Moving Average Detrending and the Analysis of Business Cycles.

This study complements the previous analyses of the effects of different types of detrending, by considering moving average detrending which is widely used in business cycle analysis by official and quasi-official agencies. Such detrending is shown to render a wide range of processes stationary, including those with up to second-order deterministic and/or stochastic nonstationartiy. At the same time, however, the medium and long run properties of the disturbance process are profoundly altered. Applied to an '1' (1) process, the component at a period equal to the moving average length is enhanced, with potentially serious consequences for business cycle analysis. If the detrending is applied to an '1' (2) process, the long run component is emphasized and the resulting (stationary) series may appear to be nonstationary. Copyright 1995 by Blackwell Publishing Ltd