Exhibit 5 addresses the issue that the macroeconomic variables used to formulate real-time monetary policy decisions are sometimes poorly measured. There are two potentially important sources of measurement error. The first is that initial releases of key macroeconomic data are imprecise and subject to revision. To quantify the magnitudes of such revisions, we use a data set maintained by the Federal Reserve Bank of Philadelphia that records the values of major macroeconomic time series as they were available to policymakers at specific dates in the past.
The top left panel illustrates this type of uncertainty by showing the distribution of revisions to the quarterly growth rate of real output over the quarter following the initial data release for the period from 1965 to 2002. Clearly, revisions can be substantial. As shown in the first line in the top right panel, the average absolute size of this revision is about ⅔ percentage point of output. Moreover, as indicated in the remaining lines, subsequent revisions can also be large, in part reflecting rebasing and other methodological changes. The revisions to other NIPA variables show broadly similar patterns. As noted in the middle left panel, the second source of measurement error arises because many of the variables that have prominent roles in our economic models are not directly observed and must instead be estimated. These variables include potential output, expected inflation, and the equilibrium real interest rate. Inevitably, estimates of these variables are subject to significant error that can be highly persistent, given the substantial lags involved in detecting important structural changes in the economy. These two sources of measurement error generate considerable uncertainty about many of the variables that are integral to monetary policy decisions. Some of the strongest policy implications likely come from errors in the measurement of the output gap. The dotted line in the middle right panel shows the staff’s real-time assessment of the output gap since 1980—that is, the estimate of the contemporaneous output gap made at the time shown. Those estimates differ considerably from the staff’s current assessment of the output gap for that period, the solid line. As indicated in the inset, the real-time errors implied by the difference between these two estimates have a standard deviation of 1¾ percentage points of output and a high degree of serial correlation. As summarized in the bottom left panel, mismeasurement of the output gap could, in principle, have no effect on policy. This would be the case if the real-time estimate of the output gap were uncorrelated with subsequent revisions to that estimate, which might occur if the revisions were based on information not available at the time of the original estimate. In practice, however, that condition has not been met, and large initial estimates have often been revised to be smaller. Under such circumstances, the optimal policy should attenuate its response to the real-time output gap measure, in order to reduce movements in the interest rate in response to the noise in that measure.
To quantify the effect of measurement error on monetary policy, we return to the FRB/US exercise from exhibit 2, only now assuming that the real-time estimate of the output gap available to the FOMC contains a random error that has the same properties as those observed from 1980 to 1998. As the bottom right panel indicates, measurement error leads to some attenuation of the policy response to the measured output gap (the middle column). However, that coefficient remains well above its estimated value. Moreover, the other coefficients of the policy rule, including that on the lagged policy rate, are largely unaffected. Of course, mismeasurement of variables other than the output gap might also have important policy implications that are not captured by this exercise. Indeed, some recent research suggests that incorporating uncertainty about the equilibrium real interest rate might result in further attenuation and some additional inertia for the optimal policy rule.
The top panel of exhibit 6 summarizes our findings. A simple analysis of optimal policy—one that uses the backward-looking version of FRB/US and assumes that uncertainty enters only through additive shocks—indicates that monetary policy should move more forcefully in response to changes in macroeconomic variables and be less inertial than observed on average since the mid-1980s. We have investigated the sensitivity of that conclusion to three factors—forward-looking behavior, parameter uncertainty, and measurement error. These factors move the optimal policy in the direction of the estimated policy rule, but none of the factors alone seems to fully explain the observed smoothness of the federal funds rate. An important caveat to this finding is that we consider each of the factors separately, owing to the analytical difficulty involved in combining them. These factors likely interact in ways that could affect the desirable degree of interest rate smoothing.
Even with the extensions considered, our analysis surely fails to capture important aspects of the policymaking environment. For example, as noted in the bottom left panel, policymakers face considerable uncertainty about the structure of the model itself. All models are approximations and therefore ignore specific variables that could at times become relevant for policy decisions. Also, the models we have used for this briefing are essentially linear, whereas the economy may under some conditions demonstrate large, discrete responses to monetary policy or other events. Policymakers’ concerns about uncertainty may be exacerbated by the fact that some of the optimal policy rules considered called for large swings in the federal funds rate that are well outside of historical norms. The problems involved with significant nonlinearities, for example, might be reflected in a concern by the FOMC about financial fragility. Such a concern could generate a smoother path for the federal funds rate if large policy changes were viewed as having adverse effects on the functioning of financial markets. Transcripts of FOMC meetings show that members of the Committee have at times argued for smaller interest rate changes based on concerns about financial fragility.
The smoothness of the federal funds rate could also result from various institutional aspects of the policymaking process, as summarized in the bottom right panel. For example, the fact that policy decisions are made by a committee, and thus require building a consensus, could generate some inertia in realized policy actions. Alternatively, the FOMC might seek to avoid reversals in the direction of the policy instrument. Such an approach would presumably involve sacrificing some macroeconomic performance, but it might be perceived to have other benefits, such as allowing Committee members to more easily explain their policy choices to the public. Judging from simulations of the FRB/US model, reversals would occur much more often under the optimal policy rule than under the estimated policy rule.
Overall, while no model can encompass all the issues that pertain to interest rate smoothing, the analysis above at least provides some benchmarks against which to gauge the appropriate pace of monetary policy adjustment. Because the observed policy is much smoother than would be prescribed by these benchmarks, it is of interest to determine what aspects of our models or of our assumed preferences may be misspecified.