Discretionary Monetary Policy Under Checks and Balances

Delegation to a central bank is traditionally analyzed within the context of the single-veto player model outlined above, where the central bank has no informational or other advantages over the government. Given this, there is no obvious reason why a single veto player would not always override any central bank decision that diverged from the veto player's preferred outcome in any period.²

There are two veto players, v₁ and v₂. Figure 1 shows their preferences relative to each other, with v₁ preferring a lower rate of inflation. Some status quo rate of inflation sq is located between the preferred rates of the two veto players. It is well understood, at least since Shepsle and Weingast (1981), that one must specify how the veto players make decisions—that is, what are the agenda-setting powers of each? We assume throughout the paper that there is one player with agenda-setting authority—the ability to make a take-it-or-leave-it offer to the other player. For example, in some countries, the executive branch can propose a candidate for central bank governor to the legislature, which must vote the choice up or down.³

In the absence of a central bank, policy does not shift from sq since there is no policy that v₂ could propose that both veto players prefer to sq. The presence of a “conservative” central banker, cb, who prefers a lower rate of inflation than either veto player can therefore reduce inflation. Policy is made as follows: (i) The central bank chooses an inflation rate; (ii) v₂ can propose to override the central bank's inflation rate; and (iii) v₁ can either accept the override proposal, in which case it is implemented, or v₁ can reject it, in which case the central bank's rate is maintained. Given this series of moves and the alignment of preferences in Figure 1, the lowest rate of inflation that the central bank can propose without fear of override is the policy preferred by v₁, which is lower than the status quo.

What if v₁ or v₂ were the only veto player? Then policy would always be at their preferred outcome, with or without the central bank; any attempt by the central bank to propose a different inflation rate would be immediately overridden by the single veto player. Clearly, then, the presence of a central bank can change policy outcomes only in the presence of multiple veto players.

This model begs two basic questions that are key to analyzing delegation and credibility. The first is where the status quo rate of inflation originates. In this simple illustration the status quo is exogenous and given. Consistent with standard models of monetary policy, the default outcome we employ in the models below—the inflation that prevails in the absence of government action in the second period—is the set of price increases programmed by the private sector into its contracts in the first period.⁴

The simple model depicted in Figure 1 also provides no reason for v₁ to delegate to a conservative central banker in the first place. This paradox emerges because the model assumes away economic shocks. Again, a simple example confirms this point. Delegation would occur if, in some previous unspecified period of play, v₁ controlled both veto points. Veto player v₁ would then have a choice: to retain future control of monetary policy or to delegate to a conservative central banker. Veto player v₁ would then confront one of three scenarios in the future. This veto player could retain control of both veto points, in which case the presence of a central bank would make no difference, and his/her preferred inflation outcome would prevail. Or v₁ could retain control of one veto point, but again, the presence of a central bank would make no difference, since v₁ could veto any attempt to shift policy away from his/her preferred outcome. Finally, v₁ could lose control of all veto points. In this case, policy would shift to the outcome preferred by v₂, regardless of whether there was a conservative central bank. None of the three cases provides a justification for delegation.

The possibility of unpredictable economic shocks can, however, motivate the inflation-averse veto player to delegate authority. These shocks are fundamental to the analysis of political decision making in many policy areas, including Rogoff's (1985) analysis of monetary policy. To see how they would motivate v₁ to delegate, one can imagine an example in which v₁ prefers a 2% rate of inflation and, given that v₁ controls all veto points, simply decides to set inflation at that rate. Then, v₁ loses control of one veto point to a policymaker who prefers a 4% rate of inflation. Inflation does not change, however, since v₁ can still block all attempts to move away from the 2% rate. Finally, however, a shock occurs, such as an increase in oil prices that leads simultaneously to higher inflation and slower economic growth. At that point, both veto players prefer a higher inflation rate, say 4% for v₁ and 6% for v₂, that they believe would best allow them to minimize the impact of the oil price shock on economic growth.

It is straightforward to see that in this situation v₁ could have done better by delegating to a conservative central bank. Assume that v₁ is indifferent between inflation at 2% and inflation at 6%—both are two percentage points away from her preferred, postshock rate of 4%. Following the shock, and absent a central bank, v₂ therefore proposes a rate just less than 6%, which v₁ would accept. What if v₁ had previously delegated authority to a conservative central bank? The conservative central bank would have responded to the shock by setting inflation at 4%, v₁'s new preferred rate. Delegation in the presence of economic shocks therefore yields inflation outcomes closer to the preferred outcome of v₁.

This heuristic example demonstrates the importance of integrating two elements into a convincing model of delegation: the emergence of the status quo from the interaction of the private sector and government actors and the presence of economic shocks. The formal models we develop here do precisely this. We first analyze policy outcomes with multiple veto players in the absence of delegation.⁶

Given the decision-making rules and the default outcome, the policy game with two veto players is otherwise the same as in the Barro–Gordon model: Private-sector actors establish their inflation expectations and then write contracts that fix prices; a supply shock ε occurs, distributed uniformly over the range [−c, c]; and two political actors decide what inflation policy to pursue after observing expected inflation and the shock. Now there are two political actors, A and N (the agenda setter and the non-agenda setter), who seek to minimize their respective loss functions:

which is just equivalent to (1) after substituting in (2).

The more inflation-averse a veto player is, the lower is b_i. In analyzing when an additional veto player will make a difference to policy outcomes, Tsebelis (2002) emphasizes the importance of taking into account the orientation of veto player preferences with respect to the status quo outcome, the outcome that prevails if veto players are unable to make a decision. Where additional veto players have policy preferences identical to those of some current veto players, their presence does not change policy outcomes. The preferred outcomes of the two veto players diverge here because they have different preferences regarding the trade-off between income and inflation, b_A, b_N > 0 and b_A/b_N. We examine the cases where the agenda setter is both more and less inflation-averse than the non-agenda setter (b_A < b_N and b_N < b_A, respectively).

Since private actors write their contracts anticipating the actions of government, we proceed as usual by backward induction to establish what rate of price increase (expected inflation or status quo inflation) will be built into the private sector's contracts. In the last period, the political actors observe the price increases written into private-sector contracts and the supply shock. If they do nothing, the price increases, which are (once again) the default or status quo outcome confronting the veto players, will constitute final inflation. If the private sector were sure that the government would not act, then it would set expected inflation, or default inflation, equal to zero. However, the private sector does not observe the random supply shock and, therefore, cannot know for sure what decision the veto players will make upon observing the private sector's contracts.

Altogether there are four possible situations that the two veto players might confront, depending on the orientation of the status quo to the preferred inflation outcomes of the veto players. Assuming for now that the agenda setter is more inflation-averse (b_A < b_N), the agenda setter's preferred inflation outcome must always be less than that of the non-agenda setter's.⁷

Case I: π_SQ < π_A < π_N

If the supply shock is large and negative, both veto players seek to stimulate output with an expansionary monetary policy; both therefore prefer a rate of inflation that is higher than the status quo rate established in private-sector contracts. The non-agenda setter prefers the agenda setter's most preferred inflation policy to the default outcome. Therefore, the agenda setter makes an all-or-nothing proposal of his/her most preferred outcome and the non-agenda setter accepts it. Final inflation is therefore π_A.

Case II: π_A < π_SQ < π_N

In Case II, the supply shock leaves the output short of what the non-agenda setter prefers. His/her preferred policy is therefore a stimulative monetary policy leading to higher inflation than given by the status quo or default outcome. The agenda setter, on the other hand, believes that the output after the supply shock is high enough that some output can be sacrificed to attack inflation more aggressively; he/she prefers a lower inflation outcome than the status quo. Therefore, no inflation outcome is preferred by both the agenda setter and the non-agenda setter to the status quo outcome. The status quo therefore prevails.

Case III: π_A < π_N < π_SQ < 2π_N-π_A

If the supply shock is large and positive, both players prefer to sacrifice some output and pursue a more restrictive monetary policy that reduces inflation below the status quo inflation outcome. If, as in Case III, status quo inflation outcome is closer to the non-agenda setter's preferred inflation outcome than is the agenda setter's preferred outcome (that is, if π_SQ < 2π_N-π_A), the agenda setter must split the difference and propose an inflation policy that lies between her preferred outcome and the outcome most preferred by the non-agenda setter. Specifically, the agenda setter successfully proposes inflation outcome π such that π_N-π = π_SQ-π_N, or 2π_N-π_SQ = π.⁸

Case IV: π_A < π_N < 2π_N-π_A < π_SQ

Under Case IV the supply shock is sufficiently large and positive that the preferred inflation outcomes of both veto players are far to the left of (much lower than) the status quo inflation outcome and the non-agenda setter prefers even the agenda setter's preferred inflation outcome to the status quo. The agenda setter therefore proposes π = π_A and the non-agenda setter accepts.

The four cases describe the possible government actions in the second period of the game conditional on the decisions of the private sector in the first period. Since the supply shock affects which case will emerge, and since the private sector does not observe the supply shock prior to writing its contracts, the private sector must therefore take into account the possibility that any of the cases can occur. Private actors must therefore assess, for any inflation rate that they factor into their contracts, the probability of each of the four cases emerging. They calculate expected inflation as the solution to (5), where the q_i's are the probabilities that the government will choose each of the four different inflation outcomes after the shock is realized:

The solution to (5) is complex, since the preferred inflation rates and the q probabilities are all themselves a function of expected inflation. Appendix 2 derives the solution for expected inflation when the agenda setter is less inflation-tolerant than the non-agenda setter, or b_A < b_N (Eq. [A.12]), and when the agenda setter is more inflation-tolerant (Eq. [A.13]). Although the solution is highly nonlinear and closed-form solutions do not exist, numerical simulations using plausible parameter values allow us to make three propositions.

First, the larger is the variance of the supply shock, and therefore the greater the uncertainty surrounding the default or status quo outcome, the less influential is the non-agenda setter in setting policy. Veto players matter less in a more uncertain environment. This is intuitively clear, since the larger is a shock, the more likely the non-agenda setter will prefer the agenda setter's preferred inflation outcome to the status quo.

Second, the addition of a second, non-agenda-setting veto player increases policy credibility, with the effect rising in the magnitude of the preference differences between the two veto players. Of course, if the two veto players have identical preferences (b₁ = b₂), the second veto player has no influence at all on outcomes. Multiple veto players therefore can increase policy credibility as well as policy stability.

Third, policy credibility is sensitive to the decision making rules. If the agenda setter is the veto player who has the least incentive to renege on policy commitments (in this case, is most inflation-averse), the addition of a second veto player has little effect on credibility. In the reverse situation, multiple veto players have a significant effect on policy credibility.

The numerical simulations generating the second and third propositions, which are most relevant to the delegation case, are illustrated in Figure 2. Differences in preferred outcomes of the two veto players are modeled as mean-preserving polarization of the preferences, where “mean-preserving” is defined as an increase in |b_N-b_A| holding (b_N-b_A)/2 constant. The addition of a second veto player always reduces expected inflation provided that the interests of the two veto players are divergent; as the third result predicts, this effect is much stronger when the non-agenda setter is less inflation tolerant than the agenda setter. This two-veto player outcome is the benchmark against which to compare the effect of an independent central bank subsequently.

The Introduction of an Independent Central Bank

The key attribute of an independent central bank is that it, rather than the private sector, sets the default rate of inflation that political veto players must accept if they choose not to override it. After the private sector has acted and the supply shock has occurred, the central bank determines an inflation policy and it is this inflation policy that prevails if the two veto players do not act. The two veto players determine whether to accept the central bank's policy. If they overturn it and then do nothing, then the private sector's contractual price increases become final inflation. Consequently, the policy that prevails should they revoke the central bank's independence is precisely the policy derived earlier when there is no central bank. This policy is labeled π_CH, the inflation policy under checks and balances. The central bank has an effect on policy, therefore, to the extent that it can change the default or status quo rate of inflation relative to what would prevail in its absence.

The central banker's loss function and preferred inflation policy are, respectively,

and

Three assumptions make the analysis more compelling. First, the central bank is assumed to be more inflation-averse than either of the political actors (b_CB < b_A,b_N). Second, for simplicity, the central banker always prefers any inflation outcome lower than the inflation that would prevail were he/she to be overridden, π_CH. Third, to avoid trivial cases, the most inflation-averse veto player prefers the outcome under checks and balances to the most preferred outcome of the central banker. The third assumption means that central banker's preferred inflation outcome, π*, is such that π_A − π*≥ π_CH-π_A when b_A < b_N and π_N − π*≥ π_CH-π_N when b_N < b_A.⁹

Since π_CH, the inflation policy that the two veto players can agree on if they override the central bank, is a key constraint on central bank decision making, the same four cases that affect the calculation of π_CH also influence the policy choice of the central bank. Private actors therefore set expected price increases, establishing expected inflation, using Eq. (8), for b_A < b_N, following the same logic as for Eq. (5).

In the earlier section, the inflation rate contracted by the private sector was labeled π_SQ, to denote that it constituted the status quo outcome. Since this is no longer the case in the presence of a central bank, Eq. (8) labels the private sector's inflation rate π^e_CB, to denote that it is the expected inflation calculated by private actors in the presence of a central bank.

The key difference in Eq. (8) is that the inflation policy that prevails under each of the four cases is simply the override-proof policy chosen by the central bank after having observed the shock. For example, consider a supply shock such that Case I is realized, π^e_CB < π_A < π_N. If the central bank were overridden, the agenda setter would propose and the other veto player would accept the agenda setter's preferred inflation outcome, or π_CH = π_A. Knowing this, the central banker can do no better than to propose the agenda setter's preferred inflation outcome, π = 2π_A-π_CH = π_A. In Case II, the inflation outcome in the event of an override of the central bank's proposal is the default outcome, or π_CH = π^e_CB, so that the central bank proposes π = 2π_A − π^e_CB, the inflation outcome that is no further from the agenda setter's preferred outcome than the default outcome.

If the non-agenda setter is less tolerant of inflation, b_N < b_A, the central bank then sets policy such that the non-agenda setter has no incentive to agree to reverse the central bank. The cases remain the same as before, but the override-proof policy is now the one that leaves the non-agenda setter just indifferent between the central bank's policy and the pursuit of an override. The derivation of expected inflation in this case is given in Appendix 3. The solutions for expected inflation in both cases (the agenda setter less and more tolerant of inflation) are derived in Appendix 3 and given by Eqs. (A.14) and (A.15).

As before, there is no simple closed-form expression for expected inflation with an independent central bank, but numerical simulations can be used to generate three relevant propositions. First, delegation of monetary policy to a more conservative central banker always lowers expected inflation when there are multiple veto players. Second, the central banker is able to pursue a lower inflation policy as polarization of the political actors increases. Third, just as the addition of a second veto player makes the most difference when the agenda setter is more tolerant of inflation, so too does the addition of an independent central bank.

The numerical simulations illustrated in Figure 3 yield each of these predictions. In each case, an independent central bank achieves lower inflation than two veto players alone, and the difference in inflation outcomes grows as polarization between the two veto players grows, but the biggest impact of an independent central bank occurs when the agenda setter is more inflation-tolerant. The latter claim can be verified by comparing the distances between the top and the bottom lines and between the two middle lines in Figure 3.

Data

Following standard practice, in our tests of hypotheses 1 and 2 we use average levels of inflation as our dependent variable and, more specifically, the log of average inflation.¹⁴

Our measure of legal central bank independence (CBI) was developed by Cukierman, Webb, and Neyapti (1992), based on 16 characteristics of central bank statutes such as the term of office for the governor, provisions for his or her replacement, limits on central bank lending to government, and procedures for resolution of conflicts between central bank and government.¹⁶

These authors have also collected data on replacement of central bank governors. Cukierman, Webb, and Neyapti (1992) argue that high turnover of central bank governors is indicative of low independence, and they show that the rate of turnover is positively and significantly correlated with inflation in a sample including both developed and developing countries.¹⁷

With respect to checks and balances in government, one would ideally like to have separate measures on the number of political actors who exercise veto power over monetary policy, their inflation preferences, and their respective agenda-setting power. These variables are not available in the cross-country setting that we seek to examine, so we instead employ two recently recently developed measures of checks and balances.¹⁹

A second measure of checks and balances, developed by Keefer (2002), also has the advantage of being based on objective criteria and of capturing the existence of coalition governments or divided control of two chambers in a bicameral system. In addition, unlike political constraints, the checks variable is explicitly incremented when a party in the government has an economic policy orientation closer to that of the main opposition party than to that of the party of the executive. More generally, checks is based on a formula that first counts the number of veto players, based on whether the executive and legislative chamber(s) are controlled by different parties in presidential systems and on the number of parties in the government coalition for parliamentary systems (as described in greater detail in Beck et al. 2001). The index is then modified to take account of the fact that certain electoral rules (closed list vs. open list) affect the cohesiveness of governing coalitions. The variable is constructed based on variables in a new database of political institutions assembled by Beck et al. (2001) and is available for the last two decades of our sample (1972–89). Since the effects of checks and balances hypothesized in the model are likely to be strongest at lower levels of checks than at higher levels, we use a log version of checks, log checks, in our regressions.²¹

The measures developed by Keefer (2002) and Henisz (2000) capture the number of veto players as well as can be done in a cross-country setting that includes both OECD and developing countries. Both of these indicators rise with the number of veto points (depending upon the number of legislative chambers) and fall when the veto points are occupied by the same political party (depending on whether majorities are multiparty coalitions).

There is no measure available that precisely assesses the policy distance between veto players. However, to test Prediction 2 we use a reasonable substitute, the political polarization variable from the Database of Political Institutions (Beck et al. 2001). The database reports the economic orientation of the four largest parties and the executive in each country, based on whether the data sources indicated parties as having an economic orientation that was left, center, or right. Pairwise comparisons of economic orientation are made; the difference in economic orientation between the pair of veto players exhibiting the largest difference in orientation is the political polarization measure. Given the way it is constructed, the political polarization measure is a reasonable proxy for the degree to which the preferences of veto players diverge. In addition, however, since it takes a value of zero for systems with only one veto player, it is simultaneously a proxy for the number of veto players in a country.

We use one additional control variable, openness, measured as imports of goods and services divided by GDP (International Financial Statistics). This follows Romer (1993), who argues that as imports increase as a share of total consumption, policymakers have less of an ex post incentive to inflate. We also follow Cukierman, Webb, and Neyapti (1992) and include decade dummy variables to control for unobserved characteristics specific to each time period.

Testing Predictions 1 and 2: Checks and Balances, Polarization, and Central Bank Independence

In order to examine whether legal CBI has a stronger negative impact on inflation in countries with multiple veto players, and in cases where veto players are more polarized, we use a model with interaction terms that allows the marginal effect of CBI on inflation to vary across political systems. The general form of Regressions 2 and 3 in Table 2 is shown in Eq. (9), where pol. inst. represents our alternative proxies for the number of veto players in government and the degree of polarization between veto players.

The interaction term is predicted to have a negative coefficient. The net effect of CBI, given by β₁ + β₃ * (pol. inst.), should be to reduce inflation only at high levels of checks and balances (or at high levels of polarization).

Regression 1 in Table 2 suggests that, although there is some theoretical ambiguity about the effects of multiple veto players on inflation, in practice the introduction of multiple political veto players may be to reduce inflation: The coefficient on log checks is negative and significant. As in previous research, legal CBI has no significant impact on inflation when it enters linearly in a sample including both OECD and developing countries, and openness is significantly and negatively correlated with inflation. The variable political constraints is similarly significant, economically and statistically, when substituted for log checks in the Regression 1 specification.

Regression 2 suggests that the effect of CBI on inflation is conditional on the presence of checks and balances in government. Consistent with Prediction 1, the interaction term CBI *log checks is negative, and an increase in CBI is estimated to have a negative effect on inflation only at high levels of checks and balances. Regression 3 produces very similar results. The interaction term CBI * polcon is negative and statistically significant.

In Regression 4 we extend the inquiry to consider whether the effect of CBI is influenced by both the number of veto players and their degree of polarization. The variable political polarization captures exactly this effect, since it is zero when there is only one veto player, while in systems with more than one veto player political polarization can take on a positive value ranging from zero to two as veto player preferences over economic policy diverge.

The coefficient on the interaction term CBI * political polarization in Regression 4 is negative and statistically significant, supporting Prediction 2. This result in Regression 4 is also robust to the linear inclusion of log checks to the regression. Figure 4 plots the estimated effect of an increase in CBI, together with its standard error, and different levels of political polarization. The model predicts that an increase in CBI will have a negative impact on inflation only at relatively high levels of polarization (political polarization > 0.6). This corresponds roughly to the mean value of political polarization for the group of countries in which political polarization is not equal to zero. This result is consistent with the predictions of our theoretical model, as it suggests that multiple veto players with divergent preferences must be present for CBI to have a significant impact.

The results in Regressions 2 and 3 bear on an unresolved puzzle in empirical work on CBI. A number of papers have found a statistically significant relationship between legal measures of CBI and inflation in advanced industrialized economies.²²

Table 3 reports the estimated effect of a one π standard deviation increase in CBI (0.13) at different levels of each of our proxies for checks and balances used in Regressions 2 and 3. It also reports the conditional standard errors of these effects, which depend on the variance of both the linear and the interactive CBI terms, as well as their covariance. To facilitate comparison, we have chosen values for each of the two variables that would reflect a political system with one, two, or three veto players. Table 3 provides additional evidence that in each case an independent central bank reduces inflation only where multiple veto players are present, as the theory predicts.

Cases of Conflict from the Bundesbank and the Fed

Qualitative evidence emerging from specific instances of confrontation between central bankers and political decision makers provides further support for the analysis here. For example, during the Gürzenich Affair of 1955–56, the German Bundesbank was able to insist on a course of action strongly opposed by Chancellor Konrad Adenauer precisely because members of his cabinet, Ludwig Erhard and Fritz Schäffer, supported the central bank. As Berger and de Haan (1999) document, although they were not formal veto players, these two were highly influential within the Christian Democratic Party and, more to the point, represented different interests. Adenauer sought a looser money supply to satisfy the demands of major industrial interests who were important contributors to the party; the two cabinet members were able to rely on broader public support for the autonomy of the Bundesbank. Two elements of this story are especially important. First, the two cabinet ministers were de facto veto players. Although public opposition to restricting the authority of the Bundesbank eventually emerged, internal cabinet opposition to Adenauer's actions seems to have been key to representing public opinion against the interests of industrialists. Only if they had de facto veto power could these two ministers have accomplished this. Second, although it is inaccurate to speak of the “agenda-setting” authority of Adenauer versus his cabinet ministers, there is little doubt that his bargaining power was greater than that of his two ministers, given his status in the party and as Chancellor. Despite this—but as the model predicts—the Bundesbank was able to get nearly everything it wanted in the conflict with the support of these two ministers.

Wooley (1984, 144–47) documents several instances of conflict among the U.S. Federal Reserve, Congress, and the President. In 1975, at the trough of a two-year-old recession, Federal Reserve policy was markedly more conservative than either the Republican President Ford or the Democratic Congress desired. At the outset of the 1975 Ninety-Fourth Congress, several committee chairmen were replaced and the more liberal Henry Reuss took over the House Banking Committee, assuming a position of agenda control. He immediately introduced legislation requiring the Fed to expand the money supply at a rate of 6% per year. Within days, however, he agreed to replace this proposal with a substantially weaker one, calling only for the Fed to increase the money supply in the first half of 1975 and to maintain long-run growth in the money supply consistent with long-run potential output growth. This bill was passed and signed by the President. Since the bill had no binding effect (Wooley [147] describes it as “toothless and compromised”), one is left to ask why stronger legislation did not succeed. The analysis here suggests that the move toward a less inflation-averse agenda setter had little effect on policy because other, more conservative veto players remained the same (e.g., the Republican president). These veto players were constrained by concerns about the inflation outcome that would have prevailed should they have overturned Fed policy. Knowing this, the Fed was able to pursue a more aggressive policy than the veto players preferred.

Testing Prediction 3: Checks and Balances, Polarization, and Political Influence on Central Banks

If our primary thesis is correct, then in addition to influencing observed policy outcomes, checks and balances and polarization should also have an impact on the degree to which a central bank is subject to political influence. We examine this issue by proxying for political influence with a variable collected by Cukierman and Webb (1995), which measures the frequency with which central bank governors are replaced in the six months following elections. As argued above, while even this variable is likely to be an imperfect proxy for political influence, it is a better proxy than measures that focus on the overall rate of turnover for central bank governors. We test two alternative versions of the hypothesis.

First, to the extent that the decision to replace a central bank governor must be agreed to by more than one veto player, then we should expect more infrequent turnover of central bank governors in countries with multiple veto players in government, in particular, in countries where there is evidence that there are multiple veto players with divergent preferences. This might be the case, for example, in a country with a coalition government where all coalition members might have a say in any decision to replace the central bank governor. Because our variable governor turnover is a direct measure of the frequency of policy change, this is a straightforward prediction.²³

We also test a second version of Prediction 3. Logically it should be the case that the replacement of governors will become more difficult when a central bank's statutes offer tenure guarantees and when there are multiple political veto players with different preferences over economic policy. The most relevant tenure guarantees would involve a long term of office and provisions making it difficult for the central bank governor to be replaced in the case of a conflict over policy. In countries where decisions over replacing the central bank governor would normally be made by a single veto player, then legal guarantees of tenure can have the effect of increasing the number of veto players whose accord is necessary to replace the central bank governor. So, for example, in the absence of tenure guarantees, the executive alone might be able to dismiss the central bank governor, but once such guarantees are in place, both the executive and the legislature would need to agree to pass a law removing the tenure guarantees before the governor could be removed from office. In contrast, in countries where there is only one veto player, it may be as easy to alter legal restrictions on dismissal as it would be to remove a central bank governor in a country without such restrictions. Given that we lack accurate data on the number of veto players that would normally be able to block an attempt to remove the central bank governor, in what follows we test both variants of Prediction 3.

Equation (10) is the specification of a test of this hypothesis (recalling that CEO measures legal protection of the tenure of the central bank governor and governor turnover measures the frequency with which central bank governors are replaced following changes in government).²⁴

Table 4 reports the estimates from six versions of (10). We begin in Regressions 1–3 by asking only whether governor turnover following elections is less frequent in countries with multiple veto players in government. These regressions therefore test the first variant of Prediction 3, that checks and balances in government should directly reduce the rate of governor turnover. The coefficients on both proxies for the number of veto players are negative and significant, as is the coefficient on political polarization. Political interference with central banks is less frequent in countries with multiple veto players having divergent preferences.

In Regressions 4–6 we consider the second variant of Prediction 3: Tenure guarantees should be more effective in reducing governor turnover in countries where there are multiple veto players in government and where veto players are polarized. That is, we expect the interaction term pol. inst. * CEO to be negative. In each of the three regressions the interaction term has the expected negative sign and is statistically significant for log checks and political constraints. An increase in tenure protections CEO reduces governor turnover only in the presence of multiple veto players (or polarized veto players in the case of political polarization). However, the conditional standard error of this effect is quite sizable in all regressions—even at very high levels of checks and balances, it is impossible to reject the hypothesis that the effect is zero. We therefore find strong evidence to support the conclusion that the presence of multiple polarized veto players reduces political influence on central banks and significant, but weaker evidence that multiple veto players increase the effectiveness of legal protections for central bank governors.

Robustness

A number of possible concerns could be raised about our core empirical results. These include the possibilities that they are driven by serial correlation, by the endogeneity of legal CBI to inflation or lagged inflation, by the effect of omitted variables, or by outliers. The evidence presented in this section indicates that these do not provide plausible alternative explanations of our empirical findings.

Results of Lagrange multiplier tests suggest that serial correlation of the error terms might have led to inconsistent estimates of standard errors in the tests in Table 1. However, the results in Table 1 are robust to controls for serial correlation (that is, the inclusion of log inflation, lagged one period). At the same time, it is possible that CBI is endogenous to inflation. Empirically, in our sample we find little evidence that central bank independence is endogenous to lagged inflation; nor is CBI endogenous to contemporary inflation.

A potential omitted variable in our regressions is income per capita. A country's level of income might simultaneously influence both inflation and the efficacy of an independent central bank. However, when entered into our Table 1 regressions, the log of real GDP per capita is never significant and leaves the estimates of the interaction terms nearly unchanged. Given that our checks and balances measures are highly correlated with levels of per capita GDP, we also consider whether the significance of our interaction terms CBI * checks simply reflects the previous finding that legal CBI has a more negative effect on inflation in wealthy countries. To assess this possibility, we compare the explanatory power of the specifications in Table 1 with a specification that substitutes logGDP and logGDP * CBI for the respective checks variables and checks * CBI variables. Using J tests first proposed by Davidson and MacKinnon (1981), we rejected the GDP specification in favor of the original specifications in Table 1, Regressions 2–4. In Regression 5 we rejected the political polarization specification in favor of the GDP specification.²⁵

Our results are also robust to two other variables omitted in Table 1, and to the exclusion of influential outliers.²⁶