Empirical Evaluation of X‐11 and Model‐Based Seasonal Adjustment Methods

Stuart Scott, Richard B. Tiller, and Daniel Chow

Abstract

X-11 and ARIMA model-based seasonal adjustments are compared for 82 series from three Bureau of Labor Statistics programs, 1) establishment employment, hours, and earnings, 2) consumer prices, and 3) producer prices. When both methods are applied automatically, the model-based approach exhibits more flexibility; its seasonal component can be virtually deterministic or rapidly varying. The model-based method tends to estimate a more stable seasonal component for series which are noisy or mildly seasonal. For some series, comparisons are made for "analyst adjustments," in which analysts attempt to improve on the automatic adjustments. Often "automatic" results from one method lead to improvements in analyst adjustments with the other method. With the current state of the art, it seems easier to improve X-11 adjustments.