Why Are Target Interest Rate Changes so Persistent?

The Fed’s target interest rate is highly persistent (an AR(1) coefficient of about 0.9-0.99 in standard Taylor-rule regressions), and the paper asks whether this reflects deliberate “interest rate smoothing” (partial-adjustment policy inertia) or serially correlated monetary policy shocks (persistent deviations from the Taylor rule). The two are observationally similar in a simple regression but imply “starkly different” things for policy analysis: under smoothing the Fed can commit to and credibly signal a gradual path, while under persistent shocks the target is being driven by serially correlated disturbances that may reflect a systematic but unmodeled response to omitted variables (pp. 126-129).

The paper uses real-time Federal Reserve Greenbook forecasts of inflation and the output gap – released publicly only after a five-year lag – as the best available measure of what the Fed actually knew and expected at each FOMC meeting, rather than relying on final, revised data that were unavailable to policymakers in real time. This real-time information set is combined with the discrete federal funds rate target and, in later tests, Eurodollar futures as a private-sector benchmark. The quarterly sample runs 1987:IV-2006:IV, a window set by Greenbook availability and ending before the 2008 zero-lower-bound episode, so the results do not speak to the post-2008 forward-guidance era (Section II, p. 133).

When the paper nests both a smoothing parameter and a persistent-shock parameter in a single AR(1) specification, both come out statistically significant (smoothing coefficient about 0.81, shock persistence about 0.46), and the paper finds that this simple nested specification cannot overwhelmingly differentiate between the two explanations. Estimated separately, a pure interest-smoothing specification (Eq. 2) fits with a smoothing coefficient of about 0.83 and R-bar-squared of 0.99, while a pure persistent-shock specification (Eq. 3) fits about as well with shock persistence of about 0.96 and R-bar-squared of 0.97 – “both specifications fit the data similarly well” (Table 1, p. 137; p. 135).

Once the paper allows higher-order dynamics – AR(p) smoothing lags and AR(q) shock lags simultaneously – and lets BIC select among the resulting grid of specifications, the selected model is robustly IS(2), AR(0): two lags of interest smoothing and zero persistent-shock lags, across the full sample and the pre-2000 and post-1990 subsamples. Under an AR(2)-smoothing-plus-AR(2)-shock specification, the two smoothing coefficients are large and significant (about 1.28-1.39 at the first lag and -0.45 to -0.62 at the second, summing to roughly 0.75-0.95), while the shock-persistence terms become insignificant at the first lag and negative and significant at the second – persistent shocks “fail to explain the data” once smoothing is given enough flexibility (Table 3, p. 141; Table 4, p. 142).

To address endogeneity of the lagged rate, the paper instruments it with non-monetary shocks (oil prices, technology shocks, fiscal policy shocks); the IV-estimated smoothing coefficient remains 0.70-0.87 across specifications, which the authors read as confirming that “interest rate smoothing is not simply an artifact of monetary policy shocks.” The paper also checks whether financial conditions (a financial-uncertainty measure, credit spreads, S&P 500 returns), lagged Greenbook forecast revisions, or time-varying-inflation-target measures (following Cogley-Sbordone-Yun and Ireland) can account for the estimated persistence; none fully does – the time-varying-target measures reduce residual persistence somewhat but the smoothing coefficient “remains significant,” and the paper concludes it “cannot fully account for the interest rate smoothing using measures of omitted variables” (Table 5, pp. 143-145; Table 7, pp. 153-157).

The paper directly tests Rudebusch’s predictability critique: Greenbook-based forecasts predict future target-rate changes at 2- and 3-quarter horizons with R-squared of 20% and 12% respectively, within the range the authors judge consistent with genuine smoothing, whereas Eurodollar futures predict less well (R-squared of 10% and 2%), which is what motivated Rudebusch’s skepticism. The authors’ reconciliation is that the relevant predictor is the Fed’s own real-time information set (Greenbook), not public market prices, and they note private-sector predictive power converges toward Greenbook-implied levels after 2000, suggesting markets came to understand Fed gradualism better over time (Table 6, pp. 145-148).

The preferred interpretation is that the Fed deliberately moves the target gradually across meetings rather than adjusting in large discrete steps, for reasons the authors link to robustness under model uncertainty, financial-market stability, credibility and signaling, and learning as new information arrives. This is reinforced with narrative evidence: at the March 22, 1994 FOMC meeting, Greenspan expressed a preference for 25-basis-point increments over larger moves as a way of preserving credibility and option value, and Bernanke’s May 20, 2004 speech “Gradualism” explicitly endorsed gradual adjustment as reducing risks from uncertainty and allowing the Fed to learn from market reactions (Section V, pp. 149-150).

Under the interest-smoothing interpretation, the implied long-run response of the target rate to inflation exceeds one, satisfying the Taylor principle; under the no-smoothing persistent-shock interpretation, the implied contemporaneous inflation response can fall below one, a “pathological” result inconsistent with equilibrium determinacy. The authors read this as an additional point favoring the smoothing interpretation, since it restores an economically sensible reaction function rather than requiring the Fed to have violated the Taylor principle (p. 134).

All results are confined to 1987:IV-2006:IV, a sample bounded by the five-year Greenbook publication lag and ending well before the 2008 zero lower bound, so the findings say nothing about how gradualism operated once the funds rate hit its floor and forward guidance became the Fed’s main tool. The analysis is also at quarterly frequency, aggregating within-quarter FOMC decisions rather than modeling meeting-by-meeting dynamics, and the Greenbook real-time data used to characterize the Fed’s information set cover only inflation and the output gap, not other variables (financial conditions, exchange rates, credit spreads) that could independently enter the reaction function (p. 133).