Seasonal Heteroskedasticity and Signal Extraction in Time Series

• Activity Number: 480
• Type: Invited
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Thomas Morgan Trimbur*+ and William Bell
• Companies: Federal Reserve Board and U.S. Census Bureau
• Keywords: seasonal adjustment ; trend ; unobserved component

Abstract

Seasonal heteroskedasticity, defined by regular changes in variability over the calendar year, occurs in a range of economic time series. For instance, systematic increases in variation can arise in winter due to an increased likelihood of severe weather adversely affecting economic activity. In current analysis of the economy, the treatment of seasonal heteroskedasticity can affect the kinds of signals taken from key indicators. For example, while data on housing starts are often taken as a leading indicator for the path of GDP growth, it remains unclear, for the purposes of measuring a useful signal, whether to down-weight an unusually large decrease in the number of starts that may be due to a large winter storm. In this paper we explore the treatment of seasonal heteroskedasticity in signal extraction for trend-cycle analysis and seasonal adjustment. An illustration shows results for estimating signals in the time series of total housing starts, using models that account for the different patterns of seasonal heteroskedasticity in the four Census regions.