Signal Extraction: Efficient Estimation, 'Unit Root'-Tests and Early Detection of Turning Points

The book provides deep insights into the signal extraction problem - especially at the boundary of a sample, where asymmetric filters must be used - and how to solve it optimally. The traditional model-based approach (TRAMO/SEATS or X-12-ARIMA) is an inefficient estimation method because it relies on one-step ahead forecasting performances (of a model) whereas the signal extraction problem implicitly requires good multi-step ahead forecasts also. Unit roots are important properties of the input signal because they generate a set of constraints for the best extraction filter. Since traditional tests essentially rely on one-step ahead forecasting performances, new tests are presented here which implicitly account for multi-step ahead forecasting performances too. The gain in efficiency obtained by the new estimation method is analyzed in great detail, using simulated data as well as 'real world' time series.