Signal Extraction: How (In)efficient Are Model-Based Approaches? An Empirical Study Based on TRAMO/SEATS and Census X-12-ARIMA

Abstract

Estimation of signals at the current boundary of time series is an important task in many practical applications. In order to apply the symmetric filter at current time, model-based approaches typically rely on forecasts generated from a time series model in order to extend (stretch) the time series into the future. In this paper we analyze performances of concurrent filters based on TRAMO and X-12-ARIMA for business survey data and compare the results to a new efficient estimation method which does not rely on forecasts. It is shown that both model-based procedures are subject to heavy model misspecification related to false unit root identification at frequency zero and at seasonal frequencies. Our results strongly suggest that the traditional modelbased approach should not be used for problems involving multi-step ahead forecasts such as e.g. the determination of concurrent filters.