Populism

Lars Frisell∗

February 5, 2005

Abstract

In their pursuit of being elected, politicians might not provide their con-stituents with independent viewpoints, but just try to outguess popular opinion.Although rational voters see through such populism, candidates can not resist re-sorting to it when the spoils of office are too large. For an intermediate parameterrange, both populism and its opposite, “candor”, can be sustained as equilibria.This means that the public’s trust or distrust in politicians may be self-fulfillingprophecies. Importantly, the more informed politicians are about public opinion,the more likely it is that populism can be avoided.Keywords: Popular opinion, pandering, political trust, electoral competition,

candidate motivation.JEL Classification: D72, D82.

Populism A political strategy based on a calculated appeal to the interests or prej-

udices of ordinary people.

— The Collins English Dictionary (2000)

Democracy thrives on popular support and withers in its absence.

— David Easton (1965)

∗Sveriges Riksbank; lars.frisell@riksbank.se. I thank Sven Feldman, Roger B. Myerson, LeonardNakamura, Rikard Nilsson, Kasper Roszbach, Jonas Vlachos, and seminar participants at the HarrisSchool of Public Policy Studies and the Federal Reserve Bank of Philadelphia for helpful comments. Iam especially grateful to Yaron Leitner, Torsten Persson, and Mattias Villani for rewarding discussions.This paper was begun while I was visiting the WZB, Berlin. I am grateful to them for their hospitalityand to the Jan Wallander and Tom Hedelius Foundation for financial support. The views expressed inthis paper are those of the author and should not be interpreted as reflecting the views of the Executiveboard of Sveriges Riksbank.

1

1 IntroductionOriginally denoting members of The People’s Party — an agrarian movement formed

in the US in 1890 — the term populist now broadly refers to any politician or political

party that, regardless of ideological bias, conforms to popular opinion in order to gain

political advantage. By definition, populist policies appear favorable for significant

factions of the electorate (“ordinary people”), but are inferior, since they are not chosen

on basis of all available information.1

Political candidates are naturally drawn towards public opinion for, whether a can-

didate agrees with it or not, it is the yardstick by which her chances to win office —

and thus the opportunity to shape policy at all — are determined. Although populist

tendencies may be a necessary ingredient in representative democracies, in certain envi-

ronments a political climate forms where politicians resort to demagoguery, and suppress

any concerns about efficiency or sustainability.2 One of the most long-lived instances of

deficient political cultures is the “classic” (left-wing) populist doctrine, which prevailed

in Latin America from the Great Depression until well in the 1990s. In country after

country, protectionism, reckless deficit spending and macroeconomic expansion led to

hyperinflation and economic crisis. Dornbush and Edwards (1991, p. 12) laconically

conclude that “although the final outcome of these experiments was not always the

total collapse and destruction of the economy (as in Chile, Peru, and Nicaragua, for

example), in all cases there were disastrous effects for those groups who were supposed

to be the beneficiaries of the policies.”

The question is how to explain these and other self-destructive policy experiments

in a world where voters are bestowed with rational expectations.3 “Ordinary people”

1An alternative term is “opportunism”, but this usually refers to policy decisions taken after anelection, and hence have not been approved of by the electorate. A more recent term is “pandering”(see Canes-Wrone et al. 2003 and Maskin and Tirole 2004). For a synthesis of the various connotationsof populism, see Canovan (1981). Canovan’s own terminology comprises seven categories, where thecurrent definition corresponds best to what she denotes “politicians’ populism”.

2See Hillman and Swank (2000) for a discussion of the significance of political cultures.3Although Dornbush and Edwards (1991, p. 8) assert that they have “no doubt...about the sincerity

of the policymakers who embarked on these programs”, Alberto Alesina expresses a different view (p.42 in the same volume): “Why is it that certain countries keep repeating the same ’mistakes’ and neverlearn? In fact, once the political and institutional incentives and constraints are taken into account,policies that appear to be mistakes are perfectly rational responses to distorted or imperfect politicalincentives.”

2

may certainly be poorly informed about the long-term consequences of different policies,

but this does not imply that they will be biased towards politicians that adopt poor

policies. Alternatively put, uninformed but rational voters will know better than to put

much faith in their own (and thus any populist’s) unreliable opinion. By consequence,

a politician known just to submitting to popular belief will be punished in terms of

election chances. The ultimate question is then why politicians would ever resort to

populist behavior; why, to paraphrase Wittman (1989), do not political entrepreneurs

emerge and provide the information to voters, win office, and reap both the direct and

indirect returns of holding office?

In this paper I employ a stylized model to explore the populist phenomenon. An

incumbent politician is about to propose a policy platform. The efficiency of a given

policy depends on the state of the world, which is unknown, but politicians and voters

have some private information on what the best policy would be. In addition, the in-

cumbent has (imperfect) information on what voters think, for example through opinion

polls. Upon observing the incumbent’s policy proposal, voters use their own informa-

tion to assess its quality. If the expected quality is sufficiently high, the incumbent

is re-elected and the proposed policy is carried out. If not, the electorate ousts the

incumbent and draws a new politician.4 While voters are solely concerned with policy

quality, politicians are also motivated by office rewards (“ego rents”), whose size are

common knowledge.

To focus on informational aspects, I assume that voters have common interests and

can be modelled by a representative voter. This does not necessarily mean that partisan

conflict is absent, but that informational quality is pivotal for the electoral outcome.

My results are as follows. I classify political environments in terms of how office-

motivated politicians are. I show that when the spoils of office are large, the incumbent

will not propose informed policies as she is too tempted to pander to her constituents’

opinion. However, since voters anticipate this, populist behavior is self-defeating. Still,

the electorate is also worse off as they are deprived of high-quality policy alternatives,

4This sequential election structure simplifies the analysis and puts the focus on the interplay betweenpolitician and voter.

3

so populism constitutes a commitment problem for both politician and voters.

Importantly, for a range of office spoils both populism and its opposite, “candor”,

are self-fulfilling expectations. This means that once a reputation for populist behavior

(a populist “culture”) has been established, politicians and voters may be stuck in

this equilibrium.5 The voter’s proficiency has ambiguous effects on welfare. More

competent constituents comprise a better source of information for politicians, but may

make populism inevitable as the temptation to mimic public opinion becomes stronger.

Interestingly, the more the politician knows about public opinion, the more likely it is

that populism can be avoided. This result challenges the view that opinion polls are an

instrument that feeds populist behavior (see, e.g., Mansfield 1994). Briefly explained,

voters will judge a discrepancy between their own opinion and the proposed policy

more harshly the lower is the precision of the policy. Hence, as long as voter opinion

has some informational value, less knowledge of public opinion, not more, may impel

the politician to increase her election chances by shifting to populist behavior.

Several recent papers have studied politicians’ incentives to conform to popular

opinion (Harrington 1993, Canes-Wrone et al. 2001, Chiu 2002, Heidhues and Lagerlöf

2003, Maskin and Tirole 2004).6 However, all of these assume binary policy spaces and

are therefore unable to capture the crucial component in the current paper: how voters

use their own information to evaluate policy proposals. Obviously, in a binary policy

space a politician’s decision whether to pander to popular opinion is all-or-nothing,

which makes the voter’s decision problem uninteresting.

In Harrington (1993) an incumbent chooses between two policies whose economic

effects will affect his election chances. The incumbent has private information on which

policy is the more efficient, which may or may not coincide with the (median) voter’s

prior belief. Harrington shows that the stronger is the voter’s prior, and the more

office-concerned is the incumbent, the more likely it is that the incumbent chooses

policy according to the voter’s prior despite this being the less efficient policy.

5Hence, in contrast to Easton’s (1965) classic input-output model of political performance, votersatisfaction will be low because they have low expectations on the system’s output, not high ones.

6Somewhat less related is a series of papers that shows that imperfect knowledge of an agent’spreferences may lead her to distort her messages in order to signal a certain bias — or lack thereof. See,e.g., Bernheim (1994), Cukierman and Tommasi (1998), and Morris (2001).

4

Like the current paper, Maskin and Tirole (2004) assume that politicians desire office

per se, and care about policy outcomes only when they shape policy themselves (what

they denote a legacy motive).7 The authors show that when the office-holding motive is

strong relative to the legacy motive, representative democracy is inferior to other forms

of government — precisely because of politicians’ pandering to popular opinion.

Several papers have demonstrated that officials’ opportunistic behavior (such as

taking bribes or seeking rents) may be strategic complements, which, although for a

very different reason, gives rise to multiple equilibria of the same flavor as in the current

paper. Aidt (2003) surveys this literature.8

The model I employ owes much to Prendergast (1993). Prendergast shows that if

superiors use subjective performance evaluation to reward workers, workers will tend to

conform to their superiors’ opinion. In an electoral setting, “performance evaluation”

is necessarily subjective since, arguably, constituents can hardly commit to vote for a

candidate irrespective of how they perceive his or her policy platform.9

Of course, in a partisan (Downsian) model where voters possess all policy-relevant

information, basing policy on popular opinion would imply no inefficiency. Indeed, in

such a setting the presence of large office spoils is only desirable, as it pulls biased

politicians towards the (expected) median voter (Calvert 1985). Only in a world where

politicians have a role as information sources can populist behavior be detrimental to

voter welfare.

The remainder of the paper is organized as follows. Section 2 sets up the model

and presents some fundamental results. Section 3 uses a log-linear approximation and

characterizes the set of perfect Bayesian equilibria (PBE) in pure strategies. Section 4

7While Maskin and Tirole allow for the possibility that a politician’s preferred policy differs fromthat of the electorate (the politician is “non-congruent”), in the current paper politicians are (knownto be) congruent. Also, since politiciancs have no unknown traits in the current model, selection effectsis not an issue. For a review of the literature on candidate selection, see Besley (2004).

8A recent contribution in this vein is Beniers and Dur (2004). In their paper, bureaucrats mayinefficiently continue projects in order to signal quality, something that becomes more tempting ifother bureaucrats behave in the same way. An analogous mechanism is present in Boot and Thakor(1993).

9Prendergast shows that “aloofness” between different layers in the organization is desireable as itreduces subordinates’ possibility to mimick their superiors. To the contrary, the voter in the currentmodel, who can not affect the proposed policies, always benefit from the incumbent’s being moreinformed about his opinion.

5

performs comparative statics, and section 5 concludes. All proofs are in the Appendix.

2 The ModelAn incumbent politician i is running for re-election. She will present her policy

platform pi ∈ < to the electorate, modeled by a single representative voter, v. After

observing her policy proposal, the voter decides whether to re-elect the incumbent, or to

replace her with a challenger, c. If re-elected, the incumbent is committed to implement

her proposed policy;10 if she is ousted the challenger implements her preferred policy,

pc ∈ <.

Utility The voter cares exclusively about policy quality, which depends on an un-

known state of nature π. Let Iv denote the voter’s information set after observing the

incumbent’s proposal. Let bi = E[(pi − π)2] and bc = E[(pc − π)2] denote the expected

squared bias (ESB) of the incumbent’s and challenger’s policy, respectively, given the

incumbent’s information, and let bi |Iv and bc |Iv denote the ESB of the respective pol-

icy given the voter’s information. Successful implementation of a policy may depend on

other factors than informational precision, such as a politician’s bureaucratic skills, the

experience of her staff, etc., where the incumbent may have a considerable advantage

over the challenger. The incumbent is re-elected if bi |Iv ≤ αbc |Iv, where α ∈ [1, α], α

finite, reflects the incumbency advantage. The ex ante probability that the incumbent

is re-elected is

φi ≡ Pr (bi | Iv < αbc | Iv) .

I shall assume that α is sufficiently high so that the incumbent has positive election

probability in equilibrium. Politicians care about policy quality, but only when they

get to shape policy themselves. This means that politicians only take policy quality

into account conditional on the fact that they are elected. Aside from this, politicians

rank policies in the same way as voters do. Politicians are also motivated by the spoils

of office per se, the importance of which is common knowledge.11 The incumbent’s ex

10For example, commitment to policy may be necessary if incumbents who are not committed canbe “bought” by special interests after the election.11However, note that in a “citizen-candidate” world where politicians’ motives differ and are unob-

6

ante utility is thus a function

Ui(pi) = f(φi,bi), (1)

which is strictly increasing in φi and strictly decreasing in bi. In the next section I

employ a simple specification of (1).

Information structure Politicians and voter have some private information on

the state of the world. Specifically, the incumbent gets a signal πi = π + ²i and the

voter gets a signal πv = π+ ²v, where ²i and ²v are normally distributed with zero mean

and variances σ2i and σ2v, respectively. In addition, the incumbent has some private

information on the voter’s signal, for example through opinion polls, πz = πv + ²z,

where ²z is normally distributed with zero mean and variance σ2z. The signals πi and πv

are independent conditional on the state. All signal variances are positive and finite.

For simplicity I assume that the precision of the challenger’s policy is given at the outset

so that bc |Iv = bc is invariant to the incumbent’s proposal.12 Except for the signal

realizations, all aspects of the game are common knowledge.

The voter’s updating The set Iv consists of two pieces of information, pi and πv,

which the voter can use to form his ex post expectation of the state. Without loss of

generality, assume pi is a convex combination of πi and πz, so that pi is an unbiased

estimate of the state. This must be optimal whether the incumbent wants to shape good

policies or just wants to get elected. It follows that the voter’s ex post estimate of π is

unbiased if and only if it is a convex combination of pi and πv. Let µ, (1 − µ) ∈ [0, 1]

denote the weights of πi and πz in pi, respectively, according to the voter’s belief. Of

course, in a perfect Bayesian Equilibrium we require that this belief is correct.

LEMMA 1. The voter’s expectation of the state is

servable, those citizens that choose to run for office will by definition be susceptible of having a lotto gain from winning it, which may give rise to the same credibility problem as when λ is high andknown. Caselli and Morelli (2004) demonstrates a selection effect of this flavor.12This assumption is not critical but makes life easy. If bc was affected by pi, the incumbent’s

optimal strategy could depend on the inference the challenger would make upon observing pi (as wellas the voter’s decision to reject pi), an inference that itself would depend on the incumbent’s strategy.Essentially it is assumed that the challenger has sufficiently much (independent) information so thatthis inference problem is negligible. (In particular, this should imply that bc always is less than bi.)

7

bπ = δ−pi + (1− δ−)πv, (2)

where δ− is the minimum of 1 and

δ∗ =µσ2v

µ2(σ2i + σ2v) + σ2z(1− µ)2. (3)

In the Appendix it is shown that bπ has the least variance among all (unbiased)estimators. The slightly complex mapping in (3) is due to the interdependence between

πz and πv. For example, if µ = 0 the policy (supposedly) equals πz. Since πz is a garbled

signal of πv the voter should put no weight on pi in his estimate, so that bπ = πv. If

µ = 1 the policy exclusively reflects the candidate’s independent information and we

get the familiar result bπ = σ2i πv+σ2vπi

σ2i+σ2v.

The role of δ∗ The ESB of the incumbent’s policy, conditional on the voter’s

information, can be decomposed into two parts: the variance of π plus the policy’s

squared bias to π. Using (2), this can be written as

bi | Iv = Var(bπ) + (pi − bπ)2 = Var(bπ) + (1− δ−)2 (pi − πv)2 . (4)

Denote the expected squared deviation between the incumbent’s policy and the voter’s

signal (given the incumbent’s information) by di = E£(pi − πv)

2¤ . The ex ante expec-tation of (4) then reads

E[bi | Iv] = Var(bπ) + (1− δ−)2di.

Two things are important to note. First, taking the voter’s belief µ as given, E[bi |Iv],

hence the incumbent’s election chances, only depends on di.13 Except for the case δ− =

1, the voter’s prior opinion will affect his ex post expectation of the state, which means

that the more a policy deviates from the voter’s opinion, the lower is its assessed quality.

Second, δ− is a crucial variable, for it determines how sensitive election chances are to

13The fact that Var(bπ) and δ− are independent of pi is due to normality. See the proof of Lemma 1.8

di: the lower is δ−, the costlier it becomes to deviate from public opinion. Indeed,

the way in which δ− changes with parameters is key to the main results of the paper.

Lemma 2 below shows that, in equilibrium, the voter always attaches some weight to

his own opinion when forming bπ.LEMMA 2: In a PBE, δ− = δ∗ < 1.

The probabilistic election function φi : E[bi |Iv] −→ (0, 1) has no explicit functional

form and must be solved numerically. In Figure 1 the incumbent’s election probability

is plotted as a function of E[bi |Iv] and α for a given bc.

α = 3

α = 2

α = 1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

2 3 4 5 6 7 8 9 10 11

φ

][ vi IbE

α = 3

α = 2

α = 1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

2 3 4 5 6 7 8 9 10 11

φ

][ vi IbE

Figure 1. The incumbent’s election probability as a function of E[bi |Iv] when Var(bπ) = 1,

δ−= 0, and bc = 2. Probability estimates are based on 500.000 random draws.

LEMMA 3: δ∗ is

(i) decreasing in σ2i ,

(ii) increasing in σ2v,

(iii) decreasing in σ2z,

(iv) increasing in µ in an open neighborhood around µ = 0.

Recall that δ∗ denotes the weight the voter attaches to the incumbent’s policy when

forming his ex post expectation of the state. Intuitively, a less informed politician

9

(higher σ2i ) or a more informed voter (lower σ2v) implies that this weight decreases.

A less precise opinion poll (higher σ2z) means that — since the voter’s signal itself is

informative — the incumbent becomes less informed, why δ∗ decreases. Part (iv) of

Lemma 3 says that, as long as µ is not too large, δ∗ is increasing in the voter’s belief of

how much weight the incumbent put on her own signal when she formed pi. Intuitively,

the more information on πi a policy contains, the more it adds to the voter’s knowledge

of π, why the weight of pi in bπ increases with µ.143 Equilibrium Analysis: A Log-Linear ApproximationConsider the following specification of (1):

Ui(pi) = φi (·)λ ∗ ui(·)1−λ,

where

φi (·) = exp−γE[bi | Iv],

and

ui(·) = exp−bi.

Hence, I assume here that the incumbent approximates φi (·) with an exponential func-

tion, an approximation that, as suggested by Figure 1, is very close.15 By assuming that

ui(·) also is exponential, we get a simple log-linear utility function, which guarantees the

existence of pure-strategy PBE. Again, assume the incumbent’s (equilibrium) election

probability is positive but less than unity, which implies γ ∈ (0, γ), γ finite. Intuitively,

an increase in γ corresponds to a decrease in α or bc : the smaller is the incumbency

advantage, and the higher is the quality of the challenger’s policy, the more sensitive

are the incumbent’s election chances to her policy’s quality. The parameter λ ∈ (0, 1)

reflects how much politicians value office spoils relative to shaping good policies (see

below). Taking logs and substituting for E[bi |Iv] gives

14However, when µ is sufficiently large a further increase makes the information in πv relatively moreuseful, why δ∗ decreases.15Exponential functions explained 95, 98, and 99 per cent of total variation in the data for the case

α = 1, 2, and 3, respectively, in Figure 1. (Calculated over the range [2, 12] with grid 0.25.)

10

ln Ui(pi) = −λγVar(bπ)− λγ(1− δ∗)2di − (1− λ)bi. (5)

DEFINITION 1:

µ0 =σ2v + σ2z

σ2i + σ2v + σ2z

µ1 =σ2z

σ2i + σ2v + σ2z

Candid politicians When λ approaches zero, (5) reduces to −bi. This kind of

politician is only interested in shaping high-quality policies, and is unconcerned with

election consequences. Although probably rare in reality, such idealism (or naïveté) is

clearly optimal for voters.16 The equilibrium policy is

p0 = µ0πi + (1− µ0)πz,

which has ex ante ESB

b0 =σ2i (σ

2v + σ2z)

σ2i + σ2v + σ2z.

In p0 the signals πi and πz enter in inverse proportions to their respective variances. It

is easily verified that p0 has the smallest possible ESB.

Office-concerned politicians When λ approaches one, the incumbent is virtually

indifferent as to which policy she pursues, so long as she gets elected. She will therefore

minimize the policy’s perceived bias, di. The equilibrium policy is

p1 = µ1πi + (1− µ1)πz,

which has ex ante ESB

16Someone suggested that Ralph Nader might fit this description. Note that the result that candidbehavior benefits the (median) voter does not hold in all circumstances. (For example, if there aremore than two candidates, there is distributive conflict among constituents, and some voters are non-strategic.) According to many observers, Nader’s candidacy in the presidential election in 2000 tiltedthe outcome in favor of George W. Bush.

11

b1 =σ2i (σ

2v + σ2z) + σ4v

σ2i + σ2v + σ2z.

It can be seen directly that b1 > b0, i.e., the precision of an office-concerned politi-

cian’s policy is lower than that of a candid one’s. In analogy with previous studies, the

politician’s desire to increase her election chances leads her to compromise the quality

of her policy. However, a rational voter realizes whether the incumbent biases her plat-

form in this manner, and adjusts his beliefs accordingly. This means that, although

p1 (on average) better accords with πv than does p0, in equilibrium this is more than

compensated for by the tougher yardstick by which p1 is evaluated. In other words, a

populist incumbent both adopts a poorer policy and, precisely because of this, faces a

smaller election probability. Nonetheless, voters are also worse off under populism, as

they are deprived of a more qualitative policy alternative.

PROPOSITION 1: Suppose that both p0 and p1 are sustainable as PBE for some

set of parameters. If the incumbent’s equilibrium policy is p1 rather than p0,

(i) the expected quality of her policy is lower,

(ii) her election probability is smaller,

(iii) the voter’s expected utility is lower.

The general case Due to the log-linear utility function, in a pure equilibrium the

policy is always either p0 or p1. What is important however, is that the two equilibria

coexist for a range of λ. The reason is that voter expectations are, to a certain extent,

self-fulfilling (cf. part (iv) of Lemma 3). Recall that µ denotes the weight the incumbent

(supposedly) assigns to her own signal when forming pi. If µ is small, for example, the

voter should put a low weight on pi in π, which implies that δ∗ is small. This increases

(1− δ∗)2, which means that a given deviation from the voter’s opinion now has larger

election consequences. As a result, the incumbent’s incentive to decrease di, i.e., to

increase the weight of πz in pi, becomes stronger as µ decreases — which indeed fulfills

the voter’s belief.

12

When λ is close to zero, p0 is the unique equilibrium, and when λ is close to one, p1 is

the unique equilibrium. For intermediate values of λ both equilibria (and one mixed) are

possible, and the equilibrium policy must be jointly determined with the voter’s beliefs.

Loosely put, a “skeptic” voter attitude tends to generate desperate politicians, while a

trusting attitude tends to generate confident ones. These findings are summarized in

Proposition 2 and illustrated in Figure 2.

DEFINITION 2:

λ0 =1

1 + γ(1− δ∗(µ1))2

λ1 =1

1 + γ(1− δ∗(µ0))2

PROPOSITION 2: 0 < λ0 < λ1 < 1. For λ < λ0, p0 is the unique equilibrium, for

λ > λ1, p1 is the unique equilibrium, and for λ0 ≤ λ ≤ λ1 both equilibria are possible.

Figure 2. Equilibrium policy as a function of λ.

1

p1 uniquep0 unique Both equilibriapossible

λ0 0λ 1λ

Figure 2. Equilibrium policy as a function of λ.

1

p1 uniquep0 unique Both equilibriapossible

λ0 0λ 1λ

13

4 Comparative StaticsA shift from p0 to p1 implies a behavioral change: instead of estimating the unknown

state the incumbent estimates the voter’s perception of the state, which, depending on

parameters, may lead to a sizeable deterioration in policy quality. Intuitively, the higher

is the candidate’s informational precision and the lower is the voter’s, the larger is the

difference. Also, better knowledge of voter opinion increases the weight the incumbent

(inefficiently) attaches to πz, which further reduces the quality of p1 relative to p0. Note

that this holds even if the voter is better informed than the politician; lower σ2z always

leads to a “too large” increase in the weight of πz in p1. This is summarized below.

PROPOSITION 3: The difference in policy quality, b1 − b0, is larger the

(i) lower is σ2i ,

(ii) the higher is σ2v,

(iii) the lower is σ2z.

Proposition 3 seems to imply that populism poses a more severe threat in environ-

ments where voters are poorly informed vis-à-vis politicians, and where candidates have

easy access to voter opinion. The latter result would corroborate the view that opinion

polls have a detrimental effect on politicians’ behavior. However, these conclusions are

unwarranted. Given a certain equilibrium, the (hypothetical) loss from populism in-

creases with σ2v and decreases with σ2z, but, more importantly, the equilibrium set itself

may depend on these parameters. In particular, a parameter change that causes λ1 to

decrease may make populism inevitable.

PROPOSITION 4: The set of λ for which populism is the unique equilibrium,

(1− λ1), is

(i) increasing in σ2i ,

(ii) decreasing in σ2v,

(iii) increasing in σ2z.

(iv) increasing in γ.

14

Part (iii) of Proposition 4 implies that if the society originally is in the candid

equilibrium, less knowledge of voter opinion may decrease welfare for two reasons. First,

policy quality decreases (b0 increases) directly as σ2z increases, but this does not affect

the incumbent’s incentive to shift from p0 to p1. More interestingly, λ1 decreases, which

means that p0 may become unsustainable. This result seems paradoxical. How can less

knowledge of voter opinion increase the risk that politicians try to mimic it? Again, the

intuition stems from how voters (rationally) evaluate deviations between the proposed

policy and their own opinion. When σ2z increases, the precision of p0 decreases, which

leads δ∗ to decrease. This implies that deviations from public opinion become costlier

in terms of election chances. Hence, the incentive to decrease di — by shifting to p1 —

increases as σ2z increases, which implies that λ1 decreases. Conversely, if the incumbent

had perfect knowledge of voter opinion, populism would always be avoidable (see Figure

3).17

The same argument applies to a change in σ2i but not σ2v. An increase in σ2v certainly

decreases policy quality (through its effect via the opinion poll) but it decreases voter

precision more. Hence, a decrease in voter proficiency makes π more dependent on

the proposed policy, which attenuates the temptation to mimic voter opinion. Roughly

expressed, although the increase in σ2v results in policies deviating more from voter

opinion, the yardstick by which the voter evaluates such deviations softens even more,

so that candor becomes easier to sustain.

Finally, a decrease in γ, either due to an increase in bc (a less informed challenger)

or an increase in α (greater incumbency advantage), reduces the sensitivity of election

chances to policy quality and increases the scope for candid behavior. However, de-

pending on the cause of the shift (e.g., whether the incumbent became more capable,

or the challenger less so), and whether it actually affects the incumbent’s behavior, this

may or may not benefit the voter.

17Specifically, if σ2z = 0, the belief µ = µ0 implies δ∗ = 1. Then, since pi alone determines π,

election chances are independent of pi, why the incumbent would maximize informational precision.The optimal policy, p0, fulfills the belief µ0.

15

2zσ

0.84

0.86

0.88

0.9

0.92

0.94

0.96

0.98

1

0 1 2 3 4

1λ

2zσ

0.84

0.86

0.88

0.9

0.92

0.94

0.96

0.98

1

0 1 2 3 4

1λ

Figure 3. λ1 as a function of σ2z (σ2i = σ2v = γ = 1).

5 Concluding RemarksThough difficult to measure, the welfare costs associated with politicians’ conforming

to popular opinion are probably significant. Due to electoral competition, candidates

will abstain from taking stands that seem controversial but are possibly very informative

for voters. For an economist, the populist phenomenon comprises somewhat of a puzzle:

even if voters are poorly informed, so long as they are rational they should on aggregate

disfavor poor policies. But if sincere behavior implied better election chances, why would

a politician ever resort to populism?

In this paper I show that populist behavior may constitute a genuine commitment

problem. Although voters can anticipate whether a candidate is distorting her messages,

when the spoils of office are too large she can not resist doing it. As a result, both voters

and the populist politician are worse off.

For a range of parameters, populism and its opposite, “candor”, can both be sus-

tained as equilibrium behaviors. The reason is that voter expectations and the politi-

cians’ incentives are mutually reinforcing. For example, if voters expect politicians to

adopt high-quality policies, then any evidence pointing against the proposed policies

will bear lightly in the voters’ assessment of them. This gives politicians more “leeway”

in choosing informed policies, and attenuates their incentive to mimic voter opinion.

This result may help explaining the existence of diverse political cultures, and begs the

intriguing question if the virtually world-wide decline in political trust since the 1950s

(see, e.g., Norris 1999) may in part be a self-fulfilling prophecy.

16

The fundamental mechanism at work, that is, the self-fulfillment of evaluators’ ex-

pectations, should apply to numerous contexts. The crucial feature is that the “sender”,

at least on the margin, is willing to compromise the quality of her message in order to

attain approval. To take just one other example, consider a researcher writing an ar-

ticle for a journal whose editorial board has a certain view on what research fields are

interesting and not. The incentive for a junior scientist to follow the current research

trend — even though she finds it uninteresting herself — should be larger than that of

senior scientist’s with an established reputation; not only because their desire to get

published may differ, but because a deviation from the trend is more likely to be viewed

as a “mistake” if it origins with the junior scientist.

To focus on informational efficiency the paper ignores partisanship, and models the

electorate by a representative voter. Adding partisan conflict to the current model may

be complicated, but suggests an interesting trade-off: In the Downsian analysis, large

office spoils are an effective means to make candidates ignore their own biases and adjust

to voter opinion, whereas here it impairs information transmission from politicians to

voters. Which effect will dominate in a hybrid model should depend on parameter

choices.

Finally, as noted in the introduction, there are other definitions of the term populism

than the one provided here. In particular, according to Harvey Mansfield (1994), what

many refer to as populist movements today are exactly what James Madison used to

denote “majority factions”. A populist regime would then correspond to his “tyranny of

the majority”, where minorities become the victims of the “passions” of the majority; a

prospect that prompted Madison to insist that decisions should be delegated to a body

of wise men who could “refine and enlarge” the public view (Madison, Hamilton, and

Jay 1987, p. 126). Exploring the circumstances when the majority would choose to

instigate such institutions could be a useful venue for future research.

17

Appendix

For later reference, let

d0 ≡ E[(p0 − πv)2] =

σ4v + σ2z(σ2i + σ2v)

σ2i + σ2v + σ2z,

and

d1 ≡ E[(p1 − πv)2] =

σ2z (σ2i + σ2v)

σ2i + σ2v + σ2z.

PROOF OF LEMMA 1: Let π = δpi+(1−δ)πv, δ ∈ [0, 1], be an arbitrary estimator

of π. If the incumbent assigns weight µ ∈ [0, 1] to πi and weight (1− µ) to πz, we have

bi = µ2σ2i + (1− µ)2(σ2v + σ2z). (A1)

In turn, the variance of π reads

Var(π) = δ2bi + (1− δ)2σ2v + 2δ(1− δ)(1− µ)σ2v. (A2)

The last term follows from the fact that the covariance between πz and πv is σ2v. Using

(A1) in (A2) and differentiating w.r.t. δ gives that δ∗ solves the first-order condition.

The second derivative reads

d2π

d2δ= 2µ2(σ2i + σ2v) + 2σ

2z(1− µ)2 > 0,

so that δ∗ solves for a unique interior minimum. Finally, it can be seen directly that δ∗

is never negative, however, it may be larger than unity.

PROOF OF LEMMA 2: Suppose first that φi does not depend on the choice of pi.

Since by assumption φi is positive, the incumbent would then minimize

bi = m2σ2i + (1−m)2(σ2v + σ2z),

with the unique solution m∗ = σ2v+σ2z

σ2i+σ2v+σ

2z≡ µ0. Substituting µ0 for µ in (3) gives

18

δ∗(µ0) =σ4v + σ2vσ

2z

σ4v + σ2vσ2z + σ2iσ

2z

< 1.

Suppose now that, contrary to the Lemma, there was a PBE such that δ− = 1, i.e.,

π = pi. Then bi |Iv and φi would be independent of pi, and the incumbent would strictly

prefer to set m = µ0. But then δ− must equal δ∗(µ0), a contradiction.

PROOF OF LEMMA 3: For notational convenience, let

ϕ =1

(µ2σ2i + µ2σ2v + µ

2σ2z − 2µσ2z + σ2z)2 > 0.

(i) dδ∗

dσ2i= −ϕµ3σ2v < 0.

(ii) dδ∗

dσ2v= ϕµ (σ2z(1− µ)2 + σ2iµ

2) > 0.

(iii) dδ∗

dσ2z= −ϕµ(1− µ)2σ2v < 0.

(iv) dδ∗

dµ= ϕσ2v (σ

2z − µ2(σ2i + σ2v + σ2z)) . Evaluating the derivative at µ = 0 gives

σ2vσ2z> 0.

By continuity, the derivative is positive in an open neighborhood around µ = 0.

PROOF OF PROPOSITION 1: Let F0(bi |Iv) and F1(bi |Iv) denote the distribution

functions corresponding to policy p0 and p1, respectively. Since voter utility is weakly

decreasing in bi |Iv, and strictly decreasing over the range bi |Iv ∈ (0, αbc), all parts

of the proposition follow if F1 first-order stochastically dominates (FOSD) F0. To see

that FOSD holds, note that p0 and p1 are normally distributed (unbiased) estimators

of πv, so that (pi − πv)2 ∼ Gamma(1

2, 2di), where 1

2is the shape parameter and 2di is

the scale parameter (see, e.g., Kenney and Keeping 1951.) Hence, bi |Iv = Var(bπ) +(1 − δ∗)2 (pi − πv)

2 is distributed as Gamma(12, 2di(1 − δ∗)2), shifted to the interval

(Var(bπ), ∞). Since the shape parameter is less than unity, f(bi |Iv) is monotonicallydecreasing and a sufficient condition for FOSD is that both Var(bπ (µ1)) > Var(bπ(µ0))

19

and d1 (1− δ∗(µ1))2 > d0 (1− δ∗(µ0))

2 . Using (A1), (A2) and Definition 1 gives

Var(bπ (µ1)) = σ2vσ4v + 2σ

2iσ2v + σ4i + σ2iσ

2z

(σ2i + σ2v) (σ2i + σ2v + σ2z)

,

Var (bπ(µ0)) = σ2iσ2v

σ4v + 2σ2vσ2z + σ2iσ

2z + σ4z

(σ4v + σ2iσ2z + σ2zσ

2v) (σ

2i + σ2v + σ2z)

,

and

Var(bπ (µ1))−Var (bπ(µ0)) = σ6v (σ4v + σ2iσ

2v + σ2vσ

2z + 2σ

2iσ2z)

(σ2i + σ2v) (σ2i + σ2v + σ2z) (σ

4v + σ2iσ

2z + σ2vσ

2z)> 0.

Using (3), Definition 1, and substituting for d0 and d1 we get

(1− δ∗(µ1))2 d1 =

σ4iσ2z

(σ2i + σ2v) (σ2i + σ2v + σ2z)

,

(1− δ∗(µ0))2 d0 =

σ4iσ4z

(σ2i + σ2v + σ2z) (σ4v + σ2iσ

2z + σ2vσ

2z),

and

(1− δ∗(µ1))2 d1 − (1− δ∗(µ0))

2 d0 =σ4iσ

4vσ2z

(σ2i + σ2v) (σ2i + σ2v + σ2z) (σ

4v + σ2iσ

2z + σ2vσ

2z)> 0.

PROOF OF PROPOSITION 2: Without loss of generality, let the incumbent’s

policy choice be a convex combination of p0 and p1. Denote the weight she puts on p0

and p1 by (1− w) and w, respectively. (5) becomes

−λγVar(π)− λγ (1− δ∗)2 ((1− w)d0 + wd1)− (1− λ)((1− w)b0 + wb1). (A3)

The expression is clearly linear in w. Hence, except for knife-edge cases, w = 1 or w = 0

is the unique solution to the incumbent’s problem. Suppose first that the voter believes

that w = 1, so that µ = µ1. If the incumbent indeed sets w = 1, (A3) becomes

−λγVar(π(µ1))− λγ (1− δ∗(µ1))2 d1 − (1− λ)b1. (A4)

20

If the incumbent instead sets w = 0, (A4) becomes

−λγVar(π(µ1))− λγ (1− δ∗(µ1))2 d0 − (1− λ)b0. (A5)

p1 is an equilibrium if and only if (A4) ≥ (A5). This holds so long as

λ ≥ (b1 − b0)(b1 − b0) + γ (1− δ∗(µ1))

2 (d0 − d1),

which, using the fact that (b1 − b0) = (d0 − d1), reduces to λ0.

Suppose instead that the voter believes that the incumbent sets w = 0. Doing so

gives the incumbent utility

−λγVar(π(µ0))− λγ (1− δ∗(µ0))2 d0 − (1− λ)b0, (A6)

while deviating to w = 1 gives

−λγVar(π(µ0))− λγ (1− δ∗(µ0))2 d1 − (1− λ)b1. (A7)

Using the definitions gives that (A6) ≥ (A7) so long as λ ≤ λ1.

It can be seen directly that λ0,λ1 ∈ (0, 1). Using Definition 1 and (3) gives

δ∗(µ0) =σ2v (σ

2v + σ2z)

σ2iσ2z + σ2v (σ

2v + σ2z)

,

and

δ∗(µ1) =σ2v

σ2i + σ2v.

The difference δ∗(µ0)− δ∗(µ1) =

σ2iσ4v

(σ2iσ2z + σ2vσ

2z + σ4v) (σ

2i + σ2v)

> 0,

which implies that λ1 > λ0.

PROOF OF PROPOSITION 3: Using the definitions gives b1 − b0 = σ4vσ2i+σ

2v+σ

2z.

21

Differentiating this term w.r.t. σ2i , σ2v, and σ2z, respectively, gives

(i) − σ4v

(σ2i+σ2v+σ2z)2 < 0.

(ii) σ2v(σ2v+2σ

2i+2σ

2z)

(σ2i+σ2v+σ2z)2 > 0.

(iii) − σ4v

(σ2i+σ2v+σ2z)2 < 0.

PROOF OF PROPOSITION 4: Part (i), (ii), and (iii) follow immediately from

Lemma 3. Since δ∗ < 1, the derivative of λ1 w.r.t. γ is positive, which proves part (iv).

ReferencesAidt, T.S., “Economic Analyses of Corruption: A Survey.” Economic Journal, Vol.

113 (2003), pp. 632-652.

Benier, K. and Dur, R., “Politicians’ Motivation, Political Culture, and Electoral

Competition.” CESifo Working Paper Series No. 1228.

Besley, T., “Paying Politicians: Theory and Evidence.” Journal of the European

Economic Association, Vol. 2 (2004), pp. 193—215.

Bernheim, D., “A Theory of Conformity.” Journal of Political Economy, Vol. 102

(1994), pp. 841-877.

Boot, A. and Thakor, A.V., “Self-Interested Bank Regulation.” American Economic

Review, Vol. 83 (1993), pp. 206-212

Canes-Wrone, B., Herron, M.C. and Shotts, K.W., “Leadership and Pandering: A

Theory of Executive Policy-Making.” American Journal of Political Science, Vol. 45

(2001), pp. 532-550.

Canovan, M., Populism. Harcourt Brace Jovanovich, New York 1981.

Calvert, R., “Robustness of the Multidimensional Voting Model: Candidate Moti-

vations, Uncertainty, and Convergence.” American Journal of Political Science, Vol.

29 (1985), pp. 69-95.

22

Caselli, F. and Morelli, M., “Bad Politicians.” Journal of Public Economics, Vol 88

(2004), pp. 759-782.

Chiu, S.Y., “On the Feasibility of Unpopular Policies under Re-Election Concerns.”

Southern Economic Journal, Vol. 68 (2002), pp. 841-858.

Cukierman, A. and Tommasi, M., “When Does it Take a Nixon to Go to China?”

American Economic Review, Vol. 88 (1998), pp. 180-197.

Dornbush, R. and Edwards, S. (eds.), The Macroeconomics of Populism in Latin

America. The University of Chicago Press, Chicago 1991.

Easton, D., A Systems Analysis of Political Life. Wiley, New York 1965.

Harrington, J.E., “Economic Policy, Economic Performance, and Elections.” Amer-

ican Economic Review, Vol. 83 (1993), pp. 27-42.

Heidhues, P. and Lagerlöf, J., “Hiding Information in Electoral Equilibrium.” Games

and Economic Behavior, Vol. 42 (2003), pp. 48-74.

Hillman, A.L. and Swank, O., “Why Political Culture Should be in the Lexicon of

Economists.” European Journal of Political Economy, Vol. 16 (2000), pp. 1-4.

Kenney, J. F. and Keeping, E. S., Mathematics of Statistics, Pt. 2. Van Nostrand,

Princeton (NJ) 1951.

Madison, J., Hamilton, A., and Jay, J., The Federalist Papers, edited by I. Kramnick,

Penguin Books, London 1987. First published in 1788.

Mansfield, H., “Democracy and Populism,” pp. 27-30 in Leiken, R. (ed.), A New

Moment in the Americas. Transaction, New Brunswick (NJ) and London 1994.

Maskin, E. and Tirole, J., “The Politician and the Judge: Accountability in Gov-

ernment.” American Economic Review, Vol. 94 (2004), pp. 1034-1054.

Morris, S., “Political Correctness.” Journal of Political Economy, Vol. 109 (2001),

pp. 231-265.

23

Norris, P. (ed.), Critical Citizens: Global Support for Democratic Governance. Ox-

ford University Press, New York 1999.

Prendergast, C., “A Theory of Yes Men.” American Economic Review, Vol. 83

(1993), pp. 757-70.

Wittman, D., “Why Democracies Produce Efficient Results”. Journal of Political

Economy, Vol. 97 (1989), pp. 1395-1424.

24