Evolving Perceptions of Central Bank Credibility: The European Central Bank Experience

Abstract

Previous articleNext article FreeEvolving Perceptions of Central Bank Credibility: The European Central Bank ExperienceLinda S. Goldberg and Michael W. KleinLinda S. GoldbergFederal Reserve Bank of New York and NBER Search for more articles by this author and Michael W. KleinTufts University and NBER Search for more articles by this author PDFPDF PLUSFull Text Add to favoritesDownload CitationTrack CitationsPermissionsReprints Share onFacebookTwitterLinked InRedditEmailQR Code SectionsMoreI. IntroductionWhat are the preferences of a central bank over inflation and output-gap stabilization objectives, and what is its preferred long-run inflation rate? While statements of priorities and goals are important, the credibility of such statements and the market perception of the policy reaction function of a central bank play a key role in determining economic outcomes. This point, early on described as the “credibility” of central bank policies, is a standard theoretical result with recent interpretation in the New Keynesian paradigms.1 It is also received wisdom among practitioners.The importance of the market perception of the central bank’s acceptable trade-offs between inflation and output goals as well its specific targets naturally leads to the question of how the market acquires this perception and whether and how it evolves over time. One view is that establishing an appropriate institutional structure is the key element in insulating the monetary authority from political pressure and thereby convincing markets that a central bank has a strong and unvarying aversion to inflation. A second, more dynamic, view focuses on the role that actual policy conduct plays in building the reputation of a central bank. These two different views have distinct implications for the relative importance of the institutional structure of a central bank, as compared to its conduct, for attaining and maintaining its credibility.2A survey of the heads of central banks and prominent monetary economists reflects a belief that the credibility of a central bank is based more on its past actions than on institutional structures that afford it independence by insulating it from political concerns, although there is also a consensus that structure matters (Blinder 2000). Empirical research has found that institutional features related to central bank independence are associated with economic performance in cross sections of countries, perhaps because these features indicate the ability of an institution to “tie its hands” and commit to a policy that may cause short-term pain in the pursuit of longer-run gain.3 There is less evidence, however, as to whether and how the credibility of a particular central bank’s policy stance evolves over time in response to the conduct of policy and other related decisions. This is a particularly timely issue. Questions were raised about the commitment of the Federal Reserve to price stability after its response to the financial crisis of 2007–9. Even more recently, similar concerns were raised about the European Central Bank (ECB) to its primary mandate of price stability after it introduced a Securities Markets Program to purchase euro area debt securities to “ensure depth and liquidity in those market segments which are dysfunctional” in May 2010.These episodes raise a relevant question of how to determine whether market perceptions of central bank policies are changing. We show how high-frequency data from asset markets can be used to address this by providing a methodology for tracking the evolution of related market perceptions. Our analysis focuses on the experience of the ECB during its early years of operation, which is a natural experiment for this issue.The ECB offers an example of both structure and conduct aimed toward achieving policy credibility. Its architects were mindful of lessons from economic theory concerning the importance of a structure that provided independence from political considerations.4 The role of conduct was also clearly apparent. At its inception, the directors of the ECB were acutely aware that their policies were closely scrutinized for indications of general tendencies. This is, of course, a specific example of a more general tendency for relatively large updating of market priors when there is the establishment of a new central bank, an adoption of new policies (such as inflation targeting or extraordinary emergency lending at the time of a crisis), or a change in leadership (Blinder 2000).5We begin by developing insights about the response of asset prices to inflation news in Section II. The key point is that these responses reflect market perceptions of the policy reaction function of the central bank, as shown in the work of Gürkaynak, Sack, and Swanson (2005) and Gürkaynak et al. (2007).6 Their calibration exercises demonstrate the responses of short interest rates, long interest rates, and the yield curve to output and inflation shocks. We build on their work by demonstrating that the patterns in these calibrations are closely tied to the public’s perception of the policy reaction function. In particular, we use this model to illustrate the change in the relationship between economic news and the term structure of interest rates with changes in the perceived anti-inflation stance of the policy reaction function parameters. These results also hold for changes in the perceived inflation target.7In Section III we propose and implement a novel method for measuring the market’s view of evolving central bank “credibility.” The method applies newly developed econometric tests for persistent time variation in regression coefficients (from Elliott and Müller [2006]) to high-frequency financial market data.8 Specifically, we use these tests to explore the evolution of the effects of news announcements on the yield curve for euro area countries for the period beginning January 1999, the time the ECB began its operations, through mid-2005, using hourly data on the term structure of bonds of euro area countries and the United States, as well as the euro-dollar exchange rate. These econometric techniques are especially informative in this context since they allow for a gradual evolution of estimated parameters rather than an abrupt change at a single moment. This evolution will capture the consequences of an ongoing updating by market participants of their views of the central bank reaction function. This type of updating can occur as market participants gradually learn through observing central bank actions and communications.The results in Section IV show significant and persistent parameter instability in the effects of economic news on European term structures and on the euro-dollar exchange rate. The identified patterns are consistent with market participants updating their views of the policy reaction function of the ECB. Additional support for our updating hypothesis is provided by considering the smoothed time path of the estimated parameters of the coefficient on the news announcement, estimated through another new, and related, econometric technique (Müller and Petalas 2010). Parameter values evolved in a manner consistent with the perception of an increasing aversion to inflation by the ECB as it tightened its monetary policy or, alternatively, as consistent with a perceived decline in the inflation target of the ECB. These results on time-varying consequences of economic news for the yield curve are complemented by results of discrete structural break tests (Andrews 1993) that demonstrate the robustness of our findings.Overall, these empirical results support the view that actions, and not just institutional structure, influence market perceptions of the policy stance of a new central bank. Benchmark test results for the term structure of U.S. interest rates present no evidence of persistent parameter instability in the response of the U.S. term structure to news, a result consistent with stable perceived weights in the Fed’s reaction function over this same interval.9 The results demonstrate that the tools introduced can capture evolving views of central bank preferences and credibility.II. Central Bank Policy Reaction Functions and Market Responses to NewsIn this section we present a model in order to demonstrate the effects of changing perceptions on the actual response of interest rates to economic news.10 The insights from this model inform our interpretation of the empirical results on the evolution of the response of the yield curve to news that are presented in Section IV. The basic argument is that market perceptions of a central bank’s stance on policy have an important impact on the performance of an economy and the consequences of policy decisions. While these market perceptions are, by nature, unobservable, we argue that actions such as changing views of the relative weight a central bank places on inflation versus the output gap in its monetary policy reaction function should be identifiable through analysis of the high-frequency response of asset prices to economic news.The model that we use to frame our analysis follows from Gürkaynak et al. (2005). This standard New Keynesian approach allows for a significant fraction of backward-looking agents (who, equivalently, can be assumed to act in a rule-of-thumb manner) and also allows for forward-looking agents.11 The model consists of six basic equations:The first two equations represent the macroeconomic structure of the economy. Equation (1) specifies current inflation, πt, as a function of expected future inflation and lagged inflation, which contributes to inflation persistence through the lag function Aπ(L). The parameter μ describes the balance of these forward-looking and backward-looking pressures. Current inflation also depends on yt, which captures the stance of current output relative to its potential (i.e., the output gap). Equation (2) describes this output gap as also having a forward-looking component and a persistent lagged effect. The output gap (if negative) declines as monetary policy is more expansionary, as reflected in declining real interest rates. The latter are introduced via the difference between the nominal rate it and inflation expectations over a comparable maturity horizon.Shocks to inflation and output are represented by and in equations (1) and (2), respectively. The source of these shocks may be either domestic or foreign in origin. Owing to international transmission, either foreign or domestic shocks could be a source of domestic inflation and lead to policy reactions (Clarida, Galí, and Gertler 2002).12Equation (3) specifies the policy reaction function of the central bank. The equation augments concerns about the output gap, which enters with weight b, and inflation, , relative to its target, , which enters with weight a, with policy rate smoothing. Greater values of the parameter c indicate a relative unwillingness of the central bank to deviate sharply from the prior period’s policy rate. Equation (3) also affords a role to recent inflation history with representing a 4-quarter moving average of inflation and includes , an independently and identically distributed shock.Equations (4) and (5) describe updating of views of the central bank inflation target, , by the central bank (eq. [4]) and by private agents as denoted by the hat notation of equation (5). As shown by Erceg and Levin (2003) and Gürkaynak et al. (2005), the inflation target of the central bank has a tendency to rise when inflation goes above the prior target, with this effect depending on the size of the θ parameter, and can be subject to exogenous changes captured by . Private-sector agents infer such changes by observing the deviation of it from their prior expectation for policy and update their view of the target according to the strength of a Kalman gain parameter κ and when the observed policy rate is higher or lower than what they would have expected given the prior perceived target. This approach provides a mechanism for central banks to update their policy reaction function and for the private sector to assess and learn about this change. The model also imposes the expectations hypothesis, as in Gürkaynak et al. (2007), in which long interest rates of maturity m, it(m), are the cumulated sum of short interest rates captured by 1-year forward rates j years ahead, fj(i, 1).Consider the consequences of varying values of a and b for the long rate, the short rate, and the slope of the yield curve (i.e., the difference between the long rate and the short rate). The impulse examined is news about inflation, which occurs through a realization of either or .13 The experiment explores the effects of this news over time and the changes in the effects under parameterizations of the central bank reaction function. In particular, suppose that there is an increase in a relative to b in equation (3), which Fuhrer and Hooker (1993) describe as greater central bank credibility in fighting inflation. There are marked differences in the response of interest rates to news with changes in the perceived values of a and b. Indeed, when a and b are varied, the entire time path of adjustment to shocks is altered in the model.14 Clarida and Waldman (2008) show that the larger the weight on the output gap, the slower the economy’s convergences to the central bank’s output and inflation targets.The graphs presented in figure 1 present the impact effects on the long interest rate, the short interest rate, and the difference between the two (i.e., the slope of the yield curve) to a positive shock to inflation, (in the top three graphs), and a positive shock to output, (in the bottom three graphs), for a wide set of pairs of the parameters a and b. The surfaces in the third graph in each of the two rows show that an increase in the relative importance of the weight the central bank places on fighting inflation (a) versus the weight it places on stabilizing output (b) decreases the yield curve slope (i.e., short rates rise by more than long rates rise) after an inflationary shock. The central bank moves more aggressively to combat inflation, placing less importance on output and employment goals. Under higher a (or lower b), there still are consequences for both and , since a positive realization of either shock will have a smaller positive effect on the long rate than on the short rate. Indeed, the model’s quantitative result is that the yield curve response is negative for all cases except when output stabilization has a very high weight compared to inflation stabilization in the central bank’s objective function. Another set of quantitative results also can be constructed for exchange rates, as analyzed more broadly in Engel and West (2005), Clarida and Waldman (2008), and much earlier by Hardouvelis (1988).Fig. 1. Impact effects of inflation and output news. Note: a represents the weight on the inflation gap, and b represents the weight on the output gap.View Large ImageDownload PowerPointEmpirically, however, it has long been recognized that there is excess sensitivity of the long end of the yield curve to news, as discussed in Ellingsen and Söderström (2001). While Gurkaynak et al. (2005) provide a range of explanations for this excess sensitivity in the data, the implication is that the unexplained excess sensitivity of the long end of the yield curve would shift up the contours shown in the far right graphs of figure 1, potentially locating that plane in more positive space. For our purposes, the main point is not whether quantitatively there is a net positive or a net negative impact effect of news for a given set of values of a and b. Rather, the key issue is that an increase in central bank credibility results in an inflationary shock moving the economy to a different point on the contour, so that credibility improvements reduce the response of the long interest rate to a greater degree than they reduce the response of the short rate. In particular, an increase in credibility alternatively could be interpreted as a decrease in the expected inflation target of the central bank, as in Erceg and Levin (2003) and Gürkaynak et al. (2005). Our approach can be viewed as complementary, although it is noteworthy that the pattern of yield curve response to news does not vary substantively with the κ or θ parameters indicating updating of the perceived central bank inflation target, .III. Empirical ApproachA. Testing StrategyThe model presented in the previous section provides a framework interpreting a time-varying response of the immediate effect of economic news on the term structure of interest rates. An empirical application requires a method for testing for instability in this relationship. In this section we discuss the method we use to test for persistent variation in the immediate response of the term structure to news.We use a linear specification linking the surprise component of news to the change in an asset price, as is standard in research on the high-frequency response of asset prices to news (see, e.g., Andersen et al. 2003). The specification for the effect of news on any asset price qt, allowing for the possibility of a time-varying coefficient on the news variable, iswhere is the change in the term structure over the short period of time between t−, just before an announcement, and t+, just after that announcement (i.e., ); represents the announced value of a variable, which is known at time t+; represents the expected value of that variable before the announcement (so is the surprise component of the announcement); and is a white-noise error term. This parsimonious specification is most appropriate when the time horizon between t− and t+ is short, for example, when it is measured in minutes rather than in days, and when news about the variable x does not become available at the same time (i.e., within the span t− to t+) as announcements about some other relevant variable.In our application, we define the term structure as , with L and S denoting long-term and short-term interest rates, respectively. Our theoretical motivation argued that a significant evolution in the response of the difference between 10-year and 2-year European interest rates would be expected for a central bank with changing market perceptions of its policy reaction function priorities.The challenge, in this context, is that there is not a single, widely recognized dramatic change in policy that was clearly a watershed that led to a change in the public’s perceptions of the ECB. Thus, we cannot perform a simple Chow test over γi. Moreover, it is unlikely that there was a discrete change in expectations, in response to either a single event or a small number of events. Rather, it is more reasonable to think of perceptions changing gradually as market participants learned about the ECB through its pronouncements and, especially, its actions.15 Thus, we would like to test for persistent change in the estimated parameter γi.16Elliott and Müller (2006) have developed a test for the presence of persistent time variation in one or more regression coefficients. Their quasi local level (qLL) statistic provides asymptotically equivalent tests for a large class of persistent breaking processes against the alternative of structural stability. The test does not require the specification of an exact breaking process, such as breaks that occur in a random fashion, serial correlation in the changes of coefficients, or a clustering of break points.17 This feature of their test makes it well suited for our purposes since we do not need to test for a particular type of updating by market participants of their views on central bank inflation aversion. The qLL statistic takes a negative value, and a value smaller (i.e., more negative) than the critical value implies a failure to reject time variation in one or more coefficients for the entire sample period. This procedure tests for persistent time variation over the entire sample and, as such, does not identify a particular date as the one most likely to represent a discrete break point.18As will be shown below, we do, in fact, find evidence that there has been persistent time variation in the slope coefficient in term spread regressions for Germany, France, and Italy, but not for the United States. We also find that there is significant parameter instability in the response of the bilateral euro-dollar exchange rate to news.19 We interpret this combined finding of parameter instability for European rates and the euro-dollar rate and parameter stability for U.S. rates as reflecting an evolving view of the inflation aversion of the ECB rather than as some structural change common to financial markets across all four of these industrial countries.The Elliott and Müller qLL statistic says nothing about the direction of change of γi. Yet, following from the results presented in Section II, a decrease in γi can be interpreted as an increase in the perceived relative inflation aversion of the central bank. For this purpose, we rely on methods from Müller and Petalas (2010), who show how to calculate the smoothed time path of γi. We present these estimated time paths. For robustness, the results from these smoothed estimates are supported by the sup-Wald tests for parameter stability (see Andrews 1993, 2003), which offer a break date for γi that roughly corresponds to the peak value of the estimated smoothed time path.As will be shown, we find that γi decreases over the sample period for the cases in which there is evidence of a significant persistent change in γi (i.e., for the three European yield curves and for the euro-dollar exchange rate). Even more tellingly, the reduction in γi tends to occur in the wake of monetary tightening by the ECB. It is unlikely that other candidate explanations for changes in the responsiveness of the slope coefficients that are not linked to the perception of the ECB policy stance would map as closely to actual ECB policy changes.B. DataThe three types of data used in our analysis are various asset prices, where the assets are government bonds and foreign exchange, inflation announcements, and related market expectations of inflation. We begin this subsection with a discussion of the five different asset prices used as dependent variables in our estimation. We then describe our construction of inflation surprises.Asset price data. Five different dependent variables are used in the regressions. In each case, the dependent variable, , represents the change in q between 30 minutes before and 30 minutes after each monthly inflation announcement over the period January 1999 to June 2005. The changes in the term spread between 10-year and 2-year interest rates for French, Italian, German, or U.S. government bonds are four of the dependent variables, with a robustness section (Sec. IV.D) considering the 2-year and 10-year rates separately.The fifth dependent variable is the case in which represents the change in the logarithm of the euro-dollar exchange rate 30 minutes before and 30 minutes after the news announcement. Through short-run interest rate parity, the exchange rate move should reflect the relative effects of news on interest rates in the euro area versus in the United States (see the appendix). In this case, a positive value of rather than indicating an increase in the premium of the long rate relative to the short rate indicates a depreciation of the euro. Evidence that γi decreases over the sample period in the exchange rate specification is consistent with a situation of more of an increase in the perceived anti-inflation stance of the ECB, as compared to the U.S. Federal Reserve.Inflationary announcements and expectations. To capture the economic news ηt that lead to asset price updating, we restrict our attention to inflation announcement measures. Candidate data releases for our study potentially include indicators of consumer price inflation for the full euro area, for individual countries in the euro area, and for the United States. The construction of the “news” variable, which is the appropriate variable to be employed in the specification, also requires measures of market expectations for the full sample period. While some earlier studies use vector autoregressions to generate measures of