Do conservative central bankers weaken the chances of conservative politicians?

Abstract

In this paper, we challenge the claim that a conservative central bank strengthens the likelihood of a conservative government. In contrast, if an election is based on the comparative advantages of the candidates, an inflation-averse central banker can deter the chances of a conservative candidate because once inflation is removed, its comparative advantage in the fight against inflation disappears. We develop a theory based on a policy-mix game with electoral competition, predicting that a tighter monetary policy reduces the chances of a conservative (i.e., inflation-adverse) party while enhancing the chances for a liberal party. To test these predictions, we examine monthly data of British political history between 1987 and 2015, and show that an increase in the interest rate in the 10 months preceding a national election decreases the popularity of a Tory government. Our analysis on a panel of six OECD countries reveals that a pre-election increase of 1 percentage point in the main targeted interest rate rises the popularity of liberal parties by around 3.43 percentage points relative to its trend.

Appendices

Appendix A: The complete model with supply shocks

In this appendix we develop the complete model with a supply shock x. Under rational expectations, denoting by $\tilde{k}$ the unexpected component of any variable k,  we have $\pi =\pi ^e+\tilde{\pi }$ and $y=\tilde{y}=\alpha \tilde{\pi }+x.$ Expected components of variables are the same as in the main text, namely $(y^j)^e=0$ and $(\pi ^j)^e=(A(g^j)^e-Br^e)/(1-C)$ (see Eq. 12); and their unexpected counterparts are

$$\begin{aligned} \tilde{\pi }^j&=A\tilde{g}^j-B\tilde{r}-\eta x,\end{aligned}$$

(A.1)

$$\begin{aligned} \tilde{y}^j&=\alpha A\tilde{g}^j-\alpha B\tilde{r}+(1-\alpha \eta ) x. \end{aligned}$$

(A.2)

The first-order condition of politician j’s programme (15) is unchanged

$$\begin{aligned} \frac{\partial L^j}{\partial g^j}=0 \Leftrightarrow A\pi ^{j}+\mu g^j-\alpha A\lambda ^j =0. \end{aligned}$$

(A.3)

By taking expectations, we find the same solution as in the main text for expected public spending, namely

$$\begin{aligned} (g^j)^e = AW[\alpha (1-C)\lambda ^j + Br^e], \end{aligned}$$

(A.4)

while the unexpected component of public spending is such that

$$\begin{aligned} A \tilde{\pi }^j+\mu \tilde{g}^j=0. \end{aligned}$$

(A.5)

Using (A.1) and (A.5), it follows that

$$\begin{aligned} \tilde{g}^j:=\tilde{g}=\frac{A}{\mu +A^2}[B\tilde{r}+\eta x]. \end{aligned}$$

Note that the unexpected component of fiscal policy is independent of the type of politician who is elected. The same feature is true for the unexpected component of inflation and output-gap

$$\begin{aligned} \tilde{\pi }^j&:=\tilde{\pi }=-\frac{\mu }{\mu +A^2}[B\tilde{r}+\eta x],\end{aligned}$$

(A.6)

$$\begin{aligned} \tilde{y}^j&:=\tilde{y}=-\frac{\mu }{\mu +A^2}\left[ \alpha B\tilde{r}-\left( \frac{\mu +A^2}{\mu }-\alpha \eta \right) x\right] . \end{aligned}$$

(A.7)

The election probability of type-R politician is now

$$\begin{aligned} \hat{p}^R=\frac{1}{2}+\frac{h}{2}[((\pi ^D)^e+\tilde{\pi }^D)^2-((\pi ^R)^e+\tilde{\pi }^R)^2] -h\bar{\lambda }(\tilde{y}^D-\tilde{y}^R). \end{aligned}$$

(A.8)

As $\tilde{y}^D=\tilde{y}^R$ and $\tilde{\pi }^D=\tilde{\pi }^R,$ we rewrite

$$\begin{aligned} \hat{p}^R=\frac{1}{2}+\frac{h}{2}[((\pi ^D)^e)^2-((\pi ^R)^e)^2]+h\tilde{\pi }((\pi ^D)^e-(\pi ^R)^e), \end{aligned}$$

(A.9)

hence;

$$\begin{aligned} \hat{p}^R=p^{R}+\alpha h A^2W\varepsilon \tilde{\pi }. \end{aligned}$$

(A.10)

where $p^{R} = \frac{1}{2} + \alpha A^{2}W^{2}h \varepsilon [\alpha A^2\bar{\lambda }-\mu B r^e]=:p(r^e)$ as in the main text, and $\tilde{\pi }$ is defined in (A.6). As the interest rate policy of the central bank is now stochastic, we cannot compute directly $\frac{\partial \hat{p}^R}{\partial r},$ but we can compute the effect of changes in the expected and unexpected components of interest rate, namely $\frac{\partial \hat{p}^R}{\partial r^e}=p'(r^e)<0$ as in the main text, and $\frac{\partial \hat{p}^R}{\partial \tilde{r}}=\alpha h A^2W\varepsilon \frac{\partial \tilde{\pi }}{\partial \tilde{r}}<0.$ Therefore, both expected and unexpected increases in the interest rate reduce the chances of the conservative candidate in the election. This result generalizes those obtained in the main text. In particular, even if the central bank is a pure inflation hawker (namely, if $\tilde{\lambda }=0),$ such that the expected interest rate is $r^n$ and $p(r^n)=1/2,$ the election probability does depend on the stochastic component of interest rate policy, and the chances of the conservative candidate increase when the economy faces unexpected cuts in the interest rate.

Appendix B: United Kingdom: database description

See Tables 5, 6, 7 and 8.

Table 5 Vote share in general elections: difference between United Kingdom and Great Britain

Table 6 Political events and the month in which they occurred

Table 7 Incumbent government characteristics

Table 8 Summary statistics (1987M1-2015M12)

Appendix C: United Kingdom: robustness checks

As a robustness checks, we estimate our model using alternative measures of our explanatory variables and a methodology to control for a potential long memory process of the variable dGovernment_Approval in Appendix C.1. Moreover, we estimate our model with alternative pre-electoral periods in Appendix C.2. Finally, in Appendix C.3 we estimate our model on a broader time period (1987M1-2021M12 versus 1985M1-2015M12) and we show that the popularity of Liberal Democrats is not impacted by the studied mechanism.

1.1 C.1 Alternative measures of explanatory variables and long memory process within popularity ratings

Table 9 implements two robustness tests. First, it takes into account the potential endogeneity of $\textit{Unemployment}.$ To do so, we use the output gap as a proxy of the unemployment rate. The latter is computed by applying a Hamilton (2018)’s filter on monthly industrial production data provided by the OECD. Then, we compute the variable Output_Gap as the difference between the cyclical component and the trend component of industrial production obtained. As this variable is nonstationary, we consider its first difference (dOutput_gap). Regressions (13) and (16) show that our results are unchanged. Second, as the main interest rate may be correlated with inflation and/or unemployment, we use a more exogenous measure of the orientation of monetary policy. To this end, we regress the main interest rate on dInflation and dUnemployment in t and $t-1.$ Then, we consider the residuals of this estimation (variable RESID) as a more exogenous measure. Once again, our main results are not significantly different (see regressions 14 and 17), although the magnitude of the pre-election effect is smaller.

Table 9 Robustness—alternative explanatory variables (1987M1-2015M12)

Looking at Fig. 4, we can posit that our variable dGovernment_Approval is characterized by a long memory process. To control for the potential impact of this long memory process, we compute a heterogeneous autoregressive model following Corsi (2009). More precisely, we introduce two variables: Government_Approval_quarter and Government_Approval_year, which represent the moving averages of dGovernment_Approval for the last quarter and the last year, respectively. As described by Corsi (2009), these variables measure the past behavior of our popularity ratings on the medium run (Government_Approval_quarter) and on the long run (Government_Approval_year). Regressions (15) and (18) in Table 9 show that these variables do not significantly affect our results. Hence, the suspected long memory process does not drive the significance of the variable PreElection10XdBase_Rate.

1.2 C.2 Alternative pre-electoral periods

We perform our estimations with alternative measures of the pre-electoral period (with length measured in terms of months). Figure 6 depicts coefficients of the interaction terms between the first difference of the main interest rate and 12 different pre-electoral measures with 99% confidence intervals. More precisely, each point of each subfigure represents the coefficient of the interaction term between $dBase\_Rate$ and a dummy taking the value of 1 in the N months preceding the election (i.e., the $PreElectionNXdBase\_Rate$ with $N = 1,2,\ldots ,12).$ The subfigures represent estimations considering the full sample and two subsamples, depending on the party in office.

Fig. 6

Effect of an increase in the main targeted interest rate before a national election. Every coefficient is presented with 99% confidence intervals

These estimations confirm that, from 6 to 10 months before a general election, an increase in the interest rate significantly and negatively impacts the popularity of a Conservative incumbent.²⁴

In addition, we perform the estimations presented in Table 1 with these different pre-electoral measures from 1 to 10 months (see Table 10).

Table 10 Robustness: coefficients and standard errors of several interaction variables (1987M1-2015M12)

Table 11 Robustness: coefficients and standard errors of several interaction variables (1987M1-2021M12)

Regardless of the length of the pre-election period, the interest rate has a significant negative impact on the right-wing incumbent’s popularity. This feature is robust to the sample period used (see Table 11 for the period 1987M1-2021M12). We still note a negative impact of a rise in the interest rate on the conservative incumbent’s popularity.

1.3 C.3 Additional robustness checks

As additional robustness checks, we question the impact of the policy rate on the Liberal Democrats’ popularity and consider a broader timeframe (1987M1-2021M12). Table 12 shows that the popularity of the Liberal Democrats is never significantly impacted by the orientation of monetary policy before an election. This result holds on the 1987M1-2015M12 and the 1987M1-2021M12 study periods. Tables 13 and 14 present the estimation of Tables 1 and 2, respectively, using an alternative period (1987M1-2021M12). Our results are qualitatively unchanged and we still observe a significant negative impact of the variable PreElection10XdBase_Rate when the incumbent is Conservative.

Table 12 Robustness—Liberal Democrats’ popularity

Table 13 Robustness—main results—government popularity, cyclical components [Government_Approval] on an alternative sample period (1987M1-2021M12)

Table 14 Robustness—main results—political parties’ popularity, cyclical components [hamConservative and hamLabour] on an alternative sample period (1987M1-2021M12)

Appendix D: Panel database

See Tables 15, 16, 17, 18 and 19 and Figs. 7 and 8.

Table 15 Interest rate considered as the main targeted interest rate by country—panel database

Table 16 National elections considered—panel database

Table 17 Conservative and Labour parties by country—panel database

Table 18 Summary statistics—panel database

Table 19 Data sources used to compute popularity ratings—panel database

Fig. 7

Evolution of government popularity, cyclical components—panel database (2011M2-2021M12). Horizontal lines represent periods in which general elections were held. Dashed blue lines represent general elections won by a right-wing party (a type-R party) and plain red lines represent elections won by a left-wing party (a type-D party) (colour figure online)

Fig. 8

Evolution of the main targeted interest rate—panel database (2011M2-2021M12). Horizontal lines represent periods in which general elections were held. Dashed blue lines represent general elections won by a right-wing party (a type-R party) and plain red lines represent elections won by a left-wing party (a type-D party). The main targeted interest rate for each country is detailed in Table 10 (colour figure online)