Politics and the stock market: Evidence from Germany

European Journal of Political Economy 22 (2006) 925 – 943 www.elsevier.com/locate/ejpe

Politics and the stock market: Evidence from Germany Jo¨rg Do¨pke a,*, Christian Pierdzioch b a

Deutsche Bundesbank, Wilhelm-Epstein-Strasse 14, 60006 Frankfurt a.M, Germany b Saarland University, Im Stadtwald, 66123 Saarbru¨cken, Germany

Received 13 December 2004; received in revised form 1 June 2005; accepted 22 November 2005 Available online 20 March 2006

Abstract We analyze the interaction of stock market movements and politics in Germany. Evidence from popularity functions and VAR-based evidence suggests that stock market returns have affected the popularity of German governments. We only find weak evidence that the political process has had an impact on the stock market. In contrast to empirical evidence for the U.S., we do not find that German stock market returns tend to be higher during left-wing than during right-wing governments. Also in contrast to results for the U.S., we find no evidence for an election cycle in German stock market returns. D 2006 Elsevier B.V. All rights reserved. JEL classification: E32; E44; G12 Keywords: Political business cycle; Stock market; Germany

1. Introduction In this paper we use data on the German stock market to test the economic theory of political business cycles (PBC). This theory, as pioneered by Downs (1957) and Nordhaus (1975), predicts that government popularity depends on the state of the economy. In empirical applications, the state of the economy is usually measured by macroeconomic variables such as output growth, inflation, and the unemployment rate (for surveys, see Nannestad and Paldam, 1994; Ga¨rtner, 1994a; Olters, 2001). However, there are good reasons to use stock market variables instead. Stock markets are in the focus of media coverage on economic issues. Day by day, newspapers and television channels report stock market movements. Hardly any other economic variable enjoys a similar degree of public attention. It is, thus, plausible to conjecture

* Corresponding author. Tel.: +49 69 9566 3051; fax: +49 69 9566 4317. E-mail address: [email protected] (J. Do¨pke). 0176-2680/$ - see front matter D 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.ejpoleco.2005.11.004

926

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

that voters make up their minds on the state of the economy at least partly by considering information on stock market movements. Besides these plausibility-based reasons, using stock market data to test the implications of PBC theory is also justified by economic reasoning. For example, Ga¨rtner and Wellershoff (1999) make a case for using stock market data by proposing that the stock market is in general thought to be a leading indicator for real economic activity. Though one can, of course, debate on the indicator properties of the stock market. Early empirical evidence on a four-year presidential election cycle in U.S. stock market returns was reported by Umstead (1977), Allvine and O’Neill (1980), and Huang (1985). The empirical results reported by these authors indicate that stock market returns are higher in the third and fourth year of a presidency, and lower in the first and second year of a presidency. Moreover, the empirical results indicate that stock market returns tend to be higher on average during Democratic presidencies than during Republican presidencies. More recent empirical results documented by Ga¨rtner and Wellershoff (1995, 1999) and Santa-Clara and Valkonov (2003) support, in general, the empirical results reported in the earlier literature. Ga¨rtner and Wellershoff report that, during the first half of a presidency, the U.S. stock market tends to be bearish. In contrast, during the second half of a presidency, the U.S. stock market tends to be bullish. Santa-Clara and Valkanov have shown that stock market returns are higher during Democratic than during Republican presidencies. They have found that this difference is not explained by business-cycle variables, that it is not concentrated around election dates, and that it cannot be explained by differences in the riskiness of the stock market across Democratic and Republican presidencies. Thus, there seems to be a presidential election cycle in U.S. stock market returns, with statistical properties of stock market returns during Democratic presidencies differing from those during Republican presidencies.1 Against the background of the efficient market hypothesis discussed in the finance literature (see Fama, 1991), these results are puzzling. Given these interesting and puzzling results, evidence from countries other than the U.S. can yield important insights into the nature of the link between the political process and stock market movements. We provide empirical evidence on the link between the political process and stock market movements in Germany. Our results for the German stock market differ significantly from those reported in the literature for the U.S. For example, in contrast to the empirical results available for the U.S. stock market, our empirical results indicate that in Germany stock market returns do not tend to differ during right-wing and left-wing governments. Moreover, using the technique suggested by MacCallum (1978), we do not find strong evidence of a pronounced political cycle in German stock market returns. This latter result, however, does not imply that the political process in Germany and stock market movements are completely independent of each other. In fact, we report, that stock market movements significantly affect the government’s popularity as measured by its approval rate: the approval rate tends to increase when the stock market is bullish, and tends to decrease when the stock market is bearish.

1

Others have investigated whether significant predictable patterns in stock market movements around election dates are detectable. Niederhofer et al. (1970) and Riley and Luksetich (1980), for example, use data for the U.S. stock market to analyze this question. In an international context, Pantzalis et al. (2000) provide empirical evidence on fluctuations of stock market returns around election dates. See also Peel and Pope (1983), among others, for an empirical study of the link between election dates and U.K. stock market returns. Vuchelen (2003) presents empirical evidence for Belgium. For further evidence, see also the literature cited by these authors.

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

927

We organize our paper as follows. In Section 2 we begin by studying the impact of stock market movements on the political process in Germany. In Section 3 we test for the presence of a political cycle in German stock market returns. In Section 4 we discuss whether differences between structures of political systems account for the differences between Germany and the U.S. In Section 5 we offer concluding remarks. 2. The impact of the stock market on the political process Because stock market movements enjoy a high degree of public attention in the media, it is interesting to study whether they are correlated with the political process in Germany. In order to study this possibility, we estimated popularity functions and vector autoregressive (VAR) models. Popularity functions have often been used in the theoretical and empirical literature to study the factors that influence the popularity of a government. A popularity function is based on the notion that a governments’ popularity can be treated as the dependent variable and stock market movements and other control variables as independent explanatory variables. A VAR model goes beyond a popularity function in that it allows more complex dynamic links between stock market movements and the political process to be recovered. 2.1. Tests based on popularity functions Empirical evidence on a possible link between stock market movements and the political process can be obtained by estimating a popularity function. Popularity functions can be justified theoretically. See for example Kirchga¨ssner (1986), who suggests a popularity function based on the seminal work of Fair (1978). The basic idea is that voters maximize utility and have preferences in a politico-economical space where competing political parties also have positions. Voters take the positions of competing political parties into account and form expectations of the utility that can be expected from voting for the competing political parties. In this regard, voters use information about the positions of the competing political parties as reflected in either the actual record of government parties or from concepts, programs and promises of competing political parties. Voters choose to vote for government parties if their utility from doing this is larger than some benchmark and otherwise vote for an opposition party or do not vote. Imposing some additional assumptions that ensure that the behaviour of individual voters can be aggregated, it is possible to write down a collective voting equation that links the proportion of votes for government parties to the proportion of votes for government parties in a previous election and the economic situation during the electoral term. Using a discrete-time formulation of this link (because the state of the economy is observable in discrete time only) and employing a Kock-lag operator, the linear baseline specification of a popularity function can be written as pt ¼ b0 þ b1 pt1 þ

n X

bi xt;i þ bðt Þ þ ut ;

ð1Þ

i¼1

where p t denotes the popularity of the government, the vector x t denotes the state of the economy, and b(t) denotes a function of time which accounts, for example, for cyclical effects.2 2

In the case of more than two competing political parties, it is possible to write down a similar system of equations for the popularity of each party (Kirchga¨ssner, 1986, page 425).

928

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

Standard popularity functions have frequently been shown to fit German data quite well (see Kirchga¨ssner, 1985). According to the popularity function we estimated, the approval rate ( p t ) of the government depends upon the unemployment rate (URt ) and the inflation rate (IRt ). We also included the lagged approval rate in the vector of regressors. In addition, we took into account that the approval rate may have been linked to stock market returns (r t ). Hence, we estimated the following equation:3 pt ¼ b0 þ b1 pt1 þ b2 URt1 þ b3 IRt1 þ b4 rt1 þ ut :

ð2Þ

We expect the coefficient of both the unemployment rate and the inflation rate to be negative and the coefficient of stock market returns to be positive. We used survey data on the government’s popularity with voters compiled by the bForschungsgruppe WahlenQ, Mannheim, to estimate Eq. (2). Data are available in electronic form from 1977 onwards. In each survey, potential voters are asked whether they approve of the politics of the government. The answers can vary from +5 (bhighly approveQ) to  5 (bhighly disapproveQ). For the periods for which no survey data are available, we imputed values using linear interpolation.4 In order to measure stock market movements, we used excess returns on the DAX. We computed excess returns by subtracting the overnight bank lending rate (Tagesgeld) from the continuously compounded returns on the DAX.5 We collected data on the DAX and on the overnight bank lending rate from the Monthly Reports of the Deutsche Bundesbank. Data on the (seasonally adjusted) unemployment and inflation rate are from the OECD databanks bMain Economic IndicatorsQ and bEconomic OutlookQ. We used quarterly data to estimate Eq. (2). Our sample period ends 2003:4. We also estimated a variant of Eq. (2) in which we used the change in the approval rate. We did so because, according to the results of unit-root tests summarized in Table 1, it is difficult to decide whether the approval rate is a stationary time series.6 In order to assess the robustness of our results, we decided to present results for both an equation in which the approval is used in levels and an equation in which the approval rate is used in first-differences.

3 Our decision to use lagged regressors is in line with our VAR evidence in Section 2.2, which shows a delayed response of the approval rate to innovations in stock market returns. Therefore, it is not surprising that contemporaneous stock market returns would be insignificant in Eq. (2). Our decision to estimate a specification of Eq. (2) that uses lagged stock market returns and a specification that uses four lags of stock market returns as regressors accounts for a delayed response of the approval rate to innovations in stock market returns. It should also be noted that using the lagged unemployment rate and the lagged inflation rate as regressors allows accounting for the fact that macroeconomic data are generally published with a time lag. Furthermore, using lagged stock market returns ensures that this regressor is weakly exogenous even if changes in the governments popularity effect the stock market. We did not include real GDP in the vector of regressors because of its high correlation with the unemployment rate. Our results do not change if we use real GDP or the OECD output gap as an additional regressor, respectively. 4 We have no data for 36 out of 290 months (=12%). 5 DAX, end of 1987 = 100. In the working paper version of this paper, we used alternative measures of stock market returns in order to assess the robustness of our results (Do¨pke and Pierdzioch, 2004). For example, we used continuously compounded real returns, where we used consumer price inflation to deflate nominal returns. Results are qualitatively similar to the results for excess returns. 6 For a discussion of the implications of a unit root in the context of estimations of popularity functions, see also Kirchga¨ssner (1985, 1986).

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

929

Table 1 Testing for a unit root in the approval rate Test statistic

Result

Augmented-Dickey–Fuller test Phillips–Perron test KPSS test

3.49** 3.31* 0.07

Note: Tests for a unit root in the approval rate following Dickey and Fuller (1979), Phillips and Perron (1988), and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) (1992). In the cases of the Augmented-Dickey–Fuller test and the Phillips–Perron test, the SIC criterion was used to determine the number of lags of the endogenous variable. In the case of the KPSS test, we used the Newey–West procedure using the Bartlett kernel. ** (*) denote rejection of the null hypothesis of non-stationarity (stationarity) in the cases of the Augmented-Dickey–Fuller test and the Phillips– Perron test (KPSS test). All test regressions were estimated by including a constant and a trend in the vector of regressors.

Unlike the unemployment or inflation rate, stock market returns are known to be a very volatile variable. Because voters know this, we also took into account the possibility that voters use past values of stock market returns to make up their minds about economic conditions. Therefore, we estimated both a version of Eq. (2) in which only lagged stock market returns enter into the vector of regressors and a version of Eq. (2) in which we include up to 4 lags of stock market returns in the vector of regressors. Table 2 summarizes the estimation results. All coefficients have the expected signs. Moreover, tests for well-behaved residuals reveal that the equations are correctly specified. As concerns the coefficients that capture the correlation of stock market movements with the approval rate, they turn out to be significantly different from zero. This result is robust across the different versions of Eq. (2) we estimated. The insignificance of the coefficients of the inflation rate and the unemployment rate in the variant of Eq. (2) in which the first-difference of the approval rate is used can be attributed to the fact the change in the approval rate is a very volatile variable which is difficult to model. Given that the attention a broader audience pays to stock markets movements may have increased during the new-economy boom in the 1990s, it cannot be ruled out that the estimated parameters are unstable over time. Thus, we tested for parameter stability. The test results are plotted in Fig. 2.7 The results show that–despite some large one-quarter forecast errors–the popularity function is stable. Neither the CUSUM nor the CUSUMQ tests on overall instability show any indicators of parameter instability (Fig. 1). As a final robustness check, we estimated a variant of Eq. (2) in which we included U.S. (excess) returns, calculated based on the logarithmic changes of the Dow Jones Industrial Average, in the vector of regressors. We included U.S. returns in the vector of regressors because developments may matter for expected economic conditions in Germany and, hence, for the popularity of the German government. For the sake of brevity, we do not report the estimation results.8 We found that including the U.S. returns does not affect the significance of the coefficient of German excess returns in our popularity functions.

7

The results in Fig. 1 are for a variant of Eq. (2) in which the level of the approval rate is modeled. A similar figure, available from the authors upon request, can be obtained when a variant of Eq. (2) is estimated in which the first difference of the approval rate is modeled. 8 Results are available from the authors upon request.

930

Variable

Dependent variable: approval rate

Dependent variable: first-difference of approval rate

Stock market variable

Stock market variable

(–)

Excess returns

Excess returns/ sum

(–)

Excess returns

Excess returns/sum

Lagged stock market returns

0.62 (2.32)** 0.82 (14.66)*** 0.06 (2.48)** 0.04 (1.29) 1.15 (18.73)*** (–) 0.84 0.15 7.67*** 3.09

0.83 (3.19)*** 0.81 (13.25)*** 0.08 (3.33)*** 0.02 (0.82) 0.91 (9.95)*** 2.17 (15.80)*** 0.88 0.06 2.87* 14.00

b0.00 (0.00) 0.10 (1.21) b0.00 (0.13) b0.00 (0.11) 1.17 (16.14)*** (–)

Adjust. R 2 Test for normalitya) Test for autocorrelationb) Test for heteroskedasticityc)

0.66 (2.96)*** 0.84 (18.22)*** 0.06 (3.07)*** 0.03 (1.26) 1.04 (17.11)*** 0.96 (3.87)*** 0.86 0.76 1.86 8.52

0.13 (0.77) 0.08 (0.96) 0.01 (0.84) b0.00 (0.04) 1.04 (15.31)*** 1.06 (3.85)*** 0.21 0.55 1.20 7.07

0.22 (1.30) 0.03 (0.30) 0.02 (1.43) 0.02 (0.65) 0.95 (11.98)*** 2.56 (19.86)*** 0.31 0.77 1.89 9.78

Constant Lagged approval rate Lagged unemployment rate Lagged inflation rate Dummy for change of governmentd)

0.09 0.03** 2.58 4.91

Note: *** (**, *) denote rejection of the null hypothesis at the 1 (5, 10) percent level. a) P-value of Jarque/Bera test for normality. b) Breusch/Godfrey (1988) test for autocorrelation of order 1. c) White test (without cross terms). Numbers in brackets are t-values calculated with robust standard errors based on the method of Newey and West (1987). d) The dummy for change in government assumes the value one in 1998:4 and zero else.

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

Table 2 Estimates of popularity functions

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

931

Fig. 1. Tests for stability of the parameter in the popularity equation. Note: This figure shows stability tests of the popularity function that includes excess returns in the vector of regressors. The upper left panel of the figure shows the Pvalues of a one-step forecast Chow-test on overall parameter constancy. The upper right panel shows the results of a CUSUM test on parameter stability. The lower left panel shows the results of a CUSUMQ test of on parameter stability. The lower left panel shows recursive estimates of the coefficient representing the impact of excess stock market returns on the popularity of the government.

2.2. Evidence from VARs and from tests for causality We used four alternative VAR models to study whether stock market movements and the political process are dynamically linked. We started with a bivariate VAR model that includes the approval rate and stock market returns as endogenous variables (VAR model 1). We then included other variables in the VAR in order to be sure that a potential link between stock market returns and the approval rate is not merely due to the influence of another variable like, e.g., inflation. Thus, the second model (VAR model 2) includes excess stock markets returns and the approval rate as endogenous variables. As an exogenous (i.e., conditioning) variable, it includes both the unemployment and the inflation rate. The third model (VAR model 3) is identical to VAR model 2, but treats the inflation rate and unemployment rate as endogenous variables. The fourth model (VAR model 4) is identical to VAR model 3, but contains lagged returns on the U.S. stock market as a conditioning variable. We determined the lag length of all VAR models by means of the minimum Schwarz information criterion. In addition, we performed lag-exclusion tests based on a VAR with one additional lag. Furthermore, we conducted tests on well-behaved residuals, i.e., tests on

932

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

Table 3 Specification of the VAR models

Minimum Schwarz criterion (at lag) Lag exclusion testa) (at lag) Test for normalityb) Test for autocorrelation at lag 1 Test for heteroskedasticityc) Log likelihood Lag length

VAR model 1

VAR model 2

VAR model 3

VAR model 4

0.46 (1) 6.92 (2) 23.37*** 3.91 41.63 41.58 1

0.38 (1) 6.24 (2) 27.16*** 2.18 59.31 40.18 1

3.61 (1) 25.88 (2) 91.47*** 73.19*** 333.51 105.81 1

3.96 (1) 27.12 (2) 96.29*** 90.40*** 210.49 357.89 1

Note: a) Wald test for joint significance of all variables at the lag reported in brackets. b) Test for normality of the residuals based on Doornik and Hansen (1994). c) Test for heteroskedasticity based on White (1980). *** (**, *) denote rejection of the null hypothesis at the 1 (5, 10) percent level.

autocorrelation, normality, and heteroskedasticity. Details regarding these tests can be found in Table 3. Given the results of these tests, we included one lag of the endogenous variables in our VAR. In order to analyze the dynamics of our VAR models, we computed generalized impulse response functions as described by Pesaran and Shin (1998).9 An impulse response function measures the time profile of the effect of a one-time structural shock to the variables included in a VAR model. It, thereby, shows how a shock affects current and expected future values of the endogenous variables of a VAR model through the dynamic lag structure of the model. Fig. 2 shows the estimated impulse response functions. It turns out that the response of government’s approval rate to a one-standard-deviation shock in stock market returns is positive and, as indicated by the reported standard error bands, significant (Panel A of Fig. 2). This result is robust across the VAR models we estimated. Thus, we conclude that stock market movements are a non-negligible determinant of the approval rate and, thus, of the popularity of the government. The impact of a shock to the approval rate on stock markets returns is significant only in VAR model 1 and VAR model 2 (Panel B of Fig. 2). This result suggests that there might be a dynamic interplay between stock market movements and the approval rate. However, as evidenced by the impulse response functions for VAR model 3, this result is sensitive to the specification of the VAR model. In fact, if one controls for the influence of inflation and unemployment, the effect of a shock to the approval rate on stock market returns is only borderline significant. In VAR model 4, the effect is also only borderline significant.10 We conclude that, in contrast to the results that have been reported for the U.S. in the earlier literature, there are no political cycles in German stock market returns. It is also worth mentioning that this result is in line with the efficient market hypothesis.

9 Impulse response functions based on a unique Choleski decomposition of the residuals of a VAR model can be obtained by imposing a specific ordering of the endogenous variables included in the system. This ordering implies that a shock to a variable placed in a lower position of this ordering exerts no contemporary effect on the variables placed in a relatively higher position of the ordering. Unlike impulse response functions based on the Choleski decomposition, generalized impulse response functions are based on an orthogonal set of innovations that does not depend on the ordering of the endogenous variables included in the VAR. Interested readers can find impulse response functions based on Choleski decompositions in the working paper version of this paper (Do¨pke and Pierdzioch, 2004). 10 It is also interesting to note that, in VAR models 2, 3, and 4, the effect of a shock to the approval rate on stock market returns is insignificant if we use two lags of the endogenous variables rather than just one lag.

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

Fig. 2. Impulse response functions. Note: Panel A depicts impulse response functions of approval rate to a one standard deviation shock to excess stock markets returns. Panel B depicts impulse response functions of excess stock market returns to a one standard deviation shock to the approval rate.

933

934

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

Table 4 Tests for Granger-causality Null hypothesis

F-value

Tests using the level of the approval rate Excess returns do not Granger cause the approval rate Excess returns do not Granger cause the approval rate instantaneously The approval rate does not Granger cause excess returns

15.70*** 4.90** 6.55**

Tests using first differences of the approval rate Excess returns do not Granger cause the approval rate Excess returns do not Granger cause the approval rate instantaneously The approval rate does not Granger cause excess returns

17.78*** 0.05 0.01

Note: *** (**, *) denote rejection of the null hypothesis at the 1 (5, 10) percent level. All tests were performed based on a lag length of 1. The tests for instantaneous Granger-causality are based on the Sims form of tests for Granger-causality (Hamilton, 1994: 304) and use the robust standard errors using the approach developed by Newey and West (1987).

In addition to the evidence based on impulse–response functions we have also tested for Granger-causality between stock markets returns and the approval rate. Following Kirchga¨ssner (1986), we considered three cases. Case 1: Stock market returns Granger cause the approval rate if lagged stock market returns help to forecast popularity. Case 2: Stock market returns instantaneously Granger cause the approval rate if current stock market returns help to forecast popularity.11 Case 3: There is a feedback relation between stock market returns and the approval rate if one finds evidence for bidirectional Granger-causality in the sense of Case 1. To test for Granger-causality, we used both the level and the first differences of the approval rate (Table 4). The test results we obtained when we use the level of the approval rate indicate that there may be a feedback relation between stock market returns and the approval rate. In contrast, the test results do not provide evidence for an instantaneous Granger-causality of excess stock market returns for the approval rate. The results we obtained when we used the first difference of the approval rate are similar with one exception. The test results provide evidence for Granger-causality running from stock market returns to popularity rather than a feedback relation. 3. The impact of the political process on the stock market The results we have derived so far indicate that the political process in Germany, as measured in terms of changes in the approval rate, did not systematically affect stock market returns. This is interesting because a number of authors have reported that stock market returns in the U.S. tend to be correlated with the political process. Given that our results for Germany differ from the results for the U.S., we now analyze in detail the robustness of our results. 3.1. Returns during left-wing and right-wing governments Results reported in the empirical literature suggest that the average U.S. stock market returns during Democratic presidencies differ from stock market returns during Republican presidencies. In particular, many authors have reported that stock markets returns are on average higher during Democratic presidencies than during Republican presidencies (see, e.g., Santa-Clara and 11

We used the Sims-formulation of a Granger-test (Hamilton, 1994: 304).

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

935

Table 5 Average excess stock market returns by political orientation of central government (1961 Q1 to 2003 Q4) Political orientation of government

Average returns

Average returns otherwise

Test for equal meana)

Test for equal medianb)

Right-wing (CDU chancellor) Left-wing (SPD chancellor) Neutral (bGrand coalitionQ)

0.026 0.069 0.006

0.058 0.022 0.046

1.78* 2.60** 1.49

2.64*** 3.36*** 1.34

Note: *** (**, *) denote rejection of the null hypothesis at the 1 (5, 10) percent level. Wilcoxon/Mann–Whitney test for an equal median.

a)

t-test for an equal mean.

b)

Valkonov, 2003; Huang, 1985). In order to analyze whether a similar stylized fact can be obtained for Germany, we analyzed whether excess stock market returns are different during leftwing than during right-wing governments. To this end, we extended our sample period to cover the years 1960–2003.12 Our results are summarized in Table 5. Our results are in contrast to those reported in the literature for the U.S. stock market. In fact, our results indicate that, if anything, in Germany stock market returns have been on average higher during right-wing governments than during left-wing governments. The results summarized in Table 5 should be interpreted with some caution for at least two reasons. First, while in the U.S., political power shifts relatively often, in Germany only three major shifts between more left-wing and more right-wing governments have occurred. Second, when computing the results we report in Table 5, we did not control for the influence of other variables that may have had an impact on stock market returns. In order to control for the influence of other variables, we followed Santa-Clara and Valkonov (2003) and regressed stock market returns on dummy variables that represent the political orientation of the government and a set of control variables. We controlled for the influence of the short-term interest rate, the spread between the long-term and short-term interest rate, lagged U.S. stock market returns, and a dummy variable that captures large stock market slumps (like the one observed in 1987). We counted every quarter in which stock market returns exceed  20% as a bcrashQ period.13 We estimated the following equation: rtþ1 ¼ b1 DCt þ b2 DLt þ b3 Controlt þ ut ;

ð3Þ

where DCt and DLt denote dummy variables for a right-wing and a left-wing governments, respectively. The estimation results are given in Table 6. It turns out that the differences between the average stock market returns across left-wing and right-wing governments are relatively small. In statistical terms, the difference is significant only at the 10% level. This indicates that the political orientation of the government is hardly visible in the magnitude of stock market returns. Moreover, the results given in Table 4 clearly suggest that, in contrast to what has been 12 Though general elections took place in Germany before 1960, the postwar reconstruction period after World War II certainly was a rather special period in both political and economic respects. Moreover, stock market data for the years before 1960 must be interpreted with some caution because during that time the German stock market was subject to regulations. Moreover, in the years after the war, German newspapers did not report stock market movements on a regular basis, indicating that in those years the Germans had other things to do than to scrutinize stock market movements. A detailed discussion of these issues can be found in Gielen (1994). 13 We downloaded end-of-quarter U.S. stock market data from the webpage of Dow Jones Company. In order to compute excess returns, we collected data on a U.S. short-term interest rate (discount rate) from the International Financial Statistics of the International Monetary Fund (series code 11160. . .ZF. . .).

936

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

Table 6 Tests for equal mean in excess stock market returns during left-wing and right-wing governments Specification

Dummy for left-wing government

Dummy for right-wing government

Control variable coefficient

Test for equal coefficientsa)

Without control variable Short-term interest rates Interest rate spread Crash dummy Crash dummy and AR(1) process Lagged U.S. excess returns

0.065 0.001 0.082 0.047 0.047

0.030 0.023 0.054 0.015 0.013

(–) 0.01 (3.42)*** 0.013 (3.55)*** 0.32 (12.03)*** 0.06/0.33 (0.78)*/(11.9)*** 0.55 (2.59)**

2.86* 1.69 2.46 3.15* 3.45*

(4.15)*** (0.04) (6.14)*** (3.21)*** (3.15)***

0.024 (1.39)

(2.32)** (1.10) (4.11)*** (1.30) (1.11)

0.007 (0.48)

0.66

Note: *** (**, *) denote rejection of the null hypothesis at the 1 (5, 10) percent level. a) Wald test. Numbers in brackets are t-values calculated with robust standard errors based on the method of Newey and West (1987). The dummy for rightwing governments assumes the value one in the periods 1960:3–1966:3 and 1982:4–1998:3, and zero else. The dummy for left-wing government assumes the value one in the periods 1969:4–1982:3 and 1998:4–2003:4. AR(1) denotes an autoregressive process of order one.

found for the U.S., in Germany stock market returns during left-wing governments have not been significantly higher than during right-wing governments. Thus, we conclude that a stylized fact that is well-established for the U.S. cannot be confirmed with German data: stock market returns are not systematically higher during left-wing governments than during right-wing governments. 3.2. Tests for a political cycle in stock market returns The result that stock market returns are hardly significantly higher during left-wing than during right-wing governments does not rule out the possibility that, irrespective of the political orientation of the government, the political process induces political cycles in stock market returns. How can such political cycles arise? In the PBC literature, answers to this question have been offered by adherents of at least two schools, namely the bopportunisticQ and the bpartisanQ school. 3.2.1. Implications of PBC theory Adherents of the opportunistic school argue that incumbent governments use expansionary policy measures to improve the economic situation just before an upcoming election. If expectations are not fully rational, the increase in overall economic activity will improve the reelection chances of the government (Nordhaus, 1975). Even if voters form rational expectations, it can be rational for governments to conduct an expansionary policy before elections if temporary information asymmetries are important (Rogoff, 1990). Such temporary information asymmetries can result in a so-called rational opportunistic cycle. According to the adherents of the school of partisan cycles (see Hibbs, 1992), political business cycles can arise if agents do not form rational expectations and more left-wing governments prefer a more expansionary and, thus, more inflationary policy, while more right-wing governments are more hawkish regarding inflation, but do not care as much about output growth and employment. We have already shown in Section 3.1 that, in contrast to what partisan PBC theory implies, there is no significant difference in stock market returns across left-wing and right-wing governments. Thus, basic partisan PBC models should have difficulties in shedding light on the

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

937

link between the political process and stock market movements in Germany. This, however, does not necessarily mean that partisan PBC theory has nothing to say about this link. In fact, the socalled rational partisan cycle theory, i.e., the partisan PBC theory extended to feature rational expectations, implies that one should not focus on the type of comparisons between left-wing and right-wing governments that formed the basis for our analysis in Tables 4 and 5. Rather, one should focus on periods in which left-wing governments are replaced by right-wing governments, and vice versa (see Berger and Woitek, 1997; Alesina and Roubini, 1992). 3.2.2. Testing for a political cycle in stock market returns In order to test for a political cycle in stock market returns, we have followed Ga¨rtner and Wellershoff (1995, 1999), who have used in their empirical analysis of political cycles in U.S. stock market returns a test developed by MacCallum (1978).14 In order to implement this test, we defined a dummy variable, E t . Starting in the quarter in which an election was held, this dummy variable assumes the values 0, 1, 2, 3, 4, 5, 6, 7, 8, 7, 6, 5, 4, 3, 2, 1, 0. Thus, for each quarter between two elections, this dummy variable assumes a different value. This dummy variable can, therefore, be used to test whether systematic and statistically significant election cycles can be detected in stock market returns. This test is not directly applicable to German data. The reason for this is that the political institutions in Germany are somewhat different from those in the U.S. Specifically, the election terms are not as regular as in the U.S. The length of an election term may differ due to unexpected elections (1972, 1983) and due to attempts to bring back the election date, after such an birregularQ election once has been held, to its traditional date in autumn. For this reason, several terms can be slightly shorter than a regular 16-quarter term. In order to account for such variations in the length of a term, we set the dummy variable equal to zero in the quarter in which an irregular election occurs. In order to test formally for the presence of an election cycle in German stock market returns, we estimated the following equation: rt ¼ b0 þ b1 DEt þ ut

ð4Þ

where r t denotes (nominal, real, or excess) stock market returns, D denotes the first difference operator, E t denotes the McCallum dummy, and u t denotes an error term. We used the first difference of the McCallum dummy because PBC theory implies that an election cycle should be detectable at the level of a stock market index. Thus, because Eq. (4) is formulated in terms of stock market returns, we must use the first difference of the McCallum dummy (see also Ga¨rtner and Wellershoff, 1995, 1999). Under the null hypothesis that an election cycle is detectable in stock market returns–and against the background of the implications of PBC theory–the coefficient of the McCallum dummy, b 1, should be statistically significant and negative. The difference between the constant,

14

A search for a PBC in stock market returns is warranted only if empirical evidence indicates a PBC in Germany, which is the case. For surveys of the empirical literature, see Ga¨rtner (1994b), Belke (2000) and Berlemann and Markwardt (2003). The general picture that emerges from the empirical literature looking for PBCs in German data is less sharp than that for the U.S. Still, even researchers who are skeptical about the existence of a PBC in Germany conclude that, with regard to the opportunistic and partisan school bof PBC theory, some support for both types of hypotheses came from German dataQ (Berger and Woitek, 1997). Hence, our reading of the empirical literature is that the evidence supporting the hypothesis of a PBC in Germany is strong enough to warrant a study of the question of whether there is a political cycle in German stock market returns.

938

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

b 0, and this coefficient captures the influence of the political process (i.e., the election cycle) on stock market returns. When estimating Eq. (4), we took into account two potential econometric problems. The first econometric problem is that it is unlikely that the error term, u t , is a white noise error term, implying that standard hypothesis tests cannot be applied. We took this into account in two different ways. First, we computed robust standard errors in order to assess the statistical significance of the coefficients. We used the technique suggested by Newey and West (1987) to compute robust standard errors. Second, we modelled the error term using an autoregressive process of order one, AR(1), to achieve white noise errors.15 The second econometric problem is that there are a number of large outliers in the time series of German stock market returns. This is a problem that is quite common in financial econometrics. In order to deal with this problem, we followed Ga¨rtner and Wellershoff (1995, 1999) and added a dummy variable to the right-hand side of Eq. (4). This dummy variable covers periods of stock market bcrashesQ. We counted every quarter in which stock market returns exceed  20% as a bcrashQ period. The estimation results are summarized in Table 7. As can be seen in Table 7, we estimated a number of alternative specifications of Eq. (4).16 Specifically, we analyzed the implications of using alternative definitions of the McCallum dummy. For example, because the linear specification of the McCallum dummy may be too restrictive to detect pronounced election cycles in stock market returns, we also estimated a specification of Eq. (4) in which the McCallum dummy raised to the power four is used as a regressor (see also Ga¨rtner and Wellershoff, 1995, 1999). The estimation results show that the bcrashQ dummy is always significant and has the expected negative sign. The coefficient of the AR(1) process we use to model the error term of Eq. (4) is statistically significant. The coefficient of the McCallum dummy is in general not statistically different from zero only at the 10% level.17 Thus, all in all, the estimation results do not indicate that there is a pronounced election cycle in German stock market returns.18 The coefficient of the McCallum dummy has the theoretically expected sign in all cases: PBC theory suggests that the sign of this coefficient should be negative. If the government tries to

15 We modeled the error term as an AR(1) process for statistical reasons. A discussion of the economic interpretation of autocorrelation in the error term would be beyond the scope of our paper. Autocorrelation of stock market returns may but need not indicate informational inefficiency of the stock market. Autocorrelation of stock market returns may also be due to, for example, transaction costs, time-variation in expected returns, or noise trading. 16 In the working paper version of this paper, we estimated, in addition to the specifications reported in Table 7, specifications with nominal and real returns on the left-hand side of Eq. (4). The results for these specifications are similar to the results reported in Table 5. Moreover, we estimated extended versions of Eq. (4) in which we used lagged U.S. excess stock market returns in order to account for the influence of developments on the U.S. stock on the German stock market. We found that, in the extended versions of Eq. (4), the McCallum dummy is insignificant. This corroborates our conclusion that there was no pronounced political cycle in German stock market returns. The estimation results for the extended version of Eq. (4) are available from the authors upon request. 17 When we used quarterly averages of the DAX index rather than end-of-quarter data, we found that the McCallum dummy is insignificant. Thus, the significance of the McCallum reported in Table 5 is not robust to slight variations in the specification of Eq. (4). This finding lends further support to our interpretation that there was no pronounced election cycle in German stock market returns. 18 This result implies that the political process in Germany does not give rise to systematic and statistically significant cyclical movements in German stock market returns. Given the high degree of international integration of financial markets, this does not imply that a potential political cycle in U.S. stock returns does not transmit onto German stock market returns. For the international transmission of the U.S. election cycle to international stock returns, see Foerster and Schmitz (1997).

Constant

Political cycle variable

Crash dummy

AR(1) process

R2

Test for autocorrelationa)

Test for ARCHb)

Test for normalityc)

0.042 (4.36)***

 0.014 (DE t ) (1.95)*

(–)

(–)

0.02

 0.014 (DE t ) (1.89)*

(–)

0.03 (0.50)

0.02

0.027 (2.89)***

 0.011 (DE t ) (1.74)*

0.33 (11.50)***

0.07 (0.85)

0.37

0.026 (2.86)***

N0.0001 (DE 4t ) (1.54)

0.33 (11.98)***

0.09 (0.93)

0.36

ARCH(1): ARCH(4): ARCH(1): ARCH(4): ARCH(1): ARCH(4): ARCH(1): ARCH(4):

46.96***

0.043 (4.32)***

Order Order Order Order Order Order Order Order

1: 0.19 4: 2.51 1: 0.66 4:0.60 1: 8.26*** 4: 10.46** 1: 8.33*** 4: 10.65**

0.04 6.37 0..09 6.69 0.18 0.12 0.12 0.59

45.39*** 6.28* 5.93*

Note: *** (**, *) denote rejection of the null hypothesis at the 1 (5, 10) percent level. a) Breusch/Godfrey (1988) test for autocorrelation (Observed R 2). b) Test for remaining ARCH-Terms (Observed R 2). c) Jarque–Bera test (Bera and Jarque, 1981) for normality. Numbers in brackets are t-values calculated with robust standard errors based on the method of Newey and West (1987). AR(1) denotes an autoregressive process of order one.

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

Table 7 Tests for political cycles in German stock returns (McCallum regressions 1961 to 2003)

939

940

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

Table 8 Tests for rational partisan cycles in excess stock market returns Equation

Dummy switch from left to right-wing

Dummy switch from right to left-wing

Test for normalitya)

Test for autocorrelationb)

Dummies with two-quarter horizon Dummies with four-quarter horizon

0.001 (0.04) 0.041 (2.02)**

0.014 (0.63) 0.001 (0.04)

1.37 1.61

0.61 0.53

Note: *** (**, *) denote rejection of the null hypothesis at the 1 (5, 10) percent level. a) Jarque–Bera test (Bera and Jarque, 1981) for normality. b) Breusch/Godfrey (1988) test for autocorrelation (Observed R 2) of order 1. See page 12 for definitions of the dummy variables. Numbers in brackets are t-values calculated with robust standard errors based on the method of Newey and West (1987). In order to model autocorrelation, we included an AR(2) term in the vector of regressors. The dummy for switches from left to right-wing assumes the value one in 1982:4–1983:1 (two-quarter horizon) and 1982:4–1983:4 (four-quarter horizon), and zero else. The dummy for switches from right to left-wing assumes the value one from 1966:4–1967:1, 1969:4–1970:1, and 1998:4–1999:1 (two-quarter horizon) and 1966:4– 1967:4, 1969:4–1970:4 and 1998:4–1999:4 (four-quarter horizon), and zero else.

induce a political business cycle and stock market returns reflect this, it should be possible to observe negative returns in the quarters following an election and positive returns in quarters preceding an election. However, given the weak level of significance we would hesitate to interpret our results as support of political cycles in excess returns. 3.2.3. Partisan cycles in stock market returns Besides examining the possibility that a pure election cycle is detectable in stock market returns, we also took into account the possibility that a partisan cycle is detectable in stock market returns. It is clear from the results reported in Tables 5 and 6 that, in contrast to what the basic partisan PBC theory implies, there is no significant difference in stock market returns across left-wing and right-wing governments. Therefore, we focused on testing the implications of the rational partisan cycle theory. In particular, to test this theory, we followed Berger and Woitek (1997) and defined a dummy variable that assumes the value 1 when there is a change Y right from a left-wing to a right-wing government (Dleft ). We also defined a corresponding t dummy variable that assumes the value one when there was a change from a right-wing to a leftY left wing government (Dright ). When defining these dummies, we took account of the historical t fact that the German political system has not experienced many clear-cut swings in either direction because the majority of changes in government were driven by changes in political coalitions. Therefore, we followed the literature and considered the following changes: 1966— change to the left (social democrats enter the government); 1969—change to the left (Christian democrats leave the government); 1982—change to the right (Christian democrats enter the government); 1998—change to the left (social democrats win chancellorship).19 We used a regression equation of the format of Eq. (4) to test the implications of the rational partisan cycle theory. Like Berger and Woitek (1997), we defined the dummy variables that account for changes in the political orientation of the government in two different ways: we used a dummy that assumes the value 1 for two quarters after an election, and we used a dummy that assumes the value 1 for four quarters after an election. Using both dummies should give us an

19 One might also wish to take into account two other events. Berger and Woitek (1997) report and criticize that some studies count 1972 as a swing from a left-wing to a right-wing government because Karl Schiller, a prominent pro-market Social democrat, left the government. In a similar vein, one could argue that in 1999 a swing from bthe left to the rightQ occurred because Oskar Lafontaine, an influential alleged left-winger among Social democrats, resigned from the office of minister of finance.

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

941

impression of the robustness of our results. In order to assure that the residuals of the regression are white noise, we included in the estimation equation an AR(2)-process for the residuals and the crash dummy that already worked in Eq. (4). The estimation results are given in Table 8. The evidence for a political cycle in stock market returns is at best limited. The dummy that accounts for switches from left-wing to right-wing government is statistically significant in one specification. The significance of this dummy indicates that the bmarket likes conservativesQ. The dummy that accounts for switches from right-wing to left-wing government is not statistically significant in both specifications. Thus, there is some, but no overwhelming evidence supporting rational PBC theory. Of course, the interpretation of the results summarized in Table 8 should not be stretched too far because shifts in the ideological composition of the government were limited. Yet, all in all, our results confirm the conclusion of Berger and Woitek that there is hardly any evidence for a partisan cycle in Germany. 4. The political system and stock market returns A question that we have not yet addressed is why our results for German stock market data are so strikingly different from those documented in the earlier empirical literature for the U.S. stock market. Why does empirical evidence indicate that there is a bpolitical cycle puzzleQ in U.S. stock market returns but not in German stock market returns? One answer to this question is that the political system in Germany is different from the political system in the U.S. A key element of the U.S. political system is that it is based on majoritarian electoral rules. In Germany, in contrast, the political system is based on proportional electoral rules. As a consequence, after World War II, governments in Germany have been formed by coalitions of two or more political parties. Perhaps even more important, in Germany, the political process has not only been shaped by the Federal parliament (Bundestag), but also by the parliament of the states of the Federal Republic of Germany (Bundesrat). Very often, the Bundesrat was dominated by right-wing parties when the Bundestag was dominated by left-wing parties, and vice versa. In many cases, this made it more or less impossible for the Federal government to reach important decisions without taking into account the interests of the governments of the states of the Federal Republic of Germany. One result of this bbalance of powerQ between the Bundestag and the Bundesrat was that political parties in Germany were not as polarized as in the U.S. This is one possible explanation for why stock market returns in Germany were not significantly different during right-wing and during left-wing governments. Recent results reported by Persson and Tabellini (2002) could indeed be interpreted to indicate that differences in the political system may be part of the explanation for the crosscountry bpolitical cycle puzzleQ we have reported in this paper. Also, Vuchelen (2003) argues that, in contrast to countries with majoritarian electoral rules, election results in countries with proportional electoral rules contain less information about future policy because it is difficult to predict the composition of a multi-party coalition from election outcomes. Hence, this could be one reason for why we do not find a significant political cycle in German stock market returns. 5. Conclusions We have established a number of interesting links between stock market movements and the political process in Germany. We have found weak evidence that, unlike in the U.S., stock

942

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

market returns in Germany tend to be higher under right-wing than under left-wing governments. Moreover, we have found no strong evidence for political or election cycles in stock market returns. Finally, VAR-based evidence as well as evidence from popularity functions has revealed that stock market returns are followed by changes in the popularity of German governments. The results of our empirical research suggest that stock market movements may have had an impact on political variables (such as, for example, the government’s popularity). The evidence for an impact of political variables on stock market movements is not very strong. We conclude from our results that, when studying whether the political process affects stock market movements, account should be taken of the possibility that political variables are not strictly exogenous. This conclusion is in line with the results documented in the empirical literature that uses economic variables to forecast election outcomes and thus assumes that the outcome of the political process is endogenous. The explanatory power of stock market movements for the government’s popularity may reflect the fact that stock market movements are forward-looking by nature and thereby summarize agents’ expectations regarding the future state of the economy. However, in our empirical analysis, we controlled for the impact of the unemployment rate and other important macroeconomic variables that also capture the stance of the economy. Therefore, it is also possible that the explanatory power of stock market movements for the governments’ popularity may reflect a sociotropic reaction, a psychological element, or a kind of bpocketbookQ voting. Acknowledgements The authors thank Matthias Jung and Annette Meyer from the bForschungsgruppe WahlenQ, Mannheim for providing the data on the popularity of the German government. The constructive comments of two anonymous referees helped us to improve the paper. The views expressed in this paper are those of the authors and do not necessarily reflect those of the Deutsche Bundesbank. References Allvine, F.C., O’Neill, D.E., 1980. Stock market returns and the presidential cycle. Financial Analysts Journal 36, 49 – 61. Alesina, A., Roubini, N., 1992. Political cycles in OECD economies. Review of Economic Studies 59, 663 – 688. Belke, A., 2000. Partisan political business cycles in the German labor market?: empirical tests in the light of the Lucascritique. Public Choice 104, 225 – 283. Bera, A.K., Jarque, C.M., 1981. An efficient large sample test for normality of observations and regression residuals. Working Papers in Economics and Econometrics. Australian National University. Berger, H., Woitek, U., 1997. Searching for political business cycles in Germany. Public Choice 91, 179 – 197. Berlemann, M., Markwardt, G., 2003. Partisan cycles and pre-electoral uncertainty. Dresden Discussion Paper in Economics, vol. 1/03. University of Dresden. Dickey, D.A., Fuller, W.A., 1979. Distribution of the estimators for autoregressive time series with a unit root. Journal of the American Statistical Association 74, 427 – 431. Doornik, J.A., Hansen, H., 1994. An omnibus test for univariate and multivariate normality. Nuffield College Discussion Paper, vol. 91. (Oxford). Do¨pke, J., Pierdzioch, C., 2004. Politics and the stock market: empirical evidence for Germany. Working Paper, vol. 1203. Kiel Institute for World Economics, Germany. Downs, A., 1957. An Economic Theory of Democracy. Harper and Row, New York. Fair, R.C., 1978. The effect of economic events on votes for president. Review of Economics and Statistics 60, 159 – 173. Fama, E., 1991. Efficient capital markets. Journal of Finance 46, 1575 – 1617.

J. Do¨pke, C. Pierdzioch / European Journal of Political Economy 22 (2006) 925–943

943

Foerster, S., Schmitz, J., 1997. The transmission of U.S. election cycles to international stock returns. Journal of International Business Studies 28, 1 – 27. Ga¨rtner, M., 1994a. Democracy, elections and macroeconomic policy: two decades of progress. European Journal of Political Economy 10, 85 – 109. Ga¨rtner, M., 1994b. The quest for political cycles in the OECD economies. European Journal of Political Economy 10, 427 – 440. Ga¨rtner, M., Wellershoff, K.W., 1995. Is there an election cycle in American stock returns? International Review of Economics and Finance 4, 387 – 410. Ga¨rtner, M., Wellershoff, K.W., 1999. Theories of political cycles: lessons from the American stock market. Rivista Internazionale di Scienze Economiche e Commerciali: RiSEC, vol. 46. pubblicazione trimestrale, Padova, pp. 613 – 630. Gielen, G., 1994. Ko¨nnen Aktienkurse noch steigen? Langfristige Trendanalyse des deutschen Aktienmarktes. Wiesbaden. Godfrey, L.G., 1988. Specification Tests in Econometrics. Cambridge University Press, Cambridge UK. Hamilton, J.D., 1994. Time Series Analysis. Princeton University Press, Princeton NJ. Hibbs, D., 1992. Partisan theory after fifteen years. European Journal of Political Economy 8, 361 – 373. Huang, R.D., 1985. Common stock returns and presidential elections. Financial Analyst Journal 41, 58 – 61. Kirchga¨ssner, G., 1985. Rationality, causality, and the relation between economic conditions and the popularity of parties. European Economic Review 28, 243 – 268. Kirchga¨ssner, G., 1986. Economic conditions and the popularity of West German parties: a survey. European Journal of Political Research 14, 412 – 439. Kwiatkowski, D., Phillips, P.C.B., Schmidt, P., Shin, Y., 1992. Testing the null hypothesis of stationary against the alternative of a unit root. Journal of Econometrics 54, 159 – 178. MacCallum, B.T., 1978. The political business cycle: an empirical test. Southern Economic Journal 44, 504 – 515. Nannestad, P., Paldam, M., 1994. The VP-function: a survey of the literature on vote and popularity functions after 25 years. Public Choice 91, 57 – 88. Newey, W.K., West, K.D., 1987. A simple, positive, semi-definite heteroscedasticity and autocorrelation consistent covariance matrix. Econometrica 55, 703 – 708. Niederhofer, V., Gibbs, S., Bullock, J., 1970. Presidential elections and the stock market. Financial Analysts Journal 26, 111 – 113. Nordhaus, W.D., 1975. The political business cycle. Review of Economic Studies 42, 169 – 190. Olters, J.P., 2001. Modeling politics with economic tools: a critical survey of the literature. Working Paper, vol. 01/10. International Monetary Fund, Washington D.C. Pantzalis, C., Stangeland, D.A., Turtle, H.J., 2000. Political elections and the resolution of uncertainty: the international evidence. Journal of Banking and Finance 24, 1575 – 1604. Peel, D.A., Pope, P.F., 1983. General elections in the U.K. in the post 1950 period and the behavior of the stock market. The Investment Analyst 67, 4 – 10. Persson, T., Tabellini, G., 2002. Do electoral cycles differ across political systems? Working Paper, vol. 232. IGIER, Bocconi University. Pesaran, M.H., Shin, Y., 1998. Generalized impulse responses in linear multivariate models. Economics Letters 58, 17 – 29. Phillips, P.C.B., Perron, P., 1988. Testing for a unit root in time series regression. Biometrika 75, 335 – 346. Riley Jr., W.B., Luksetich, W.A., 1980. The market prefers republicans: myth or reality? Journal of Financial and Quantitative Analysis 15, 541 – 559. Rogoff, K., 1990. Equilibrium political cycles. American Economic Review 80, 21 – 36. Santa-Clara, P., Valkonov, R., 2003. The presidential puzzle: political cycles and the stock market. The Journal of Finance 58, 1841 – 1872. Umstead, D.A., 1977. Forecasting stock market prices. Journal of Finance 32, 427 – 448. Vuchelen, V., 2003. Electoral systems and the effects of political events on the stock market: the Belgian case. Economics and Politics 15, 85 – 102. White, H., 1980. A heteroskedasticity-consistent covariance matrix and a direct test for heteroskedasticity. Econometrica 48, 817 – 838.