plan in (e)?

1.6.1-Q2. Suppose that there are three periods and preferences %0 , %1

, %2 in each period are respectively represented by the following utility functions:
2
U0 (c0 , c1 , c2 ) = u(c0 ) + u(c1 ) + u(c2 ), (75)
& U1 (c1 , c2 ) = u(c1 ) + u(c2 ), (76)

&
U2 (c2 ) = u(c2 ), (77) + 1
7 1

A
.

nauseated
5) 9 v0 . 8
+
10
220x0
.

.
90 .

.7
t 0
11 9 14 4 1
10 1)3
= 0
·

, .

Plans
where u(c) = 10c, = 0.9 and = 0.8. In period 0, the agent has available 1 2) %) 10 2) 10 0)
:

2 1
, ,
0 , ,
0 , ,

sav
, , , , ,

apple, which he can either consume (and thus get immediate utility u(1) = 10) authg abt

utility without thinking

naive : max

& or save (and get immediate utility u(0) = 0). If he saves, the apple will magically 20 14 .
4 17 . 2

turn into 2 apples, which will become available to him in period 1. In period t 3 (2 0) 10 2) 4 3 = 1 , , , , ,

1, he can decide whether to consume 1 apple and save 1, or to consume both,

d throw away
option
or to save both. Each apple saved is available for consumption in period 2 (no
, 6 ,0 2
0, 1 ,
0 000

magic here). In period 2 he can either consume or throw away what is available
,
,

to him (and we assume, for simplicity, that he cannot throw away anything in
I
E of consis plan at time
previous periods)3 . follow any plac future self will
through

(a) What are the consumption plans available to the agent in period 0? (e.g. 5
(1,0,0) is a consumption plan that involves eating the apple in period 0, and --
O

thus having nothing to consume for the remaining periods). * wasting consistent No
So the other
3 are

(b) If the agent is naive, then which plan does he adopt in period 0?
(c) Which plan does he adopt in period 1?
· (2 0) G 1) 10 2) ,
, , , ,

(d) In period 2, does he go through with the plan adopted in period 1?

(e) If the agent is sophisticated, what is the set of consistent plans in period
1, given that he saves the apple in period 0?
(f) If the agent is sophisticated, what is the set of consistent plans in period
0? It
(g) Which plan does the agent choose in period 0?
(h) Does he go through with this plan?
(i) If the agent had the opportunity to commit to a plan in period 0, which
plan would he commit to? BDM
B
1.6.1-Q3. Consider the static - model.
(a) Does the model exhibit Impatience?
(b) Does the model exhibit Separability?

1.6.2 Exercises on Overconfidence

1.6.2-Q1. (Value of Information) The overconfidence model suggests that
people have some vague idea of where their abilities might lie, but may maintain
a higher degree of confidence in their abilities if they feel that it would motivate
3 The apple-story is wierd, but you can think of it as a consumption-saving problem. The

agent has some wealth in period 0, and he has to decide whether to consume it all, or to
invest it and consume it over his (short) life. For whatever reason, he has no investment
opportunities in period 1, but can save.

30
them to exert e↵ort, and thus improve the likelihood of succeeding. What does
the model say about how people might feel towards information that would
accurately reveal their true abilities? For instance, if an agent could take a test
that would reveal his true probability of succeeding, would he? Or would he
prefer ignorance?

1.6.2-Q2. (Self-Handicapping) A well-documented and puzzling finding

in psychology is that people with fragile self-confidence sometimes create ob-
stacles to their own performance. People may withhold e↵ort, prepare inade-
quately, drink alcohol before a task, choke under pressure, etc. This is called
self-handicapping. Explain this phenomenon in the context of the overconfi-
dence model.

1.6.2-Q3. (Demand for Confident Partners) People tend to seek out op-
timistic, self-confident partners rather than self-doubting ones. This holds in
the context of relationships, but also in the formation of teams, etc. Provide an
explanation.

1.6.2-Q4. When it comes to preparing for a task (e.g. an exam, a speech,

etc.), there appear to be two types of people. One type adopts the strategy of
strategic optimism, which involves setting high expectations, and then actively
avoiding thinking about what could go wrong. The second type adopts the
strategy of defensive pessimism, which involves lowering expectations and men-
tally playing through all the bad things that might happen. Messing with the
strategy adopted by either type negatively a↵ects performance! For instance,
stressing out strategic optimists by making them think about all that could go
wrong hinders their performance. Similarly, making defensive pessimists com-
placent by encouraging them not to worry hinders their performance as well.
Provide a simple formal model that would explain why one might be a strate-
gic optimist or a defensive pessimist. The model should be di↵erent from, but
can be similar in spirit to the model outlined in these notes. It can involve a
di↵erent environment, di↵erent relationships between e↵ort and probability of
success, di↵erent variables that the agent can manipulate, etc.

1.6.3 Exercise on Understanding Your Parents

1.6.3-A1. If Mom could reward her child, then what would she do? Is the
prediction of the model realistic? Why or why not?

1.6.4 Solutions for beta-delta model

1.6.1-A1.
(a) Yes. The preference %0 has a (static) beta-delta representation, and we
know that beta-delta representations exhibit preference reversals.
(b) Yes. As in the proof of the dynamic consistency of the dynamic DU
model, observe that for any c, c1 , c2 , c01 , c02 ,

31
 (c, c1 , c2 ) 0 (c, c01 , c02 )
=) U0 (c, c1 , c2 ) > U0 (c, c01 , c02 )
=) u(c) + u(c1 ) + 2 u(c2 ) > u(c) + u(c01 ) + 2 u(c02 )
=) u(c1 ) + 2 u(c2 ) > u(c01 ) + 2 u(c02 )
=) u(c1 ) + u(c2 ) > u(c01 ) + u(c02 )
=) U1 (c1 , c2 ) > U1 (c01 , c02 )
=) (c1 , c2 ) 1 (c01 , c02 ), thus establishing dynamic consistency.
(c) Yes. The proof is similar to that in the previous part.
2
(d) No. Intuitively, since = > 1, the agent cares more about the
future than about the present. p p
(e) (1, 1, 1). Since , U0 (2, 0, 0) = 2, U0 (1, 2, 0) = 1+.9⇥ 2 = 2.27, U0 (1, 1, 1) =
1 + .9 + .81 = 2.71, we see that (1, 1, 1) is optimal.
(f) Yes. The agent follows a DU model since = 1, and thus he is dynami-
cally consistent. Consequently, he always follows through with his plans.

1.6.1-A2.
(a) {(1, 0, 0), (0, 2, 0), (0, 1, 1), (0, 1, 0), (0, 0, 2), (0, 0, 1), (0, 0, 0)}.
(b) (0, 2, 0). The agent chooses the plan that maximizes utility U0 . The
plans (0, 1, 0), (0, 0, 1) and (0, 0, 0) cannot maximize utility because they involve
throwing away consumption in the last period. The utility of eachremaining
plan is U0 (1, 0, 0) = 10, U0 (0, 2, 0) = 20 = 14.4, U0 (0, 1, 1) = 10 (1 + ) =
13.68, U0 (0, 0, 2) = 20 2 = 12.96. Thus, (0, 2, 0) is the optimal choice.
(c) (2, 0). At t = 1, the agent maximizes U1 . Again, the plans (1, 0), (0, 1), (0, 0)
cannot be maximizers. Since U1 (2, 0) = 20, U1 (1, 1) = 10(1 + ) = 17.2, and
U1 (0, 2) = 20 = 14.4, (2, 0) is the best plan.
(d) Yes. After choosing (2, 0), there is no apple to eat, so c2 = 0.
(e) Given c0 = 0, the consistent plans in period 1 are (2, 0), (1, 1), and (0, 2).
This is determined by checking how the period 2 self would behave, conditional
on any given action by self 1. If self 1 consumes c1 = 0, then self 2 can get
payo↵ 20 by c2 = 2, and 10 by c2 = 1, and 0 by c2 = 0. Thus, c2 = 2 is the best
choice, given that self 1 consumes c1 = 0. Similarly, if self 1 consumes c1 = 1,
self 2 finds c2 = 1 optimal. Finally, if self 1 consumes c1 = 2, self 2 must choose
c2 = 0, since that is the only feasible choice.
(f) (0, 2, 0), (1, 0, 0). It is easy to see that (0, 0) is the consistent plan given
c0 = 1, because it is the only feasible plan. On the other hand, given c0 = 0,
the period 1 utility of the plans are U1 (2, 0) = 20, U1 (1, 1) = 10 + 10 =
17.2, U1 (0, 2) = 20 = 14.4. Since U1 (2, 0) is the highest, c1 = 2 is the optimal
choice given c0 = 0. Then it must be that c2 = 0.
(g) (0, 2, 0). If we compare the two consistent plans in period 0, we can see
that (0, 2, 0) gives a higher utility in period 0.
(h) Yes.
(i) The sophisticated agent in this problem does not need to commit, since
his best consistent plan is also his very best plan. If he had to commit to some
plan, he would commit to (0, 2, 0).

32
 1.6.1-A3. (a) Yes. (b) Yes. The proofs of (a) and (b) are entirely analo-
gous to those for the DU model.

1.6.5 Solutions for Overconfidence

1.6.2-A1. In general, the model implies that people would prefer ignorance
in the interest of maintaining their motivation. Someone with low abilities who
also believes that his abilities are low (that is, less than eV ) would not be
particularly averse to information since he has nothing to lose from taking the
test. But, an agent who believes that his true ability is higher than eV and tries
to maintain a level of confidence higher than eV would prefer to be ignorant:
if he finds out that his true ability is less than eV , then he may lose motivation
to take up the project. A person who is absolutely sure that his true ability is
above eV would not mind more information.
While the desire to remain ignorant might generally be seen as pathological,
the model tells us that there is such a thing as rational ignorance: when there
is too much at stake, willful ignorance can be rational.

1.6.2-A2. The idea that self-confidence is a device for motivation can ex-
plain this. Consider a version of the overconfidence model where there are many
periods and many possible projects that can be done, but only one after another.
Failure in the first project might sap one’s motivation for future tasks – the fail-
ure may reveal information about one’s true abilities (the type of information
we were talking about in the preceding application) and the possibility that
the information will be bad gives rise to an incentive to avoid the information.
Due to self-handicapping, failure does not reveal any meaningful information re-
garding one’s abilities. One can attribute the failure to inadequate preparation
rather than an intrinsic lack of ability.

1.6.2-A3. An explanation in terms of motivation is that there are positive

spill-over e↵ects associated with surrounding oneself with people who have a
positive view of their ability to succeed, and negative spill-over e↵ects associated
with surrounding oneself with people who see all that can go wrong and place
a non-negligible probability on failure. Consequently, the desire to boost one’s
own self-confidence may create a demand for confident partners.

1.6.6 Solution for Understanding Your Parents

1.6.3-A1. Letting r denote reward, Mom will choose r and m such that

U C ( 5 + r, 10) U C (5 m, 10),

that is, 5 + r + 2 5 m 2 and thus r 6 m, where r 0. So she

would threaten the bad action, and reward the good action in any combination
satisfying r 6 m.
Since r = 1 and m = 0 is possible, the model admits the possibility that
Mom would spoil the kid rotten. The reason we may not see this very often in

33