A New Approach to Approximate the Trend-Cyclical-Component at the Current End of a Time Series.

In this paper, the authors compare a trigonometrically designed low-pass filter with the Hodrick-Prescott filter and a conventional moving average. The authors examine by means of transfer functions, some practical applications and in comprehensive stochastic simulations how well the three filters extract the current business cycle development of economic time-series. Within the framework of the simulations the correlation between the true and the estimated component is measured. The analysis show that the trigonometrically designed filter represents the current cyclical information more unbiased and reliable than the two other filters. Besides, it delivers early turning point signals. The new filter performs even better, if it is used in interaction with the estimation of a seasonal component. Especially for time-series showing not too high irregular fluctuations, the advantage of a directly approximated cyclical component increases against the blurred seasonally adjusted time-series.