Measuring monetary policy.

I. INTRODUCTION

Accurate measurement of the effects of changes in monetary policy on the economy is essential, both for good policy-making and for choosing among alternative macroeconomic theories. Unfortunately, attempts to quantify the links between central bank actions and the economy quickly run into a major roadblock: there is no consensus on how to measure the size and direction of changes in monetary policy. The traditional approach, which identifies changes in monetary policy with changes in the stock of money, is not adequate, since in practice the growth rates of monetary aggregates depend on a variety of nonpolicy influences. For example, because the Federal Reserve System's operating procedures have typically involved some smoothing of short-term interest rates, and hence accommodation of money demand shocks, observed money growth rates in the United States reflect changes in money demand as well as changes in money supply.(1) Secular changes in velocity brought about by financial innovation, deregulation, and other factors are a further barrier to using money growth rates alone as a measure of the direction of policy.

As the deficiencies of money stock growth as a measure of the stance of monetary policy have become widely recognized, many researchers have tried to find alternative indicators. Recent attempts have largely fallen into two general categories. First, following the example of Friedman and Schwartz [1963], Romer and Romer [1989] reintroduced the "narrative approach" to the study of monetary policy. based on a reading of the minutes of the Federal Open Market Committee, Romer and Romer determined a set of dates at which policy-makers appeared to shift to a more anti-inflationary stance. An appealing aspect of the Romers' approach is that it uses additional information - specifically, policy-makers' statements of their own intentions - to try to disentangle money supply from money demand shocks. A disadvantage of this approach, besides inherent problems of reproducibility and subjectivity, is that it does not clearly distinguish between the endogenous and exogenous components of policy change, which is necessary for identifying the effects of monetary policy on the economy [Dotsey and Reid 1992; Leeper 1993; Shapiro 1994; Hoover and Perez 1994; Sims and Zha 1995; Bernanke, Gertler, and Watson 1997]. The Romers' methodology also yields a rather limited amount of information: they give dates only for contractionary changes in policy, not expansionary shifts, and their method provides no indication of the severity or duration of each episode. Building on the Romers' work, Boschen and Mills [1991] used FOMC documents to rate monetary policy in each month as "very tight," "tight," "neutral," "easy," or "very easy," depending on the relative weights the policy-makers assigned to reducing unemployment and reducing inflation. Although Boschen and Mills provide a more continuous and possibly more informative measure of policy than do Romer and Romer, their indicator likely also suffers relatively more severe problems of subjectivity and commingling of endogenous and exogenous policy changes.

The second general strategy for measuring monetary policy stance - which is the focus of the present article - is to use prior information about central bank operating procedures, in conjunction with vector autoregression (VAR) estimation techniques, to develop data-based indexes of policy. For example, Bernanke and Blinder [1992] argued that over much of the past 30 years the Fed has implemented policy changes through changes in the federal funds rate (the overnight rate in the market for commercial bank reserves). They concluded that the funds rate may therefore be used as an indicator of policy stance (see also Laurent [1988] and Bernanke [1990]); in particular, they interpreted VAR innovations to the funds rate as innovations to the Fed's policy. In a similar vein, Sims [1992] used short-term rates as monetary indicators in a multicountry study. However, not all researchers working in the VAR-based literature have adopted short-term interest rates as their preferred indicator of policy. Following a suggestion of Thornton [1988b], Christiano and Eichenbaum [1992] have made the case for using the quantity of nonborrowed reserves as the primary measure of monetary policy (also see Eichenbaum [1992]). Strongin [1995] proposed as a policy indicator the portion of nonborrowed reserve growth that is orthogonal to total reserve growth. He motivated this measure by arguing that the Fed is constrained to meet total reserve demand in the short run but can effectively tighten policy by reducing nonborrowed reserves and forcing banks to borrow more from the discount window. Cosimano and Sheehan [1994] characterized Fed policy after 1984 as borrowed-reserves targeting, which suggests that borrowed reserves might be a useful indicator for the more recent period.

Both the narrative and VAR-based methods for measuring monetary policy have been widely used in applied work.(2) Unfortunately, there is evidently little agreement on which of the various measures most accurately captures the stance of policy, leading many authors to hedge by using a variety of indicators. Eichenbaum and Evans' [1995] study of the effect of monetary policy on exchange rates is fairly typical in employing three alternative policy measures: in their case, Strongin's measure, innovations to the federal funds rate, and the Romer dates. However, although using alternative measures allows the researcher to claim robustness when the results for each indicator are similar, this strategy provides no guidance for cases when the results for different indicators are inconsistent. (Indeed, we show below that alternative indicators can lead to quite different inferences.) Moreover, simply using a variety of alternative measures of monetary policy cannot guarantee that some more accurate indicator has not been excluded; that the best indicator is not perhaps some combination of the various "pure" indicators; or that the best indicator is the same for all countries or for all periods. Thus, it would be quite useful to have a systematic method of comparing alternative candidate indicators of policy.

Eichenbaum [1992, p. 1010] has stressed the importance of finding a means of choosing among indicators, noting that in his particular application "inference depends very sensitively on which of the two candidate measures [short-term interest rates or nonborrowed reserves] we work with." He also suggests that "further progress on these issues can be made only by carefully studying the institutional details of how monetary policy is actually carried out in the different countries. . . ." Following Eichenbaum's suggestion, in this article we develop and implement a general, VAR-based methodology in which the indicator of monetary policy stance is not assumed but rather is derived from an estimated model of the central bank's operating procedure. More specifically, we employ a "semi-structural" VAR model that leaves the relationships among macroeconomic variables in the system unrestricted but imposes contemporaneous identification restrictions on a set of variables relevant to the market for commercial bank reserves.

Our method has several advantages over previous approaches. First, because our specification nests the best known quantitative indicators of monetary policy used recently in VAR modeling, including all those mentioned above, we are able to perform explicit statistical comparisons of these and other potential measures, including hybrid measures that combine the basic indicators. Second, our analysis leads directly to estimates of a new policy indicator that is optimal, in the sense of being most consistent with the estimated parameters describing the central bank's operating procedure and the market for bank reserves. Third, by estimating the model over different sample periods, we are able to allow for changes in the structure of the economy and in operating procedures, while imposing a minimal set of identifying assumptions. Finally, although we consider only the post-1965 U.S. case in this paper, our method is applicable to other countries and periods, and to alternative institutional setups.(3)

A frequently heard criticism of the VAR-based approach is that it focuses on monetary policy innovations rather than on the arguably more important systematic or endogenous component of policy. We believe this criticism to be misplaced. The emphasis of the VAR-based approach on policy innovations arises not because shocks to policy are intrinsically important, but because (as we discuss further below) tracing the dynamic response of the economy to a monetary policy innovation provides a means of observing the effects of policy changes under minimal identifying assumptions.(4) However, although we disagree with the view that analysis of policy shocks is uninteresting, we also recognize that it would be useful to have an indicator of the overall stance of monetary policy, including the endogenous component. An additional contribution of this article is to propose just such an indicator, one that is both consistent with our underlying approach and, we believe, intuitively appealing.

The rest of the article proceeds as follows. Section II briefly describes our general methodology for identifying innovations to monetary policy. Section III lays out a standard model of the market for bank reserves that nests some common alternative descriptions of Fed operating procedures. Estimation of this model by GMM (Section IV) allows us both to evaluate the leading candidate indicators of policy innovations and to develop an alternative measure. Section V discusses how the choice of policy measure affects our conclusions about the impact of monetary policy on the economy. Section VI introduces our total policy measure, inclusive of both the systematic and random components of policy, and compares it with narrative measures of monetary policy. Section VII concludes.

II. METHODOLOGY

Bernanke and Blinder [1992] proposed the following strategy for measuring the dynamic effects of monetary policy shocks. Suppose that the "true" economic structure is

(1) [Mathematical Expression Omitted]

(2) [Mathematical Expression Omitted].

Equations (1) and (2) define an unrestricted linear dynamic model that allows both contemporaneous values and up to k lags of any variable to appear in any equation.(5) Boldface letters are used to indicator vectors or matrices of variables or coefficients. In particular, Y is a vector of macroeconomic variables, and p is a variable indicating the stance of policy. Note that for the moment p is taken to be a scalar measure, e.g., the federal funds rate. Equation (2) predicts current policy stance given current and lagged values of macroeconomic variables and lagged policy variables, while equation (1) describes a set of structural relationships in the rest of the economy. The vector [v.sup.y] and the scalar [v.sup.p] are mutually uncorrelated "primitive" or "structural" error terms. As in Bernanke [1986], the structural error terms in equation (1) are premultiplied by a general matrix [A.sup.y], so that shocks may enter into more than one equation: hence the assumption that the elements of [v.sup.y] are uncorrelated imposes no restriction. The assumption that the policy shock [v.sup.p] is uncorrelated with the elements of [v.sup.y] is also not restrictive, in our view; we think of independence from contemporaneous economic conditions as part of the definition of an exogenous policy shock.(6)

The system (1)-(2) is not econometrically identified in general. Bernanke and Blinder point out that to identify the dynamic …