Monetary policy surprises and interest rates: Evidence from the Fed funds futures market

Prior event studies of Fed policy and interest rates — notably Cook and Hahn (1989) and Roley and Sellon (1995) — regressed changes in market rates on the raw, undecomposed federal funds target change, without separating the anticipated portion (already priced in by efficient markets) from the unanticipated portion, producing a classic errors-in-variables problem that attenuates the estimated response toward zero. Kuttner shows the scale of this bias directly: regressing the 3-month bill-rate change on the raw target change yields a coefficient of only 26.8 basis points (R²=0.42), whereas after decomposing the target change into anticipated and unanticipated components using Fed funds futures prices, the coefficient on the surprise component alone rises to 79.1 basis points — more than twice as large (Table 3, p. 533, contrasted with Table 1).

The spot-month Fed funds futures rate on day t of month s equals the market’s expectation of the average overnight funds rate over the remaining days of the month plus a risk premium (Eq. 2, p. 528); the one-day “surprise” is the one-day change in this spot-month futures rate, rescaled by the factor m/(m−t) — where m is the number of days in the month — to convert the change from a monthly-average unit into a point-in-time target-rate unit (Eq. 7, p. 529). This rescaling also differences away the risk premium term, since it is assumed roughly constant from one day to the next. Because the m/(m−t) factor becomes very large and noisy in the final three days of the month, Kuttner switches to the one-month-ahead futures contract for changes occurring in that window. The anticipated component is then simply the actual target change minus this surprise. Timing conventions differ across the sample: before 1994, target changes are dated to the day after the announcement (when they took effect); from February 1994 onward, changes are dated to the 2:15 p.m. announcement itself; two episodes (18 December 1990 and 15 October 1998) required additional inter-day timing corrections (pp. 530–531).

In the baseline one-day event-study regression across the 42 target-change days from June 1989 to February 2000, the anticipated component’s coefficient is small and statistically insignificant at every maturity (e.g., 4.4 basis points, t=0.8, at 3 months), while the unanticipated component’s coefficient is large and highly significant everywhere — 79.1bp (t=8.4) at 3 months, 71.6bp (t=8.5) at 6 months, 71.6bp (t=7.8) at 12 months, 61.4bp (t=6.0) at 2 years, 48.1bp (t=4.3) at 5 years, 31.5bp (t=3.1) at 10 years, and 19.4bp (t=2.3) at 30 years (Table 3, p. 533). A Wald test rejects equality of the anticipated and unanticipated coefficients at the 0.05 level or better for every maturity, consistent with the efficient-markets prediction that only genuine news should move prices.

The unanticipated-component coefficient declines roughly monotonically with maturity — from about 79bp at 3 months down to about 19bp at 30 years — a pattern qualitatively consistent with the expectations hypothesis, since a persistent shift in short rates should have a proportionally smaller effect on long yields. However, the response is less than proportional to what a purely permanent level shift would imply, which Kuttner interprets as evidence that surprise target changes are read by markets largely as shifts in the timing of a change the Fed was going to make anyway, rather than as changes in the ultimate level of rates — so an “earlier-than-expected” move has a smaller long-rate effect than an “unexpected-level” move of the same size (Section 5, pp. 539–543).

Regressing changes in futures rates one to five months ahead on the same anticipated/unanticipated decomposition, the coefficient on the unanticipated component is between 0.55 and 0.64 at every horizon, and a Wald test cannot reject that these coefficients are equal across horizons (Table 7, p. 542). Kuttner reads this as showing that “surprise target rate changes have little effect on expectations of future Fed actions” (p. 541): a surprise moves the perceived level of the target for the near future roughly uniformly across horizons, rather than signaling that further, additional changes are coming.

Restricting the sample to the 86 scheduled FOMC meeting dates from June 1989 to February 2000 leaves the pattern qualitatively unchanged — the 3-month unanticipated coefficient is 62.8bp (t=9.4) versus a small anticipated coefficient of 9.0bp (t=2.5) (Table 5, p. 537) — and re-estimating at monthly frequency (Table 6, June 1989–January 2000) gives a 3-month unanticipated coefficient of 74.3bp (t=9.8) and a 30-year coefficient of 51.9bp (t=6.6), somewhat larger than the daily estimates, especially at long maturities. At monthly frequency the anticipated coefficient on bills is also larger (3-month: 58.0bp, t=3.3), which Kuttner notes is consistent with the expectations hypothesis, since a month’s bill-rate change can incorporate revisions to the expected rate path that occurred at any point during that month, not just on the announcement day.

Regressing bill-rate changes on the previous day’s target change and surprise yields R² of 0.04 or less in every specification (Table 4, p. 536), indicating essentially no delayed price response. Kuttner attributes this to increased Fed transparency after 1994: “since 1994, no guesswork whatsoever has been required, as target rate changes have been announced immediately following FOMC meetings” (p. 535), so there is little scope for a lagged reaction once the change itself has occurred.

Two dates — 17 May and 16 August 1994 — saw 50-basis-point surprise rate increases accompanied by bond-market rallies, the opposite of the expected sign, because the accompanying FOMC statements signaled that no further tightening was imminent. Kuttner notes these two observations would otherwise distort the 5-year-maturity scatterplot (Fig. 2, p. 534) and treats them as evidence that the content of the accompanying policy statement — not just the numerical size of the surprise — can drive the long-rate reaction (Section on robustness, p. 534).

The sample is necessarily confined to 6 June 1989 through 2 February 2000 because the CBOT Fed funds futures market began trading only in 1989, and the surprise measure’s derivation assumes at most one target change per calendar month, an assumption violated in three months in the sample (footnote 4, p. 529). The time-averaging correction may also be contaminated by a monthly-varying futures risk premium, and the differing pre-/post-1994 timing conventions for dating target changes add sensitivity to data-source vintage. Finally, the paper documents only the response of interest rates themselves; it does not trace the surprise’s implications through to output or inflation.