Bureaucrats or Politicians? Part I: A Single Policy Task

This paper investigates the normative criteria that guide the allocation of a policy task to an elected politician versus an independent bureaucrat. The bureaucrat is preferable for technical tasks for which ability is more important than effort, or if there is great uncertainty about whether the policymaker has the required abilities. The optimal allocation of redistributive tasks is ambiguous, and depends on how the bureaucrat can be instructed. But irrespective of the normative conclusion, the politician prefers not to delegate redistributive policies. ( JEL D72, D73, D82) Policies are chosen and implemented by both elected representatives (politicians) and nonelected bureaucrats. The view that politicians choose policies and bureaucrats implement the grey many cases bureaucrats do much more than nonelected bankers of regulatory state the bureaucracy a key player and the execution of a large amount of Fiscal policy is by and large chosen by elected representatives (governments and legislatures): bureaucrats are involved in important aspects of auditing and implementation, they do not choose tax rates or the amount of spending their department.


Bureaucrats or Politicians? Part I: A Single Policy Task
By ALBERTO ALESINA AND GUIDO TABELLINI* This paper investigates the normative criteria that guide the allocation of a policy task to an elected politician versus an independent bureaucrat. The bureaucrat is preferable for technical tasks for which ability is more important than effort, or if there is great uncertainty about whether the policymaker has the required abilities. The optimal allocation of redistributive tasks is ambiguous, and depends on how the bureaucrat can be instructed. But irrespective of the normative conclusion, the politician prefers not to delegate redistributive policies. (JEL D72,D73,D82) Policies are chosen and implemented by both elected representatives (politicians) and nonelected bureaucrats. The view that politicians choose policies and bureaucrats implement them is too simplistic; the boundaries between decision and execution are a grey area and in many cases bureaucrats do much more than executing either de jure or de facto. For instance, in most countries, nonelected central bankers conduct monetary policy, with much independence. Regulatory policies are normally the result of both political and bureaucratic intervention, but the rise of the regulatory state has made the bureaucracy a key player in both the decisions and the execution of a large amount of legislation. Fiscal policy is by and large chosen by elected representatives (governments and legislatures): bureaucrats are involved in important aspects of auditing and implementation, but they do not choose tax rates or the amount of spending for their department.
Is this division of tasks appropriate? More generally, what criteria should guide the allocation of responsibilities among politicians and bureaucrats? We explore this question from a normative perspective by asking what is the socially optimal allocation of tasks between these two types of policymakers.
Economists have emphasized one specific argument in favor of delegation of policy to a nonelected bureaucrat: time inconsistency in monetary policy. Kenneth S. Rogoff (1985) pointed out that an independent and inflationaverse central banker not subject to ex post democratic control would improve social welfare. But there is more to it. For instance, fiscal policy is also marred with a host of time inconsistency problems, but societies seem reluctant to allocate this policy prerogative to independent bureaucrats. 1 An interesting question is why this never happens.
We focus the analysis on the individuals at the top (party leaders or high-level bureaucrats such as central bank governors). Our premise is that the main difference between top-level politicians and top-level bureaucrats lies in how they are held accountable. Politicians are held accountable, by voters, at election time. Top-level bureaucrats are accountable to their professional peers or to the public at large, for how they have fulfilled the goals of their organization. These different accountability mechanisms induce different incentives. Politicians are motivated by the goal of pleasing voters, and hence winning elections. Top bureaucrats are motivated by "career concerns," that is, they want to fulfill the goals of their organization because this improves their external professional prospects in the public or private sector. 2 Armed with this premise, we analyze a model of task allocation in which a social planner exploits the different incentives of bureaucrats and politicians and assigns tasks to maximize social welfare.
In this paper, we analyze a policy environment with a single task. From a normative perspective, bureaucrats are preferable to politicians in technical tasks for which ability is more important than effort, or if there is large uncertainty about whether the policymaker possesses the required abilities to fulfill his task. For purely redistributive tasks (splitting the cake), behind a veil of ignorance voters generally prefer to delegate to a bureaucrat if he can be instructed to be "fair"; but elected politicians have an incentive to retain redistributive tasks under their direct control to build winning coalitions. In a companion paper, Alesina and Tabellini (2006), we generalize the model to multiple tasks and derive additional results: politicians are preferable to bureaucrats if flexibility is valuable and time inconsistency is unlikely to be a relevant issue; if policy complementarities and compensation of losers is important; and if vested interests do not have large stakes in the policy outcome.
A recent principal-agent literature addresses related issues in career-concerns models. Mathias Dewatripont, Ian Jewitt, and Jean Tirole (1999a, b) discuss the foundations of this approach and apply it to study the behavior of government agencies. They focus on some issues related to ours, namely the nature and "fuzziness" of the agencies' mission, but they do not contrast bureaucratic and political accountability. Eric Maskin and Tirole (2001) investigate the attribution of responsibilities between accountable and nonaccountable agents. The latter have intrinsic motivations, while the former seek to please their principals because of implicit rewards (career concerns). In our setup, instead, we neglect the role of intrinsic motivations. Both bureaucrats and politicians need to be kept accountable with implicit incentives, but the implicit incentive schemes can be of two kinds: those that define a politician (striving for reelection), and those that define a bureaucrat (career concerns). Christian Schultz (2003) contrasts direct democracy, representative democracy, and bureaucratic delegation. Like Maskin and Tirole (2001), he views bureaucrats as unaccountable and focuses on the trade-off between ideological polarization and accountability: bureaucrats are less polarized than partisan politicians, but are more inflexible since they are unaccountable and cannot be removed after shocks to the voters' policy preferences. Timothy Besley and Maitreesh Ghatak (2005) also study intrinsically motivated agents, and focus on how to combine intrinsic motivation with implicit rewards. Besley and Stephen Coate (2003) contrast appointed and elected regulators of public utilities; both policymakers' types are intrinsically motivated, but direct election allows the voters to unbundle policy issues.
The paper is organized as follows. Section I describes the basic model and justifies its assumptions. Section II analyses imperfect monitoring and discusses how the difficulty of tasks induces a preference for bureaucrats. Section III deals with redistribution. The last section concludes.

I. The Model
Consider a society that has to decide whether to assign a policy task to an elected officer or to a bureaucrat. With the generic term "policymaker," we indicate who chooses policy, either a politician or a bureaucrat. In the simplest case, we consider a single policy, the result of which is determined by the effort put in by the policymaker and by his ability. Thus, the policy outcome y is where a represents the effort of the policymaker and ϳ N( , 2 ) is his random ability. Ability and effort are additive. 3 Citizens care about the policy outcome according to a well behaved utility function, u ϭ U( y) ϭ y. We use linear utility, but in the last section, where it makes a difference, we add some discussion of strictly concave utility functions.
Effort is costly, and the strictly convex and increasing cost is labelled c ϭ C(a). The reward for the policymaker is labelled R(a) and it differs depending on whether the policymaker is a politician or a bureaucrat. Both of them maximize their utility defined as with C a Ͼ 0, C aa Ͼ 0 and R(a) to be defined below (subscripts denote partial derivatives). 4 The timing of events is as follows. At the "constitutional table," society chooses who has control rights over policy, the bureaucrat or the politician. Next, the policymaker chooses effort, a, before knowing his ability, . Finally, nature chooses , outcomes are observed, and the reward is paid. Irrespective of who has control rights over policy, only the outcome y is observed by the principals, not its composition between effort and ability. Hence the agent's reward can be based only on the policy outcome, y.
In this simple environment, an optimal contract with the policymaker based on performance would achieve the first-best level of effort (see Appendix). But the assumption that policy performance is verifiable and contractible is hard to swallow. Public policy typically pursues many goals that are often hard to measure and to reward directly through explicit and verifiable contracts. Moreover, if society could write unrestricted optimal performance contracts with its policymakers, then the question asked in this paper would be utterly uninteresting: bureaucratic delegation under an optimal contract would always dominate political delegation. But this implication does not even come close to any observed institutional arrangement.
We thus assume that policy performance, y, is observable but not contractible. Both bureaucrats and politicians are rewarded based on observed performance, but through an implicit reward scheme that contains specific restrictions rather than an optimal explicit contract. In the next two subsections we spell out our specific assumptions about the implicit rewards offered to a bureaucrat and to a politician, which give rise to two different reward functions, R B (a) and R P (a), respectively. These reward functions are taken as given throughout the analysis. Our normative question is which reward function is more appropriate, given the nature of the policy task.

A. The Bureaucrat
We posit that the bureaucrat is motivated by "career concerns." That is, he is concerned with the perception of his ability in the eyes of those who may offer him alternative job opportunities in the private or public sector, given the stated goals of the bureaucratic organization. This assumption is especially appropriate for high-level bureaucrats who have already been promoted to the top of the bureaucracy, say a central bank governor or the chairman of a regulatory agency. 5 More precisely, let x be the relevant measure of performance with which the bureaucrat is evaluated (the stated goals of his organization). We assume that the bureaucrat's reward is (the suffix B stands for bureaucrat): where ␣ is the market value of talent, E denotes unconditional expectations over the random variable x, and E denotes expectations over , conditional on the realization of x. Equation (3) contains several implicit assumptions. First, the bureaucrat cares about his talent as perceived by outside observers representing his relevant "labor market." Second, the expectation of talent is formed by conditioning on the bureaucrat's observed performance. Third, the relevant measure of performance, x, must be defined in advance. Fourth, the market value of talent is a given parameter, ␣, possibly different from one.
In the context of this simple model, it is natural to assume that the relevant measure of performance for the bureaucrat coincides with social welfare, so that x ϵ y (this assumption will be revisited in the last section that deals with redistribution). Denoting the public's perception of a by a e and using (1), we can rewrite the bureaucrat's reward function (3) as This allows us to easily compute the equilibrium level of effort. First, take the first-order condition with respect to actual effort, a, taking expected effort a e as given. Then, impose the equilibrium requirement that a e ϭ a. By (4) and (2), we obtain where a B indicates the equilibrium effort of the bureaucrat. How does equilibrium effort by the bureaucrat differ from that induced by an optimal contract? Comparing (5) with (A1) in the Appendix, we see that the bureaucrat puts in the first-best level of effort if ␣ ϭ 1, i.e., if the market value of bureaucratic talent coincides with the true value of talent for society. 6 But if the value of talent for the bureaucrat differs from that for society, and in particular if it is lower, then bureaucratic behavior is no longer socially optimal.

B. The Politician
The politicians's goal is to be reelected, which happens if the voters' utility exceeds a threshold W. Denoting by ␤ the value of office, we can write the reward function for the politician as (the suffix P stands for politician): where u ϭ y is voters' utility and P(W Ϫ a) ϭ Pr( Յ W Ϫ a). Voters are rational. Thus, they realize that the alternative to reelecting the incumbent is to get another politician with average talent, who will exert the equilibrium level of effort. It follows that Like the bureaucrat, the politician chooses effort before observing his talent, taking the voters' expectations as given. With a normal distribution for , equilibrium effort by the politician, a P , is defined implicitly by the first-order condition where n( ) ϭ 1/ ͌ 2 is the density of the normal distribution of evaluated at its mean. 7 How does the effort of the politician compare with that of the bureaucrat? Comparing (5) and (8) we find the answer is ambiguous and depends on parameters' values. A higher value of office, ␤, increases the effort of the politician; a higher market value for bureaucratic talent, ␣, increases the effort of the bureaucrat. Under the 7 This model could easily be generalized to several periods, if the politician's ability today is a signal of his ability tomorrow, but some random element of ability is present every period so that it can never be fully learned in advance. A widely studied case in the political business cycle literature is that of an MA (1) process for ability. Torsten Persson and Tabellini (2000) discuss the implications of this political model more extensively.
A more general formulation, outlined in the Appendix, would have the politician care about both reelection and, conditional on losing office, career prospects outside politics. If the value of political office is sufficiently high compared to the expected benefit of a career outside politics, then the main implication of our model would still hold. assumption that the participation constraint is always satisfied, in this simple example voters prefer whatever arrangement results in higher effort. To simplify notation, and since no additional result hinges on the value of these two parameters, in the remainder of the paper we set ␣ ϭ ␤ ϭ 1. 8

C. Discussion
The model seeks to capture a key difference between political and bureaucratic accountability. The politician is held accountable by the voters who choose whether or not to reelect him, based on their utility. The bureaucrat is held accountable by his professional peers or by the public at large, for how he fulfills the goals of his organization. These different accountability mechanisms imply a different objective function: the politician strives to achieve a threshold level of utility for the voters; the bureaucrat wants to maximize his perceived talent. Hence, the key behavioral difference between the two types of policymaker is that one maximizes an expected value, and the other maximizes a probability-both defined over the same random variable. In Alesina and Tabellini (2006) we analyze situations of multiple tasks, where politicians and bureaucrats differ also in a second dimension: for the politician, the relevant measure of performance is voters' utility; for the bureaucrat, it is whatever goals have been assigned to the bureaucratic organization.
While the assumption that politicians maximize the probability of victory at the election is now common, there is not a standard model of bureaucratic behavior. Thus, although we are not the first to use it (see in particular Dewatripont, Jewitt, and Tirole 1999a, b), our "career concerns" model of a bureaucrat needs some discussion. Consider the assumption that the bureaucrat cares about his talent as perceived by outside observers. While we have justified this assumption with reference to monetary rewards in future jobs, it can be interpreted more broadly.
Top bureaucrats may care about their perception of talent "per se," as a matter of self-image, pride, or legacy. Alan Greenspan retired after he resigned as chairman of the Federal Reserve, but he certainly cares about the perception of his ability in managing monetary policy.
How do these straw men-"politician" and "bureaucrat"-relate to real world cases? Probably the most compelling example of our "bureaucrat" is a central banker. His incentives to fulfill his task are driven primarily by the desire to appear competent, although even a central banker occasionally may bend to the electoral needs of a politician. Like our bureaucrat, a central banker sets policy without political interferences and his tasks are determined by a clear mandate to keep inflation low. An American president is, instead, the quintessential example of a politician: he seeks reelection for himself in his first term and for his party in his second, and is not constrained by preassigned or narrowly defined tasks.
Top-level bureaucrats in charge of important agencies may be preparing a leap into politics, so they may worry about their popularity, and not only their competence per se. On the contrary, politicians may look ahead to a career in the private sector. While these caveats point to a large gray area and intermediate cases between our politician and our bureaucrat, it is useful as a first step to identify clearly how career concerns and electoral incentives lead to different results depending on the nature of the policy (but see also footnote 7 above).

II. Imperfect Monitoring
We now move to the case of imperfect monitoring, that is, a situation in which talent is not perfectly observable. Thus, we add noise, , besides talent () and effort (a): with ϳ N(0, 2 ), uncorrelated with and unobservable. Only performance y is observed and can be the basis of rewards.
In this case, the reward for bureaucrats can be rewritten as (10) R B ͑a͒ ϭ E͑E͉͑y͒͒ ϭ ϩ E͑ ϩ ϩ a Ϫ a e Ϫ ͒, 8 Since we are not considering an optimal contract, both the bureaucrat and the politician could be earning rents in equilibrium (i.e., their participation constraint need not bind).
where ϭ 2 /( 2 ϩ 2 ) Ͻ 1. Given our assumption of normality of the distributions, we obtain a well-known signal extraction result. Now the perception of talent is "discounted" by a term which reflects the signal-to-noise ratio. In equilibrium the choice of the bureaucrat is given by: Not surprisingly, the bureaucrat puts in less effort the lower is the signal-to-noise ratio. 9 Next, we turn to political delegation. The politician's reward is given by the same expression as above, except that now the distribution from which the probability Pr( y Ն W) can be computed has a larger variance, which reflects the variance of both and of . It is immediate to derive the first-order condition of the politician as follows: where n( , 0) ϭ 1/( ͌ 2 ϩ 2 ͌ 2) is the density of the random variable ϩ , evaluated at the mean of both and .
We are now ready to establish the following: PROPOSITION 1: The comparison between a P and a B is ambiguous. Imperfect monitoring (high 2 ) reduces effort for both types of policymakers. Higher 2 increases a B but decreases a P . Therefore, less monitoring does not favor one or the other type of policymakers. This result is related to those obtained by Dewatripont, Jewitt, and Tirole (1999b), who also point out that performance less closely tied to talent or effort weakens the incentives of agents motivated by career concerns. But note that the same conclusions also apply to a politician. Hence, imperfect monitoring reduces the performance of both policymaker types (relative to an optimal contract), but it does not provide an argument for preferring a politician to a bureaucrat at the constitutional stage.
More uncertainty about talent, however, does favor the bureaucrat over the politician. With imperfect monitoring, a larger variance of increases the effort of the bureaucrat, while it has the opposite effect on the politician. Intuitively, an increase in the variance of increases the signal-to-noise ratio and implies that observed performance ( y) is a better indicator of ability (). This makes the bureaucrat work harder, since by assumption he fully internalizes the benefit of higher expected ability. 10 The politician, instead, wants only to overcome the reelection threshold (giving the voters more than their reservation utility is a waste). If ability is more uncertain (if 2 is high), then reelection prospects are less sensitive to effort, since more of the policy outcome is due to randomness. Hence, his incentives are weakened.
This result has a practical and sensible implication: bureaucrats are better than politicians in tasks where the dispersion of possible levels of ability is wide, that is, when there is more uncertainty over the policymaker's ability. The reason is not that bureaucrats are more gifted on average, but rather that they have stronger incentives to pretend that they are gifted. Very simple tasks are unlikely to be associated with talent uncertainty: anybody can do them. When tasks become more difficult, the variance in the level of ability is likely to go up, and 9 Note that, with imperfect monitoring, the career concern contract no longer induces the optimal amount of effort even when there is no difference between the value of ability for the bureaucrat and for society. Given risk neutrality, the optimal contract (under the assumption that the principal observes only y, and ability is evaluated equivalently by society and the bureaucrat) would still induce the same amount of effort as in (5) above (see also Section A of the Appendix). That is, imperfect monitoring would not add any distortions. But if the bureaucrat can be rewarded only implicitly through career concerns, as we assume, then imperfect monitoring entails an additional loss of welfare for the voters. 10 Here the bureaucrat is risk neutral, which means that his compensation is a linear function of expected ability (conditional on performance). A risk-averse bureaucrat would put in even more effort with more uncertainty over , if his marginal utility were convex (e.g., with iso-elastic utility function, as in the literature on precautionary savings). This would further increase his attractiveness relative to the politician. But the opposite would be true if the bureaucrat's marginal utility were concave. (In this case, more uncertainty over could weaken the bureaucrat incentives, if the effect on marginal utility outweighs the effect on the signal-to-noise ratio.) bureaucrats are preferable to politicians. One reason why more difficult tasks are likely to be associated with a higher variance of ability is that often difficult tasks are multidimensional. If each dimension is associated with a different ability, then overall variance increases as the number of dimensions goes up. 11 The implication that bureaucratic rather than political accountability works better for complex tasks is strengthened if evaluating the performance of a bureaucrat also requires special technical abilities-that is, if the extent of imperfect monitoring also depends on who does the monitoring. In the case of politicians, the ultimate judges of performance are the voters at large. The performance of bureaucrats, instead, is evaluated primarily by their professional peers. Hence, imperfect monitoring is less of a problem if politicians are given simple tasks, since bureaucrats can more easily be held accountable by their peers for more technically demanding tasks. Maskin and Tirole (2001) and David Epstein and Sharyn O'Halloran (1999) reach a similar conclusion in different models.
Is the real world attribution of task broadly consistent with this implication? If difficult tasks are also technically more demanding, then the answer is clearly positive. In many cases, technical tasks are delegated to bureaucrats, for instance, managing the financial structure of public debt, or regulating public utilities or other industries, while politicians retain the technically less demanding task of setting general targets. In the United Kingdom, for instance, politicians choose a target level of inflation; the technically demanding task of choosing interest rates to achieve such target is delegated to the Bank of England. It is not always true, however, that difficult tasks are technically more demanding. Some complex policy decisions, such as in foreign policy, require ability of a general rather than a specialized kind. According to Proposition 1, if there is large uncertainty about the policymaker's ability, these complex and yet technically undemanding tasks are also better left in the hands of bureaucrats. But here, we often observe a politician in charge. Alesina and Tabellini (2006) discuss additional reasons, related to contract incompleteness, why politicians may perform better in such complex policy environments.

III. Splitting the Cake
We now return to the model without imperfect monitoring and consider a purely redistributive policy, "cake splitting." Consider three voters, the minimum number required to make the problem interesting. The policy task delivers a "cake" that can be divided among the three voters, therefore: The utility function of the voters is linear, U(c J ) ϭ c J , J ϭ 1, 2, 3 as before. We comment below on how the results would change with risk-averse voters.
The key difference between a politician and a bureaucrat is that the former needs a majority to win and the latter simply wants to signal talent. Consider the bureaucrat first. At the constitutional stage, the bureaucrat can be given no redistributive tasks, in which case redistribution is entirely arbitrary-we call this an "unfair" bureaucrat. Alternatively, behind a veil of ignorance, he can be assigned the task of redistributing equally, that is y/3 for all three voters-we refer to this case as a "fair" bureaucrat. But irrespective of whether he is fair or unfair (i.e., of how he splits the cake), his talent is still judged by the aggregate measure of performance, x ϵ y, not by how he redistributes. His first-order conditions are thus identical to those in (5) in Section I.
Next, consider the politician. Since he needs to please only a majority, he gives y/2 to two voters and zero to the third one. Hence, his reward is: (13) R P ͑a͒ ϭ Pr͑y/2 Ն W͒, 11 We are grateful to a referee for this remark. He also pointed out that bureaucrats work harder than politicians if performance is more sensitive to ability than to effort. Rewriting (9) as y ϭ K ϩ a ϩ , where K is a parameter that captures the relative importance of ability, we obtain that a higher K increases a B but reduces a P . To the extent that ability (rather than effort) is needed in complex tasks, this reinforces our conclusion. But many complex policy decisions, such as in foreign policy, require ability of a general rather than a specialized kind.
where W is the reservation utility of individual voters. Implicit in (13) is the assumption that voters expect that the incumbent, if reelected, will maintain the same redistribution observed today, i.e., he will split the cake in half between the voters who reelect him. With forward-looking and rational voters, W equals the average expected utility they can get if the opponent is elected. If the hypothetical redistribution implemented by the opponent is unknown, then W ϭ ( ϩ a e )/3. Going through the usual steps of maximizing with respect to effort for given expectations and then imposing rational expectations, in equilibrium the politician's optimality condition implies where n(z) denotes the normal density evaluated at point z. Comparing (14) with (8) in Section I, we see that once the politician is also in charge of redistribution, he can get away with less equilibrium effort. The reason is that here he needs only please two voters out of three. He can thus reduce effort and still please two voters with the portion of the cake taken away from the minority. 12 Note the asymmetry: voters expect the incumbent to preserve the observed redistribution over time, but they are uncertain about how the opponent would redistribute. This asymmetry creates an incumbency advantage: the voters are more willing to reappoint the incumbent even if he is incompetent, because they benefit from his redistribution. Indeed, in equilibrium the probability that the incumbent is reappointed is Pr[ Ն (2 Ϫ a P /3)] Ͼ 1/2. Since the density n(z) is lower at the point z ϭ (2 Ϫ a P )/3 than at the point z ϭ , the incumbency advantage also reduces equilibrium effort. 13 Here we as-sumed a very stark asymmetry: no uncertainty at all about how the incumbent will redistribute, and maximal uncertainty about the opponent. But the nature of the results would be preserved with less stark assumptions, as long as voters are more uncertain about the redistributive policies of the opponent compared to those of the incumbent.
The assumption that the opponent's future redistributive policies are more uncertain than those of the incumbent can be derived from more primitive assumptions. For instance, suppose that politicians have lexicographic preferences. First they care about reelection, as spelled out above; second, conditional on being reelected, they also care about the welfare of specific groups of voters. Suppose further that voters ignore these redistributive preferences. Then, the incumbent's redistributive policies reveal his preferences, and voters correctly expect these policies to be continued if he is reelected. As they cannot observe what the opponent would do, voters face more uncertainty if voting for the opponent. This simple example also points to the fact that it is in the interest of politicians to pretend that they are ideologically biased in favor of specific groups or policies, even if in reality they are purely opportunistic. The ideology of politicians is like their brand name: it keeps voters attached to parties and reduces uncertainty about how politicians would act once in office. 14 Given these results, who is better for the voters behind the constitutional veil of ignorance-the bureaucrat or the politician? If voters are risk neutral, and given that they ignore the redistribution chosen by the politician, they care only about aggregate performance, y. This makes the bureaucrat more attractive for the voters for a larger range of parameter values, compared to the case of simple nonredistributive tasks in Section I. With risk-averse voters, the normative comparison between bureaucrat 12 This result is similar to that obtained in John Ferejohn (1986) and Persson and Tabellini (2000). But here, since voters are forward looking, we rule out the Bertrand competition among voters that instead features in the backwardlooking voting equilibrium of Ferejohn (1986). 13 Indeed, if the voters were certain to be included in the winning coalition by the opponent, their reservation utility would be W ϭ ( ϩ a e )/2. In this case, the probability that the incumbent is reappointed is 1/2 and his equilibrium effort would coincide with (8); thus, there would be no dilution of effort due to redistribution. 14 Allan Drazen and Marcela Eslava (2004) analyze a model of electoral policy cycles where voters infer the redistributive preferences of the incumbent from the policies he enacts. and politician also depends on whether the bureaucrat is "fair" or "unfair." A fair bureaucrat is even more attractive compared to the politician, not only because he is likely to exert more effort, but also because he is less risky-the politician exposes the voters to the risk of being in the minority. 15 But the result may be reversed if the bureaucrat is unfair and implements a totally arbitrary redistribution. In this case, political redistribution is less risky, since two voters out of three are always included in the winning majority.
The desirability of a bureaucrat, thus, ultimately depends on whether he can be instructed and trusted to be fair. In a complex world it may be difficult to precisely assign redistributive tasks to a bureaucrat. Yet, in the few cases in which a bureaucratic organization is observed to be in charge of redistribution, it appears to work well. In India, an independent nonpolitical federal agency (the Finance Commission) is in charge of distributing revenue across states according to prespecified criteria such as relative poverty or need. A similar task is also performed by a second Indian commission made up of politicians and members of government (the Planning Commission). Stuti Khemani (2005) compares the performance of both commissions and finds that the distribution of transfers by the bureaucratic commission is more consistent with ex ante equity objectives compared to that of the political commission.
We summarize this discussion in the following: PROPOSITION 2: The possibility of redistribution creates an incumbency advantage and reduces the equilibrium effort of the politician; the effort of the bureaucrat is not affected. With risk neutrality and fair bureaucrats, the latter are always strictly preferred ex ante. Risk aversion makes the bureaucrat more or less desirable ex ante depending on how easy it is to impose fair treatment of all voters in his task description.
This result can also explain why we almost never observe bureaucrats in charge of redistributive tasks. Even if voters prefer a fair bureaucrat to a politician, the latter has a strong preference to retain redistributive tasks under his direct control. As shown above, redistribution enables a politician to build winning coalitions of voters, increasing his incumbency advantage and reducing equilibrium effort. Thus, if the constitution is designed by politicians, rather than chosen by the voters behind a veil of ignorance, then the politician would never choose to delegate redistributive tasks to an independent bureaucrat. Alesina and Tabellini (2006) further discuss this positive question of when and how politicians choose to delegate to independent bureaucrats.

IV. Conclusions
Our analysis rests on a fundamental assumption. Bureaucrats want to signal their competence for career concerns, politicians for reelection purposes. This implies that bureaucrats maximize the expected value of their perceived ability, and politicians want to make sure that their perceived ability overcomes the minimum threshold needed for reelection. From a normative perspective, this difference implies that some policy tasks, but not others, ought to be delegated to independent agencies. Politicians are preferable if ability is less important than effort or if there is little uncertainty about whether the policymaker has the required abilities; bureaucrats are preferable in the opposite case. This result is consistent with the observation that highly technical tasks (monetary policy, regulatory policies, public debt management) are typically delegated to high-level bureaucrats.
In the case of redistributive policies, voters prefer a bureaucrat, if "fair" redistributive goals can be clearly specified ex ante and the bureaucrat can be trusted to implement them. If, instead, redistribution implemented by a bureaucrat is arbitrary or unpredictable, then risk-averse voters prefer a politician. But irrespective of voters' preferences, a politician always prefers not to delegate redistributive tasks, because coalition building increases his incumbency advantage and reduces equilibrium effort. This might explain why delegation to independent bureau-crats is very seldom observed in fiscal policy, even if many fiscal policy decisions are technically very demanding.
This paper focused on a single policy task, but in reality policymakers often have multiple related tasks that create policy trade-offs. In that case, optimal task allocation depends on how delegation can be designed, and in particular on whether bureaucrats can be assigned state contingent tasks. Alesina and Tabellini (2006) study task allocation between a politician and a bureaucrat with multiple tasks, both from a normative and a positive perspective (the positive case corresponds to a politician facing reelection who chooses what to delegate to a careerconcerned bureaucrat).

A. The Optimal Contract
Consider the simple model of Section I. If effort a is verifiable and contractible, then the optimal contract induces the first-best level of effort, a*, defined implicitly by (A1) 1 ϭ C a ͑a*͒.
Next, suppose that effort is unobservable, but performance y is verifiable and contractible. Given risk neutrality of principal and agent, the first-best can still be achieved by an optimal explicit contract rewarding the agent with a simple linear payoff based on performance: where the constant w is defined by the agent's (ex ante) participation constraint, namely by the condition that Under the optimal performance contract, the participation constraint must bind, and given (1) and (A3), this implies w ϭ ϩ a* Ϫ C(a*).

B. More General Objective Function for the Politician
As mentioned in Section I, the politician's objective function could be written more gen-erally by assuming that he cares about both reelection and, conditional on losing office, his career prospects outside politics. In this case, his reward function could be written as where, as before, P(.) is the probability of losing the election. The first-order condition for effort evaluated at the equilibrium is (B2) n͑ ͒͑␤ Ϫ ␣ ͒ ϩ 1 2 ␣ ϭ C a ͑a P ͒. If the value of political office is sufficiently high compared to the expected benefit of a career outside politics (if ␤ is sufficiently higher than ␣ ), the main implication of our model would still hold.