Political Economy - Lecture 2: Political agency and ...marcsangnier.free.fr/.../politicaleconomy_m2_2015-2016_lec2.pdf · PoliticalEconomy-Lecture2: Politicalagencyandelectoralcontrol

Political EconomyLecture 2: Political agency and electoral control

Marc [email protected]

2015-2016, Fall semesterAix Marseille School of Economics

Political Economy - Lecture 2: Political agency and electoral control

1 Introduction

2 Canonical model of political agency

3 Empirical evidence

4 Extended model accounting for clientelism and targeted policies

5 Extended model accounting for voters’ information

6 Conclusion

2 / 45

Political Economy - Lecture 2: Political agency and electoral controlIntroduction

1 IntroductionA principal-agent frameworkAsymmetric information

3 / 45


A principal-agent framework

A principal-agent framework

• We analyzed how voting might help to aggregate heterogeneouspreferences.

• In democracies, policies are implemented by elected politicians.• Once elected, politicians have an informational advantage overcitizens.

• Models of “political agency” were politicians are agents andvoters are principals.

4 / 45


Asymmetric information

Asymmetric information

• Issues due to asymmetric information:• Politicians’ characteristics, i.e. adverse selection: As quality

is unobserved, how to ensure that the “best” politicians areselected?

• Politicians’ behavior, i.e. moral hazard: As effort is unobserved,how to ensure that elected politicians implement policies thatare consistent with voters’ preferences?

• Elections are a way for voters to punish a politician who wouldhave not implemented satisfying policies.

5 / 45

Political Economy - Lecture 2: Political agency and electoral controlCanonical model of political agency

2 Canonical model of political agencySetup of the modelOptimal decisionsImplications for the quality of government

6 / 45


Setup of the model

Setup of the model

• A very simple agency framework that includes adverse selec-tion and moral hazard, and allows to investigate how electionmitigate (or not) these issues.

7 / 45


Setup of the model

• Two time periods: t ∈ {1, 2}.• In each period, two possible states of the world may occur:

st = {0, 1}. Each state occurs with probability 12 .

• The state of the world is not observed by voters, it is onlyobserved by the incumbent politician.

• In each period, the elected politician takes a single decisionet ∈ {0, 1}.

• The politician’s action is not directly observed by voters.• Voters’ payoff UV

t is ∆ if et = st and 0 if et 6= st .• Both voters and politicians discount the future at rate β < 1.

8 / 45


Setup of the model

• Politicians get a payoff E > 0 from being in power (ego rent orwage).

• Politicians can be congruent or dissonant: i ∈ {C ,D}. Theprobability that a randomly picked politician is congruent is π.

• Congruent politicians share voters’ preferences, i.e.:

UCt =

{E + ∆, if et = st ,E + 0, if et 6= st .

• Dissonant politicians get (private) benefit from taking decisionsthat are not favorable to citizens, i.e.:

UDt =

{E + 0, if et = st ,E + rt , if et 6= st ,

where rt ∈ [0,R], the dissonance rent, is drawn at each periodfrom a cumulative distribution function G(r) with mean µ.

• Politicians’ type is unobserved by voters.9 / 45


Setup of the model

Timing

1 Nature determines the state of the world s1 and the type of theincumbent politician.

2 If the incumbent politician is dissonant, r1 is drawn from G(r).3 The incumbent politician chooses e1.4 Voters observe their payoff UV

1 .5 Voters decide whether or not to reelect the incumbent against

a randomly chosen challenger.6 Nature determines the state of the world s2.7 If the incumbent politician is dissonant, r2 is drawn from G(r).8 The incumbent politician chooses e2.9 Voters observe their payoff UV

2 .

10 / 45


Optimal decisions

Equilibrium

• Let us denote the incumbent’ decision as:

et (s, i) , with s ∈ {0, 1} and i ∈ {C ,D} .

• Optimal decisions in period 2:

e2(s,C) = s2,e2(s,D) = 1− s2.

• Optimal decisions in period 1:• A congruent incumbent always chooses e1(s,C) = s1.• Optimal decision by dissonant politicians depends on the elec-

toral best-response by voters between the two periods.

11 / 45


Optimal decisions

• Electoral best-response depends on how voters update their be-liefs about politicians’ type.

• Voters only observe ∆ or 0. What do they learn from thisobservation about politicians’ type?

• Assume voters observe ∆ at the end of period 1:

P(i = C |UV = ∆) = P(i = C)× P(∆|i = C)P(∆) = π × 1

π + (1− π)λ,

where λ is the probability that a dissonant incumbent takes acongruent decision in period 1.

• Since:∀λ : P(i = C |UV = ∆) > π,

voters will always reelect an incumbent that delivers ∆.

12 / 45


Optimal decisions

• Assume voters observe 0 at the end of period 1:

P(i = C |UV = 0) = P(i = C)× P(0|i = C)P(0) ,

i.e.:

P(i = C |UV = 0) = π × 0π × 0 + (1− π)(1− λ) ,

where 1−λ is the probability that a dissonant incumbent takesa dissonant decision in period 1.

• Since:∀λ : P(i = C |UV = 0) = 0 < π,

voters will never reelect an incumbent that delivers 0.

13 / 45


Optimal decisions

• A dissonant incumbent will behave congruently if short-termbenefits from dissonance are lower than long-term benefits ofcongruence:

E + β (E + E(r2)) > E + r1 + β × 0,

⇔ r1 < β(µ+ E ).

• Thus:λ = G (β(µ+ E )) .

• Bad politicians behave well in period 1 if dissonance is suffi-ciently low.

14 / 45


Implications for the quality of government


• Ex ante voters’ welfare is:

W = V1 + βV2,

where:V1 = [ π︸︷︷︸

P(i=C)

+ (1− π)︸︷︷︸P(i=D)

λ︸︷︷︸P(∆|i=D)

]∆,

and,V2 = [ π︸︷︷︸

P(i=C)

+ (1− π)︸︷︷︸P(i=D)

(1− λ)︸︷︷︸P(0|i=D)

π︸︷︷︸P(i=C)

]∆

15 / 45



• W is increasing in π (intuitive).• W is increasing in λ.• But V1 is increasing in λ, while V2 is decreasing in λ:

• Interpreting λ as politicians’ discipline, more discipline improvespoliticians’ performance in period 1 but makes harder to detectdissonant incumbents.

16 / 45



Model’s testable predictions

• Term limit effects:1 Politicians behave differently when they can or cannot run for

reelection.2 If reelection is possible, rent extraction will be higher after the

reelection than before (no discipline in period 2).

17 / 45

Political Economy - Lecture 2: Political agency and electoral controlEmpirical evidence

3 Empirical evidenceTerm limit effect and politicians’ decisionsTerm limit effect and corruption

18 / 45


Term limit effect and politicians’ decisions


Timothy Besley & Anne Case, 1995. “Does Electoral Accountability Affect EconomicPolicy Choices? Evidence from Gubernatorial Term Limits,” The Quarterly Journal ofEconomics, vol. 110(3), pages 769-798, August.

• US states’ governors can hold office at most twice: Is there adifference between governors’ first and second term? I.e. whenthey can or cannot run for reelection?

• Estimation strategy:

Policy outcomes,t = β0 + β1Second terms,t + δt + δs + . . .

• Policy outcomes: taxes and state government spending.

19 / 45



Impact of term limit on policy outcomes (t-statistics in parentheses).Source: Besley and Case (1995)

20 / 45


Term limit effect and corruption


Claudio Ferraz & Frederico Finan, 2011. “Electoral Accountability and Corruption:Evidence from the Audits of Local Governments,” American Economic Review,American Economic Association, vol. 101(4), pages 1274-1311, June.

• Are politicians less corrupt if they can run for reelection?• Municipalities in Brazil:

• Brazilian mayors can be reelected at most once.• Since 2003, the central government randomly select 60 munici-

palities per month to be audited. Reports are publicly availableand provide an objective measure of corruption.


Corruptioni = β0 + β1First termi + . . .

21 / 45



Differences in corruption by first- and second-term mayors.Source: Ferraz and Finan (2011)

22 / 45



• How to ensure that being in the first or the second term israndom?

• Alternative estimation strategy:• Compare new mayors who defeated the incumbent by a small

margin to reelected incumbents who won the election by a smallmargin.

23 / 45



Random assignment of first- or second-term.Source: Ferraz and Finan (2011)

24 / 45

Political Economy - Lecture 2: Political agency and electoral controlExtended model accounting for clientelism and targeted policies

4 Extended model accounting for clientelism and targeted policiesSetup of the modelOptimal decisionsImplications for the quality of government

25 / 45


Setup of the model

Setup of the model

• What if politicians can set targeted policies such that rents areonly extracted from some groups of citizens?

26 / 45


Setup of the model

• Same assumptions as before (see slides 8–10)• But:

• 3 groups of voters (g = 1, 2, 3) of equal size 1/3.• The incumbent can take different decisions eg

t ∈ {0, 1} regard-ing the three groups:

Ugt =

{0, if eg

t 6= st∆, if eg

t = st

• Voters of each group only observe their own payoff Ugt ∈ {0,∆}.

• A dissonant politicians gets 13 of the rent rt ∈ [0,R] for each

non-congruent targeted policy.

27 / 45


Optimal decisions

Equilibrium

• Optimal decisions in period 2 (no change):

∀g , eg2 (s,C) = s2,

∀g , eg2 (s,D) = 1− s2.

• Optimal decisions in period 1:• A congruent incumbent always chooses ∀g eg

1 (s,C) = s1 (nochange).

• Again, optimal decision by dissonant politicians depends on theelectoral best-response by voters between the two periods.

28 / 45


Optimal decisions

• Since groups are identical an of equal size, the dissonant in-cumbent must choose N, the number of groups too please.

• Voters know that they only observe their own payoff.• Assume group g observe ∆ at the end of period 1:

P(i = C |Ug1 = ∆) = P(i = C)× P(∆|i = C)

P(Ug1 = ∆) ,

where:

P(Ug1 = ∆) = π+(1−π)

[P(N = 3) + 2

3P(N = 2) + 13P(N = 1)

].

• Since P(i = C |Ug1 = ∆) > π, a group who observe ∆ will

always support the incumbent.

29 / 45


Optimal decisions

• Knowing this, a dissonant incumbent has no interest to pleaseall groups, nor to please only one group, so:

P(N = 3) = P(N = 1) = 0.

• She must decide whether 2 groups or none.• A dissonant incumbent will behave congruently with 2 groupsif and only if:

E + 13 r1 + β(E + E(r2)) > E + r1 + β × 0,

⇔ r1 <32β(µ+ E ).

• Thus:P(N = 2) = G

(32β(µ+ E )

)≡ λ′.

30 / 45


Optimal decisions

• It follows that λ′ > λ (where λ is the probability from thecanonical model that a dissonant incumbent behave congru-ently in period 1).

• Targeted policies improves discipline in period 1.

31 / 45





W = V1 + βV2,

where:V1 = [π + (1− π)

{λ′ 23

}]∆,

and,V2 = π + (1− π)(1− λ′)π]∆.

• Assuming that βπ ≈ 0, ex ante welfare is higher under clien-telism if:

23λ

′ > λ,

which depends on the shape of function G .32 / 45

Political Economy - Lecture 2: Political agency and electoral controlExtended model accounting for voters’ information

5 Extended model accounting for voters’ informationSetup of the modelOptimal decisionsImplications for the quality of governmentEmpirical evidence

33 / 45


Setup of the model

Setup of the model

• In the canonical model, voters learn immediately about the con-sequences of policy choices. Quite unrealistic

• How does the political equilibrium change if voters are onlyimperfectly informed?

34 / 45


Setup of the model

• Same assumptions as before (see slides 8–10)• But:

• Voters imperfectly observe payoffs (0 or ∆). They observe pay-offs with probability χ after the incumbent has taken her/hisdecision.

• Voters get some information about the incumbent’s type. Theyobserve i ∈ {C ,D} with probability τ after the incumbent hastaken her/his decision.

• Assume that voters reelect the incumbent if no information isrevealed.

35 / 45


Optimal decisions

Equilibrium

• Optimal decisions in period 2 (no change):

e2(s,C) = s2,e2(s,D) = 1− s2.

• Optimal decisions in period 1:• A congruent incumbent always chooses e1(s,C) = s1 (no change).• Optimal decision by dissonant politicians depends on the elec-

toral best-response by voters between the two periods, which isnow modified by available information.

36 / 45


Optimal decisions

• Suppose a dissonant incumbent chooses e1(s,D) 6= s1.• Voters will know she/he is dissonant with probability τ+(1−τ)χ.• So, the incumbent will be reelected with probability

(1− χ)(1− τ).• Suppose a dissonant incumbent chooses e1(s,D) = s1.

• Voters will know she/he is dissonant with probability τ .• So, the incumbent will be reelected with probability 1− τ .

37 / 45


Optimal decisions

• A dissonant incumbent will behave congruently if short-termbenefits from dissonance are lower than long-term benefits ofcongruence:

E+(1−τ)×β (E + E(r2)) > E+r1+(1−χ)(1−τ)×β×(E + E(r2)) ,

⇔ r1 < χ(1− τ)β(µ+ E ).

• Thus:λ = G (χ(1− τ)β(µ+ E )) .

• Discipline is increasing in χ (quality of voters’ information) butdecreasing in τ (reduced incentive to take congruent decisions).

38 / 45





W = V1 + βV2,

where V1 is the same as before, but:

V2 = [π+ (1−π) {(1− λ)((1− (1− χ)(1− τ)) Z + λτZ}]∆,

where Z is the probability that the action is congruent duringa politician’s first term:

Z = π + (1− π)λ

39 / 45



• χ and τ affect W both directly and indirectly (via λ).• χ always increases λ, and hence W.• τ has an ambiguous effect. Higher quality information increaseschances of selecting a congruent politician in period 2, butreduces discipline in period 1. The second effect dominates ifπ is low.

40 / 45


Empirical evidence

Empirical evidence

Timothy Besley & Robin Burgess, 2002. “The Political Economy of GovernmentResponsiveness: Theory and Evidence from India,” The Quarterly Journal ofEconomics, Oxford University Press, vol. 117(4), pages 1415-1451.

• Is government more responsive to citizens’ needs when votersare better informed?

• Examine how Indian state governments react to falls in foodproduction and crop food damage depending on newspaper cir-culation.

41 / 45


Empirical evidence


Government’s reactions,t = β0 + β1Shocks,t + β2Newspapers,t+β3Newspapers,t × Shocks,t+δs + δt + . . .

• Shocks: Drought and flood.• Policy reactions: Calamity relief expenditure and public fooddistribution.

42 / 45


Empirical evidence

Newspaper circulation and government responsiveness (t-statistics in parentheses).Source: Besley and Burgess (2002) 43 / 45

Political Economy - Lecture 2: Political agency and electoral controlConclusion

6 Conclusion

44 / 45

Political Economy - Lecture 2: Political agency and electoral controlConclusion

Conclusion

• Other issues (not covered):• Once elected, politicians might have better information thanvoters about what is good or bad for the country. Yet, reelec-tion concerns might push politicians to take non-optimal deci-sions (from the social welfare point of view) to get reelectedsince voters are wrong about what should be done.

• If politicians can hold office more than twice, dissonant politi-cians are less disciplined as time goes. The idea is that theyhave more to loose as they can stay longer in office by behavingcooperatively. Need repeated games to model such situations.

• Wages offered to politicians can allow for endogenous qualityof candidates. Same thing for the information structure.

45 / 45

End of lecture.

Lectures of this course are inspired from those taught byD. Acemoglu, Y. Algan, R. Durante, and B. Olken.

Political Economy - Lecture 2: Political agency and ...marcsangnier.free.fr/.../politicaleconomy_m2_2015-2016_lec2.pdf · PoliticalEconomy-Lecture2: Politicalagencyandelectoralcontrol

Documents