The Sensitivity of Long-Term Interest Rates to Economic News: Evidence and Implications for Macroeconomic Models

The paper asks whether far-ahead forward interest rates respond to daily macroeconomic and monetary-policy surprises, and finds that they do – a result the authors argue is inconsistent with standard New Keynesian models. In the purely forward-looking Clarida-Gali-Gertler (2000) model and the more persistent, partially backward-looking Rudebusch (2001) model, the short-term interest rate returns to its steady state within about ten years after any macroeconomic or monetary-policy shock (Section I, Figure 1). If that is true, forward rates far enough in the future should reflect only the (constant) steady-state rate and show essentially no reaction to today’s news. The paper sets out to test this prediction directly using daily data on very-long-horizon forward rates.

Macroeconomic surprises are standardized deviations of 13 data releases from their median private-sector forecast, and the monetary-policy surprise is the change in the current-month federal funds futures rate around FOMC announcements. For each of 13 releases (capacity utilization, consumer confidence, core CPI, the employment cost index, GDP advance, initial claims, leading indicators, NAPM/ISM, new home sales, non-farm payrolls, core PPI, retail sales, and the unemployment rate), the surprise equals the released value minus the median Money Market Services survey forecast published the Friday before, divided by the surprise’s own historical standard deviation (Section II.A). The monetary policy surprise is the change in the current-month fed funds futures contract rate on the day of an FOMC announcement, scaled up to account for the announcement’s timing within the month (using the next-month contract when the announcement falls in the last seven days of the month), following the futures-based approach later associated with Kuttner (2001). Two categories of dates are excluded: the September 17, 2001 inter-meeting emergency cut, and eight 1990-1994 policy surprises that coincided with employment-report release days, due to an endogeneity concern (Section II.B, footnotes 8-9).

The regression is an OLS of the daily change in the forward rate at a given horizon on the full set of macro and monetary-policy surprises, run separately for each horizon with Huber-White standard errors. Forward rates – rather than yields – are used because far-horizon forward rates reflect expected future short-term rates at that specific horizon without mixing in near-term rate expectations the way a yield to a given maturity does (Section II). Table 1 reports results for the forward rate horizons ending 1, 5, and 10 years ahead; Figure 2 extends the coefficient plot out to 15 years ahead. Because the specification only measures the same-day revision in the market’s expectation of that future forward rate, the regression identifies a revision to expectations, not a forecast of the level of rates far in the future.

Eleven of the 13 macro surprises are significant at the 10-percent level for the five-year-ahead forward rate, and ten remain significant at the ten-year-ahead horizon, with point estimates that are economically sizable and highly persistent out to 15 years. From Table 1, a one-standard-deviation surprise moves the forward rate ending five years ahead by, for example, 4.42 basis points for the employment cost index, 4.12 bp for GDP advance, 3.48 bp for non-farm payrolls, 3.29 bp for NAPM, and 2.88 bp for consumer confidence (all significant at 1-5%); at the ten-year-ahead horizon the corresponding responses are typically smaller but still significant for most releases (e.g., 3.73 bp for the employment cost index, 1.88 bp for non-farm payrolls). The authors describe the persistence of these effects out to 15 years as “remarkable” (p. 430).

A monetary policy tightening surprise raises near-term forward rates but causes forward rates roughly 9-15 years ahead to fall significantly – moving in the opposite direction from the policy surprise – which the authors argue is not actually inconsistent with earlier findings that long-term yields move in the same direction as policy, because those studies examined yields (which mix in near-term rate expectations) rather than far-ahead forwards, and used less precise surprise measures. At the one-year-ahead horizon the policy surprise coefficient is 0.47 (significant at 1%), consistent with the persistence of the federal funds rate and Taylor-rule logic, but Figure 3 shows the sign flipping and becoming significantly negative at long horizons. This is explicitly contrasted with Cook and Hahn (1989) and Romer and Romer (2000), who report long-term yields moving in the same direction as policy actions (Section II, pp. 430-431).

Because in both benchmark models considered, the short rate is mean-reverting enough that its response to any single shock is essentially exhausted within about ten years, so the model-implied far-ahead forward rate response should be close to zero – directly at odds with the sizable responses estimated in Table 1 and Figures 2-3. Figure 1 plots the short-rate impulse responses in the CGG (2000) and Rudebusch (2001) models to representative macro and policy shocks: even in the more persistent Rudebusch specification, the short rate has essentially returned to steady state within ten years for each shock type (Section I).

The authors propose that private agents’ estimate of the Federal Reserve’s unobserved long-run inflation target is not fixed but updates over time in response to incoming data, and show that adding this single channel – with small updating parameters theta = 0.02 and kappa = 0.1 – lets an otherwise standard model reproduce the full set of empirical forward-rate patterns. Starting from the Fisher relation that the steady-state nominal short rate i* equals the real rate r* plus the inflation target pi*, and that the far-ahead forward rate approximately equals i* plus a risk premium, the model in Section III lets the central bank’s actual (unobserved) target pi* follow an AR(1) around trailing inflation (Equation 4) and lets private agents’ estimate pi-hat* update each period based on trailing inflation and the wedge between actual and expected short rates, in a rule resembling Kalman filtering (Equation 5). The authors state that “this simple modification to the baseline model can account for all of the patterns of forward rates we see in the data” (p. 433, Figure 4).

Survey-based long-run inflation expectations move in a way that closely tracks the forward-rate evidence, and TIPS-based (real) forward rates also respond to macro surprises, which the authors read as evidence against a changing real long-run rate r as the main driver – though they acknowledge they cannot cleanly separate a changing inflation-risk premium from a changing inflation target.* The Survey of Professional Forecasters’ median ten-year CPI inflation forecast fell from about 4 percent in 1991Q4 to a little under 2.5 percent by the end of 2002, a decline the authors say “matches closely” the decline evident in the forward-rate data and other survey measures (p. 434). Because TIPS-based (inflation-protected) forward rates also react to macro surprises, changes in the real rate r* are judged unlikely to be the primary explanation (citing the 2003 working paper predecessor, footnote 15). The authors note, however, that changes in the risk premium associated with long-horizon inflation uncertainty are “inherently difficult” to distinguish from changes in the long-run expected inflation rate itself, and argue this ambiguity is secondary to the paper’s main conclusion since both channels point toward imperfectly anchored long-run expectations (p. 434).

The results are estimated on daily U.S. data from January 1990 to December 2002 and therefore predate the zero lower bound and unconventional monetary policy, the inflation-target model’s key approximation is stated to hold only while inflation stays within the relatively low range observed in the sample, and identification of the monetary policy surprise before 1994 relies on inferring surprises from futures-rate changes rather than explicit FOMC target announcements. The paper also reports that results are robust to using STRIPS-based forward rates (which require no yield-curve smoothing), alternative sample periods, and alternative outlier treatment, and to using all 39 available MMS-covered releases rather than the baseline 13 (footnotes 3 and 5).