Motivations

Models in the previous lectures

Political Agency Model

Conflicts of interest

among citizens

Conflicts of interest

between citizens & politicians

e.g.

Income redistribution

Division of a pie

e.g.

Corruption

Disaster relief

Theoretically, an extension of the principal-agent model

The principal (citizens) can only replace the agent (politicians)

as an incentive mechanism

Baseline Model

We follow Section 3.3 of Besley (2006) with a slight modification

Originally(?) proposed by Banks and Sundaram (1993)

Players

Citizens as a single player

Incumbent politician & Opposition candidate

Two preference types ( $i \in \{c,d\}$ ):

Denote $\pi\equiv \Pr(i=c)$

(can be very close to zero)

We abstract from conflicts among citizens

Congruent (same preference as citizens)

Dissonant (have a personal agenda)

Time, Policy, Election

Two periods, $t \in \{1,2\}$

Politician in office in period $t$ implements a policy denoted by $e_t \in \{0,1\}$

Can be infinite horizon (see section 3.4.7 of Besley (2006))

$e_t = 1$ is a good policy (benefiting citizens)

There is an election between the two periods

Citizens' preference

\Delta

\Delta

if $e_t = 1$

0

0

otherwise

Discount the next period payoff by $\beta$

\sum_{t=1}^2 \beta^{t-1} 1(e_t=1)*\Delta

\sum_{t=1}^2 \beta^{t-1} 1(e_t=1)*\Delta

Citizens' present-value payoff

Politicians' preference (1 of 3)

Both types of politicians yield a payoff of

E

E

if being in office

0

0

otherwise

$E$ includes ego-rent and the wage as a politician

And discount the next period payoff by $\beta$

Politicians' preference (2 of 3)

The congruent type of politicians yield an additional payoff of

\Delta

\Delta

if $e_t = 1$

0

0

otherwise

They share the preference with voters

i.e.

Politicians' preference (3 of 3)

The dissonant type of politicians yield an additional payoff of

0

0

if $e_t = 1$

r_t

r_t

otherwise

They have a conflict of interest with voters

where $r_t \sim G(\cdot)$ with the mean $\mu$ and the support $[0, R]$

Examples of $r_t$

Private agenda

Capture by special interest or constituency

Disutility from implementing good policy (or incompetence)

i.e.

Information

Voters do not observe

the type of politicians (both incumbent and opposition)

the rent for dissonant politician $r_t$

This is a world where politicians know better

Timing of Events (1 of 3)

Period 1

1

Nature picks:

The incumbent politician observes these while voters do not

the type of the incumbent politician $i \in \{c,d\}$

the rent for dissonant politician $r_1$

2

The incumbent politician chooses policy $e_1 \in \{0,1\}$

Period 1 payoffs realize for the incumbent and voters

Timing of Events (2 of 3)

Election

3

Voters decide whether to re-elect the incumbent

But voters do not observe it

4

Nature picks the type of opposition candidate

Timing of Events (3 of 3)

5

The winning politician chooses policy $e_2 \in \{0,1\}$

Period 2

Period 2 payoffs realize and the game ends.

6

Nature picks:

the rent for dissonant politician $r_2$

Analysis

Guess and verify

Now we assume it's optimal for voters to

Re-elect the incumbent if and only if $e_1=1$

Then derive the optimal policy choice for the incumbent

Finally, given this optimal incumbent's behavior, prove that

Re-elect the incumbent if and only if $e_1=1$

is indeed optimal for voters

Period 1 congruent incumbent's optimization

When voters re-elect the incumbent if and only if $e_1=1$

So it's optimal for the congruent incumbent to pick $e_1=1$

E + \Delta + \beta(E + \Delta) > E

E + \Delta + \beta(E + \Delta) > E

Payoff from $e_1=1$

Payoff from $e_1=0$

\Pr(e_1=1|i=c)=1

\Pr(e_1=1|i=c)=1

Voters' optimization when observing $e_1=1$

\Pr(e_1=1|i=d)=G(\beta(E+\mu))

\Pr(e_1=1|i=d)=G(\beta(E+\mu))

Now we know

\Pr(e_1=1|i=c)=1

\Pr(e_1=1|i=c)=1

Plug these into

\Pr(i=c | e_1=1) =\frac{\Pr(e_1=1|i=c)\pi}{\pi\Pr(e_1=1|i=c)+(1-\pi)\Pr(e_1=1|i=d)}

\Pr(i=c | e_1=1) =\frac{\Pr(e_1=1|i=c)\pi}{\pi\Pr(e_1=1|i=c)+(1-\pi)\Pr(e_1=1|i=d)}

=\frac{\pi}{\pi+(1-\pi)G(\beta(E+\mu))}

=\frac{\pi}{\pi+(1-\pi)G(\beta(E+\mu))}

>\pi

>\pi

as long as $G(\beta(E+\mu))<1$

So it's optimal to re-elect the incumbent if $e_1=1$

\Pr(e_1=0|i=d)=1-G(\beta(E+\mu))

\Pr(e_1=0|i=d)=1-G(\beta(E+\mu))

We now know

\Pr(e_1=0|i=c)=0

\Pr(e_1=0|i=c)=0

Plug these into

\Pr(i=c | e_1=0) =\frac{\Pr(e_1=0|i=c)\pi}{\pi\Pr(e_1=0|i=c)+(1-\pi)\Pr(e_1=0|i=d)}

\Pr(i=c | e_1=0) =\frac{\Pr(e_1=0|i=c)\pi}{\pi\Pr(e_1=0|i=c)+(1-\pi)\Pr(e_1=0|i=d)}

=0

=0

<\pi

<\pi

So it's optimal to kick out incumbent if $e_1=0$

Voters' optimization when observing $e_1=0$

Guess and verify

Now we assume it's optimal for voters to

Re-elect the incumbent if and only if $e_1=1$

Then derive the optimal policy choice for the incumbent

Finally, given this optimal incumbent's behavior, prove that

Re-elect the incumbent if and only if $e_1=1$

is indeed optimal for voters

Are there other equilibria? (1 of 2)

What if voters always re-elect or kick out the incumbent?

$e_1=1, e_2=1$ if congruent

Then the incumbent simply picks the policy they like in both periods

$e_1=0, e_2=0$ if dissonant

But then by observing $e_1$ voters know the type of the incumbent

Re-electing the dissonant incumbent is not optimal

(0 < \pi\Delta)

(0 < \pi\Delta)

Kicking out the congruent incumbent is not optimal

(\pi\Delta < \Delta)

(\pi\Delta < \Delta)

This is not an equilibrium

Are there other equilibria? (2 of 2)

Do voters re-elect the incumbent if and only if $e_1=0$ ?

If $E > (1-\beta) \Delta$ ,

congruent incumbent prefers being re-elected than choosing $e_1=1$

In this case, both types then choose $e_1=0$

This is a Perfect Bayesian Equilibrium

as long as voters' off-the-equilibrium belief is that

the incumbent is dissonant if $e_1=1$

This equilibrium is ruled out by the Intuitive Criterion of Cho and Kreps (1987)

(cf. Section 16.3 of Tadelis (2013))

Proposition

(1) Period 1 policy is given as follows:

otherwise

e_1=1

e_1=1

if the incumbent is congruent

or if the incumbent is dissonant and $r_1 < \beta(E+\mu)$

e_1=0

e_1=0

(2) Voters re-elect the incumbent as long as

e_1=1

e_1=1

Implications

Alternative modelling approaches

Pure moral hazard (no hidden type)

Pioneered by Barro (1973) and Ferejohn (1986)

On the equilibrium path, citizens are indifferent

between the incumbent and the opposition candidate

This equilibrium is not robust

to allowing the heterogeneity among politicians (Fearon 1999)

Career Concern Model

Originally proposed by Holstrom (1999) for employees

Applied to policy-makers by Dewatripont et al. (1999)

and Alesina and Tabellini (2007, 2008)

The model assumes that policy-makers do not know their own type

Fine if the type is competence

Implausible if the type concerns preference

Evidence

Citizens punish the incumbent

if his/her performance is worse than expectation

Given this behavior of citizens

Incumbent restrains from shirking, being corrupt, etc.

Two predictions from the political agency model

1

2

Ferraz and Finan (2008, 2012) test these predictions

in the context of Brazilian mayors' corruption

Over-invoicing

Corruption by Brazilian mayors

1

2

Pay five times more than cost to a road-building firm

and get kickback

e.g.

Diversion of public funds

e.g.

Pocket fiscal transfer from the Ministry of Health

by forging fake receipt for medical equipment purchase

3

Illegal procurement

e.g.

Only one firm satisfies the qualification

to bid for the construction of a stadium

pp. 710-711 of Ferraz and Finan (2008)

Citizens punish the incumbent

if his/her performance is worse than expectation

Given this behavior of citizens

Incumbent restrains from shirking, being corrupt, etc.

Two predictions from the political agency model

1

2

Tested by Ferraz and Finan (2008)

October 2004

Municipal elections

376 municipalities

audited

300 municipalities

audited

Voters observe

mayor's corruption

before election

Voters don't observe

mayor's corruption

before election

Similar due to random assignment

Difference in

electoral outcomes

Impact of observing

mayor's corruption

=

Research design for 1st prediction

Data

Corruption: # of violations recorded in the audit report

Election outcomes & candidate characteristics

Source: Tribunal Superior Eleitoral

Municipality demographic characteristics

Source: Population census in 2000

Municipality institutional characteristics (e.g. # of radio stations)

Source: Municipality survey in 1999

Available also for those audited after the 2004 elections

Results

Source: Table 3 Column 4 of Ferraz and Finan (2008)

Pre-election audit	0.084 [0.104]
Pre-election audit x 0 violation	0.010 [0.156]
Pre-election audit x 2 violations	-0.253* [0.148]
Pre-election audit x 3 violations	-0.321* [0.192]
Pre-election audit x 4+ violations	-0.159 [0.168]
# of observations	373

* Significant at 10%

Citizens punish the incumbent

if his/her performance is worse than expectation

Given this behavior of citizens

Incumbent restrains from shirking, being corrupt, etc.

Two predictions from the political agency model

1

2

Tested by Ferraz and Finan (2011)

Source: Figure 2 of Ferraz and Finan (2011)

Finding

Term-limited

Application

Impact of Term Limits

What's the benefit of term limits?

The baseline political agency model suggests

only a negative impact of term limits on citizens' welfare

Is there any situation where term limits enhance welfare?

Smart and Sturm (2013) extend the political agency model

to tackle this issue

Here we follow Section 3.4.3 of Besley (2006)

Extending the baseline model

In the baseline model, voters know $e_t=1$ is a good policy

Now assume that voters are unsure about which is a good policy

To formally model this uncertainty, introduce a state of the world:

s_t\in\{0,1\}

s_t\in\{0,1\}

And voters' payoff is given by

\Delta

\Delta

0

0

if $e_t = s_t$

if $e_t \neq s_t$

But voters do not know $s_t$

Voters prior belief on $s_t$

\Pr(s_t=0)=1/2

\Pr(s_t=0)=1/2

At the election, voters observe $e_1$

But citizens' payoff from $e_1$ doesn't realize yet

When choosing policy, the incumbent observes $s_t$

Extending the baseline model (cont.)

The congruent politician's preference: the same as voters

The dissonant politician's preference:

\Delta

\Delta

0

0

if $e_t = s_t$

if $e_t \neq s_t$

r_t

r_t

0

0

if $e_t = 1$

if $e_t = 0$

e.g. $e_t=1$ pleases an interest group, but could also benefit the whole population

Extending the baseline model (cont.)

Timing of Events

Period 1

1

Nature picks:

The incumbent politician observes these while voters do not

the type of the incumbent politician $i \in \{c,d\}$

the state of the world $s_1\in \{0,1\}$

the rent for dissonant politician $r_1 \sim G(\cdot)$

2

The incumbent politician chooses policy $e_1 \in \{0,1\}$

Period 1 payoffs realize for incumbent, but not for voters

Timing of Events (cont.)

Election

3

Voters decide whether to re-elect the incumbent

But voters do not observe it

4

Nature picks the type of opposition candidate

Timing of Events (cont.)

5

The winning politician chooses policy $e_2 \in \{0,1\}$

Period 2

All payoffs realize and the game ends.

6

Nature picks:

the state of the world $s_1\in \{0,1\}$

the rent for dissonant politician $r_1$

Analysis

Among possibly many equilibria, focus on the one in which:

Congruent incumbent picks $e_1 = 0$ regardless of $s_1$

Dissonant incumbent picks $e_1 = 0$ if $r_1 < \beta(E+\mu)$

Voters re-elect the incumbent if and only if $e_1=0$

That is, when the re-election incentive backfires

We now derive the condition under which this is an equilibrium

picks $e_1 = 1$ otherwise

Voters optimization

They prefer electing a congruent politician for period 2

Probability of being congruent:

\frac{\pi}{\pi+(1-\pi)G(\beta(E+\mu))}>\pi

\frac{\pi}{\pi+(1-\pi)G(\beta(E+\mu))}>\pi

Opposition

Incumbent delivering $e_1=0$

It's optimal to re-elect the incumbent if $e_1=0$

\Delta > \frac{1}{2}\Delta

\Delta > \frac{1}{2}\Delta

if dissonant

if congruent

Dissonant incumbent's optimization

For the same reason as in the baseline model, it's optimal to pick

$e_1 = 0$ if $r_1 < \beta(E+\mu)$

$e_1 = 1$ otherwise

Congruent incumbent's optimization

Given that he'll be re-elected by choosing $e_1=0$ ,

it's clearly optimal to choose $e_1=0$ if $s_1=0$

If $s_1=1$ , his payoff is

if picking $e_1=0$ (and re-elected)

\beta(E+\Delta)

\beta(E+\Delta)

\Delta

\Delta

if picking $e_1=1$ (and kicked-out)

So it's optimal to choose $e_1=0$ even if $s_1=1$ when

E>\frac{(1-\beta)\Delta}{\beta}

E>\frac{(1-\beta)\Delta}{\beta}

An equilibrium

Congruent incumbent picks $e_1 = 0$ regardless of $s_1$

Dissonant incumbent picks $e_1 = 0$ if $r_1 < \beta(E+\mu)$

Voters re-elect the incumbent if and only if $e_1=0$

The following strategy profile:

is a Perfect Bayesian Equilibrium under the condition

$e_1 = 1$ otherwise

E>\frac{(1-\beta)\Delta}{\beta}

E>\frac{(1-\beta)\Delta}{\beta}

If the incumbent is term-limited in period 1

Congruent incumbent picks $e_1 = s_1$

Dissonant incumbent picks $e_1 = 1$

Probability of congruent politician in office: $\pi$

In both periods:

Voters' welfare w/ and w/o term limit

\pi \Delta + (1-\pi) \frac{1}{2}\Delta

\pi \Delta + (1-\pi) \frac{1}{2}\Delta

if term-limited

if not term-limited

Period 1 (Incentive effect)

\frac{1}{2}\Delta

\frac{1}{2}\Delta

Term limit increases welfare by

\frac{\pi}{2}\Delta

\frac{\pi}{2}\Delta

Voters' welfare w/ and w/o term limit

\pi \Delta

\pi \Delta

if term-limited

if not term-limited

Period 2 (Selection effect)

Term limit decreases welfare by

(1-\pi)(1-\lambda)\frac{\pi}{2}\Delta

(1-\pi)(1-\lambda)\frac{\pi}{2}\Delta

+ (1-\pi) \lambda \frac{1}{2} \Delta

+ (1-\pi) \lambda \frac{1}{2} \Delta

\lambda \equiv G(\beta(E+\mu))

\lambda \equiv G(\beta(E+\mu))

+ (1-\pi) (1-\lambda) \pi\Delta

+ (1-\pi) (1-\lambda) \pi\Delta

+ (1-\pi) (1-\lambda) (1-\pi)\frac{1}{2} \Delta

+ (1-\pi) (1-\lambda) (1-\pi)\frac{1}{2} \Delta

\pi \Delta + (1-\pi) \frac{1}{2}\Delta

\pi \Delta + (1-\pi) \frac{1}{2}\Delta

where

Voters' welfare w/ and w/o term limit

decreases welfare by

\beta(1-\pi)(1-\lambda)\frac{\pi}{2}\Delta

\beta(1-\pi)(1-\lambda)\frac{\pi}{2}\Delta

Term limit

increases welfare by

\frac{\pi}{2}\Delta

\frac{\pi}{2}\Delta

Net effect

[1-\beta(1-\pi)(1-\lambda)]\frac{\pi}{2}\Delta > 0

[1-\beta(1-\pi)(1-\lambda)]\frac{\pi}{2}\Delta > 0

Implications

Term limit can increase welfare when

Being in office is too attractive for congruent politicians

i.e. the equilibrium condition:

Citizens do not observe the policy outcome before elections

expect a particular policy to be better ( $e_1=0$ )

Maskin and Tirole (2004),

using a similar version of the political agency model,

discuss when policy-making should be delegated to "judges"

E>\frac{(1-\beta)\Delta}{\beta}

E>\frac{(1-\beta)\Delta}{\beta}

(Related) Evidence

L ist and Sturm (2006 QJE)

US state governors are more likely to implement

environment-friendly policies in two cases:

when term-limited in brown states

when not term-limited in green states

Other

Applications

Political business cycles

Sibert and Rogoff (1988 Restud) and Rogoff (1990 AER)

use a version of political agency model (cf. Section 3.4.4 of Besley 2006)

Policy decisions twice within Period 1

But the second policy's outcome is unobserved at the election

to derive the electoral cycle in public investment

Evidence

Shi and Svensson (2006) for cross-country evidence

Akhmedov and Zhuravskaya (2004 QJE) for Russian regions

Cole (2009) for Indian states

The role of media

Besley and Burgess (2002 QJE)

extend the political agency model in which

They show that disaster relief by Indian states is larger

for states with many people subscribing to newspapers

the probability of learning the policy outcome

increases with the presence of mass media

Optimal tax policy if politicians are corruptible

Yared (2010 Restud)

Allow government to issue debt in the presence of aggregate shocks

Acemoglu, Golosov, Tsyvinski (2008 Econometrica)

In a neoclassical growth model with non-stationary voting strategy and without any restriction on tax instruments

Theoretical applications

Incorporate with citizen-candidate model

Van Weelden (2013 Restud)

Autocracy

Padro-i-Miquel (2007 Restud)

Besley and Kudamatsu (2008)

Extend the model by replacing citizens with the ruling elite

If replacing the leader also causes the ruling elite to lose power

they cannot discipline the leader

Political agency model can incorporate

non-electoral removals of national leaders

Polarization & the disciplining effect of election

If some groups of voters always vote for the opposition

it's less rewarding for the dissonant incumbent to restrain

Disenfranchising such opposition supporters

may therefore improve the quality of government

See Besley and Kudamatsu (2008) for formal argument

e.g. Pak Chunghee in South Korea in 1963

e.g. ethnic allegience

For evidence from India, see Banerjee and Pande (2009)

Causes of democratization

Fearon (2011 QJE)

Bidner and Francois (2013 QJE)

Why politicians have an incentive to hold elections regularly

When citizens are willing to replace the incumbent with less experienced opposition candidate

Summary

Citizens punish the incumbent

if his/her performance is worse than expectation

Given this behavior of citizens

Incumbent restrains from shirking, being corrupt, etc.

Two implications from political agency model

1

2

Both are empirically supported

in the context of Brazilian mayors' corruption

The model can be extended to analyze various policy issues where citizens and politicians are in conflict

Announcement

on

Political Economics II

By 8 December (the first lecture)

Buy a copy at COOP

Read Chapters 1-5

Take note of your questions

to ask during the first lecture

Scheduling

Term Paper

Workshop

December 1: Last Lecture

December 15: Submission deadline (9 am)

December 8: Political Economics II Lecture 1 on empirical methods

December 11 (Mon) or 12 (Tue): Workshop?

18 (Mon)

December 6: My office hour (16:20-17:50)

December 13: My office hour (16:20-17:50)

Sample	2nd-term and 1st-term later re-elected
Mayor in first-term	-0.040*** [0.013]
# of observations	313

Sample	2nd-term and 1st-term later re-elected	2nd-term and 1st-term served as mayor before
Mayor in first-term	-0.040*** [0.013]	-0.038*** [0.014]
# of observations	313	287

Political Agency Model

Lecture 4, Political Economics I

OSIPP, Osaka University