Term structure evidence on interest rate smoothing and monetary policy inertia
📄 Summarized from the full manuscript · Human-reviewed for faithfulness before publication
In brief
Does the Fed deliberately move rates in small gradual steps toward a new target? Estimated policy rules find only 20 to 30 percent of a desired change happens within the quarter, widely read as deliberate inertia. This paper tests that against the term structure: real quarterly smoothing should let markets forecast much of the future path. In eurodollar futures there is essentially no predictive power beyond about three months. A rule with immediate full adjustment but persistent, serially correlated shocks fits the funds-rate data just as well, and matches the term-structure evidence smoothing cannot. So the apparent inertia is largely a statistical illusion from omitted persistent shocks, not genuine gradualism.
What this paper finds — and why it matters
Estimated Taylor-type policy rules typically find the Fed adjusts the funds rate only 20-30 percent of the way to its desired level each quarter, widely read as deliberate “interest rate smoothing.” Reviewing estimates of both a backward-looking Taylor rule and the forward-looking Clarida-Galí-Gertler variant on 1987-1999 U.S. data, the paper confirms the standard result: partial-adjustment coefficients around 0.7-0.9, fitting the data far better than a version with no lagged rate at all. The paper then tests the interest-rate-smoothing interpretation against an implication it must carry: if the funds rate genuinely adjusts only partially each quarter, a large share of its future path should be predictable from information already available, and rational financial markets should price that predictability into the term structure. Using eurodollar futures rates to construct real-time forecasts, the paper finds an R² of 0.57 for the funds-rate change one quarter ahead, but only 0.11 two quarters ahead and 0.03 three quarters ahead – essentially no forecastable variation beyond about three months, sharply at odds with what a highly inertial policy rule implies. The paper then shows that a policy rule with immediate, full adjustment (no smoothing at all) but a persistent, serially correlated shock term fits the historical funds-rate data just as well as the standard partial-adjustment rule on conventional goodness-of-fit measures, and formal tests cannot reliably distinguish the two specifications within available samples. Because only the serially-correlated-shock version is consistent with the term structure’s near-total unpredictability of the policy rate beyond a quarter, the paper concludes that the large lag coefficients found throughout the policy-rule literature are largely a statistical illusion generated by omitted persistent shocks to policy, not evidence that central banks deliberately smooth interest-rate changes over multiple quarters.
Summary of a classic paper, AI-assisted and human-reviewed. See the linked original for the authoritative claims and full conditions.
Questions & answers
Q1. What is the “conventional wisdom” about monetary policy inertia that the paper challenges?
That estimated Taylor-type policy rules with a lagged interest-rate term consistently find a large, significant coefficient on the lagged rate – commonly 0.7 to 0.9 – widely interpreted as central banks deliberately smoothing rate changes over several quarters (Section 1-2). The paper quotes Clarida, Galí, and Gertler’s (2000) own characterization directly: “the estimate of the smoothing parameter ρ is high in all cases, suggesting considerable interest rate inertia: only between 10 and 30 percent of a change in the [desired interest rate] is reflected in the Funds rate within the quarter of the change… our estimates confirm the conventional wisdom that the Federal Reserve smooths adjustments in the interest rate” (Section 1, p. 1162). The author reproduces this result on 1987-1999 U.S. data for both a backward-looking rule (ρ̂₁ = 0.73) and a forward-looking, Clarida-Galí-Gertler-style rule (ρ̂₂ = 0.79), each fitting far better than the corresponding non-inertial (ρ = 0) version (Section 2, eqs. 3-7).
Q2. Why does the paper distinguish “quarterly” from “short-term” (weekly/monthly) interest rate smoothing?
Because an earlier, separate literature on day-to-day and week-to-week Fed behavior explicitly rejects any partial adjustment at a quarterly frequency, even though it documents genuine smoothing at shorter horizons (Section 1, p. 1162-1163). Quoting Mankiw and Miron (1986): the postwar term structure suggests that “while the Fed might change the short rate in response to new information, it always (rationally) expected to maintain the short rate at its current level” beyond about a month; and Goodfriend (1991): “changes in the rate set by the Fed… are essentially unpredictable at forecast horizons longer than a month or two.” The paper argues the two literatures are “in fact largely independent” because they concern very different time frames, and that many quarterly-rule papers have wrongly assumed short-term smoothing implies quarterly smoothing (Section 1, p. 1163).
Q3. What theoretical argument makes quarterly policy inertia plausible in the first place?
In a forward-looking New Keynesian model, gradual adjustment of the policy rate can be optimal because private-sector demand and pricing decisions respond to expected future interest rates, so a central bank credibly committed to a gradual rule can “alter expectations of future interest rates that are also important determinants of current demand,” achieving more stabilization for a given current rate move (Section 3, p. 1167-1168, citing Woodford 1999, Rotemberg and Woodford 1999, Levin et al. 1999). Calibrating such a model, the paper finds “a large range of optimal lag coefficients – between 0 and 0.8 – can be rationalized for some combination of model and loss function,” with the optimal degree of smoothing turning out to depend less on the persistence parameters of inflation or output than on how forward-looking interest-rate expectations are assumed to be (Section 3, p. 1171).
Q4. What is the term-structure test, and why should genuine policy inertia leave a trace there?
If the funds rate genuinely adjusts only, say, 20 percent toward its target each quarter, “the remaining 80 percent adjustment should be expected to occur in future quarters” – a predictable future path that rational financial markets, under the expectations hypothesis, should price into the term structure (Section 5, p. 1174-1175). The paper operationalizes this with a regression of the realized future change in the funds rate on the change implied by eurodollar futures rates, which under rational expectations should show a slope near one and a high R² if the future path is genuinely foreseeable (Section 4, eq. 14).
Q5. What do the eurodollar-futures regressions actually show?
Predictability collapses sharply beyond one quarter: the R² is 0.57 for the one-quarter-ahead change in the funds rate, but only 0.11 for the two-quarter-ahead change and just 0.03 for the three-quarter-ahead change (Section 4, eqs. 15-17, p. 1172-1173). “These regressions indicate that there is little if any information usually available in financial markets for predicting the change in the funds rate 3-6 months out… and no information for predicting it 6-9 months out” (p. 1173) – a pattern the paper argues is inconsistent with the large amount of “pent-up” future adjustment that a partial-adjustment coefficient of 0.7-0.9 implies.
Q6. How does the paper show that “smoothing” and “persistent shocks” are hard to tell apart statistically?
By estimating an alternative version of the same policy rules that assumes immediate, full adjustment (no lagged rate at all) but allows the regression error to be a highly persistent AR(1) process, and showing it fits the historical data about as well as the standard partial-adjustment specification (Section 6, eqs. 18-19, p. 1178). The serially-correlated-shock version of the backward-looking rule has an AR(1) shock coefficient of 0.92, and the forward-looking version 0.77, with response coefficients on inflation and the output gap “broadly comparable” to the partial-adjustment estimates. Testing a general specification nesting both hypotheses, the paper finds “it is difficult to obtain decisive direct empirical evidence against either” – p-values of 0.18 and 0.14 for the two hypotheses in the main sample – and that “the evidence appears quite fragile to even modest changes in the sample,” with each hypothesis rejected outright in slightly different subsamples (Section 6, p. 1178-1179).
Q7. Why does the paper favor the serially-correlated-shocks interpretation over genuine smoothing, given this statistical near-equivalence?
Because the two interpretations are not equivalent in what they imply for the term structure, and only one of them matches the evidence: “persistent deviations from an output and inflation response occur because policymakers are slow to react” under smoothing, versus reflecting “the policymaker’s response to other persistent influences” under the shock interpretation, and “[o]nly the serially correlated shocks rule is consistent with the historical evidence showing that the term structure is largely uninformative about the future course of the policy rate” (Section 7, Conclusion, p. 1183-1184). The paper thus uses the term structure as an independent, out-of-sample discriminator between two specifications that are nearly indistinguishable using the funds-rate data alone.
Q8. What alternative explanations does the paper consider and set aside, and what caveat does it leave for future work?
It considers, and largely sets aside, the possibility that the rational-expectations hypothesis of the term structure fails, or that time-varying term premia are driving the low predictability results, noting that abandoning rational expectations “would undermine many aspects of any explicitly forward-looking macroeconomic modeling exercise,” and that if short rates were dominated by volatile term premia “it seems unlikely that they can communicate the subtle expectations of future monetary policy” that the optimal-inertia literature itself requires (Section 7, p. 1184). The paper allows that “some intermediate case,” with a modest smoothing coefficient (e.g., around 0.4) alongside serially correlated shocks, might not be strictly rejected, but notes that real-world Fed communication describes only week-to-week “incrementalism,” never quarterly smoothing, quoting a July 2000 New York Times account of Alan Greenspan basing a rate decision on “economic data released in coming weeks” rather than on a pre-planned multi-quarter adjustment path (Section 7, p. 1184).
Key terms in this paper
Definitions below follow the paper's own usage.
- Partial-adjustment (interest rate smoothing) policy rule
- The standard empirical specification, following Taylor (1993) and its forward-looking Clarida-Galí-Gertler variant, of the funds rate as a weighted average of a "desired" rate responding to inflation and the output gap and last quarter's actual rate: i_t = (1-ρ)(desired rate) + ρ i_{t-1} + shock. Estimated on U.S. data, ρ typically falls in the 0.7-0.9 range, which the literature "widely interpreted" as "sluggish adjustment of the policy rate to its determinants" (Sections 1-2).
- Term-structure predictability test
- The paper's central empirical test: because a partial-adjustment rule with ρ around 0.8 implies "the remaining 80 percent adjustment should be expected to occur in future quarters," rational-expectations financial markets should be able to forecast a large share of future policy-rate changes; using eurodollar futures to construct such forecasts, the paper finds an R² of 0.57 for one-quarter-ahead changes but only 0.11 two quarters out and 0.03 three quarters out -- "little if any information usually available in financial markets for predicting the change in the funds rate 3-6 months out... and no information... 6-9 months out" (Section 4).
- Near-observational equivalence of "smoothing" and serially correlated shocks
- The paper's demonstration, following the inventory-adjustment econometrics literature, that a policy rule with zero smoothing (ρ = 0) but a serially correlated shock (AR(1) coefficient of 0.92 for Rule 1 and 0.77 for Rule 2) fits U.S. data "as well as" the standard partial-adjustment rule on conventional fit statistics, and that formal "common factor" tests using a general nesting specification cannot reliably distinguish the two forms -- p-values for rejecting either hypothesis are fragile across only modestly different subsamples (Section 6).
- The illusion of monetary policy inertia
- The paper's conclusion that the large lag coefficients found in estimated policy rules "reflect serially correlated or persistent special factors or shocks that cause the central bank to deviate from the policy rule" rather than deliberate gradual adjustment, because "[o]nly the serially correlated shocks rule is consistent with the historical evidence showing that the term structure is largely uninformative about the future course of the policy rate" (Section 7, Conclusion).