Legislative budget cycles

Saporiti and Streb (2008) do study a bargaining process between the legislature and the executive, but conceive of the legislature as a unitary actor. There is also empirical work documenting a relationship between political polarization and the magnitude of political budget cycles. But to our knowledge, there are no prior models of the electoral budget cycle explicitly focusing on the game among legislators.

Studies of the election benefits derived from patronage are often clouded by endogeneity bias and unobserved heterogeneity among politicians which rarely are addressed. However, those few studies that do address these issues find that patronage does deliver electoral benefits. Levitt and Snyder (1997), which finds effects in US congressional races, is the seminal paper. More recently, Leitschig and Morrison (2010) have shown similarly positive effects in Brazil.

We are not arguing that clientelism flourishes only in a weak party system. Machine politics—client politics organized by strong parties—is a clear counter-example to such an argument. See Kitschelt (2000) for a useful summary of the role of the party and the variety of links between legislators and distracted voters. Instead, we simply argue that weak parties contribute to political budget cycles via the encouragement of clientelism. To the extent that machine politics is prevalent in our sample, this would tend to bias against our finding results (see Sect. 4) since such polities would have both strong parties and strong budget cycles. It is also important to recognize that clientelism is not simply a function of electoral rules. Kitschelt (2000: 859) puts it nicely: “Contrary to the naïve presumption that single-member districts (SMDs) offer the greatest chances for clientelist linkage building, the later pertains in multimember districts (MMDs), provided that the ballot offers personal preference votes, no vote pooling among candidates on the same list, and no party control over the nominations process.”

One could also think of legislators as characterized by their influence within the legislature, which determines the fraction of total spending captured by the district. In such a case it is easy to show that parties, by regularizing influence and making it observable, lead to a damping of the political budget cycle. In the absence of strong parties, the influence of a particular legislator may fluctuate up and down as favors are banked or called in and salient issues and relevant expertise change. But strong parties, which exercise an oligopoly over access to power, typically institute a seniority system whereby influence grows with time of service. As a result, influence is no longer a moving average but a simple and observable function of incumbency. As influence is no longer hidden, there is no longer a need to signal influence spending excessively in election years and thus strong parties dampen the political budget cycle. On closer reading of the literature discussing party institutionalization and quality of governance, we feel that the version presented in this paper better captures the role of party institutionalization; we have thus omitted the formal version of this model of influence.

There is evidence that parties target spending toward swing districts: Dahlberg and Johansson (2002) find evidence for Sweden and Arulampalam et al. (2009) find evidence for India. Fishback et al. (2003) find that several New Deal programs were strongly directed to swing voters as well as being used for other purposes.

Persson and Tabellini (2003b) find evidence that electoral budget cycles are stronger in majoritarian countries. Notice that this does not contradict our hypothesis which concerns the ability of a strong party system to dampen whatever budget cycle already exists. Indeed, we simultaneously confirm Persson and Tabelini’s results in our sample while finding support for our theory (see columns [9] and [10], Table 4).

We discuss multi-member districts and proportional representation in Sect. 2.3.

Recall that this is the baseline model without parties. The party system is developed in the following section.

Note that this means that while the PM has power during coalition building, she has no more influence than any other government member when policies are set.

In the alternate (not included in this version) influence model to which we alluded in footnote 6, the government maximizes a weighted joint utility where the weights depend on the members’ influence.

To streamline the sufficiently complex expressions that will follow, we assume a zero interest rate.

This is a critical assumption but not, we feel, a controversial one. Competence is conceived as the ability to identify projects that will be valued by constituents. This is clearly a skill that requires local knowledge, probably of a nature that is too complex to be verifiably communicated to fellow representatives from separate districts.

Elections can be made less frequent without altering the essential character of the results.

This is another critical assumption that is standard in the literature and, we feel, defensible. Voters are much more likely to seek and receive timely and complete information about projects in their own district from both direct information from their own experience and indirect information via the local media and their own personal social networks.

In his discussion of the origin of American political parties, Aldrich (1995) assumes that the conceptual alternative to parties is, rather than a randomly chosen minimal winning coalition, a universalist log-rolling such that all members receive their pork projects. Such an alternate baseline would imply both that parties reduce the size of government and magnify parties’ reduction of the political budget cycle.

Technical Appendices A, B, C, and D and the data description Appendix E are available online, doi:10.​1007/​s11127-012-0043-9.

The district subscript k has been suppressed because districts in government are symmetric and receive the same level of spending. In all cases, the government sets spending equal to zero in all districts whose representatives are not in government. The government also treats all districts in government symmetrically. Thus we could drop the district subscript, k. Governments cannot condition spending on the quality of the districts’ incumbent because this is unverifiable and perhaps unknown even to the representative himself.

Why then do low type legislators accept offers to be in government if election-year signaling will result in their ouster? Recall that a legislator cannot predict his future type and is thus ignorant during government formation. Accepting an offer is thus accepting a lottery that with chance h one is assured re-election and chance 1−h one is assured non-election. The benefit of accepting is the district-specific spending. The alternative lottery—the chance that an out-party legislator is reelected—is not specified by the model but it is trivial to show that there is a wide range of parameters under which it is optimal to accept an offer to be in government.

Our results do not depend on this assumption of a uniform distribution. However, we believe that it simplifies what is already a complex explanation.

Stimson et al. (1995) estimate the speed with which the US House and Senate respond to changes in the opinion of the electorate. They calculate a coupling constant, defined as the time over which two-thirds of the total dynamic response is felt, of 1.3 years for the House and 1.2 years for the Senate. They find that the response is roughly one-third electoral turnover and roughly two-thirds changes in platform. While limited to a single country, we feel that this study supports our assumption that platforms do respond to policy, but with a lag of one policy period.

We do not explicitly model the choice of whether the party should cheat or not, simply assuming that voters punish parties who renege on their platforms and that parties are long-lived actors who value their credibility.

Recall that the transformed policy space is the one which is, following the nationwide preference shock, renormalized such that the national median is located at 0 and thus the district ideal points are distributed uniformly across [−0.5,0.5].

Alternately, we could motivate this by arguing that legislators whose ideology is closer to that of the Prime Minister are willing to pay higher rents to be part of the government.

Notice that a PM who is more extreme than the 25^th or 75^th percentile of all representatives cannot achieve her preferred policy because there are insufficient extremists to make her ideal point the median in the government. The best she can do is achieve the 25^th or 75^th percentile. Nonetheless, she still has a choice of coalitions: in a 101-member legislature numbered from 0 to 100 according to ideology, the median of both {0–50} and {0–25,51–75} is the 25^th legislator. It is true that these extreme coalitions require the participation of all of the most extreme legislators, which could conceivably give them bargaining power in negotiations over formation. But the effect of any one legislator’s rejection would be simply to move the achievable coalition one legislator to the center; a modest change. By ruling out this behavior, we assume that the pork-barrel benefits of being in government outweigh this modest effect on policy. In essence, individual legislators have too little bargaining power to shift the equilibrium policy enough to dissuade them from accepting the transfers that come with being in government.

The results can be generalized to multiple parties assuming that parties have well-defined relative policy positions.

With party positioning and signaling, we are assuming and solving a static rather than a dynamic problem. Since the quality of members does not persist between elections, the only state variable that persists between elections is the number of the members of each party. We do take into account the identity of the incumbent party—the party that can signal with extra spending—when calculating optimal party platforms (see the Appendix B). As the results show, party platforms are such that the incumbent party is no more likely to retain power than the out-party is to seize it. With no continuation value to being in government, the dynamic problem reduces to a static one.

A district would also prefer to elect a representative who will be part of the government (thus earning pork for the district). But as shown in Appendix C, parties choose their platforms such that the ex-ante probability of each party winning the majority of seats is equal.

It has been suggested that we include the effective number of parties in our specification. We do not believe this to be advisable. A system can have institutionalized parties and stable vote shares with either few or many parties. On the other hand, weak party institutionalization can, by encouraging new parties, lead to a larger effective number of parties. Thus our measure of vote-share volatility already captures that component of the effective number of parties that is relevant to party institutionalization.

While it is easier to find seat-share data than vote-share data, vote-share volatility is preferred to seat-share volatility because votes are a more direct measure of the connection of the electorate to the party. Minimum thresholds in the translation of votes to seats, the rules for which vary across countries, mean that the latter is an inconsistent proxy for the former.

We use the Epanechnikov kernel with STATA’s default optimal bandwidth.

We would have liked to use a measure of targetable expenditures rather than the budget deficit. Unfortunately, we were unable to find a more specific measure that afforded sufficient coverage. There is also the attendant difficulty in identifying what expenditures are targetable and whether such a classification is constant across polities.

Why do we find a strong association between electoral budget cycles and democratic institutions when a previous study by Brender and Drazen failed to find any such connection? We suspect that the additional controls for budget transparency and the strength of the party system have made the difference.