A New Measure of Monetary Shocks: Derivation and Implications

The actual federal funds rate is contaminated by two distinct forces that bias estimates of monetary policy’s effects in opposite directions: endogeneity (the Fed raises rates during booms, creating a spurious positive correlation between tightening and future output) and anticipatory movements (the Fed pre-emptively raises rates in response to Greenbook forecasts of future inflation and growth, creating a spurious positive correlation between tightening and incoming price pressure that produces the “price puzzle”). The paper’s goal is to construct a policy indicator that isolates only deliberate, independent funds-rate changes – free of both contaminations – so that regressions and VARs using it recover more credible estimates of the real effects of monetary policy.

For each scheduled FOMC meeting between March 1969 and December 1996, the authors read the Record of Policy Actions and internal memos on the expected federal funds rate to identify the meeting-level intended change in the rate, producing a full series listed in Appendix Table A1. The series deliberately excludes intermeeting moves (the authors state “we do not have a record of the information that intermeeting moves are based on”), moves based on incorrect preliminary data, and meetings without accompanying Greenbook forecasts – exclusions that reduce precision slightly but, the authors argue, should not introduce bias.

The intended change in the funds rate at each meeting is regressed on the pre-meeting funds-rate level and the Greenbook forecasts (and forecast revisions since the previous meeting) of output growth, inflation, and unemployment at horizons from one quarter before to two quarters after the meeting; the residuals from this regression are the monetary policy shock series. Estimated over 263 FOMC meetings (1969:3-1996:12), the regression has R-squared 0.28 and Durbin-Watson 1.84. Selected coefficients: the sum of output-growth level-forecast coefficients is 0.04 (t=2.5); the sum of output-growth forecast-revision coefficients is 0.24 (t=4.0); the change in the current-quarter output-growth forecast is 0.15 (t=5.1); the sum of inflation level-forecast coefficients is 0.04 (t=2.3); and the current-quarter unemployment forecast coefficient is -0.048 (SE=0.021). The resulting meeting-level residuals are aggregated to a monthly shock series.

In the baseline single-equation specification (R-squared=0.86, N=324, sample 1970:1-1996:12, with 36 lags of the shock), a one-percentage-point contractionary shock produces a cumulative effect on industrial production that is slightly positive for the first four months, then declines roughly linearly, reaching an essentially flat -4.3 percent from month 22 through 27 – highly significant, with t-statistics exceeding 2.5 in every month from horizon 17 through 27 (two-standard-error interval at month 22: -7.0 to -1.4 percent). By month 48 the effect is -1.7 percent but no longer precisely estimated (two-standard-error interval -6.4 to +3.0 percent). A positive coefficient on the first lag of the shock (0.0038, SE=0.0018) is traced to a single April 1980 outlier; the Record of Policy Actions for that month shows no evidence the Fed’s aggressive easing reflected information beyond the Greenbook forecast, so the authors attribute it to sampling error. Setting that one observation to zero lowers the first-lag coefficient to 0.0023 (t=1.1) and the peak output effect to -3.4 percent.

Using PPI for finished goods (baseline specification, R-squared=0.57, 48 lags of the shock), the price level is virtually unchanged for the first 22 months after a one-percentage-point contractionary shock (a small, statistically insignificant rise of up to 0.3 percent in the first eight months), then falls substantially: -1.9 percent at month 30, -5.9 percent (t=5.5) at month 48 (two-standard-error interval -8.0 to -3.7 percent), with no statistically significant rise in prices at any horizon. This contrasts sharply with regressions using the actual federal funds rate, which show a sustained price puzzle: prices rise about one percent over the first two years and stay roughly flat through month 48 (two-standard-error interval at month 48: -1.4 to +3.3 percent). The authors write: “The fact that there is a strong price puzzle when the actual funds rate is used, but not when our new measure is used, strongly suggests that the funds rate is contaminated by endogenous and anticipatory movements.” Using CPI excluding shelter or the PCE price index instead of PPI gives somewhat smaller but still large and significant 48-month effects of about -3.6 percent (t=4.3 and t=5.0, respectively).

Adding the contemporaneous value and 12 lags of a world commodity price index changes the estimated effects only modestly: the cumulative output effect becomes -1.8 percent at 12 months (versus -2.1 percent without the control) and -3.2 percent at 24 months (versus -4.3 percent); the cumulative price effect becomes -0.2 percent at 24 months (versus -0.5 percent) and -4.4 percent at 48 months (versus -5.9 percent). The authors read these small, delayed changes as evidence the measure is largely free of anticipatory actions motivated by supply-side information. By contrast, when commodity prices are added to a regression using the actual funds rate, the price puzzle is reduced but not eliminated – the estimated price impact remains positive, though no longer significantly different from zero – leading the authors to conclude that “controlling for commodity prices is only a partial solution to the problem of forward-looking policy-making.”

A three-variable recursive VAR (log industrial production, log PPI, and the cumulated shock measure, following a variant of Christiano-Eichenbaum-Evans (1996) with three years of lags) broadly confirms the single-equation findings: output rises slightly for the first two months, falls through month 23 to a peak of -2.9 percent (maximum t-statistic over 3), then returns most of the way to its initial level; prices are small and insignificant for eight months, then fall to -0.5 percent at month 18, -2.4 percent at month 30, and -5.0 percent at month 48 (t-statistics reaching 2, 3, and 4 at months 20, 24, and 26). The smaller peak output effect relative to the single-equation estimate (-2.9 versus -4.3 percent) reflects that the VAR’s shock is transitory (falling to about half its initial size within a year) rather than the permanent shock assumed in the single-equation exercise. When commodity prices are added to the VAR, the new-measure output effect peaks at -1.9 percent after 22 months (t=2.7) and the price effect reaches -3.8 percent at 48 months (t=4.5); estimated on the same commodity-price-augmented specification, a VAR using the actual funds rate produces a maximum output effect of only -0.9 percent and a 48-month price effect of only -0.5 percent, and its price puzzle is only partly resolved. The authors summarize that their VARs “suggest larger but not faster output effects of monetary policy than found in previous VAR studies” such as Sims (1992), Christiano et al. (1996), and Bernanke and Mihov (1998), which find maximum industrial-production effects of roughly 1 to 1.5 percent versus 2 to 3 percent with the new measure.

Dropping the October 1979-May 1981 nonborrowed-reserve-targeting episode lowers the peak output effect from -4.3 to -3.4 percent and the 48-month effect from -1.7 to +0.2 percent, with standard errors rising by about 10 percent, while splitting the sample around 1983 has little effect on the coefficient sums, paralleling Orphanides (2003)’s finding that the Fed’s reaction function has been stable over time. Using 48 rather than 36 lags of the policy measure in the output regression barely changes estimates through month 36 but lowers the 48-month effect from -1.7 to -0.8 percent (more mean reversion at long horizons). Three alternative shock series considered in Section I.C – using only the intended-rate change, only residuals from regressing the actual rate on Greenbook forecasts, or estimating the Greenbook-forecast relationship separately before and after 1983 – are all correlated above 0.97 with the baseline series and give qualitatively similar output effects (for example, a peak industrial-production impact of -3.9 percent with the pre/post-1983 split).

The authors identify the lag before prices respond as the paper’s “most important uncertainty”: in some specifications the price level begins falling within six months of the shock, while in others it is unchanged for as much as 22 months. Other flagged limitations: the sample cannot extend past December 1996 because Greenbook forecasts are embargoed for five years, so the analysis omits the later Greenspan tenure and, since the whole approach relies on a meaningful “intended funds rate,” cannot be applied at all to the post-2008 zero-lower-bound period; intermeeting policy moves and moves based on incorrect preliminary data are excluded for lack of an adequate information base, reducing precision though not, the authors argue, introducing bias; and the positive first-lag coefficient in the output regression is attributed to a single April 1980 outlier rather than a genuine short-run response, a judgment based on finding no evidence in the Record of Policy Actions that the easing reflected information beyond the Greenbook forecast.