Lec Slides

MSc Microeconomics (Econm1010)
In-Uck Park
University of Bristol
Autumn 2016
I Primary reading: Mas-Colell, Whinston and Green (1995)
Microeconomic Theory, Chapter 7-9, 12.C.
I Supplementary reading: Gibbons (1997) “An Introduction
to Applicable Game Theory,” Journal of Economic
Perspectives, 11, 127-149 (up to p137).
Gibbons (1992) A Primer in Game Theory.
I Office Hours: Thur 2:30-3:30pm (Office 3B5)
I Slide file and lecture notes are available as pdf in
Blackboard.
In a nutshell, a game is
In a nutshell, a game is (cont’d)
In a nutshell, a game is (cont’d)
Games of Complete Information
To describe a game, we need to specify:

(i) The players
(ii) The rules: Who moves when and how? What do they know
when they move?
(iii) The outcomes depending on how the game gets played
(iv) The payoffs depending on the outcome of the play.
Example 7B1: Matching Pennies
1\2 H T
H (−1, 1) (1, −1)
T (1, −1) (−1, 1)
Games of Complete Information: Normal-form
Example 7B1: Coordination game
T \S ESB CS
ESB (100, 100) (0, 0)
CS (0, 0) (100, 100)
Def 7D2: A normal form representation ΓN of a game specifies

the set of players, I ; a set Si of available strategies for each player
i ∈ I ; and a vNM payoff function ui : S → R for each i ∈ I where
S = S1 × · · · × SI . That is, ΓN = [I , {Si }, {ui }].
Def 7E1: Given a normal-form game ΓN = [I , {Si }, {ui }], a mixed
strategy of player i, σi : Si → [0, 1], is a probability distribution
over Si , i.e., σi ∈ ∆(Si ).
• By the property of vNM functions, each ui is extended to be

defined on Σ = ∆(S1 ) × · · · × ∆(SI ) in a natural manner:
X
ui (σ1 , · · · , σI ) = [σ1 (s1 )σ2 (s2 ) · · · σI (sI )]ui (s).
s∈S
• Example: In Matching Pennies game, if both players mix H and

T equally (i.e, σi (H) = σi (T ) = 1/2 for i = 1, 2), then 1’s payoff is
u1 (σ1 , σ2 ) = σ1 (H)σ2 (H)u1 (H, H) + σ1 (H)σ2 (T )u1 (H, T )

+ σ1 (T )σ2 (H)u1 (T , H) + σ1 (T )σ2 (T )u1 (T , T )
= 0.
HW1: calculate the expected utility of each player when both

choose H with probability 2/3 and L with probability 1/3.
Def 8B: A strategy si strictly dominates another strategy si0 6= si
for player i in ΓN = [I , {Si }, {ui }] if
ui (si , s−i ) > ui (si0 , s−i ) for all s−i ∈ S−i
where S−i = S1 × · · · × Si−1 × Si+1 · · · × SI and

(si , s−i ) = (s1 , · · · , sI ). In this case we say that the strategy si0 is
strictly dominated (by si ). A strategy si is a strictly dominant
strategy for player i if si strictly dominates all other strategies in
Si .
Iterated Deletion of Strictly Dominated Strategies: Explain
the process.
• The process produces a “strong” solution: If this process
produces a sharp prediction, there is little reason for the players to
behave differently. However, in many circumstances the predictions
are not narrow enough.
Example 2.1
1\2 L M R
1\2 L M R U (3, 0) (0, −5) (0, −4)
U (1, 0) (1, 2) (0, 1) C (1, −1) (3, 4) (−2, 3)
D (0, 3) (0, 1) (2, 0) D (2, 4) (4, 1) (−1, 8)
1\2 L M R
U (1, 0) (1, 2) (0, 1)
D (0, 3) (0, 1) (2, 0)
Example 2.1
1\2 L M R
1\2 L M R U (3, 0) (0, −5) (0, −4)
U (1, 0) (1, 2) (0, 1) C (1, −1) (3, 4) (−2, 3)
D (0, 3) (0, 1) (2, 0) D (2, 4) (4, 1) (−1, 8)
1\2 L M R
1\2 L M R U (3, 0) (0, −5) (0, −4)
U (1, 0) (1, 2) (0, 1) C (1, −1) (3, 4) (−2, 3)
D (0, 3) (0, 1) (2, 0) D (2, 4) (4, 1) (−1, 8)
Example 2.1
1\2 L M R
1\2 L M R U (3, 0) (0, −5) (0, −4)
U (1, 0) (1, 2) (0, 1) C (1, −1) (3, 4) (−2, 3)
D (0, 3) (0, 1) (2, 0) D (2, 4) (4, 1) (−1, 8)
1\2 L M R
1\2 L M R U (3, 0) (0, −4)
U (1, 0) (1, 2) (0, 1) C
D (0, 3) (0, 1) (2, 0) D (2, 4) (−1, 8)
• Role of common knowledge

Def 8D1: A strategy profile s = (s1 , · · · , sI ) ∈ S constitutes a
Nash Equilibrium (NE) of a game ΓN = [I , {Si }, {ui }] if for every
i = 1, · · · , I ,
ui (si , s−i ) ≥ ui (si0 , s−i ) for all si0 ∈ Si . (Ne)
Informally, a NE is a strategy profile with the mutual best-response

property, i.e., each player’s strategy is a best-response to those of
all other players.
Examples: Example 2.1 above.
Justifying the NE:
(i) NE as a consequence of rational inference.
(ii) NE as a necessary condition for a unique predicted outcome.
(iii) Focal points.
(iv) NE as a self-enforcing agreement (e.g., via communication).
(v) NE as a stable social convention.
Def 8D2: A mixed strategy profile σ = (σ1 , · · · , σI ) ∈ Σ is a NE if

for every i = 1, · · · , I , we have ui (σ) ≥ ui (si0 , σ−i ) for all si0 ∈ Si .
• Note the difference from Def 8D2 of M-W-G. The two are
equivalent by Prop 8D1.
HW2: Convince yourself that the two are equivalent.
Prop 8D2: (Existence) Every game ΓN = [I , {Si }, {ui }] in which
the sets I and Si ’s are finite has a mixed strategy NE.
Games of Complete Information: Extensive-form
Components of a game tree/extensive-form game:

I Nodes: initial node, decision nodes, terminal nodes
I Branches connecting nodes
I Players: Each player makes a move at each decision node of
his/hers.
I Payoffs for all players at every terminal node
I Information sets that describe all of the possible
distinguishable ‘circumstances’ in which a player might be
called upon to make a move.
N.B. In games of complete information, if players observe all

previous moves (i.e., games of perfect information), then all
information sets are singletons (i.e., consist of a single node).
Figures 7C1 & 7C3: Extensive forms for Matching Pennies
We assume the following properties on games:

(1) Perfect recall which basically means that players do not forget
what they once knew, including their own actions taken
previously.
(2) Common knowledge of the structure of the game and the
rationality of the players (i.e., they try to maximize their
payoffs).
(3) The payoffs are the von Neumann-Morgenstern (vNM) utility
levels. (Details are in Chapter 6).
Strategies: A strategy of a player is a complete contingent plan
that specifies what the player will do at each of her information
sets if she is called on to play there.
Examples 7C1 & 7C3: Strategies in Matching Pennies
7C3 : a (mixed) choice between H and T for each player i.
7C1 : as above for player 1; for player 2, a (mixed) choice at the
left information set (i.e., when player 1 chose H) and a
(mixed) choice at the right information set (i.e., when player 1
chose T).
Def 7E2: Given an extensive-form game, a behavioral strategy for
player i specifies, for every information set of hers, a mixed
strategy over actions available at that information set.
• By the Kuhn’s Theorem (1953), for games of perfect recall, we
lose no generality by considering only behavioral strategies (rather
than mixed strategies in the full sense).
Example 9B1: Predation game

E\ I F A
Out (0,2) (0,2)
In (-3,-1) (2,1)
• There are two NE.

− In & A
− Out & F
• “Backward induction” selects

the former as the plausible one.
Backward Induction: Think ahead to foresee what would happen

in possible future contingencies, and determine the optimal action
in the current stage of the game based on the anticipation of such
later behavior.
Prop 9B1: (Zermelo’s Theorem) Every finite extensive form game

of perfect information ΓE has a pure strategy NE that can be
derived through backward induction. Moreover, if no player has the
same payoffs at any two terminal nodes, then there is a unique NE
that can be derived in this manner.
Example 9B3: A modified version of predation game
Firm E Firm E

Out In Out In

for E: 0 Firm I 0 Firm E
for I: 2 F A 2 F A

-‐ 3 2 Firm I
-‐ 1 1 F A F A

Example 9B1 -‐ 3 1 -‐ 2 3
-‐ 1 -‐ 2 -‐ 1 1
Example 9B3
Firm E Firm E

Out In Out In

for I: 2 F A 2 F A

-‐ 3 2 Firm I
-‐ 1 1 F A F A

Example 9B1 -‐ 3 1 -‐ 2 3
-‐ 1 -‐ 2 -‐ 1 1
Example 9B3
Firm E Firm E

Out In Out In

for I: 2 F A 2 F A

-‐ 3 2 Firm I
-‐ 1 1 F A F A

Example 9B1 -‐ 3 1 -‐ 2 3
-‐ 1 -‐ 2 -‐ 1 1
Example 9B3
• There are multiple NE: In one firm E chooses “In”, followed by

(Acc, Acc). In others firm E chooses “Out”, with various
specifications of behavior in the off-equilibrium path (i.e, when
“In” is chosen contrary to what the equilibrium prescribes).
• The latter set of equilibria are not sensible because they are
sustained only when the players believe that they would not play a
NE in the remaining part of the game once “In” is chosen for some
unexpected reason.
Def 9B1: A subgame of an extensive form game ΓE is a subset of
the game that “constitute an extensive form game in its own.”
Def 9B2: A profile of strategies σ = (σ1 , · · · , σI ) in an I -player
extensive form game ΓE is a subgame perfect Nash equilibrium
(SPNE) if it induces a NE in every subgame of ΓE .
Games of Complete Information: Duopoly Models
• Market demand is x(p): continuous and strictly decreasing.

• Two firms have the same constant marginal cost c > 0 with
0 < x(c) < ∞.
The Bertrand Model (Price Competition)

– The firms choose prices simultaneously, so a normal-form game.
– However, available strategies are infinite.
– See digram in the next slide.
The Cournot Model (Quantity Competition)

– The firms choose output levels simultaneously, so a normal-form
game.
– Again, available strategies are infinite.
– Construct an analogous diagram where each firm chooses
qi ∈ (0, x −1 (0)), i = 1, 2.
The Stackelberg Model (Leader-Follower) – extensive form
Static Games of Incomplete Information
• Some players are uncertain about other player’s utility levels.
Dating game CI
P1 \ P2 red white
steak (2, 1) (0, 0)
chicken (0, 0) (1, 2)
• Two pure-strategy NE: (Steak, Red) and (Chicken, White).

• One mixed-strategy NE: P1 chooses S with 2/3 prob and C with
1/3 prob, and P2 chooses R with 1/3 and W with 2/3 prob (Verify
this).
Dating game II (a discrete version of p.139, Gibbons (1997), JEP):
P1(C ) \ P2(P) red white

steak (2 + tc , 1) (0, 0)
chicken (0, 0) (1, 2 + tp )
where tc and tp are independent random variables that assume a

positive value t > 0 (interpreted as feeling a big appetite) with
probability p, and 0 (small appetite) with probability 1 − p. Set
t = 2 and p = 3/4 and find an “equilibrium” in which different
types take different actions.
Dating game II (explained differently)
P1 \ P2 red white
( )
2 if type `
steak ( , 1) (0, 0)
4 if type h
( )
2 if type `
chicken (0, 0) (1, )
4 if type h
• P1 and P2 are `-type with probability 1/4 and h-type with

probability 3/4. Their types are independent and private (i.e.,
unknown to the other player).
• A player may use a different strategy depending on her type.
• Hence, a player’s strategy specifies one strategy for each type.
• A strategy profile is a Bayesian Nash equilibrium if each player’s
strategy contingent on each type is a best response to the other
player’s strategy.
Dating game II (cont’d)
P1 \ P2 red(R)1/4 white(W )3/4

( )
2 if type `
steak(S)3/4 ( , 1) (0, 0)
4 if type h
( )
2 if type `
chicken(C )1/4 (0, 0) (1, )
4 if type h
The following is a BNE: P1 chooses C if ` and S if h and P2

if ` and W if h as verified below. (choice prob in blue above)
chooses R (
` : gets 2/4 from S and 3/4 from C
I P1 if
h : gets 4/4 from S and 3/4 from C
(
` : gets 3/4 from R and 2/4 from W
I P2 if
h : gets 3/4 from R and 4/4 from W
A static Bayesian game is [I , {Si }, {ui }, {Ti }, {Fi (·)}] with the
following interpretation (Cf. Harsanyi, 1967, 1968):
1) Nature draws a type ti ∈ Ti for each player i ∈ I

independently according to the probability distribution Fi (·)
over player i’s set of possible types Ti . (Unless otherwise
stated, we assume Ti is finite and Fi (ti ) > 0 for all ti ∈ Ti .)
2) Nature reveals ti to player i but not to other players.
3) The players simultaneously choose an action ai from their
respective action sets Si .
4) Payoffs ui ((a1 , · · · , aI ); ti ) are received by each player.
Def 8E: A pure strategy in a Bayesian game is a decision rule si (·)

that assigns an action si (ti ) ∈ Si for each possible ti ∈ Ti . A (pure
strategy) Bayesian Nash equilibrium (BNE) for the Bayesian game
[I , {Si }, {ui }, {Ti }, {Fi (·)}] is a profile of strategies
(si (·), · · · , sI (·)) such that for every i and every ti ∈ Ti ,
E−i [ui (si (ti ), s−i (t−i ); ti )] ≥ E−i [ui (ai , s−i (t−i ); ti )] for all ai ∈ Si ,
where E−i [ui (ai , s−i (t−i ); ti )] is the expected utility level of player i
of ti -type when he chooses ai and all other players choose their
actions according to their decision rules sj (·), j 6= i.
Dynamic Games of Incomplete Information
• Beliefs are elevated to the level of importance of strategies in

the definition of equilibrium.
Figure 9 (p143, Gibbons): Two equilibria (L,L’) and (R,R’); Both
are SPE, but (R,R’) is not “sequentially rational.”
• Given an extensive-form game, a belief profile specifies for each

information set one probability distribution over the nodes
contained in that information set.
Definition: Given an extensive-form game ΓE (of complete or
incomplete information), a strategy profile σ and a belief profile µ
constitute a perfect Bayesian equilibrium (PBE) if σ is a NE of ΓE
and, in addition,
(SR) the players’ strategies are sequentially rational given their
beliefs, i.e., for every information set the action of the player
(who moves there) from that information set onwards is
optimal given the player’s belief at that point (and the other
players’ strategies from then on), and
(BR) whenever possible, beliefs are determined by Bayes’ rule from
σ.
Bayes’ rule: Let H be an information set that is reached with a

positive probability
P according0 to a strategy profile σ, i.e.,
Prob(H|σ) = x 0 ∈H Prob(x |σ) > 0. Then, Bayes’ rule specifies
the posterior belief at H as
Prob(x|σ)
Prob(x|H, σ) = P 0
for all x ∈ H.
x 0 ∈H Prob(x |σ)
Exercise 2.3: For the game in Figure 9 (p143, Gibbons) find the
posterior belief µ2 for player 2’s information set from a strategy
profile σ = (σ1 , σ2 ) where σ1 (L) = a > 0, σ1 (M) = b > 0 and
σ1 (R) = c > 0, and a + b + c = 1.
Exercise 2.4: Do Exercises 8.E.1, 9.B.14 and 9.C.2 in M-W-G.
(Treat weak PBE = PBE.)
Exercise 2.5: A simple card game
Player 1 starts with a hidden card (private type) which is either
high or low with equal probabilities. First, player 1 decides whether
to bet or not. If player 1 does not bet, the game ends with both
players’ payoffs being 0. If player 1 bets, then player 2 chooses
either to call or to fold, which ends the game with the following
payoffs: 2 and −2 for players 1 and 2, respectively, if player 2 calls
and player 1 is high; −2 and 2 for players 1 and 2, respectively, if
player 2 calls and player 1 is low; 1 and −1 for players 1 and 2,
respectively, if player 2 folds regardless of player 1’s type.
(a) Represent this game as a game tree.
(b) Show that it is not possible in equilibrium that player 1 bets for
sure if high and never bets if low.
(c) Show that it is not possible in equilibrium that player 1 of
either type bets for sure.
(d) Find an equilibrium and verify it fully.
Adverse Selection, Signalling and Screening
Example: Mergers and acquisitions:
I An M&A businessman, Andy, is in negotiation with a
potential seller, Beth, of her firm, F.
I If Andy buys a firm with a current value $x, he can reorganise
and improve its market value to $1.5x and sell at that price.
I The current value of firm F, x, is drawn from a uniform
distribution on [0,$1m] and is private information of the seller.
I Naturally, Beth would sell the firm only if she can get a price
that exceeds x.
I If Beth accepts a price $p, Andy will infer that
I x is a random draw from a uniform distribution on [0, p];
I his resale price will be uniformly distributed on [0, 1.5p];
I Hence, his expected net profit is 0.75p − p < 0.
I That is, there is no price at which Andy is willing to buy.
Consequently, transaction will not take place whatever x is.
Adverse Selection, Signalling and Screening
Market failure due to Adverse Selection:
I Note that if x was known to Andy as well, a firm of any value
would have been traded and improved upon.
I Therefore, the above result of “no transaction at all” is a
market failure.
I The source of such market failure is asymmetric information:
if the information was complete (i.e., the buyers also have
access to the information) the firm would have been traded
for mutual benefit.
I Two devices that are observed in the market to overcome this
problem (at least partially) are signalling and screening.
An Economist article “Secrets and agents” at
http://www.economist.com/news/economics-brief/21702428-
george-akerlofs-1970-paper-market-lemons-foundation-stone-
information
Signalling
Stark examples of signaling:

I Drinking from a jar one has prepared is a definitive signal that
it is not poisoned (because the cost of doing so is much larger
for the one who poisoned the jar relative to the one who did
not, so much so that the former cannot afford to do so).
I In the underworld (of gangsters) where the info on the
member’s identity is of utmost importance, obtaining
membership requires a signal too costly for others (e.g.,
undercover policemen) to afford, e.g., committing murder.
Signalling
A preliminary model without education:
(1) The nature chooses a worker’s type/productivity t: t = H = 2
with prob λ ∈ (0, 1/2) and t = L = 1 with prob 1 − λ. There
are fewer high productivity workers as λ < 1/2.
(2) Two firms, A and B, simultaneously make wage offers, wA
and wB , to the worker.
(3) The worker decides whether to accept wA and work for firm
A, to accept wB and work for B, or to reject both offers and
take an outside option.
(4) If the worker works for firm i = A, B, then the worker’s payoff
is wi , firm i’s payoff is t − wi where t is the worker’s type,
and the other firm’s payoff is 0. If the worker takes an outside
option, she receives her outside option value rt and both
firms’ payoffs are 0, where the worker’s outside option value
(also known as “reservation utility/payoff”) is rL = 0 for
L-type, and is rH > 1 + λ for H-type.
Signalling
We assume that rH < 2 so that it is socially beneficial for H-type
worker to participate in the labour market. In particular, we
assume that
1 + λ < rH < 1.5.
In this preliminary model:

I Neither firm may hire H-type workers, i.e., only L-type workers
may be hired in any equilibrium. This means that there is
market failure due to incomplete information.
I Specifically, the following is a PBE: both firms offer the same
wage equal to the productivity of L-type, i.e., wA = wB = 1;
and an L-type worker accepts firm i’s offer with probability
1/2 each for i = 1, 2, and an H-type worker takes the outside
option. The equilibrium payoff is 1 for an L-type worker, rH
for an H-type worker, and 0 for both firms.
Signalling - Spence
Education signalling - Model:
(1) The nature chooses a worker’s type/productivity t: t = H = 2
with prob λ ∈ (0, 1) and t = L = 1 with prob 1 − λ.
(2) Conditional on her type,
( the worker chooses an education level
c(e, H) = e/2 if H-type
e ≥ 0 which costs her
c(e, L) = e if L-type.
(3) Observing the chosen education level e, two firms, A and B,
simultaneously make wage offers, wA (e) and wB (e)
(4) The worker decides whether to take wA (e), wB (e), or an
outside option.
(5) If the worker works for firm i = A, B, then
I the worker’s payoff is u(wi (e), e|t) = wi (e) − c(e, t).
I firm i’s payoff is t − wi (e), and the other firm’s payoff is 0.
If the worker takes an outside option, her payoff is rt − c(e, t)
and both firms’ payoffs are 0.
I Note that the education does not enhance productivity–This
is for simplicity but it also makes the signaling function of
education more apparent.
Signalling - Spence
A strategy profile specifies

(i) an education level for each type of the worker, eH and eL ;
(ii) a wage schedule wi (e) for each firm i = A, B, that specifies a
wage offer depending on all possible education level e ≥ 0
that the worker might take;
(iii) for each possible pair of wage offers (conditional on the
education level taken), the worker’s choice among the two
firms’ offers and an outside option.
I For (iii), it is obvious that she would take the higher offer so
long as it is no worse than her outside option;
I in case the two offers are the same (and no worse than her
outside option), we postulate that she accepts either with
equal probabilities.
Signalling - Spence
Lemma 3.1: Consider the stage in which a certain education level

e is observed and the posterior belief at this information set
assigns a prob µ(e) to the worker being of a H-type. Given this
belief, the NE of the “continuation game” from this information
set onwards is as follows.
(a) If 1 + µ(e) < rH , both firms offer wA (e) = wB (e) = 1 and an
L-type worker accepts either with equal probabilities while an
H-type worker takes the outside option.
(b) If 1 + µ(e) ≥ rH , both firms offer wA (e) = wB (e) = 1 + µ(e)
and both types of worker accept either firm’s offer with equal
probabilities.
Signalling - Spence
Separating PBE: eH∗ 6= eL∗ .
I Since µ(eL∗ ) = 0 by Bayes’ rule, wA (eL∗ ) = wB (eL∗ ) = 1.
I However, since wi (e) ≥ 1 for all e ≥ 0 by Lemma 3.1, it is
optimal for L-type worker to choose eL∗ only if eL∗ = 0. Thus,
eL∗ = 0.
I Since µ(eH∗ ) = 1 by Bayes’ rule, wA (eH∗ ) = wB (eH∗ ) = 2 and
the worker accepts either with equal probability by Lemma 3.1
(b).
I For the two types’ e choices to be optimal, the following
Incentive
( Compatibility (IC) conditions are necessary:
wi (eH ) − eH∗ /2 ≥ wi (eL∗ ) − eL∗ /2 for H-type
∗
wi (eL∗ ) − eL∗ ≥ wi (eH∗ ) − eH∗ for L-type.

∗ ∗ ∗
Given eL = 0, wi (eL ) = 1 and wi (eH ) = 2, these IC conditions
determine 1 ≤ eH∗ ≤ 2.
Signalling - Spence
Separating PBE: eH∗ 6= eL∗ (cont’d)
I Also necessary are the following Participation Constraint (PC):
(
wi (eH∗ ) − eH∗ /2 ≥ rH (the outside option value) for H-type
wi (eL∗ ) − eL∗ ≥ 0 for L-type.
∗
These are satisfied if eH ≤ 4 − 2rH . Note that 4 − 2rH > 1 as
rH < 1.5.
I Summarizing, there is a continuum of separating equilibria
that vary in eH∗ : eL∗ = 0, 1 ≤ eH∗ ≤ 4 − 2rH ,
wA (eL∗ ) = wB (eL∗ ) = 1 and wA (eH∗ ) = wB (eH∗ ) = 2.
I To complete the description of the equilibrium, we need to
take care of “off-equilibrium” paths by specifying strategies
and a “legitimate” belief profile for contingencies that does
not arise according to the equilibrium strategies, in particular,
after e 6= eL∗ , eH∗ :
I wi (e) = 1 and µ(e) = 0 for 0 < e < eH∗ , and
I wi (e) = 2 and µ(e) = 1 for e ≥ eH∗ .
I Illustrate these with diagram and explain “single-crossing
property”.
Signalling - Spence
w uL = w − e = 1
H uH = w − e /2 = rH
2 €
€

rH
1
€ L
0 1 4 − 2rH e
Signalling - Spence
w uL = w − e = 1
H
2 €
wi (e) = 2 supported by µ (e) = 1.

rH € €
1
€ L wi (e) = 1 supported by µ (e) = 0.
€ €
0 e
*
H e
Signalling - Spence
Equilibrium refinement using the Intuitive Criterion
I Suppose a separating equilibrium with e ∗ > 1, and notice
H
i) that a L-type worker would never do better than following the
equilibrium by choosing some e ∈ (1, eH∗ ) instead, because no
firm would ever offer more than 2 in any circumstances,
ii) whereas a H-type worker could possibly do better than the
equilibrium by choosing e ∈ (1, eH∗ ) in case the subsequent
wage offer is (sufficiently close to) 2.
I Hence, it would be sensible for the firms to think that the
worker would be of H-type in case they see such e ∈ (1, eH∗ )
chosen unexpectedly, hence make a wage offer of 2.
I Foreseeing this, a H-type worker would indeed choose such e,
thereby upsetting the supposed equilibrium.
I Consequently, the separating PBE with eH∗ = 1 is the only one
that survives this refinement criterion, introduced by Cho and
Kreps (1987) and called the Intuitive Criterion.
Signalling - Spence
Pareto improvement by signaling? Comparing the separating

PBE’s with the equilibrium without signaling opportunity, we get
the following implications:
I Signalling alleviates market failure as H-type also participates
in the market.
I Signalling can bring about Pareto-improvement by benefiting
the worker (strictly for at least one type). For example, an
H-type worker’s utility in the separating PBE’s is
2 − eH∗ /2 ≥ rH because 1 < eH∗ ≤ 4 − 2rH , in particular, it is
1.5 > rH in the PBE that satisfies the Intuitive Criterion; and
an L-type worker has the same utility in the two outcomes.
I However, relative to the case of complete information, market
participation of H-type comes with a signalling cost which is
borne by an H-type worker.
Signalling - Spence
Second-Best Market Intervention: Pareto improvement is
possible if cross-subsidization is introduced by a central authority
when λ is large.
w

2

1
0 eH* = 1 e
Expected Utility Function and Risk Aversion
St. Petersburg Paradox

I You get $2n if H comes up for the first time in the n-th toss.
I Would you enter this lottery for a price of $10, $100, $1mil?
St. Petersburg Paradox

I You get $2n if H comes up for the first time in the n-th toss.
I Would you enter this lottery for a price of $10, $100, $1mil?
I The expected income from this lottery is
∞
X 1
2n × = 1 + 1 + 1 + 1 + · · · = ∞.
2n
n=1
I Yet, almost no one wants to play this lottery for a price≥$100.

I This is because people are concerned about the anticipated
utility level they will derive from spending the income, rather
than the amount of income itself.
I That is, they try to maximise the expected utility level, rather
than the expected income level.
Expected Utility Theory

I A (money) lottery is a pair of m = (m1 , · · · , mN ) ∈ RN and
p = (p1 , · · · , pN ) ∈ (0, 1)N such that p1 + p2 + · · · + pN = 1
where mn is a monetary amount and pn is the probability of
that amount.
I Expected Utility Theory: If a consumer’s preference over
possible lotteries satisfy certain standard conditions, his
preference is represented by a vNM expected (or Bernoulli)
utility function u : R → R in the sense that for any two
money lotteries (m, p) and (m0 , p0 ),
N N0
X X
0 0
(m, p) (m , p ) if and only if u(mn )pn ≥ u(mn0 )pn0 .
n=1 n=1
Risk attitude
I An agent who has a concave (and increasing) u is risk averse
in the sense that he prefers receiving the expected income of a
lottery with certainty to the lottery itself. Most individuals are
risk averse.
I An agent who has a convex (and increasing) u is risk loving in
the sense that he prefers a lottery itself to receiving the
expected income of the lottery, i.e., enjoys the risk.
I An agent who has a linear (and increasing) u is risk neutral in
the sense that he is indifferent between the two. Most firms
are assumed to be risk neutral.
Screening–Rothschild and Stiglitz (1976, QJE)
Insurance
I Customers: Risk-averse individuals with an initial wealth w
and an expected utility function u.
I He/she will have an accident (or become ill) with probability P
incurring a loss of d.
I So, without insurance, this agent is holding a “lottery” (an
uncertain income prospect) of w with probability 1 − P and
w − d with probability P.
I A customer’s indifference curves in the space of lotteries
(x, y ) can be drawn, where x and y denote the wealth of the
insured in case of accident and no accident, respectively.
I Insurance company: Risk-neutral. Offers a policy (p, b) where
p is the premium and b is the indemnity in case of accident.
I The lotteries that can be reached by purchasing actuarially
fair policies (that satisfy p − bP = 0) can be drawn in the
−P
same graph, as a straight line with slope 1−P .
Screening–Rothschild and Stiglitz
Model
I The pool of potential customers consists of two types denoted
by t = G or B.
I t-type has a probability Pt of becoming ill where PB > PG > 0.
I Each customer has a wealth w and the cost of recovery is d.
I The portion of G is λ.
I Each customer’s type is private information, i.e., known to the
individual but not to insurance firms.
I Multiple firms offer multiple insurance policies (p, b), called a
menu. The firms are competitive and will only reap the
normal expected profit of 0 (competitive insurance market).
Insurance under full information
Equilibrium: An equilibrium consists of two policies, one for each

type, such that
(i) each type chooses the policy designed for it over the other,
(ii) each insurance policy is actuarially fair (conditional on being
purchased only by the type it is designed for), and
(iii) there is no other policy that will make positive profit if offered
(i.e., entry is free but no profitable entry is possible)
Result (to be illustrated by diagrams):

I A pooling equilibrium (in which both types buy the same
policy) does not exist.
I There is a unique possibility of a separating equilibrium.
I There may not be an equilibrium at all if λ is large enough.
A pooling equilibrium does not exist.
A uniquely possible separating equilibrium is as below:
... because the following allows a profitable entry
Possibility of no equilibrium at all (for large λ)
Interpretation of the results:

(a) Under asymmetric information, there is no equilibrium without
screening (i.e., if a single policy is offered to everyone, rather
than a menu to choose from), resulting in market failure.
(b) Hence, screening tends to alleviate the market failure
stemming from adverse selection.
(c) Compared to full info situation, though, B-type has the same
policy while G -type has a worse policy not to attract its
inferior counterpart (leading to a welfare loss for G-type).
(d) Thus, efficiency is not fully restored because the “better”-type
bears a “separation cost” to distinguish from the inferior
counterpart.
Comparison to signalling:
I These findings generally hold for signalling as well, in
particular, (b) and (d).
I The main difference is that signalling is initiated by the
informed party whereas screening is initiated by the
un-informed party.
I Another difference is that free entry condition is not imposed
in Spence’s signalling model. The results will be closer to
each other if this condition is imposed in the signalling model
as well.

Lec Slides

Uploaded by

Copyright:

Available Formats

Lec Slides

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lec Slides

Uploaded by

Copyright:

Available Formats

MSc Microeconomics (Econm1010)

To describe a game, we need to specify:

Example 7B1: Matching Pennies

Example 7B1: Coordination game

Def 7D2: A normal form representation ΓN of a game specifies

• By the property of vNM functions, each ui is extended to be

• Example: In Matching Pennies game, if both players mix H and

u1 (σ1 , σ2 ) = σ1 (H)σ2 (H)u1 (H, H) + σ1 (H)σ2 (T )u1 (H, T )

HW1: calculate the expected utility of each player when both

ui (si , s−i ) > ui (si0 , s−i ) for all s−i ∈ S−i

where S−i = S1 × · · · × Si−1 × Si+1 · · · × SI and

• Role of common knowledge

ui (si , s−i ) ≥ ui (si0 , s−i ) for all si0 ∈ Si . (Ne)

Informally, a NE is a strategy profile with the mutual best-response

Def 8D2: A mixed strategy profile σ = (σ1 , · · · , σI ) ∈ Σ is a NE if

Components of a game tree/extensive-form game:

N.B. In games of complete information, if players observe all

We assume the following properties on games:

Example 9B1: Predation game

• There are two NE.

• “Backward induction” selects

Backward Induction: Think ahead to foresee what would happen

Prop 9B1: (Zermelo’s Theorem) Every finite extensive form game

• There are multiple NE: In one firm E chooses “In”, followed by

• Market demand is x(p): continuous and strictly decreasing.

The Bertrand Model (Price Competition)

The Cournot Model (Quantity Competition)

• Some players are uncertain about other player’s utility levels.

• Two pure-strategy NE: (Steak, Red) and (Chicken, White).

Dating game II (a discrete version of p.139, Gibbons (1997), JEP):

P1(C ) \ P2(P) red white

where tc and tp are independent random variables that assume a

• P1 and P2 are `-type with probability 1/4 and h-type with

P1 \ P2 red(R)1/4 white(W )3/4

The following is a BNE: P1 chooses C if ` and S if h and P2

1) Nature draws a type ti ∈ Ti for each player i ∈ I

Def 8E: A pure strategy in a Bayesian game is a decision rule si (·)

• Beliefs are elevated to the level of importance of strategies in

• Given an extensive-form game, a belief profile specifies for each

Bayes’ rule: Let H be an information set that is reached with a

Stark examples of signaling:

In this preliminary model:

A strategy profile specifies

Lemma 3.1: Consider the stage in which a certain education level

wi (eL∗ ) − eL∗ ≥ wi (eH∗ ) − eH∗ for L-type.

wi (e) = 2 supported by µ (e) = 1.

€ L wi (e) = 1 supported by µ (e) = 0.

Pareto improvement by signaling? Comparing the separating

St. Petersburg Paradox

St. Petersburg Paradox

I Yet, almost no one wants to play this lottery for a price≥$100.

Expected Utility Theory

Equilibrium: An equilibrium consists of two policies, one for each

Result (to be illustrated by diagrams):

Interpretation of the results:

You might also like