Political and economic determinants of interest rate behavior: are central banks different?

Johnson, David R. ; Siklos, Pierre L.

I. INTRODUCTION

Do politicians manipulate the instruments of macroeconomic policy to improve their electoral prospects or in response to the preferences of their own party? These questions are addressed, but not yet resolved, by the literature on political business cycles.(1)

To generate a political business cycle, politicians must control the instruments of macroeconomic policy. Monetary policy plays an important role in the determination of inflation, output and employment, the objects of concern to politicians and voters. Yet, politicians do not directly control the instruments of monetary policy in industrialized countries. Instead, these instruments are controlled by the central banks, which vary widely across countries in their degree of statutory independence. We wish to examine empirically whether monetary policy is influenced by the timing of elections or a change in the party in power after controlling for unexpected changes in economic fundamentals. Our paper assesses, therefore, one aspect of independence, namely whether a central bank makes monetary policy decisions without measurable political influence.(2)

In this study we attempt to look at what central banks actually do as opposed to what they are legislated to do, to distinguish between the form and substance in discussions of central bank independence as in Mayer [1976].(3) Our paper enlarges and improves on Cowart [1978] and Woolley [1983], both of whom studied the role of political factors in monetary policy for a sample of industrialized countries. Our methodology is to compare the responsiveness of interest rate changes to economic and political factors in seventeen countries. As we discuss below, there are differences in opinion about whether an interest rate variable or a monetary aggregate is a better representation of monetary policy actions.

We compare the results of qualitative measures of differences in central bank design - e.g., rankings of banks in order of independence as determined by legislative criteria - with the results of a quantitative comparison of differences in central bank behavior. This exercise should help identify the important qualitative features in the reform of existing central banks or in the design of new central banks.(4)

In our empirical work the relevant relationships are estimated with a recursive vector autoregression technique to control for temporal instability in the economy between 1960 and 1990 and the endogeneity of the variables in question. This methodological innovation has important consequences for the understanding of reaction functions and, more generally, for measuring the impact of macroeconomic shocks on monetary policy. We also stress results for separate subsamples of fixed and flexible exchange rates as well as for the full sample. Our methodology measures the influence of politics on monetary policy relative to the choices and information actually available to the central bank when the monetary policy decision was made.

Our study does not measure the influence of politics on the instruments of fiscal policy as in Alesina et al. [1991]. Our study also does not measure the direct influence of fiscal variables on the instruments of monetary policy, although there is some evidence on this issue in, for example, Parkin [1986] and Burdekin and Wohar [1990]. Burdekin and Wohar estimate the relationships between deficits, money base growth and inflation for a sample of countries and time period comparable to ours. All these studies find some evidence that central banks do accommodate fiscal policy by financing deficits via money growth, especially since the end of Bretton Woods. We return to the issue of fiscal policy in considering the interpretation of our results but stress that the effects of fiscal policy are partly captured in our specifications because we measure monetary policy relative to the state of the economy.(5)

A brief review of the literature which measures statutory central bank independence is given in section II. We discuss issues in measuring the role of political factors in monetary policy in section III. The lessons of that literature are used in specifying the econometric model in section IV. Results are presented and discussed in section V. Section VI concludes.

II. QUALITATIVE MEASURES OF CENTRAL BANK INDEPENDENCE: A REVIEW

The aim of this literature is to describe institutional arrangements of central banks and how these arrangements influence economic performance. Fair [1980] lists, for twenty central banks, the wide variation in legislated goals, procedures to resolve conflicts with politicians, terms of office of central bank governors and board of directors, accountability to the legislature, and ownership of the shares. Bernanke and Mishkin [1992] also use the case-study approach to examine central bank behavior in six industrialized countries that are included in our sample. They conclude that central banks do not follow consistent targets in the short-run for either money or interest rates, a result similar in spirit to our results.

Several studies quantify the differences between central bank institutional arrangements. In the most comprehensive study of central bank institutional arrangements, Cukierman, Webb and Neyapti [1992] create an index of independence based on differences in central bank laws for a very large sample of countries. The work continues in Cukierman [1992, ch. 19] to encompass even more countries and to refine the earlier indices of central bank independence. Walsh [1993] criticizes this work for the lack of robustness in the supposed (negative) correlation between inflation and central bank independence and the failure to adequately define what is meant by independence. Goodhart [1994] criticizes Cukierman's work because it suggests that central bank independence is determined by a large number of characteristics when, empirically, this does not seem to be the case. Moreover, Fischer [1994] finds an apparent conflict between expert opinion which tends to place less emphasis on the role of legislation in influencing inflation rates and the independence index which stresses the negative correlation between inflation and statutory independence.

Parkin [1986], Grilli, Masciandaro and Tabellini [1991], and Burdekin and Willett [1991] also create indices of central bank independence based on institutional characteristics. Grilli et al. define independence as the degree to which a central bank enjoys the freedom to accomplish a specified objective of monetary policy free from political influence.(6) Four such rankings appear in Table I. These rankings have been used to consider two important questions. Parkin [1986] and Burdekin and Laney [1988] ask: Is a more independent central bank associated with a lower government deficit? Cukierman [1992], Burdekin and Willett [1991], Alesina [1988], and Alesina and Summers [1993] use rankings to argue that average inflation rates are lower when central banks are more independent.

There is a significant literature which discusses the usefulness of these rankings (see the survey in Pollard [1993]). Fischer [1994] argues for a distinction between goal independence, which is a function of how precisely defined central bank objectives are, and instrument independence, which is obtained when a central bank has full discretion over the conduct of monetary policy. These papers argue that the absence of goal independence is a key determinant of inflation performance for the same set of countries considered by Grilli et al. Thus, for example, the Reserve Bank of New Zealand has no goal independence, because it must adhere to a rigid inflation objective, while the Swiss National Bank has considerably more such independence because its objectives are much broader. Although Alesina and [TABULAR DATA FOR TABLE I OMITTED] Summers [1993] conclude that more independent central banks deliver lower inflation, they also find no relationship between statutory independence and real economic performance. The latter result is similar in nature to our findings. Alesina and Grilli [1991] stress the importance of such qualitative factors in designing a European central bank. However, Cargill [1989], after comparing the U.S. and Japanese experiences, points out the dangers of focusing too narrowly on the legislation governing central bank behavior as a guide to predicting economic performance. However formal the mechanism which grants some degree of independence to a central bank, there exist less formal but equally powerful mechanisms to override such constraints. The Coyne affair in the history of the Bank of Canada is a particularly fascinating example (Gordon [1961]). A second example, the Bundesbank, has its independence constrained by virtue of the decentralized nature of its policymaking body rather than by its legal structure vis-a-vis the government (Hochreiter [1990]). This feature has implications for the institutional design of the proposed European Central Bank.

As Table I makes clear, there is relatively little agreement in the rankings across the four studies, except that Switzerland and Germany are viewed as the two most independent central banks in all these studies.(7) An additional difficulty with rankings dominated by inflation performance is that statutorily dependent central banks such as Japan's perform almost as well as legally independent banks such as those of Germany and Switzerland with respect to inflation.(8) These issues motivate our study to try to measure what central banks do in response to political events as a means of providing information about central bank independence.

III. POLITICAL INFLUENCE ON CENTRAL BANK BEHAVIOR: THE REACTION FUNCTION APPROACH

There is a long history, beginning with Reuber [1964], of estimating and interpreting central bank reaction functions (see also Alt and Chrystal [1983, ch. 6]). A reaction function measures how the instruments of monetary policy, the vector [I.sub.t], react to the central bank's information about the economy as of period t-1, a vector labelled [Z.sub.t-1], and political factors, [P.sub.t]. The equation

(1) [I.sub.t] = B(L)[Z.sub.t-1] + C(L)[P.sub.t]

can be derived formally from an optimal control problem where the policymaker has quadratic preferences over economic outcomes and a linear model of the economy. It is important to note that the coefficients in the reduced form B(L) are a combination of coefficients incorporating the policymaker's model (the relationships between [Z.sub.t] and [I.sub.t], the central bank's information and its policy instruments), as well as the policymaker's weights on achieving different objectives, and even the values of the objectives. Early researchers (Christian [1968] or Havrilesky [1967]) stressed the temporal instability of the reaction function. Also, since (1) is a reduced form, its estimation does not directly reveal policymakers' preferences. This identification problem, and the problem of temporal instability, makes reaction functions difficult to interpret. Khoury [1990], who examines forty-two previously estimated reaction functions for the Federal Reserve, found them to be sensitive to arbitrary details of the specification. Beck [1990] finds that although a politically motivated monetary cycle exists, it is not easy to find an econometric specification which can consistently be used to determine the significance of such cycles.

In spite of these difficulties, reaction functions have been used to ascertain the impact of political influence on central bank behavior. Our study combines two approaches used by others in measuring the influence of political factors, in reaction functions.(9) Nordhaus [1975] argues that monetary policy is looser before an election to increase the probability of re-election. Hence, [P.sub.t] is a dummy variable active several quarters before elections. A second approach, suggested by Hibbs [1977] asks: Does a politically influenced central bank react more to increases in unemployment and less to increases in inflation during periods of "left-wing" government than in periods of "right-wing" government? Such a possibility presumes both that the central bank is open to political influence and that left-wing and right-wing parties are different so that the coefficients in the political function vary with the party in power. Chappell and Keech [1986] estimate and simulate such a model for the United States.

The significance of electoral and political variables in equations like (1) varies widely. Most of the literature deals with the behavior of the U.S. Federal Reserve and its response to political factors. Allen [1986] finds no influence from a variety of political and electoral variables on the growth of the monetary base. Beck [1984; 1987] considers both the behavior of the federal funds rate and the growth of base money and finds little evidence of pre-election loosening of the instruments of monetary policy.

There is slightly more evidence of looser monetary policy during Democratic (left-wing in the American context) presidential administrations. Havrilesky [1987; 1988] finds higher monetary growth under Democratic presidents. It is important to be cautious in interpreting reaction functions where the instrument of monetary policy is money growth. For most central banks, money growth is not the direct instrument of monetary policy - at most it is a target.(10) Beck [1987], Hamburger and Zwick [1981], and Fair [1978] provide some evidence that the appearance of a political cycle in the money supply is a result of Federal Reserve accommodation of fiscal policies which differ by presidential administration. Hakes [1990], in studying the minutes of the Federal Open Market Committee, presents evidence that Federal Reserve policy is influenced by both the Federal Reserve chairperson as well as the U.S. President.

Overall, the evidence in favor of political influence on monetary policy variables directly controlled by the Federal Reserve is mixed. Woolley [1983, 335] refers to this finding as the "macropolitics paradox." Thus, while there is considerable belief in political influence on macroeconomic outcomes, empirical evidence for direct political influence on monetary policy is not conclusive.

There are relatively fewer studies which measure the extent of political influence on central banks other than the Federal Reserve. A sampling of such works includes Frey and Schneider [1981] who find some evidence of Bundesbank accommodation of an expansionary fiscal policy.(11) Cowart [1978] finds some evidence that, when measured by changes in the discount rate, socialist (left-wing) European governments respond more to unemployment. Hodgman and Resek [1983] estimate comparative reaction functions for Germany, France, Italy and the United Kingdom. Political factors are not explicitly considered, but there is some analysis of politically motivated subsamples. Woolley [1983] does add political factors to this analysis, and he finds more effect of political factors in France and the United Kingdom than in Germany. Our study extends the non-American evidence to a larger sample of countries, a longer time period, and a wider variety of exchange rate arrangements. We consider the interest rate as the instrument of monetary policy, are more careful about informational considerations, and control for the state of the economy.

IV. ECONOMETRIC SPECIFICATION

Background

Three issues stand out in estimating a central bank reaction function: the choice of an instrument of monetary policy and the objectives of monetary policy; the issue of temporal instability; and the specification of political and electoral variables. One of our special concerns is to have specifications which allow comparability across countries.

A market-determined short-term interest rate is chosen as the central bank's instrument of monetary policy. Our second choice was the central bank discount rate, used in two of the seventeen countries, Ireland and New Zealand, where no market-determined rate was available for the whole sample.(12) The central bank is assumed to achieve its desired average interest rate within a quarter. We considered and rejected several other monetary instruments as possibilities for this study. Narrow or broad monetary aggregates which include the liabilities of the commercial banking system are clearly not controlled by the central bank within the one-quarter sampling frequency used in this study. In many countries, fluctuations in interbank clearings, bank crises, and seasonality mean that even the monetary base is not fully under central bank control in the short term. Studying the monetary base would also require considerable institutional knowledge of all changes in reserve requirements or reserve accounting procedures in each country, an awkward task in a comparative study. The short-term interest rate is, however, generally comparable across countries.(13) Using the level or the rate of change of the exchange rate as a measure of monetary policy across countries was also rejected. There is no adequate model of the "normal" exchange rate, from which to assess the tightness of policy. Although central banks often manage exchange rate intervention, governments choose the exchange rate regime. The importance of the exchange rate varies with the openness of the economy, making intercountry comparisons difficult. Since we control for the influence of world interest rates on the domestic short-term interest rate, this may also give some information about the desired exchange rate. Therefore we believe that the signal-to-noise ratio on the stance of monetary policy is highest for an interest rate instrument. Put differently, individuals, interest groups and politicians are likely to focus on interest rate behavior, and not the state of money supply growth, in evaluating central bank or government actions (see also Bernanke and Blinder [1992]).(14)

We hypothesize that the central bank responds to unexpected changes in the unemployment rate and the rate of inflation. Both objectives appear in the legislated objectives for most central banks and are of concern to politicians. The instrument of policy, the short-term nominal interest rate, affects both unemployment and inflation as well as the electoral prospects of the party in power. In addition, studying interest rate behavior allows us to measure of the influence of the rest of the world on national monetary policy, an influence which we would expect to be stronger if nominal exchange rates are fixed. More importantly, perhaps, the dynamic relationships between inflation, unemployment and nominal interest rates are unlikely to be stable over time. Our response to this concern is the estimation of the model with a recursive vector autoregression. In each recursion a new observation is added and the model is reestimated but old observations are not discarded. Thus, the coefficients describing the time-series behavior of unemployment, inflation and interest rates are not assumed to be constant over the sample. The advantage of this methodology is that, unlike in previous studies of this kind, the central bank reacts to unexpected inflation and unemployment shocks using only information available at the time policy was made.

The question we consider empirically then is whether changes in interest rates are related to unexpected changes in inflation or unemployment rates, as well as to purely political events such as elections or partisan changes in government. We develop these ideas formally below.

The Conceptual Framework

We assume that the following vector autoregressive representation (VAR) characterizes the central bank's information about the economy:

(2) [Mathematical Expression Omitted].

The VAR generates reduced-form forecasts of the inflation rate, [Mathematical Expression Omitted], the unemployment rate, [Mathematical Expression Omitted], and the short-term interest rate, [Mathematical Expression Omitted], as the fitted values of the VAR using coefficients estimated with information available as of period t - 1. Seasonal variables and a trend were also added to the VAR. Political effects are omitted precisely because we ask how economic variables would be forecast from only their past history. The VAR represents the central bank's use of past information in a way which imposes relatively few restrictions on the data. The VAR methodology also recognizes the endogeneity of the variables in (2). To the extent that changes in lagged fiscal policy change lagged interest rates, unemployment rates and inflation rates, the effects of fiscal policy (other than its contemporaneous effects) are incorporated into the study. More direct fiscal policy proxies, comparable across countries and at the quarterly frequency, are not available for most of the countries in our sample (see also note 5 above). The impact of lagged monetary policy through lagged interest rates changes on inflation and on unemployment is also captured. More importantly, the coefficient estimates in (2) are permitted to vary over time and are produced only from the sample of information known by the monetary authority at the time policy is carried out.(15) This is the principal attraction of the recursive regression approach since the estimates provide evidence about how central banks react to information they actually had and not to information which was unavailable. Most reaction functions are estimated simply using the full sample period, a procedure which gives the central bank too much information.(16)

The reaction function estimated is

(3) [Mathematical Expression Omitted].

Equation (3) specifies that the nominal interest rate is a difference-stationary process and its innovation, the change in short-term interest rates, measures changes in central bank policy. To the extent that the inflation process exhibits considerable persistence - a phenomenon widely agreed to exist - changes in short-term rates are changes in both real and nominal interest rates.

In equation (3) central banks react to lagged forecast errors because contemporaneous inflation and unemployment rate data are not available at the time of a central bank decision on current interest rates. If [Delta][R.sub.t] [greater than] 0, the central bank raises interest rates in response to the new information. Since the model is estimated with quarterly data, restricting the information set for the forecasts to t-1 provides less information than is actually available to market participants from monthly observations of inflation, unemployment and interest rates; some of this data would be available within the quarter. Whether central banks and economic agents act immediately upon receiving new information is a separate question. Thus, we preferred to err on the side of providing perhaps slightly less information than would actually be available to central banks.(17)

If the lagged forecast errors from the VAR, [Mathematical Expression Omitted], are greater than zero, inflation would be higher than expected given the information at the time. The sum of the coefficients in the lag operator [Alpha](L) would be positive if the central bank reacts to higher-than-expected inflation by tightening monetary policy. The treatment of unemployment forecast errors is similar. If lagged forecast errors in unemployment, [Mathematical Expression Omitted], are greater than zero, the sum of the coefficients in [Beta](L) is expected to be negative if interest rates are lowered in response to increases in unemployment. Moreover, equation (3) permits political events, captured by the variable [P.sub.t], to possibly influence interest rates in a manner to be described below. Finally, the response of domestic interest rates to world interest rates is measured by [Delta](L). There are three exchange rate regimes in our sample. Following 1973 most European countries have operated under a quasi-fixed exchange rate system; the remaining countries have operated under a managed float regime (the amount of management varies widely). Before 1973, under the Bretton-Woods regime, fixed but adjustable exchange rates are the rule. Initially, in all three regimes, we proxy influence of the "world" interest rate ([R.sup.w]) by the simple average of the German, U.S., and Japanese interest rates. For the latter three countries the world interest rate is defined as the simple average of the interest rate in the other two countries.(18) For the European countries we then considered the possibility that the German interest rate alone might be a better proxy for the world rate, while for Canada and Australia, the U.S. interest rate alone was used. Estimates were generated for both measures of the world interest rate, only the second set of results is included in the empirical section. The size of [Delta](L), the response of domestic interest rates to the world interest rate, should, of course, depend on the exchange rate regime.(19)

Political Variables

Political and electoral variables are then added to the reaction functions. For each of the specifications we ask: Do the political or electoral variables help explain the behavior of the monetary authorities beyond their reaction to lagged forecast errors in inflation and unemployment? If so, there is some evidence that the monetary authority is influenced by political forces and is, in this sense, less independent than a monetary authority not influenced by political events.

Two types of political events are considered. In the first type [P.sub.t] is a dummy variable active around elections. In the Nordhaus [1975] model, everything else being equal, we should find that a dependent central bank would reduce interest rates in the months before an election to aid in the re-election of the governing party. This is implemented with a dummy variable (Eprior) equal to 1 in the election quarter and in the two previous quarters. For a politically motivated, less-independent central bank the coefficient on such a dummy should be negative. A second dummy variable (Etmp), equal to 1 in the election quarter and in the two previous quarters and equal to -1 in the two subsequent quarters, was also added to the reaction functions. A politically motivated central bank may defer an increase in interest rates until after an election. This variable partly handles the issue of a pre-election fiscal expansion, if such an expansion raised inflation and lowered unemployment, since the politically motivated central bank would try to defer the necessary interest rate increases. A significant positive sign on this variable would capture the temporary nature of the election-oriented monetary policy. The election dummy Etmp is distinguished from the election dummy Eprior, for with the latter variable the election effect is permanent. A zero or a significant positive sign on either election dummy is evidence of independence, while a negative sign on either is evidence of dependence. We experimented with versions of these variables active for more quarters than described above with no impact on the conclusions.

The second type of political influence focuses on the political preferences of the party in power. Alesina and Roubini [1990] argue that a switch from a right-wing to a left-wing government would be associated with expansionary monetary policy.(20) A dummy representing this switch measures, given past unexpected inflation and unemployment, whether a new left-wing government lowers nominal interest rates. If a new right-wing government dislikes inflation more than its left-wing predecessor, monetary policy will be tightened and interest rates increased.(21) A dummy variable Drpta2 (Drpta4) takes on a value of +1 in the election quarter and two (four) quarters after the election of a new right-wing government (we also specified a dummy eight quarters in length with no effect on the conclusions). The variable is set equal to -1 for the same time period following the election or appointment of a new left-wing government. Consequently, the reaction of interest rates to governing party is positive since a right-wing government (+1) would want to raise interest rates if the central bank were dependent while a left-wing government (-1) would prefer lower interest rates with a dependent central bank, other things being equal.

One complication is the fact that, in several of the countries in our sample, election timing is endogenous. The case in which a dependent central bank reduces interest rates before an election and the case where an election is called because of a series of interest rate reductions are observationally equivalent and cannot be distinguished with our test. We find little evidence, however, of election-timing effects on interest rate behavior in any country. Central banks are equally independent across countries in this sense. To the extent that elections are called for reasons beyond favorable economic circumstances, ends of terms, scandals, votes of confidence or leadership changes, these exogenous events should provide inference concerning the effect of political timing on central bank activity. We return to the endogeneity issue in the discussion of the empirical results.

Considerations of fiscal policy can also produce complications. In the case of partisan changes, if a new left-wing government expanded fiscal policy, and if the central bank accommodated the expansion so that interest rates did not rise, a zero coefficient on the party dummy is evidence of dependence. A negative coefficient could, in principle, reflect a fiscal contraction by a newly elected right-wing government. Nevertheless, it is important to recognize that this issue may not be of much practical importance, since it is difficult to imagine actual changes in fiscal policy enacted within a quarter of a new government taking office.

We also implement a second test of the partisan model that is more straightforward and perhaps more telling. Reaction function (3) is estimated separately where in one sample the data are taken from the period where the left-wing government is in power; in a second sample we use data only from the period when the right-wing government is in power. If the coefficients of the reaction function are significantly different by partisan regime, then this may be viewed as substantial evidence in favor of the joint hypothesis that the central bank is influenced by the party in power and that the two parties do in fact have different preferences over their response to the forecast errors in inflation and unemployment. If the coefficients do not differ significantly, it is not possible to know which part of the joint hypothesis is rejected - either the central bank is independent or the political parties do not differ.

V. EMPIRICAL RESULTS

Forecast Errors and Reaction Functions

Summary statistics by country for the full sample, as well as for two subsamples defined as the Bretton Woods or fixed exchange rate period (denoted BW in the tables) and the post-Bretton Woods sample (denoted PBW) of managed floating exchange rates for some countries, or quasi-fixed exchange rates for others, revealed some interesting results as did data for each country where the full sample (1960-1990) is split to compare the inflation, unemployment and interest rate records of left- and right-wing governments.(22) Dates for the end of the Bretton Woods period were found from information contained in various issues of the International Monetary Fund's International Financial Statistics. Partisan change and electoral data were taken from Alesina, Cohen, and Roubini [1991] and updated.

Average inflation rates in all countries except Japan are higher during the post-Bretton Woods period. For all but Japan and Switzerland average nominal interest rates are also higher. It appears that virtually all countries took advantage of flexible exchange rates to run higher inflation rates. The average unemployment rate in the post-Bretton Woods period is higher for all seventeen countries.

Comparing the partisan records, average inflation is higher under left-wing governments in ten of the fifteen countries shown (Japan and Switzerland are defined as not having experienced partisan changes); average nominal interest rates are higher under right-wing governments in five of fifteen countries. This includes two cases, Canada and the United Kingdom, where inflation under right-wing governments was lower, so right-wing governments had higher ex post real interest rates. Unemployment rates are higher under right-wing governments in nine of fifteen countries.

Separate estimates of equation (3) according to the exchange rate regime as well as estimates for the full sample are presented. In traditional macroeconomic models fixed exchange rates can generate a loss of independence for monetary policy except, of course, for the reserve currency country. However, if capital controls exist between financial markets, interest rate differentials can arise and persist. Our sample, however, covers a period of rapid developments in capital markets and improving capital mobility and, therefore, reinforces the need to use recursive estimation to address the issues considered here. By contrast, under managed or flexible exchange rates, there is more scope for independent monetary policy. Thus, exchange rate regimes may produce differences in central bank behavior, a possibility generally ignored or downplayed in discussions of central bank independence. Neither exchange rate regime is pure of course. The Bretton Woods period incorporates several discrete changes in the "fixed" exchange rate; the post-Bretton Woods period incorporates a wide variety of exchange rate arrangements including some periods where there were important limits on exchange rate changes within the European Community. The null hypothesis that the coefficients in each equation of the VAR are stable between the Bretton Woods and post-Bretton Woods samples (not shown) is strongly rejected. Each VAR (equation (2)) was estimated with six lags of inflation, unemployment, the interest rate, a constant, seasonal dummies and a trend.(23)

Other preliminary statistical work revealed the following other important conclusions. First, having all the information matters. The standard deviations of the regression residuals using data for the whole sample (or for either exchange rate regime) are significantly lower than those of either a recursive VAR, a random walk forecast or a rolling regression forecast (see the discussion in note 19). Whatever the means by which a central bank generates "forecasts" from the full sample, it clearly uses information not available when the actual policy decisions were made. Second, the forecast errors for all three forecasting models possess means that are not significantly different from zero. It is also the case that these forecast errors have substantial and significant positive serial correlation in most cases. Finally, the recursive VAR forecasts always outperform the random walk forecasts, while the rolling VAR forecasts do not always outperform the random walk forecasts. Space limitations lead us to present results for reaction function (3) using the forecast errors generated from the recursive VAR.

Table II presents the estimates of equation (3) omitting the political variables. Estimates for the first lag only of the inflation and unemployment estimates are shown to conserve space. If all central banks are equally independent, then the response of interest rates to the forecast errors in inflation, unemployment, and the response to the exogenous change in the world interest rate provide an objective description of central bank behavior in these countries. Table II shows that the change in the world interest rate is a statistically significant determinant of changes in the domestic interest rates in many countries - in six of seventeen during the Bretton Woods period, and seven of seventeen during the post-Bretton Woods period. In the latter period, the world interest rate variable matters for France, Germany, Belgium, and the Netherlands, all key participants of the European Monetary System. World interest rate changes also played a role in Canada, Sweden and Switzerland in the post-Bretton Woods period. While it may be surprising that, particularly in the Bretton Woods period, the world interest rate did not play a larger role in the determination of the timing of domestic interest rate changes it must also be remembered that many countries had capital controls. The results also suggest that some of the countries successfully sterilized capital flows under fixed exchange rates.(24) It is clear then that some countries were able to use domestic monetary policy to manoeuvre around the world interest rate while other countries chose not to, were unable to (e.g., Canada), or were simply not integrated with leading financial markets during the Bretton Woods period (e.g., Japan, Australia and New Zealand), at least in the short run.

The coefficients which measure the reaction of monetary policy to inflation or unemployment forecast errors do yield some useful information, particularly when compared across exchange rate regimes. First, only in Austria is the coefficient which measures the reaction of interest rate changes to an inflation forecast error or unemployment forecast error significant during the Bretton Woods period. This suggests that the general adoption of fixed exchange rates did restrict the policy activities of other central banks. Some caution should be taken with the results for the Bretton Woods sample for two reasons. First, there are some serious measurement problems with 1960s data, especially for the unemployment rate. Second, because each recursive regression must begin with some minimum sample, we have provided central banks with more information than they would have had in the early portion of the Bretton Woods sample. The estimates for the post-Bretton Woods period avoid this problem entirely.

The post-Bretton Woods period does yield more information on the comparison of central bank behavior across countries. A central bank is often thought to be "independent" if it reacts to an unexpected increase in inflation by increasing nominal interest rates.(25) By this measure, central banks in Japan, Switzerland and Sweden are independent. Central banks in the United States, Japan and Austria respond to an increase in unemployment with looser monetary policy. There is, however, a perverse case, since the central bank in the Netherlands responds to an increase in unemployment by raising interest rates. How can our findings be squared with observed differences in inflation performance in the countries in our sample? Based on the data in Table II there is a close association between inflation differentials vis-a-vis the U.S. and interest rate differentials relative to U.S. interest rates. Nevertheless, the correlation between these differentials is substantially higher in the post-Bretton Woods sample (.88 versus .76), suggestive perhaps of greater interdependence in interest rate policies among the OECD countries in our sample. This view is reinforced by the findings in Table II which show that the responsiveness of domestic interest rate changes to world interest rate changes is substantially higher in the post-Bretton Woods period in five countries (Canada, France, the Netherlands, Ireland and Germany) out of the fourteen for which we can make such comparisons; only two countries (Austria and Norway) show less dependence on the world interest rate in the post-Bretton Woods era. There is no statistical difference for the remaining seven countries, although the coefficients are larger in the post-Bretton Woods period for four of the seven in this category. It should also be pointed out that whereas only one central bank (Austria) reacts significantly to either inflation or unemployment shocks in the Bretton Woods period, seven central banks (in the U.S., U.K., Denmark, Sweden, Japan, Switzerland, and New Zealand) react to such shocks in the post-Bretton Woods period, an indication that central banks became more responsive to macroeconomic conditions after exchange rates floated more freely.

Overall, however, forecast errors in inflation and unemployment have comparatively little explanatory power for interest rate changes for most countries as one can tell by looking at the [R.sup.2]. This may suggest some weakness in our approach to modelling monetary policy. However, this weakness occurs over a wide variety of institutional and legal arrangements for monetary policy and over a variety of specifications.(26) Despite the weakness of these results, we find little cross-country difference in central banks' reactions to inflation or unemployment shocks. Thus the results in Table II cannot, in the very influential Barro-Gordon [1993a; 1993b] framework (a model that was part of our prior), explain the differences between average national inflation rates through measurable ways in which the different central banks reacted to unexpected changes in inflation and unemployment.(27)

Electoral Variables

Table III investigates the effect of election timing on interest rate changes in the context of reaction function (3) through the election dummies (Eprior and Etmp). In eight countries - Canada, Sweden, Ireland, Austria, the Netherlands, Germany, Japan and Norway - for either the full sample or one of the subsamples, the central bank allows an election to affect the [TABULAR DATA FOR TABLE II OMITTED] timing of interest rate changes. In four countries - Sweden, Ireland, Canada and Italy - the sign is not inconsistent with that of an independent central bank; that is, the pre-electoral period involves either a temporary or permanent increase in interest rates as opposed to the decrease predicted by the political business cycle literature. In nine countries - the U.S., the U.K., France, Italy, Switzerland, Belgium, Australia, New Zealand and Denmark - there is no statistical effect of election timing on interest rate changes, a result which could be interpreted as evidence of independence. In Germany and Austria, interest rate reductions take place before elections and are reversed after elections. Only in the Netherlands are elections preceded by permanent reductions in interest rates. It is noteworthy in this context that the central banks in Germany, Austria and the Netherlands appear in Table I among the most independent central banks. However, when the separate Bretton Woods and post-Bretton Woods samples are examined, the pre-electoral effect for the Netherlands disappears, and the temporary effect on German and Austrian interest rates is seen to be a feature of the post-Bretton Woods sample only. In the Austrian case, the post-Bretton Woods effect is significant only at the 10 percent level of significance. Overall then, the effect of elections on interest rate changes is weak for all the countries considered.

It is interesting to compare these results with those reported by Alesina, Cohen and Roubini [1991] who consider the effect of the pre-electoral period on monetary growth. They find, for a similar set of countries in a panel, that unexpected monetary growth does increase before elections. In individual country results, they find only New Zealand and Australia exhibit significant excessive monetary growth before elections. They also find that in New Zealand, Austria and Norway monetary growth before elections differs [TABULAR DATA FOR TABLE III OMITTED] by political party (more for left-wing than for right-wing governments). There is at least one difficulty in comparing our results with theirs. In using monetary growth as the measure of monetary policy, the dependent central bank could accommodate expansionary pre-election fiscal policy and not allow interest rates to rise. In their study this would be expansionary monetary policy (central bank dependent), in our study this would not be expansionary monetary policy (central bank independent). A positive pre-election effect on interest rates would be revealed as evidence of independence in our study but not theirs. These issues highlight the need not only to be careful about defining what is meant by an independent central bank (see also Goodhart [1994] in this connection) but also the complications [TABULAR DATA FOR TABLE IV OMITTED] that arise because of the connection between fiscal and monetary policies even if central banks are primarily responsible for monetary policy only.

Partisan Variables

Tables IV and V present results which investigate the role of partisan politics in the reaction functions. Partisan variables are defined as in Alesina and Roubini [1990]. The simplest description of partisan effects argues that a new right-wing (left-wing) government tightens (loosens) monetary policy because it inherits a higher-than-desired rate of inflation (unemployment). This result is found only in the full sample for the United States and in Sweden in the post-Bretton Woods sub-sample. The effect is not found in the United States in the Bretton Woods sub-sample. The Federal Reserve is the most politically dependent central bank over the full sample by this measure. In four other countries, France, Ireland, Italy, and New Zealand, there is some evidence of a temporary partisan effect. But the effect is not the effect described above. In these countries the election of a new right-wing government lowers interest rates and the election of a left-wing government raises interest rates. This is more difficult to interpret as a political event. One story would have the central bank accommodating a reduction in inflationary expectations and allowing nominal interest rates to fall in the right-wing case and rise in the left-wing case. By this measure six central banks exhibit some political dependence, four from the bottom of the usual rankings.(28) Even with five banks indicating some partisan response for at least one subsample or the full sample, that leaves nine other central banks all equally independent by this measure. Two central banks, Japan and Switzerland, did not face partisan changes in government in the sample.

TABLE V

Test of Coefficient Stability in the Reaction Function, Equation
(3), across Partisan Classification of Government(a)

 Full Sample Post-Bretton Woods
Country R L Prob. R L Prob.

Canada 36 82 .05(*) 29 54 .16
United States 73 45 .40 59 16 .24
United Kingdom 76 41 .65 56 19 .31
France 105 13 .00(*) 55 13 .00(*)
Germany 94 24 .19 INS INS INS
Denmark 39 31 .62 29 31 .62
Sweden 24 94 .02(*) 24 48 .10(**)
Belgium 59 59 .30 33 39 .53
Netherlands 90 28 .06(**) 58 21 .04(*)
Australia 50 44 .49 29 39 .76
Austria 37 53 .60 31 41 .24
New Zealand 79 34 .00(*) 35 31 .13
Norway 53 52 .74 26 45 .97
Ireland 65 53 .77 19 46 .98
Italy CT CT CT CT CT CT

a For sample details, see notes to Table II. The null hypothesis
that the coefficients of the reaction function estimated in Table
II are constant between samples drawn from right-wing and left-wing
governments is tested. This tests the joint null hypothesis that
political parties have different preferences over monetary policy
and that the central bank is politically dependent. R denotes the
number of quarters the right-wing government was in power, L the
number of quarters the left-wing government was in power.

* indicates statistically significant at the 5% level; + at the
10% level.

Table V presents estimates, when available, of reaction function (3) for subsamples with different political parties in power. Left-wing (denoted L) and right-wing (denoted R) governments are assumed to have the same preferences over time for their preferred responses to the unemployment, inflation, and world interest rate shocks. Under the null hypothesis that the central bank is independent, coefficients in the reaction functions should be the same across partisan regimes. The probability values reported in Table V present the significance levels for the test of coefficient stability across partisan regimes. Thus, a low probability value indicates rejection of the null. In the post-Bretton Woods period, two probability values, those of France and the Netherlands, are less than 5 percent. The probability value for Sweden is significant at the 10 percent level. For the full sample, central banks in New Zealand, Sweden, Canada and France are found responsive to partisan changes at the 5 percent level while the Netherlands central bank would reject the joint null at 6 percent. No other central bank changes its behavior significantly with a change in the party in power. The five central banks which appear dependent by this measure never appear in the top three most independent banks in Table I. This indicates some correlation between behavior and the rankings according to legislative provisions. There are no partisan changes in Switzerland or Japan to consider. There are only short periods of partisan change in Italy.

VI. CONCLUSIONS

'The goal of this research was to provide a quantitative assessment of the independence of central banks from political influence for a sample of seventeen OECD countries, as measured by central bank interest rate responses. We also compared our model estimates to the usual rankings based on legislative factors. However, it proved difficult to generate consistent reactions of central banks in any country to forecast errors from VAR's estimated with the information the central bank actually had at the time the policy was made. One possibility, which cannot be disproved, is that the methodology simply does not identify any substantial economic shocks to which monetary policy would have reacted. Central banks are doing something else or the models specified in this paper are too simple to capture what central banks are actually doing.

If the framework is accepted as identifying interesting economic shocks to inflation, unemployment and world interest rates to which central banks might or should have responded, the evidence in this paper is suggestive in four ways. First, central bank responses to these shocks are not closely associated with their position in the usual qualitative rankings of central bank behavior. Of the central banks ranked as very independent (see Table I), that is, Germany, Switzerland and the United States, only the Swiss central bank responds to an unexpected increase in inflation with tighter monetary policy. However, similar behavior is exhibited by the central banks in Japan and Sweden, which appear lower in the usual rankings. There is no obvious correlation between responses to unemployment shocks and a bank's position in the qualitative rankings for the central banks in other countries considered.

Second, for Germany and Austria there is some evidence that interest rate increases are deferred until after elections in the post-Bretton Woods period. These two central banks are frequently considered to be relatively independent of political influence.

Third, the strongest quantitative evidence for the usual rankings of central bank independence based on qualitative factors is found for New Zealand, France, Sweden, Canada and the Netherlands. In these countries a change in the partisan classification of the party in power does influence the reaction function of the central bank. These same five countries are not listed among the most independent central banks in the sample based on central bank laws.(29) This is our strongest support of the usual rankings, and our interpretation is that this is modest but not overwhelming support of the hypothesis of partisan political influence on central bank interest rate policies.

Fourth, for many of the central banks around the world, only the change in the world interest rate has a strong impact on interest-rate-setting behavior. Neither electoral nor partisan factors were found to have widespread or large effects on interest rate policies among the central banks we considered. The design of a central bank's constitution and the degree of its statutory independence may thus not be as important as commonly considered, particularly when there are close international links between financial markets. External factors have tended to be ignored or downplayed in most discussions of central bank independence.

At this point it is not possible to conclude that the legislated differences in central bank independence generate easily observed variations in interest-rate-setting behavior of central banks in either their reaction to economic events, their pre- and post-electoral behavior or their responses to a change in government. The search for any simple characterization of central bank behavior with respect to political influence continues. Future research, however, might expand the model of this paper to include a fiscal policy measure or possibly consider other instruments of monetary policy, a monetary aggregate and even the exchange rate. After all, just as reaction functions are known to be temporally unstable, as has been demonstrated again in this paper, similarly the choice of a monetary policy instrument can and does change over time in ways that cannot be captured completely in the short-run via the interest rate instrument alone.(30)

1. See Alesina [1988] for a survey, or recent contributions by Alesina and Roubini [1990], and Alesina, Cohen and Roubini [1991].

2. Our study parallels, to some extent, the discussion of monetary policy and political influence in Alesina, Cohen and Roubini [1991]. They study monetary growth (relative to its immediate history) rather than interest rate changes, our proxy for central bank policy actions.

3. An additional problem is the possibility of two-way causation. A population which fears inflation imposes an effective constraint on the political system through the central bank, which manifests itself in a low-inflation performance but appears to be "caused" by central bank independence. Cukierman, Webb and Neyapti [1992] also recognize this problem and indeed find evidence of two-way causality between inflation and the turnover rate of central bank governors, which they use as a proxy for independence.

4. New Zealand recently undertook a substantive reform of its central bank towards more independence. Similar proposals are being contemplated for Canada and the United Kingdom. The European Community is in the process of designing an independent European central bank.

5. One important reason we do not include in this study the interaction of fiscal variables and monetary policy instruments is that we have no confidence in the consistency or comparability of data on government debt for the full sample of seventeen countries from 1960 to 1984. The OECD Main Economic Indicators does not publish a consistent data series for either government debt or the deficit back to 1960. The deficit data in the International Financial Statistics (usually line 80) are subject to numerous changes in definitions, and are available since 1960 for only nine of the seventeen countries. The accounting conventions differ across countries where there is a full sample of data. Finally, the wide differences in political structures in the countries examined in this paper (e.g., federal versus unitary) complicate greatly the proper measurement of fiscal policy. In our methodology, if expansionary fiscal policy were to reduce unemployment and increase inflation, the central bank would take this into account in setting monetary policy.

6. They also construct a separate index which measures the freedom a central bank has to choose the instruments of monetary policy.

7. Rank correlations between the rankings are low and fall rather substantially when Switzerland, Germany, and the U.S. are omitted.

8. As a referee pointed out, however, the negative correlation between inflation and an index of central bank independence remains even if Japan is classified as dependent (e.g., as in the Burdekin and Willett ranking).

9. A third variant of the political variables, found in Frey and Schneider [1981], asks if policy responds to the popularity of the government as measured by polls. This kind of political variable is not considered in the present study.

10. The experience of different countries with monetary growth targets is discussed in Bernanke and Mishkin [1992].

11. Vaubel [1994] conducts non-parametric tests of the connection between M1 growth and the monetary policy preferences of the government in Germany. He concludes that there is a pre-electoral acceleration of monetary growth when the government has a political majority in the Bundesbank Council.

12. This choice is less desirable because the role of the discount rate varies from country to country. It is clear from the reports of the central bank of Ireland that the discount window in Ireland is active and that the discount rate varies with market rates. There was less information on the role of the discount window in New Zealand. There was no market-determined short-term interest rate for New Zealand available for the full sample period.

13. If a central bank or other branch of the government operates through credit allocation, then a shortage of credit to a sector should increase the interest rate in the market-determined sector. This was the reason for using a market-determined rate where possible. See Aftalion [1983] for a description of the Bank of France's activities in credit allocation. Such effects may also occur in other countries.

14. Beck [1987] and Willett and Keen [1990] also used an interest rate as a measure of monetary policy actions. Indeed, our approach can be said to take up Willett's suggestions about exploring the connection between interest rates, economic fundamentals and political effects on central bank behavior.

15. We do not consider the issue of data revisions. For the variables in (2), data revisions are less important than for some other variables such as, for example, measures of output or monetary aggregates. Estimation of (2) also proceeds as if each series is stationary. This is an uncontroversial assumption for all the series except possibly the interest rate. However, in most sub-samples for all countries, individual series did not appear to contain a unit root once a trend is permitted.

16. This may be a slight exaggeration since central banks probably receive economic information in advance of the public. The question is whether this information is used immediately in evaluating interest rate policy. Clearly, we cannot control for this possibility since we are using quarterly data.

17. The two-step procedure employed here could be open to the "generated regressor" problem despite the resort to recursive estimation. However, as the original model (2) is a VAR, the generated regressor problem does not arise. In any event, in a few cases we estimated reaction functions (3) using Generalised Least Squares with no impact on our conclusions.

18. There is a clear problem with simultaneity for the three countries. The reaction functions for these three countries were also estimated without the world interest rate and yield similar results. Other methods could have been implemented to capture world interest rate effects, such as using a proxy for the world real interest rate, but our results would then be open to the criticism that the ex post real interest rate does not adequately measure ex ante real interest rates.

19. A second type of reaction function (4) is representative of the existing literature on reaction functions, as in Woolley [1988]. The VAR is constrained to be a random walk and the reaction function becomes

(4) [Mathematical Expression Omitted].

We found that the recursive VAR clearly outforecast the random walk specification.

In addition we also estimated the reaction function

(5) [Mathematical Expression Omitted].

In (5) the policy variable is [Mathematical Expression Omitted], the deviation of the short-term nominal interest rate in period t from the value which would have been forecast with nonpolitical information available in t - 1. This variable is the error term, [Mathematical Expression Omitted], from the VAR (equation (2)). The information sets in (5) and in the VAR (2) are not identical; the innovation in interest rates in period t uses slightly more information than the forecast errors dated t-1 and earlier. The innovation in interest rates as of period t-l, the proposed policy instrument in (5), is uncorrelated with the least squares residual with six lags in these same variables, as of information in period t-1. It need not, in theory, be uncorrelated with six lags of the same variables as of period t-2, t-3 up to the length of the [Alpha](L) and [Beta](L) coefficients. In practice the correlations between this version of the policy variable and the lagged forecast errors were very low and the reaction function proposed in (5) simply did not yield many interesting results.

20. The definitions of "right-wing" and "left-wing" follow Alesina and Roubini [1990] and Alesina, Cohen and Roubini [1991]. A general difficulty with the specification of both versions of the political dummy variable is that previous studies do not use a statistical criterion to determine how long the dummies should remain active. If, for example, monetary policy acts with long and variable lags, then existing specifications may be misspecified.

21. A referee correctly points out that a central bank could be viewed as being (somewhat) independent if the interest rate change is less than "desired" by politicians. Unless we have a model of notional interest rates by, say, political party we cannot, strictly speaking, measure degrees of independence.

22. It has been pointed out to us that since we are estimating (2) recursively it might be interesting to examine how the coefficients in the reaction functions change over time. Given the number of countries and coefficients to be estimated it is impractical to display all such results. However, we found that the exchange rate regime had the greatest effect on the recursive coefficient estimates.

23. We also estimated VARs with four lags and the conclusions were unaffected. Although it may be preferable to select some optimal lag length for each country's data based on some criterion such as Akaike's Information Criterion (AIC), there is the danger that different VARs for different countries would produce results which are not comparable. Moreover, we would be left asking why information sets differ across countries, a task beyond the scope of this study. Also, we are restricted in the choice of lags because of considerations of degrees of freedom. Finally, an econometric model should not be chosen on the basis of a single criterion. Thus, a model chosen according to Akaike's criterion may be inferior according to another criterion. There are tradeoffs when estimating any econometric model. Nevertheless, we experimented with a variety of specifications and lag lengths. The conclusions reached in this study are generally robust to these choices.

24. The evidence on the ability of central banks to sterilize is mixed.

25. This assumes, of course, that the government prefers to defer an increase or disagrees with the belief that higher future inflation is imminent. However, the complete freedom to act and to do so based on anticipations of inflation rather than past inflation does serve to reinforce the perception that the central bank is acting independently of government influence. Nevertheless, even an independent central bank might occasionally postpone an interest rate change for nonpolitical reasons (e.g., wait and see if a perceived increase in inflation is likely to occur). Even in this case, however, what is important for our purposes is that interest rate effects be independent of political or partisan factors.

26. It is possible that we chose the wrong instrument of monetary policy and that money growth, deficits or exchange rates would have produced "better" results than interest rate changes. It is also possible that other specifications with different information structures could have produced "better" results. We do not explore these questions further in this paper.

27. See Canzoneri [1993, 143] who argues that "the Barro-Gordon model is just not up to the task. It does not discuss governments or political processes."

28. There could be, as usual, a contemporaneous fiscal policy effect if actual or expected fiscal policy changes with the change in government. A left-wing (right-wing) fiscal expansion (contraction) which is accompanied by unchanged monetary policy would raise (lower) interest rates. This would represent evidence of independence for the central banks of France, Ireland, Italy and New Zealand.

29. It has been suggested to us that partisan changes as usually defined in the literature are well-defined only for the U.S. It is possible that for other countries partisan changes are far more subtle than can be captured by the dummy variables used in this study.

30. As one referee put it, interest rate policies might be just inconsistent enough to reveal the macroeconomic responses reported in this paper, as opposed to ones which might otherwise have been expected.

REFERENCES

Aftalion, Florin. "The Political Economy of French Monetary Policy," in The Political Economy of Monetary Policy, edited by D. Hodgman. Boston: Federal Reserve Bank of Boston, 1983, 7-25.

Alesina, Alberto. "Macroeconomics and Politics," in NBER Macroeconomics Annual, edited by Stanley Fischer. Cambridge, Mass.: MIT Press, 1988, 13-52.

Alesina, Alberto, G. D. Cohen, and N. Roubini. "Macroeconomic Policy and Elections in OECD Democracies." National Bureau of Economic Research Working Paper No. 3830, September 1991.

Alesina, Alberto, and Vittorio Grilli. "Establishing a Central Bank." Paper presented at the European Central Bank Conference, Reshaping Monetary Politics in Europe. Georgetown University, 1991.

Alesina, Alberto, and Nouriel Roubini. "Political Cycles: Evidence from OECD Economies." Review of Economic Studies, October 1990, 663-88.

Alesina, Alberto, and Lawrence H. Summers. "Central Bank independence and Macroeconomic Performance." Journal of Money, Credit and Banking, May 1993, 151-62.

Allen, Stuart D. "The Federal Reserve and the Electoral Cycle." Journal of Money, Credit and Banking, February 1986, 88-94.

Alt, James E., and K. Alex Chrystal. Political Economics. Berkeley: University of California Press, 1983.

Barro, R. J., and D. B. Gordon. "A Positive Theory of Monetary Policy in a Natural Rate Model." Journal of Political Economy, August 1983a, 589-610.

-----. "Rules and Discretion and Reputation in a Model of Monetary Policy." Journal of Monetary Economics, July 1983b, 101-22.

Beck, Nathaniel. "Domestic Political Sources of American Monetary Policy 1955-82." Journal of Politics, August 1984, 787-817.

-----. "Elections and the Fed: Is There a Political Monetary Cycle?" American Journal of Political Science, February 1987, 194-216.

-----. "Political Monetary Cycles," in The Political Economy of American Monetary Policy, edited by Thomas Mayer. Cambridge: Cambridge University Press, 1990, 115-30.

Bernanke, Ben, and Allan S. Blinder. "The Federal Funds Rate and the Channels of Monetary Transmission." American Economic Review, September 1992, 901-21.

Bernanke, Ben, and Frederic Mishkin. "Central Bank Behavior and the Strategy of Monetary Policy: Observations from Six Industrialized Countries," in NBER Macroeconomics Annual 1992, edited by Olivier J. Blanchard and S. Fischer, Cambridge: The MIT Press, 1992, 183-227.

Burdekin, Richard C. K., and Mark E. Wohar. "Monetary Institutions, Budget Deficits and Inflation." European Journal of Political Economy, 1990, 531-551.

Burdekin, Richard C. K., and Thomas D. Willett. "Central Bank Reform: The Federal Reserve in International Perspective." Public Budgeting and Financial Management, no. 4, 1991, 619-50.

Canzoneri, Matthew B. "Central Bank Independence, Growth, Investment, and Real Rates: A Comment." Carnegie-Rochester Conference Series on Public Policy, December 1993, 141-46.

Cargill, Thomas F. "Central Bank Independence and Regulatory Responsibilities: The Bank of Japan and the Federal Reserve." Solomon Brothers Center for the Study of Financial Institutions, Monograph Series in Finance and Economics, Monograph 1989-2, 1989.

Chappell, Henry, and William Keech. "Party Differences in Macroeconomic Policies and Outcomes." American Economic Review, May 1986, 71-4.

Christian, James W. "A Further Analysis of the Objectives of American Monetary Policy." Journal of Finance, 23(3), 1968, 465-77.

Cowart, Andrew T. "The Economic Policies of European Governments, Part I: Monetary Policy." British Journal of Political Science, July 1978, 285-311.

Cukierman, Alex S. Central Bank Strategy, Credibility and Independence. Cambridge: The MIT Press, 1992.

Cukierman, Alex S., S. B. Webb, and B. Neyapti. "The Measurement of Central Bank Independence and Its Effect on Policy Outcomes." World Bank Economic Review, 6(3), 1992, 353-98.

Fair, D. E. "Relations Between Government and Central Bank: A Survey of Twenty Countries," in Appendices, Committee to Review the Functioning of Financial Institutions. London: Her Majesty's Stationery Office, 1980.

Fair, R. C. "The Sensitivity of Fiscal Policy Effects to Assumptions about the Behavior of the Federal Reserve." Econometrica, September 1978, 1165-79.

Fischer, Stanley. "Modern Central Banking." Presented at a Symposium on Central Banking Held in Honour of the Tercentenary of the Bank of England, June 1994.

Frey, B., and F. Schneider. "Central Bank Behavior: A Positive Empirical Analysis." Journal of Monetary Economics, May 1981, 291-315.

Goodhart, Charles A. E. "Game Theory for Central Bankers: A Report to the Governor of the Bank of England." Journal of Economic Literature, March 1994, 101-14.

Gordon, H. S. "The Bank of Canada in a System of Responsible Government." Canadian Journal of Economics and Political Science, February 1961, 122.

Grilli, Vittorio, Donato Masciandaro, and Guido Tabellini. "Political and Monetary Institutions and Public Financial Policies in the Industrial Countries." Economic Policy, October 1991, 342-92.

Hakes, David R. "The Objectives and Priorities of Monetary Policy under Different Federal Reserve Chairmen." Journal of Money, Credit and Banking, August 1990, 327-37.

Havrilesky, Thomas. "A Partisanship Theory of Fiscal and Monetary Regimes." Journal of Money, Credit and Banking, August 1987, 308-25.

-----. "A Test of Monetary Policy Action." Journal of Political Economy, June 1967, 299-304.

Hibbs, Douglas A., Jr. "Political Parties and Macroeconomic Policy." American Political Science Review, December 1977, 1464-87.

Hochreiter, Eduard. "The Austrian National Bank Act: What Does It Say About Monetary Policy?" Konjunkturpolitik, 36(4), 1990, 245-56.

Hodgman, Donald R., and Robert W. Resek. "Determinants of Monetary Policy in France, The Federal Republic of Germany, Italy and the United Kingdom: A Comparative Analysis," in The Political Economy of Monetary Policy, edited by D. R. Hodgman. Boston: Federal Reserve Bank of Boston, 1983, 147-70.

Khoury, Salwa S. "The Federal Reserve Reaction Functions: A Specification Search," in The Political Economy of American Monetary Policy, edited by Thomas Mayer. Cambridge: Cambridge University Press, 1990, 27-50.

Mayer, Thomas. "Structure and Operations of the Federal Reserve System," in Compendium of Papers Prepared for the Financial Institutions and the Nation's Economy Study, Committee on Banking, Currency and Housing, 94th Congress, Second Session. Washington: Government Printing Office, 1976.

Nordhaus, W. "The Political Business Cycle." Review of Economic Studies, April 1975, 169-90.

Parkin, Michael. "Domestic Monetary Institutions and Deficits," in Deficits, edited by James M. Buchanan, Charles K. Rowley, and Robert D. Tollison. London: Basil Blackwell, 1986, 310-37.

Pollard, Patricia S. "Central Bank Independence and Economic Performance." Review of the Federal Reserve Bank of St. Louis, July/August 1993, 21-36.

Reuber, Grant L. "The Objectives of Canadian Monetary Policy 1949-61: Empirical 'Trade-Offs' and the Reaction Functions of the Authorities." Journal of Political Economy, April 1964, 109-32.

Vaubel, Roland. "A Public Choice Analysis of the German Bundesbank." Manuscript, University of Mannheim, 1994.

Walsh, Carl E. "Central Bank Strategies, Credibility, and Independence: A Review Essay." Journal of Monetary Economics, November 1993, 287-302.

Willett, Thomas D., and Edward Keen. "Studying the Fed: Toward a Broader Public-Choice Perspective," in The Political Economy of American Monetary Policy, edited by Thomas Mayer. Cambridge: Cambridge University Press, 1990, 13-25.

Woolley, John T. "Partisan Manipulation of the Economy: Another Look at Monetary Policy with Moving Regression." Journal of Politics, May 1988, 335-60.

-----. "Political Factors in Monetary Policy," in The Political Economy of Monetary Policy, edited by D. R. Hodgman. Boston: Federal Reserve Bank of Boston, 1983, 177-203.