Rational partisan theory: empirical evidence for the United States.

• Author: Carlsen, Fredrik

1. Introduction

   Partisan theories of macroeconomic policy postulate that political parties promote policies that are consistent with the preferences of their core constituencies. Supporters of left-wing parties are particularly averse to unemployment because they hold a large part of their wealth as human capital, whereas supporters of right-wing parties are averse to inflation, which creates uncertainty about the return on financial and residential capital. Left-wing parties are therefore more inclined to stimulate aggregate demand than right-wing parties.

   A number of empirical studies have confirmed that left-wing governments tend to conduct more expansive monetary and fiscal policies.(1) Whether different demand policies translate into partisan effects on output depends on the economy's supply side. Alesina (1987) and Alesina and Sachs (1988) were the first to formulate the theory of partisan preferences within a rational expectations framework with forward-looking, optimizing agents.(2) Their rational partisan theory (RPT) features the idea that the economy exhibits postelection output cycles when nominal wage contracts are signed before elections because wage setters hedge against the possible election outcomes. Unless the election outcome is known with certainty, a left-wing party will succeed in raising output growth after an election victory, whereas the economy will experience a recession after a right-wing election victory. The amplitude of postelection output fluctuations is larger the more surprising is the election outcome.

   Most studies undertaken to test the RPT have not attempted to estimate the degree of uncertainty associated with each election. For instance, the multicountry study by Alesina and Roubini (1992) tests the RPT by including lags of an intervention variable that is + 1 or -1 in the N quarters (N = 4, 6, or 8), starting with that of change of government and otherwise zero in output growth and unemployment regressions. Paldam (1991) applies yearly data for a sample of OECD countries to compare the average output growth and unemployment rates during the first year(s) of the government term with the average growth and unemployment rates of the two years preceding the election. Alesina and Rosenthal (1995) estimate a growth equation for the United States on the basis of yearly data, including as regressor a dummy assuming the value of + 1 in the second year of a Republican administration, -1 in the second year of a Democratic administration, and zero otherwise.

   These and other studies have established that output growth indeed responds to partisan changes but do not examine whether partisan effects are caused by election surprises. An exception is Cohen (1993), who constructs a partisan intervention variable on the basis of U.S. presidential trial-heat polls. Cohen reports results favorable to the RPT; however, he does not check whether his intervention variable registers partisan effects because it is correlated with simple partisan intervention terms included in other studies.

   This paper employs estimates of the degree of uncertainty associated with U.S. presidential elections to test the RPT. Two series of election-win probabilities are computed. First, I run a set of regressions across the elections from 1956 to 1992, explaining the vote share of the incumbent party as a function of lags of Gallup presidential approval rates and GDP growth rates. Quarterly estimates of the incumbent party's election-win probability are computed from the fitted vote shares and estimated forecasting errors of the respective regressions. The second series is constructed from Gallup presidential trial-heat polls.

   Since the impact of demand policies on output growth depends crucially on the supply side, it is not obvious how to conduct a proper test of the RPT. This paper presents two approaches. First, I test the "strict" version of the RPT - that partisan effects are due to the intersection of sticky wages and electoral uncertainty - using data about actual wage contracts; this analysis extends earlier work by Hibbs, Carlsen, and Pedersen (1996). Next, I test a "flexible" version of the theory - that partisan effects are stronger after relatively surprising election outcomes - by interacting standard partisan dummy variables with indices characterizing election surprises.

   The original RPT was formulated in the context of a stylized supply-side model in which all wage contracts are signed an instant before the election and expire simultaneously. The theoretical part of the paper extends the supply-side model of the original RPT in ways that are more realistic for empirical analysis by introducing staggered and overlapping wage contracts. I consider both the case in which contracts that carry over an election may assign different wage growth rates to pre- and postelection periods and the case in which a contract assigns uniform wage growth rates to each period covered by the contract.

   Combining the two models and the two probability series gives four alternative partisan intervention variables. When entered in output growth regressions, I find that all four variables are dominated by a simple partisan intervention term, which is + 1 during the first half of Democratic administrations, -1 during the first half of Republican administrations, and zero otherwise.

   Although the strict version of the RPT is not supported by the data, we cannot conclude that election surprises have no effect on output growth as the partisan intervention terms are computed from theoretical models involving very specific supply-side assumptions. The last part of the paper presents a test of the RPT that is probably more robust to alternative supply-side assumptions.

   For each election, a set of surprise indices are computed; the indices measure the probability assessments held by agents prior to an election that the actual winner would lose. The indices are interacted with partisan dummy variables to examine whether partisan effects are stronger the more surprising is the election outcome.

   When included in output growth regressions, I find that the interaction term is always statistically insignificant. Thus, the two approaches produce the same conclusion: Partisan output growth differences are not caused by election surprises.

   The rest of this paper is organized as follows. Section 2 develops the theoretical RPT models. The probability series are computed in section 3. Output growth regressions using intervention terms derived from the theoretical models are reported in section 4. Section 5 presents tests involving surprise indices, and section 6 concludes.

2. Theory

   Consider an economy that consists of spot market wage contracts and staggered, overlapping long-term wage contracts. Spot contracts last one period. Long-term wage contracts of type i, i = 1, ..., I, last for [N.sub.i] periods, where 1 [less than] [N.sub.1] [less than] ... [less than] [N.sub.1].(3)

   At the beginning of each period t, some workers sign spot contracts specifying the spot market nominal wage growth rate, [Mathematical Expression Omitted]. Other workers re-sign long-term contracts that expire at the end of period t - 1. The rest of the workers are covered by long-term contracts signed in earlier periods that carry over period t. Long-term contracts of type i signed in period t, [Mathematical Expression Omitted], i = 1, ..., I, specify nominal wage growth rates for [N.sub.i] periods, denoted [Mathematical Expression Omitted].

   The average nominal wage growth rate in period t, [W.sub.t], is a weighted average of the spot market nominal wage growth rate and the time t nominal wage growth rates specified by the long-term contracts that cover period t:

   [Mathematical Expression Omitted], (1)

   where [Mathematical Expression Omitted] is the worker-weighted density of spot contracts and [Mathematical Expression Omitted] is the worker-weighted density of long-term contracts of type i signed in period [Tau].(4) Since all workers are on either spot or long-term contracts, we have that

   [Mathematical Expression Omitted]. (2)

   Following the traditional RPT approach, the demand side of the economy is not explicitly modeled; it is assumed that inflation is directly controlled by the government.(5) The government sets the period t inflation rate, [[Pi].sub.t], after period t spot market contracts have been signed and long-term contracts that expire at the end of period t - 1 have been re-signed.

   Let T denote an election period. The election is held after workers have signed period T wage contracts and the old administration has set [[Pi].sub.T]. The new administration assumes control of the inflation starting from period T + 1. Election terms last N periods; elections take place in T - N, T, T + N, and so on.

   Output growth, [Y.sub.t], is assumed to depend on the growth rate of average real wages:

   [Mathematical Expression Omitted], (3)

   where [Mathematical Expression Omitted] is the natural output growth rate.

   Two parties compete for office. Democrats (D) are relatively more concerned with output growth, whereas Republicans (R) emphasize fighting inflation. We write the parties' objective functions as

   -[([[Pi].sub.t]).sup.2]/2 + [[Beta].sup.D][Y.sub.t]

   and

   -[([[Pi].sub.t]).sup.2]/2 + [[Beta].sup.R][Y.sub.t],

   where [[Beta].sup.D] [greater than] [[Beta].sup.R].(6)

   Inserting for [Y.sub.t] from Equation 3 gives the two parties' preferred choices of inflation, [Mathematical Expression Omitted] and [Mathematical Expression Omitted].

   [Mathematical Expression Omitted],

   and

   [Mathematical Expression Omitted],

   where [[Pi].sup.D] [greater than] [[Pi].sup.R].

   Consistent with the original RPT model, I assume that workers attempt to keep their real wages constant. If the administration's preferences are known with certainty, workers set nominal wage growth equal to the actual inflation rate. This implies that neither spot nor long-term contracts that start and expire between...