Stylized Facts Test for the Signal-Extraction Techniques

One of the important tools of the business cycle research are the signal-extraction techniques (SETs). They allow to study both the stylized facts and the turning points of the business cycles. However, these are highly sensitive to the SETs. In this paper we try to see how some of the SETs affect the stylized facts and to compare the performance of several detrending techniques in terms of the distortions they introduce into the first four moments of the extracted business cycles. To accomplish this, the Monte Carlo experiments for various DGPs, including deterministic and stochastic, common and individual trend specifications of the observed time series, were undertaken. The results allow to rank different SETs according to their performance and to reveal the sources of distortions. Finally, we try to improve upon performance of the SETs by constructing mixed, mutlivariate and mixed multivariate filters using univariate detrending techniques as building blocks. It turns out that linear combination of the filters behaves better than the best of SETs of which it is comprised. Multivariate filtering also leads to improvements of the SETs performance.