Forecast Quality and Simple Instrument Rules: A Real-Time Data Approach

Abstract

We start from the assertion that a useful monetary policy design should be founded on more realistic assumptions about what policymakers can know at the time when policy decisions have to be made. Since the Taylor rule – if used as an operational device - implies a forward looking behaviour, we analyze the reliability of the input information. We investigate the forecasting performance of OECD projections for GDP growth rates and inflation. We diagnose a much better forecasting record for inflation rates compared to GDP growth rates, which for most countries are almost uninformative at the time a Taylor rule should sensibly be applied. Using this data set, we find significant differences between Taylor rules estimated over revised data compared to real-time data. There is evidence that monetary policy seems to react more actively in real time than rules estimated over revised data suggest. Given the evidence of systematic errors in OECD forecasts, in a next step we attempt to correct for these forecast biases and check to which extent this can lower the errors in interest rate policy setting. An ex-ante simulation for the years 1991 to 2001 supports the proposal that correcting for forecast errors and biases based on an error model can lower the resulting policy error in interest rate setting for most countries under consideration. In addition we investigate to what extent structural changes in the policy reaction behaviour can be handled with moving instead of expanding samples. Our results point out that the information set available needs a careful examination when applied to instrument rules like those of the Taylor type. Limited forecast quality and significant data revisions recommend a more sophisticated handling of the dated information, for which we present an operational procedure that has the potential of reducing the risk of severe policy errors.