The Response of Term Rates to Fed Announcements

The February 3-4, 1994 FOMC meeting “marked the first change in the Federal funds rate target since September 4, 1992” and concluded with “the most important press release since Chairman Volcker announced the beginning of the now infamous nonborrowed reserves targeting experiment of the 1979-1982 period”: the Fed publicly disclosed that the target had risen 25 basis points, and has continued disclosing target changes immediately after every FOMC meeting since (Introduction, p. 1, quoting Chairman Greenspan: “the main issue here is that, as far as I am concerned, I would like us to be counted. We are the central bank and we are making a major move.”). The authors argue this procedural change “has added a new policy tool to the Fed’s arsenal: the announcement effect,” distinct from “the conventional view of the monetary transmission mechanism,” which relies on the liquidity effect and the Fed’s “monopolistically held supply of bank reserves” (p. 1).

Guthrie and Wright document that the RBNZ can move its official cash rate by announcement alone because it operates a “channel” system with standing lending and deposit facilities that make the announcement “a credible threat” – and because in New Zealand the announcement typically constitutes a genuine surprise to the market (Introduction, pp. 2-3). The authors identify two key departures for the U.S. case: the Fed does not pay interest on reserves and its discount window is “rarely used by banks” for institutional reasons, so “one needs to explain where is the necessary slack built into the reserves market to absorb the announcement effect”; and unlike New Zealand’s official cash rate announcements, “the announcement of a new level for the Federal funds rate target does not necessarily constitute a surprise to the market” – these changes are often “widely anticipated,” which is why measuring the exogenous, surprise component of an announcement requires modeling expectations explicitly, via the ACH methodology, rather than simply treating every announcement as a shock.

The paper documents a sharp decline in required reserves – from about $60 billion in 1994 to below $45 billion in 1999 – driven by retail sweep programs that dynamically reclassify customer deposits out of reservable accounts, alongside reserve-requirement cuts in 1990 and 1992, and a shift in the ratio of non-transaction to transaction deposits from 2:1 in 1994 to 4:1 in 1999 (Section 2, pp. 4-5). It also documents that discount-window borrowing, despite the discount rate typically sitting below the funds rate, has been “essentially reduced…to zero” in the 1990s owing to reputational stigma, and that the Fed adapted its operating practices (moving daily open-market operations earlier in the day, dropping contemporaneous reserve accounting in 1998) in ways consistent with “the Fed’s disposition to accommodate reserve demand shocks” documented by Strongin (1995) – together suggesting “the traditional view of the reserves market depicted in earlier papers needs to be reformulated” (pp. 6-7).

On November 22, 1989, the Fed added reserves at a time the funds rate was below its 8-1/4% target because of “an erroneous borrowed reserves forecast,” but the market misread this as a signal that the target had been lowered, prompting business-press reports of a policy change that had not actually occurred (Section 2, pp. 7-8). The authors show, via the daily spread between the prime rate and the funds-rate target, that “after November 22, 1989 there is virtually no uncertainty in reading Fed moves: banks adjusted the prime rate, often within the same day of a target change, and with certainty.” They adopt 1989 as their primary sample break (rather than 1994) partly because the Federal funds futures market – needed for an alternative, Kuttner-style surprise measure – was also created in 1989, and because it affords more target changes for identifying the announcement effect.

Their identification strategy rests on a simple observation: “in an environment in which there is no announcement effect, the magnitude of the liquidity effect that one can compute with a VAR…is invariant to the inclusion or exclusion of dates in which the target was changed,” whereas if an announcement effect is genuinely at work, small movements in nonborrowed reserves around target-change dates should be associated with disproportionately large funds-rate movements driven by the (unobserved) announcement effect, biasing the estimated liquidity effect upward when target-change dates are included (Section 3, pp. 10-11). Estimating structural VARs (generalizing Strongin’s 1995 model to allow banks to buffer shocks through excess reserves) separately for 1984-1989 and 1989-1999, and comparing single-state to two-state (target-change versus no-change) specifications, they find “the results are quite striking”: in the pre-1989 sample the liquidity effect is actually more pronounced during target-change periods, but in the post-1989 sample “the liquidity effect measured by the two-state VAR is significantly more tenuous than the full sample response,” a gap the authors read as direct evidence of an announcement effect operating since 1989 (Section 3.1, pp. 13-14).

The Fed’s target-change process is a “marked point process”: irregular in timing and typically moving in discrete 25-basis-point increments, requiring a joint model of when a change occurs (the “point process”) and how large it is (the “mark”) (Section 4, pp. 15-16). The authors adapt the autoregressive conditional hazard (ACH) model of Hamilton and Jorda (2000) to forecast the maintenance-period-ahead hazard of a target change conditional on past changes and exogenous variables (production, prices, unemployment), paired with an ordered-probit (OP) model forecasting the size of any change conditional on one occurring; because the two models’ likelihoods can (under weak conditions) be maximized separately, this avoids the difficult numerical integration or Monte Carlo methods that comparable duration models usually require (Section 4, pp. 16-17). A parallel one-day-ahead logit model using federal funds futures prices, following Kuttner (2000), serves as a robustness check for the post-1989 period, correctly classifying target changes with 89% specificity and 81% sensitivity.

Regressing Treasury rates (3-month through 30-year) on the ACH/OP-forecast expected component and the surprise component of target changes, “the response of term rates to target surprises is quite strong during periods when the target was changed,” though “the significance of the response decays with T-notes of increasing maturity, particularly after 1989” (Section 5, pp. 25-26). Critically, “the pre-1989 responses [to surprises during periods when the target was not changed] are statistically insignificant at all maturities, suggesting that the market had little understanding on how the Fed operated in setting new targets,” whereas post-1989 “the impact of the unanticipated component of target changes…monotonically increases with the maturity of the T-note and is statistically significant” even when the target does not change – consistent with the market extracting information about the policy stance “both when the Fed changes interest rates unexpectedly as well as when it remains surprisingly inactive.” As predicted, “the responses to the expected component of target changes are all insignificant at all maturities for both the pre- and post-1989 samples” (p. 26).

Splitting observations into four cases – target changed/expected, changed/unexpected, unchanged/expected-to-change, and unchanged/expected-to-stay – lets the authors test whether the market reacts to surprising inaction, the “non-announcement effect” of Roley and Sellon (1998) (Section 5.1, pp. 28-31). They find “there appears to be no significant response to the Fed maintaining interest rates when a target change was expected during the pre-1989 sample” (coefficients even turn negative for maturities over a year), but “a significant response of medium to long term maturities when measured with ACH/OP forecasts” after 1989 – a result “close to a 90% confidence level” even using Kuttner’s alternative surprise measure. The authors interpret this as further evidence that “prior to 1989, the market had a hard time forming expectations on the timing of policy moves,” while after 1989 the market’s improved understanding of “the schedule of FOMC meetings” and the routine of announcements let it register surprise even when the Fed does nothing.

The authors note “the Fed initially viewed the announcement effect negatively” and that “the motivation behind the initial public disclosure of a target change in 1994 was taken with extreme caution and was not initially viewed as a permanent change in operating procedures” (Conclusions, p. 33). Yet they argue their evidence “support[s] the usefulness of this practice since it appears that term rates are more responsive to policy movements” once announcements are made – an implicit case that the announcement policy, adopted somewhat reluctantly, ended up strengthening rather than weakening the Fed’s control over the term structure that ultimately matters for the real economy.

The authors frame their whole exercise as testing the “second link” in monetary transmission – how movements in the overnight rate affect longer-term rates – which “typically relies on some type [of] rational expectations mechanism that explains long-term rates as a weighted average of expected future short-term rates” (Conclusions, pp. 33-34). Because only surprise (not anticipated) target changes and surprise inaction move term rates in their results, and because this pattern strengthens once the market can be shown (via the 1989 and 1994 institutional breaks) to have a clearer read on Fed behavior, the authors conclude that “the rational expectations hypothesis seems to be well and alive: prices incorporate expectations on future overnight rates and in the long run, the policy stance determines the long term rate.”