Expectations, Open Market Operations, and Changes in the Federal Funds Rate

The puzzle is that “the market reacts very quickly and sometimes without any immediate open market purchases or sales by the Trading Desk” once the FOMC publicly announces a change in its federal funds rate target (Introduction), quoting Meulendyke (1998): “the rate has tended to move to the new, preferred level as soon as the banks knew the intended rate.” Taylor’s resolution is that traders, knowing the Trading Desk follows a systematic, credible reaction function that will adjust reserve supply the next day if the rate has not converged, revise their expectation of tomorrow’s funds rate upward or downward immediately – and because banks’ demand for Fed balances today depends on that expectation, the rate moves today even with zero change in the actual supply of balances.

Three changes matter: the FOMC’s 1994 shift to publicly announcing federal funds rate target changes (later expanded in December 1999 to include the FOMC’s assessment of balance-of-risks), the July 1998 switch from contemporaneous to lagged reserve accounting, and the sharp decline in required reserve balances (from about $30 billion in 1990 to $5-6 billion by 2001) caused by banks’ ability to “sweep” consumer accounts out of reserve-requiring categories (Introduction). Together these changes make the Trading Desk’s communicated intentions, rather than the mechanical scale of its reserve operations, central to how the funds rate actually moves – which is precisely what the paper’s model is built to capture.

The federal funds market is a “double auction” in which traders at financial institutions bid and ask on overnight loans of Fed balances, with most trades intermediated by federal funds brokers; the commonly reported daily effective rate is “a weighted average of the rates on trades reported by brokers” (Sections “Trading in the Federal Funds Market,” “The Daily Effective Federal Funds Rate”). Taylor stresses this average masks substantial intraday and cross-sectional dispersion – citing a documented instance of “more than a 100 basis point difference between the rate on early morning trades and the daily effective rate” on December 30, 1999 – and that the Trading Desk itself enters the market only briefly each day, in a randomly timed ten-minute window around 9:30 a.m., using repurchase agreements and matched sale-purchase agreements to adjust the supply of balances.

Over the daily sample from August 1998 to September 2000, the average deviation of the effective rate from target was close to zero (0.1 basis points) with a standard deviation of 18 basis points, but that standard deviation fell from 20 basis points (Aug. 1998-July 1999) to 16 basis points (July 1999 onward), even including the unusually large deviation around the century date change (Section “The Recent Time Series Properties…”). At the same time, the estimated first-order autocorrelation of the deviation series rose from 0.3690 in the earlier subperiod to 0.4932 in the later one – Taylor reports “on average only about 6 percent of a deviation persists beyond two days” overall, but notes the increase in persistence “parallels the reduced volatility and could reflect greater smoothing of shocks.”

Volatility around target-rate changes fell sharply within the sample: the standard deviation of the funds-rate/target-rate gap was about 28 basis points for the first four of the nine target changes examined (1998-2000) but only about 12 basis points for the latter five (Section “Target Rate Changes,” Table 2). Granger-causality tests show the target rate’s future path is uniformly a significant predictor of the funds rate (the hypothesis that target does not Granger-cause funds rate is rejected in all nine episodes), while the reverse hypothesis – that the funds rate does not Granger-cause the target – is rejected in only three of the nine cases, which Taylor reads as evidence the target change was anticipated by the market in those particular episodes.

Equation (2), the Trading Desk reaction function, sets b_t = b_(t-1) + beta(rho_(t-1) - r_(t-1)) – the Desk raises the supply of Fed balances (b) the day after the funds rate (r) was below the target (rho), and lowers it the day after the rate was above target, with the response necessarily lagged because the Desk’s single daily intervention occurs each morning before that day’s effective rate is determined (Section “A Trading Desk Reaction Function”). Equation (3), the demand for Fed balances, sets b_t = alpha(gamma * E_t[r_(t+1)] - r_t) + epsilon_t, building on Furfine’s (2000a) intertemporal optimization and Guthrie and Wright’s (2000) New Zealand-based analysis of balances demand driven by the opportunity cost of borrowing from the central bank; gamma (between 0 and 1) captures a less-than-full-arbitrage down-weighting of the expected future rate, and alpha (finite, reflecting transactions costs and steep overnight-overdraft penalties) keeps demand from being infinitely elastic to the current-versus-expected-future rate gap.

For a simulated 50-basis-point target increase with parameters gamma = 0.9, alpha = 0.3, beta = 0.1, “the federal funds rate increases by 41.7 basis points on the day of the…announcement, even though there is no open market operation on that day designed to bring about this change,” with the remaining gap closing geometrically over subsequent days (Section “Simulations of a Change in the Target Federal Funds Rate”). The mechanism: if balances are expected to fall on day t+1 (per the reaction function), then the expected rate for day t+1 must rise (per the demand equation); but if tomorrow’s rate is expected to rise, today’s demand for balances shifts out today, raising today’s rate immediately, purely on the strength of that expectation.

He acknowledges two gaps: first, “the timing is not as precise as the model” – funds-rate movements around actual target changes sometimes lead and sometimes lag the announcement date itself, unlike the model’s clean day-of-announcement jump (Section “Simulations…”); second, “there is evidence from federal funds futures markets that changes in the federal funds rate target are sometimes anticipated by the market many days or even many weeks before they actually take” place, yet the model would predict funds-rate movement tracking that anticipation, “and such movements are hard to detect” – which Taylor flags as a case where “the model may be too successful,” leaving open whether the Trading Desk is offsetting such pressure or whether the demand-side mechanism needs refinement.

His conclusion is that the two are not separate channels but that the credible threat of future open market operations is the fundamental mechanism underlying observed “open mouth” effects: “traditional ‘open market operations’ are the fundamentals that underlie these announcement or expectations effects,” and the apparent power of announcements alone depends entirely on the Trading Desk’s reaction function being expected to actually govern its future behavior – “its actual policy must be consistent with the announced policy,” since otherwise “funds traders will soon begin to expect that there is some other policy reaction function at work” (Conclusion).