1 Appendix: Exercises and Solutions

1.1 Abstract Choice Theory

1.1.1 Preference
1.1.1-Q1. Consider the preference % over A = {a, b, c} defined by
a b (1)
b c. (2)
Is this preference complete? Is it transitive?

1.1.1-Q2. Suppose that an agent prefers hot chocolate to ice cream in the
winters but flips her preference in the summer. How can this agent’s preference
be summarized in terms of the basic framework?

Exercise 1 Suppose an agent has a preference % over the infinite set of all
possible bundles of two commodities. A bundle is a pair (x, y) denoting the
quantity x of the first commodity and quantity y of the second.. Suppose that %
is defined as follows: for any (x, y) and (x0 , y 0 ),
if x > x0 then (x, y) (x0 , y 0 ); (3)
0 0
if x = x and y > y then (x, y) (x0 , y 0 ); (4)
0 0 0 0
if x = x and y = y then (x, y) ⇠ (x , y ). (5)
The agent makes a choice only according to the first good x, unless the quantity
of x is the same in both bundles in which case he then chooses according to the
second good y. This is an example of what is called a lexicographic preference.
(a) Show that this lexicographic preference is complete and transitive.
(b) An interesting fact about lexicographic preferences is that it cannot have
any utility representation (the formal proof for this is beyond the scope of this
course). Can you see how this is possible given our last proposition?
Solution 2 (a) For any (x, y) and (x0 , y 0 ) one of the three conditions in the
question will necessarily hold, and therefore a ranking will exist. Thus the pref-
erence is complete. For transitivity, suppose (x, y) % (x0 , y 0 ) % (x00 , y 00 ). Given
the definition of the preference, we see that x x0 x00 . If any one of these
inequalities is strict then we get x > x and thus (x, y) (x00 , y 00 ). If there are
00

no strict inequalities then x = x0 = x00 , and the preferences imply y y 0 y 00 .

It follows that x = x00 and y y 00 and therefore (x, y) % (x00 , y 00 ). In either
case, we have shown that (x, y) % (x0 , y 0 ) % (x00 , y 00 ) implies (x, y) % (x00 , y 00 ),
as desired.
(b) The proposition states that, when A is finite, an equivalence exists.
We defined the preference for an infinite domain A, so the proposition does not
apply.

1
1.1.2 Choice
1.1.2-Q1.
(a) Suppose A = {a, b, c} and an agent’s preference % is given by

a b, (6)
b c, (7)
c a. (8)

(i) Does % satisfy Completeness?

(ii) Does it satisfy Transitivity?
(iii) Determine C% ({a, b}), C% ({a, c}), C% ({b, c}) and C% ({a, b, c}).

(b) Redo for the preference % defined by

a ⇠ b (9)
a ⇠ c (10)
b c (11)

1.1.2-Q2. Let A = {a, b, c} and suppose that all you know about an agent’s
preference % is that it satisfies Transitivity, and that

a b, (12)
b c. (13)

What can you infer about the agent’s preference between a and c?

1.1.2-Q3. Let A = {a, b, c}. Suppose that you don’t know the agent’s pref-
erences, but that you observe him making the following choices:

C({a, b}) = {a, b}, (14)

C({a, c}) = {c}, (15)
C({b, c}) = {c}. (16)

Using the notion of revealed preference, deduce the agent’s preferences. Do his
preferences satisfy Order?

1.1.2-Q4. Redo question 3 assuming that the choices are

C({a, b}) = {a}, (17)

C({a, c}) = {c}, (18)
C({b, c}) = {b}. (19)

1.1.2-Q5. (Important) Suppose that you are told that an agent’s choices
maximize some (not necessarily complete and transitive) preference %.

2
 (i) Prove that if a 2 C({a, b}) then a % b.
(ii) Prove that if C({a, b}) = {a} then a b.
(iii) Prove that if C({a, b, c}) = {a} then a % b. Confirm by example that
C({a, b, c}) = {a} is possible even if a ⇠ b.

1.1.2-Q6. Suppose you know that the agent maximizes some preference and
all you observe is the following choice:

C({a, b, c}) = {b}.

What can you deduce about the agent’s preferences? Do they have to be com-
plete? Do they have to be transitive?

1.1.2-Q7. Suppose we observe the following choices:

C({a, b}) = {a} (20)

C({a, b, c}) = {b}. (21)

Can these choices be generated by any (not necessarily complete and tran-
sitive) preference % under the hypothesis of preference maximization?

1.1.3 Utility
1.1.3-Q1. Suppose A = {a, b, c} and consider the utility function defined by

U (a) = 10, (22)

U (b) = 8, (23)
U (c) = 4. (24)

(a) If this function U is a utility representation of a preference %, then what

can you say about %?
(b) Determine the choices generated by U .

1.1.3-Q2. Suppose A = {a, b, c} and consider the preference % defined by

a b, (25)
b c, (26)
c a. (27)

Does % satisfy Completeness? Does it satisfy Transitivity? Can there possibly

exist a utility representation for this preference? Why or why not?

1.1.3-Q3. Suppose A is the set of positive integers, that is A = {1, 2, 3, ....}.

For concreteness imagine that each integer indicates the number of pancakes
the agent may consume. Consider the utility function

U (a) = 10 a2 ,

3
defined for each a 2 A. That is, the utility from 1 pancake is U (1) = 10 1 = 9,
the utility from 2 pancakes is U (2) = 10 4 = 6, etc.
(a) What is the agent’s choice from the menu {4, 6, 8, 10}? I.e., what is
C({4, 6, 8, 10}) =?
(b) Does the agent regard pancakes as a good or a bad? (Recall from EC201
that a commodity is a good if the agent prefers more of it to less, and it is a
bad if the agent prefers less of it to more).

1.1.3-Q4. Suppose A = {1, 2, 3, 4} and consider two utility functions U and

V defined by

U (a) = 2a + 10, (28)

2
V (a) = (U (a)) (29)

for each a 2 A. Are U and V ordinally equivalent?

1.1.3-Q5. Suppose A is the set of positive integers (that is, A = {1, 2, 3, ...})
and suppose that there is a utility function U given by

U (a) = ( 2)a ,

for every number a 2 X. That is, U (1) = 2, U (2) = 4, U (3) = 8, etc.

Consider also the utility function V given by

V (a) = U (a)2 .

Are U and V ordinally equivalent?

1.1.3-Q6. (a) Suppose A is the set of weakly positive real numbers, and
consider a preference % over A that is ‘monotone’ in that ‘more is better’,
a b =) a % b. Show that this preference admits concave representations and
convex representations as well. This demonstrates that the ‘law of diminishing
marginal utility’ (which corresponds to concave utility) is not a meaningful
property of the preference.
(b) Suppose that a preference is defined over bundles (a, b), where a bundle is
a pair of weakly positive real numbers. Suppose that the utility of a bundle is a
sum of utilities U (a, b) = a+u(b). Show that the shape of any indi↵erence curve
depends on whether u is concave or convex. (For instance, draw the indi↵erence
curve for u(b) = b0.5 and u(b) = b2 ). This demonstrates that in some scenarios
the law of diminishing marginal utility does meaningfully describe a property
of preference.

1.1.4 Revealed Preference

1.1.4-Q1. We know that WARP implies preference maximization and that pref-
erence maximization implies Sen’s ↵ (among other properties). It follows that
WARP must imply Sen’s ↵. Provide a direct proof for this. That is, without
referring to preference maximization, prove that WARP implies Sen’s ↵.

4
 1.1.4-Q2. Suppose that C is rationalized by a preference % that is complete
but not transitive. Does it satisfy Sen’s ↵?

1.1.4-Q3. Suppose you observe two agents, exhibiting choices C1 and C2

respectively given by:

C1 ({apples, oranges}) = C2 ({apples, oranges}) = {apples}, (30)

C1 ({apples, grapes}) = C2 ({apples, grapes}) = {apples}, (31)
C1 ({oranges, grapes}) = C2 ({oranges, grapes}) = {oranges},(32)
C1 ({apples, oranges, grapes}) = {apples}, (33)
C2 ({apples, oranges, grapes}) = {apples, oranges}. (34)

Use the Propositions from the notes to determine whether each of the agents
can be thought of as maximizing a complete and transitive preference. For each
agent that can be thought of in such terms, derive the preference he maximizes.

1.1.4-Q4. Suppose we observe only that C({a, b}) = {a}, C({b, c}) = {b}
and C({a, c}) = {c}. Do these choices satisfy WARP? Can they be rationalized
by a complete and transitive preference? If your answers seem inconsistent with
a proposition in the notes, how can the apparent inconsistency be resolved.

1.1.4-Q5. Suppose A = {a, b, c} and that all we observe is that C({a, b}) =
{a}. Does this agent satisfy WARP (with respect to the choice data we have)?
What are the di↵erent kinds of preference relations that could rationalize the
choice data?

1.1.4-Q6. Prove that if C is rationalizable by a complete and transitive

preference, then it satisfies Sen’s : If B 0 ⇢ B, then

a, b 2 C(B 0 ) and b 2 C(B) =) a 2 C(B).

Describe what Sen’s is saying intuitively.

1.1.4-Q7. Provide a direct proof for the claim that WARP implies Sen’s .

1.1.4-Q8. Prove that if choices C are nonempty then the revealed preference
relation must be complete.

1.1.5 Solutions for Preference

1.1.1-A1. Since there is no preference ranking between a and c, the prefer-
ence % cannot be complete. Furthermore, it is not transitive either because by
transitivity, a b and b c implies a c, but for the preference %, it is not
the case that a c.

5
 1.1.1-A2. The agent’s preferences depend on the season, thus she cannot
be modelled by a single preference relation. In the winter she has a preference
%W over A = {H, I} defined by H W I and in the summer she has the
preference %S over A = {H, I} defined by I S H.

1.1.6 Solutions for Choice

1.1.2-A1.
(a) - (i) Yes
(ii) No. For instance, observe that Transitivity requires that b c and c a
imply b a, whereas for the preference, a b.
(iii) C% ({a, b}) = {a}, C% ({a, c}) = {c}, C% ({b, c}) = {b}, C% ({a, b, c}) =
. The latter holds because there is no best element in the choice set {a, b, c}:
a is better than b, b is better than c, c is better than a. Thus, our theory of
choice predict that the agent’s choice from {a, b, c} will be empty.
(b) (i) Yes.
(ii): No. For instance, if the preference satisfied Transitivity then the first
two indi↵erences would imply that b is indi↵erent to c, which is not the case.
(iii): Here’s just one answer: C% ({a, b, c}) = {a, b}. Remember that the
preference maximization states that an element is chosen if it is weakly preferred
to every other alternative in the choice set. Since a % b and a % c,1 a must be
a choice. Moreover, since b % a and b % c,2 b must also be a choice. However,
c cannot be a choice since, although c % a it is not the case that c % b. Thus
C({a, b, c}) = {a, b}.

1.1.2-A2. By Transitivity, a b and b c implies a c.

1.1.2-A3. Since choice reveals preference, we can deduce from C({a, b}) =
{a, b} that a ⇠ b, from C({a, c}) = {c} that c a, and from C({b, c}) = {c}
that c b. That is, the agent’s underlying preference % is defined by:

a ⇠ b (35)
c a (36)
c b. (37)

This preference satisfies Completeness and Transitivity, and thus Order.

1 Since weak preference means "indi↵erent OR strictly preferred", the indi↵erence a ⇠ b

imples that a is weakly preferred to b, that is, a % b. Similarly, a ⇠ c implies a % c. This is

why we can write a % b and a % c. Make sure you are clear on this.
2 The indi↵erence a ⇠ b implies not only a % b (as in the previous footnote), but also b % a.

That is, the fact a and b are indi↵erent means that each is weakly preferred to the other. This
follows by definition of weak preference (again, make sure you are clear on this). Also, since
weak preference means "indi↵erent OR strict preference", b c implies b % c. This is why we
can write b % a and b % c.

6
 1.1.2-A4. As in Q3, you can deduce that the agent’s underlying prefer-
ence % is defined by:

a b (38)
c a (39)
b c. (40)

This is obviously Complete. But it is not Transitive since, for instance, by

Transitivity, a b and b c would imply a c, but in this case it is c a.
Thus % does not satisfy Order.

1.1.2-A5.
(i) By the definition of preference maximization, a 2 C({a, b}) =) a % b.
(ii) Observe that
C({a, b}) = {a}
=) a 2 C({a, b}) and b 2 C({a, b}) (by high school math)
=) a % b and b 6% a (by definition of preference maximization)
=) a b (by definition of weak preference).
(iii) By preference maximization, C({a, b, c}) = {a} implies that it must be
that a % b. To see that a ⇠ b cannot be ruled out, note that for a preference
% defined by a ⇠ b and a ⇠ c (so that b, c are incomparable), then preference
maximization implies C({a, b, c}) = {a}.

1.1.2-A6. All that can be deduced is that the agent’s underlying prefer-
ence % satisfies:

b % a (41)
b % c. (42)

We cannot deduce the preference between a and c from the information given
in the question. Consequently, the underlying preference may or may not be
complete. They do not have to be transitive either. For instance the preference
defined by b ⇠ a, b c and c a will rationalize the choice.

1.1.2-A7 No. If there was a preference % that generated these choices, then
the first choice implies a b whereas the second choice implies b % a. But, a
standard preference cannot have both a b and b % a (it must be one or the
other, by Minimal Rationality), and so there cannot be any (minimally rational)
preference generating the choices. (If you’ve already looked at chapter 2 then
you can more simply arrive at this answer by noting the violation of Sen’s ↵.

1.1.7 Solutions for Utility Representations

1.1.3-A1.

7
 (a) The agent’s preference % is defined by:

a b (43)
b c (44)
a c. (45)

This preference % satisifes Order. (This is obvious without any calculations,

actually: if the preference % did not satisfy Order, it could not have had a
utility representation in the first place).
(b) C({a, b}) = {a}, C({a, c}) = {a}, C({b, c}) = {b} and C({a, b, c}) = {a}.

1.1.3-A2. This preference % is complete but not transitive. The prefer-

ence % cannot have a utility representation since it does not satisfy Order.

1.1.3-A3.
(a) C({4, 6, 8, 10}) = {4} since 4 has the highest utility.
(b) It is a bad since the higher the value of a, the lower the value of U (a).
That is, the agent prefers less to more.

1.1.3-A4. Yes. There are two ways to show this. One is to work out the
rankings according to both preferences and to observe that both rankings are
the same (the ranking is monotone, that is, 4 strictly better than 3, 3 strictly
better than 2, etc.). The other way is to prove it formally by using the definition
of ‘ordinal equivalence’ and showing that for any a, b 2 A,

U (a) > U (b) () V (a) > V (b) (46)

U (a) = U (b) () V (a) = V (b). (47)
p
To do this, note that V = U 2 and thus, U = V . So, if U (a) > U (b) then
2
U
p(a) > p U (b)2 and thus V (a) > V (b). Conversely, if V (a) > V (b), then
V (a) > V (b) and thus U (a) > U (b). This establishes that U (a) > U (b) ()
V (a) > V (b). A similar argument establishes that U (a) = U (b) () V (a) =
V (b), that is, if a and b are indi↵erent according to U then they are also indif-
ferent according to V , and vice versa.

1.1.3-A5. No. For instance, U (2) > U (3) but V (3) > V (2), that is,
both utility functions rank the alternatives 2 and 3 di↵erently and thus do
not generate the same ranking of alternatives. Consequently they cannot be
ordinally equivalent.

1.1.3-Q6. (a) See text.

(b) The indi↵erence curve is described by the
equation U (a, b) = k for any constant k. Therefore in a diagrammatic repre-
sentation, the equation of an indi↵erence curve is a = k u(b) (if a is on the
vertical axis). The indi↵erence curve a = k b0.5 is a bowed toward the original
whereas a = k b2 is bowed out.

8
1.1.8 Solutions for Revealed Preference
1.1.4-A1. The direct proof is as follows.

Proposition 3 If C satisfies WARP then it also satisfies Sen’s ↵.

Proof. We need to show that if a 2 B 0 ⇢ S then a 2 C(B) =) a 2 C(B 0 ).

So suppose a 2 B 0 ⇢ B and a 2 C(B). Since C(B 0 ) is nonempty, it must
contain at least one element, so pick any b 2 C(B 0 ). Note that since b 2 C(B 0 )
and a 2 B 0 ⇢ B, it follows that a, b 2 B, B 0 . We are thus in a position to
invoke WARP. We have determined that a, b 2 B, B 0 and that a 2 C(B) and
b 2 C(B 0 ). WARP then implies directly that a 2 C(B 0 ), as desired.

1.1.4-A2. Yes it does. See proposition 5.

1.1.4-A3. C1 satisfies WARP whereas C2 violates it.
1.1.4-A4. WARP is satisfied, but C is not rationalizable. Proposition
?? is not contradicted however because it presumes complete choice data while
here choice from {a, b, c} is missing.
1.1.4-A5. WARP cannot be contradicted because there isnt enough data:
it is evident from the definition of WARP that at least two choice observations
(some C(S) and C(B 0 )) are needed in order to establish a violation. Therefore,
C satisfies WARP vacuously.
All preference relations over A (not necessarily complete and transitive) that
have a b can rationalize the choice.
1.1.4-A6. The interpretation of Sen’s is that if a is revealed indi↵erent
to b in a small menu B and if b is chosen in a large menu, then a must also be
chosen in the large menu. If it wasnt, then b would be strictly revealed preferred
to a. In a spirit consistent with WARP, Sen’s is saying that choice (revealed
preference) should not contradict itself.
The proof that rationalizability implies Sen’s is as follows.

Claim 4 If C is rationalizable by a complete and transitive preference, then it

satisfies Sen’s .

Proof. We want to show, using the fact that C is rationalizable, that if B 0 ⇢ B

and a, b 2 C(B 0 ) and b 2 C(B) hold then a 2 C(B) must hold. So suppose that
C is rationalizable by a complete and transitive % and that B 0 ⇢ B. Then
a, b 2 C(B 0 ) and b 2 C(B)
=) a ⇠ b and b % c for all c 2 B (by pref maximization)
=) a % c for all c 2 B (by transitivity of %)
=) a 2 C(B) (because a 2 B and pref maximization).

1.1.4-A7. The proof is as follows..

Claim 5 If C satisfies WARP then it must satisfy Sen’s .

9
Proof. Suppose B 0 ⇢ B and a, b 2 C(B 0 ) and b 2 C(B). These imply in
particular that a, b 2 B \ B 0 , a 2 C(B 0 ) and b 2 C(B). But then WARP
implies that a 2 C(B) must hold. Hence proved.

1.1.4-A8. The revealed preference relation can be defined as follows: a %

b if and only if there is a menu B containing a, b and a 2 C(B). That is, a is
revealed preferred to b if it is chosen from a menu when b was available. Recalling
that C gives us the choice C(B) for every menu B, we know in particular that
C tells us the choice from all two-element menus {a, b}. If C is nonempty, then
for any a, b, we know that C({a, b}) is nonempty, and thus (using the definition
of the revealed preference relation we just gave) we can determine that either
a % b or b % a (or both).

1.2 Non-Standard Abstract Choice

1.2.1 Exercises
1.2.1-Q1. (a) Suppose a Satisficer exhibits C({a, b, c}) = {a, b}. What can
be inferred about the agent’s aspiration level and the utilities of the alternatives?
(b) Suppose that an agent who respects the Satisficing model is observed to
choose C({a, b}) = {a, b} and C({a, b, c}) = {a}. What can you deduce about
her preference rankings and aspiration levels in each menu?

1.2.1-Q2. Construct an example to demonstrate that the subjective menus

model does not necessarily satisfy WARP.

1.2.1-Q3. Suppose u over A = {a, b} is such that the utility of each

alternative is positive. Suppose ↵(i) = 1i . Show that if C(a, b) = {b} then
C(b, a) = {b}. That is, if b is chosen in a list then it will also be chosen if the
list is rearranged so that b is available earlier.

1.2.1-Q4. Consider the motivated reasoning model. Suppose a, b 2 B 0 ⇢

B and that C(B) = {a}.
(i) What can you deduce about J(B), J(B 0 ) and J({a, b})?
(ii) If C({a, b}) = {a}, can we deduce that a b?
(iii) If C({a, b}) = {a} and b 2 J(B 0 ), can we deduce that a b?
(iv) Suppose C({a, b}) = {a}. Argue that it must be that b 62 C(B 0 ), whether
or not b is justified in B0.

-Q5. Construct an example of Diminishing Attention model with negative

utilities that shows that the agent may choose a worse of two bad options.

1.2.2 Solutions
1.2.1-A1. (a) Either v ⇤ > u(a) = u(b) > u(c) or u(a), u(b) v ⇤ > u(c).
(b) Answer: We can deduce that (i) u(a) > u(b), u(c) (ie a has higher utility
than both b and c), (ii) u(a) > u(b) v ⇤ ({a, b}) (ie both a and b meet the

10
aspiration level in the menu {a, b}), and (iii) only a meets the aspiration level in
{a, b, c} or none do (ie either u(a) v ⇤ ({a, b, c}) > u(b), u(c) or v ⇤ ({a, b, c}) >
u(a), u(b), u(c).
Reason: Note that C({a, b, c}) = {a} implies either that a is the only alter-
native that meets the aspiration level or that no alternative meets the aspiration
level and a is in fact the unique preference maximizer. This determines (iii), but
note also that in each case we determine that u(a) > u(b), u(c), yielding (i).
Finally, since C({a, b}) = {a, b} and since we know u(a) > u(b), it follows that
both a and b must meet the aspiration level in the menu {a, b}, and so (ii) must
hold.

1.2.1-A3.
Proof. By the hypothesis,
C(a, b) = {b}
Proof. =) vL (b) > vL (a)
=) ↵(2) · u(b) > ↵(1) · u(a)
=) 12 u(b) > u(a)
1
=) u(b) > u(a) (since u(b) 2 u(b))
=) u(b) > 12 u(a) (since u(a) 1
2 u(a))
=) ↵(1) · u(b) > ↵(2) · u(a)
=) vL0 (b) > vL0 (a)
=) C(a, b) = {b}

1.2.1-A4. (i) By the hypothesis, we can deduce that a is justified in B. Since

there is some reason that makes it the best in B, it follows that the reason makes
it the best in a subset B 0 as well. By the same token, a is justified in {a, b}.
Therefore,
a 2 J(B), a 2 J(B 0 ) and a 2 J({a, b}).
(ii) No. It may be that b a but that b is not justified.
(iii) Yes. If b is justified in B 0 then it is also justified in the subset {a, b} (by
the same argument as in (i)). Consequently J({a, b}) = {a, b}, given that we
also know that a is justified in the menu. Thus by the hypothesis of the model,
C({a, b}) = {a} implies that a b.
(iv) If b is not justified in B0 then there is no way it can be chosen, and
thus b 62 C(B 0 ). If on the other hand if b is justified, we saw in part (iii) that it
cannot be a preference maximizer in J(B 0 ) (since a beats it). Therefore, again,
it cannot be chosen in that menu, b 62 C(B 0 ), as desired.

1.2.1-A5. Use the fact that the fraction of a negative number is a larger
number.

1.3 Choice under Objective Uncertainty

1.3.1 Exercises on Expected Utility Theory
1.3.1-Q1. Can an agent who respects Expected Value theory be risk averse?

11