Abstract:
|
Seasonal heteroskedasticity may occur in a range of economic time series. Where present, such regular changes in variability over the calendar year can affect the signals, taken from key indicators, useful for understanding historical patterns and for discussions related to current analysis and policy making. In this paper, we investigate the role of seasonal heteroskedasticity in signal extraction and explicitly treat the areas of trend estimation and seasonal adjustment. In particular, we consider three standard time series models expanded to include a seasonally heteroskedastic irregular component, which form the basis for estimating trends more robust to non-economic forces - such as severe weather - linked to the annual cycle. In an application to time series of US housing starts for the four Census regions, measures of seasonal noise are compared across the models; we also examine their relative statistical performance and the properties of extracted signals that they produce, including the precision of the trend and seasonally adjusted series and their robustness to the weather-related volatility in construction activity during the winter months.
|