Do Actions Speak Louder Than Words? The Response of Asset Prices to Monetary Policy Actions and Statements

The paper asks whether the effect of FOMC announcements on asset prices is “adequately characterized by a single factor, namely the surprise component of the change in the current federal funds rate target” (Introduction, p. 56) – and answers no: two factors are required. A motivating example makes the point vividly: on January 28, 2004, two- and five-year Treasury yields jumped 20 and 25 basis points in the half-hour around the FOMC announcement – “the largest movements around any Federal Open Market Committee announcement over the fourteen years for which we have data” – even though the funds-rate decision itself (no change) was fully anticipated; the market was reacting to a wording change in the statement. The authors formalize and generalize this observation: alongside the familiar “current federal funds rate target” factor, a second “future path of policy” factor, closely tied to FOMC statements, is needed to describe how asset prices respond.

The authors use intraday tick data around every FOMC announcement from January 1990 to December 2004, comparing a “tight” 30-minute window (10 minutes before to 20 minutes after) and a “wide” one-hour window (15 minutes before to 45 minutes after) against a daily window for comparison. Data sources are tick-by-tick federal funds futures (Genesis Financial Technologies, traded at CBOT since October 1988), two-, three-, and four-quarter-ahead Eurodollar futures (ED2-ED4), tick-by-tick on-the-run Treasury yields (3-month, 6-month, 2-, 5-, 10-year) from GovPX available from June 1991, and 5-minute S&P 500 quotes. The funds-rate surprise itself is measured using the Kuttner (2001) method applied to current-month fed funds futures, scaled by D/(D-d) to account for the FOMC’s timing within the month, with the next-month contract substituted for meetings in the last seven days of a month. Narrowing the window to 30 minutes around the announcement is what neutralizes the simultaneity and omitted-variable bias (e.g., contemporaneous employment reports) that plagues daily- or monthly-frequency regressions of asset prices on policy surprises.

A Cragg-Donald (1997) matrix-rank test applied to a latent-factor model of the futures responses rejects both the zero-factor and one-factor hypotheses, but does not reject two factors. For the futures-only sample (138 announcements) the rejection p-values are approximately .00007 (zero factors) and .004 (one factor); for the Treasury-plus-stock sample (120 announcements) they are approximately .00004 and .011. In both samples the two-factor hypothesis is not rejected (p = .30 and .36, respectively). The authors summarize: “surprise changes in the federal funds rate alone are not sufficient to describe the response of asset prices” (Section 2.1). The two-factor finding is robust across both asset sets tested.

The two principal components estimated from futures rates with up to one year to expiration are rotated by an orthogonal matrix so that the second factor has no effect on the current-month funds futures rate; the first factor (“target”) then corresponds to surprise changes in the current funds-rate target, and the second (“path”) captures whatever in an announcement moves year-ahead rate expectations without moving the current funds rate. The target factor Z1 correlates 95% (R-squared = .91) with the standard funds-rate-surprise measure, confirming it recovers essentially the same object as prior single-factor approaches (e.g., Kuttner 2001, Cochrane-Piazzesi 2002). Scale is normalized so that a change of .01 in Z1 corresponds to a 1-basis-point funds-rate surprise, and Z2 is scaled so that its effect on the four-quarter-ahead Eurodollar rate matches Z1’s effect (about 55.1 basis points) – making the two factors’ coefficients directly comparable in the results below.

Regressing the absolute value of the path factor on a dummy for whether the FOMC released a statement yields a coefficient of 0.070 (statistically significant), with a constant of 0.044 and R-squared = .18, strongly rejecting the hypothesis that statements and the path factor are unrelated. Nine of the ten largest path-factor observations, and 21 of the top 25, fall on FOMC statement dates; the single largest is the January 28, 2004 episode described above (Z2 = 42.7 basis points). The authors are careful to note the path factor is nonetheless “to some extent a residual that is subject to various interpretations” (p. 78) – a few large realizations occur on non-statement dates (e.g., December 20, 1994), apparently driven by other events such as speeches.

In a joint two-factor regression, a one-percentage-point target surprise raises 2-/5-/10-year Treasury yields by 49/28/13 basis points and lowers the S&P 500 by about 4.3%, while a one-percentage-point path innovation raises 5-/10-year yields by 37/28 basis points – a larger effect at the long end than the target factor – while moving the S&P 500 by only about -1%. Comparing the R-squared from one-factor versus two-factor specifications, the path factor accounts for roughly two-thirds of the explainable variation in two-year yields, three-fourths in five-year yields, and nine-tenths in ten-year yields – summarized in the abstract as “75 to 90 percent” of the explainable variation in long-term yields being attributable to statements rather than to funds-rate actions. Separately, in the single-factor (target-only) event-study regressions reported in Table 1, the funds-rate surprise loads strongly on short rates and declines with maturity: tight-window coefficients of 0.537 (3-month bill, R-squared=.80), 0.455 (2-year), 0.264 (5-year), and 0.125 (10-year, R-squared=.08), with a negative coefficient of -0.087 on the five-year-ahead five-year forward rate – consistent with the companion GSS (2005, AER) finding that far-ahead forward rates move inversely with policy surprises.

An out-of-sample check using the first FOMC minutes released on the newly accelerated release schedule (January 4, 2005) finds yield movements broadly in line with the path-factor model’s predictions, though not exact. The minutes moved the one-year-ahead Eurodollar rate by +6.5 basis points and 2-/5-/10-year yields by +4.9/+5.0/+4.1 basis points, compared with path-factor-implied predictions of 4.8/4.4/3.3 basis points from the main-sample estimates – reasonably close. The S&P 500 fell 0.49% versus a predicted 0.11%, and the five-to-ten-year forward rate rose 3.0 basis points versus a predicted 1.9 – larger misses. The authors characterize the overall fit as “broadly in line” with the model rather than as precise confirmation.

The authors argue FOMC statements are “not an independent policy tool” but instead work “through their influence on financial market expectations of future policy actions” (p. 57), and that this gives the FOMC substantial room to influence long-term rates even near the zero lower bound. They stress the finding does not make policy actions secondary – rather, “their influence comes earlier, when investors build in expectations of those actions in response to FOMC statements (and perhaps other events, such as speeches and testimony).” A secondary, more tentative interpretation is that the path factor partly reflects revisions to investors’ expectations of future output and inflation (the Romer-Romer 2000 central-bank-information channel), which the authors suggest could explain why the stock market’s response to the path factor is muted relative to its large long-end yield effect – though they note Faust, Swanson, and Wright (2004b) question this information-effect story, at least for the target factor. The policy implication drawn is that the FOMC can commit to a state-contingent path for the funds rate years into the future and so remains “largely unhindered” even at a low or zero nominal funds rate, which the authors read as directly supporting Reifschneider-Williams (2000) and Eggertsson-Woodford (2003).

The authors flag four main limitations: the path factor’s residual, interpretation-dependent character; slower market assimilation of path-factor information than target-factor information; a data-coverage gap in the Treasury sample; and an unresolved puzzle in the stock-market response to the path factor. Specifically: (1) the path factor is described as “to some extent a residual that is subject to various interpretations” (p. 78); (2) markets appear to take time to digest path-factor information – a wide-window measure of the path factor regressed on its tight-window counterpart gives R-squared = .83, versus .98 for the target factor, implying assimilation of statement content is not instantaneous; (3) intraday Treasury data begin only in June 1991, about 18 fewer observations (and 5 rather than 8 employment-report dates) than the funds-rate-only analysis; and (4) the roughly -1% stock-market response to the path factor is called “somewhat surprising” given the factor’s large long-end yield effect, and the proposed output/inflation-expectations reconciliation is explicitly conjectural. Standard errors on the generated-regressor (factor) results were bootstrapped (1,000 repetitions) and found “essentially identical” to asymptotic standard errors, and wide-window results are reported as “in line with” the tight-window results throughout.