The Taylor rule and real-time data – a critical appraisal

Abstract

In a number of recent papers, it has been argued that the use of ex post data can distort the picture when trying to analyse monetary policy reaction functions. This paper aims to establish whether the Taylor rule has been a reasonable representation of US monetary policy using both ex post and real-time output gap data. Results show that real-time data generate only minor differences to ex post data and, more interestingly, that the Taylor rule appears to be a questionable tool for evaluation of the Federal Reserve during the investigated samples.

Acknowledgements

I am grateful to Paul Söderlind and Simon van Norden for providing data. Financial support from Sparbankernas Forskningsstiftelse is gratefully acknowledged.

Notes

See for instance Orphanides (2001) and Orphanides and van Norden (2002).

Interest rate smoothing has been employed in empirical work by for instance Clarida et al. (1998, 2000), Gerlach and Schnabel (2000) and Doménech et al. (2002). In general,

in the estimated equations has been significant, which has been interpreted as evidence for the hypothesis that central banks adjust the interest rate gradually towards its target rate. However, the large estimates found in different studies generally imply that the adjustment speed of the process is implausibly slow. Accordingly, this interpretation of the parameter ρ has lately been questioned by for instance Rudebusch (2002), Österholm (2003), Gerlach-Kristen (2004) and Söderlind et al. (2004).

There are indications in Österholm (2003) that the Taylor rule performs fairly well in the 1960s and 1970s but not thereafter.

Phillips (1988) defined near integration as a case in which an AR process have roots close to, but not exactly on, the unit circle. Both processes that have roots smaller than unity (strongly autoregressive) and larger than unity (mildly explosive) are discussed in Phillips’ analysis; in this paper, however, the concept near integration will refer to processes with roots smaller than unity. For the AR(1) process

, this can be described as setting the AR parameter to ϕ = 1 − c/T where c is a small positive number and T the sample size.

Inference based on t-statistics from the regressions should be interpreted with some caution even when the standard errors have been Newey–West corrected. The reason for this is twofold: when there is no cointegration, the t-statistics diverge. When there, on the other hand, is cointegration, the distribution of the t-statistics is asymptotically standard normal only when the regressors are independent of the error term.

Lag length in the Augmented Engle–Granger (AEG) test and the bandwidth parameter in the KPSS test are chosen using the same procedures as in the ordinary unit root tests, that is by the Akaike information criterion and the Newey–West method.

Note that if the variables actually are cointegrated, estimation of Equation 1 yields a consistent estimate of the cointegrating vector even if the true data generating process is of interest rate smoothing type.