Robust Estimation of Trend and Seasonal Components in the Presence of Outliers and Level Shifts

• Activity Number: 108
• Type: Topic Contributed
• Date/Time: Monday, August 9, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economics Statistics Section
• Author(s): Thomas Trimbur*+
• Companies: U.S. Census Bureau
• Keywords: level shifts ; unobserved components ; state-space ; heavy-tailed density ; non-Gaussian model ; robust filter

Abstract

This paper develops methods for estimating trends and seasonal components in time series subject simultaneously to both level shifts and additive outliers. The approach is based on using heavy-tailed densities for the innovations in unobserved components models. In this framework, the timing and magnitude of large changes in the level of a process, or of outlying observations, is stochastic. Such flexibility may be helpful in modeling the range of behavior found in real time series; more informative and robust estimation becomes feasible, and this avoids the risk of major change in assessment of outlier/break points, that may have major impact on results in procedures based on binary categorization. After setting out the econometric methodology, the paper provides empirical illustrations. It is shown how the model-based assessment of the sample observations is updated optimally with the arrival of new data. Outliers and level shifts are rated on a continuous scale and distinction is made between them on the basis of available information. Trend and seasonal estimates are obtained which are less subject to revision with future data.