New Seasonal Adjustment and Signal Extraction Methods in the Manufacturers’ Shipments, Inventories, and Orders (M3) Survey

• Activity Number: 160 - Time Series Methodology: Modern Practices in Seasonal Adjustment and Software
• Type: Topic-Contributed
• Date/Time: Tuesday, August 10, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Section for Statistical Programmers and Analysts
• Author(s): James A Livsey* and Colt Viehdorfer
• Companies: U. S. Census Bureau and US Census Bureau
• Keywords: seasonal adjustment; signal extraction

Abstract

Signal extraction, specifically seasonal adjustment, is ubiquitous in establishment survey collection and dissemination. As data becomes available at higher frequencies and lower levels of disaggregation, it is prudent to explore modern signal extraction techniques. This work investigates two model-based signal extraction methods with applications to the U.S. Census Bureau’s M3 survey; signal extraction in ARIMA time series (SEATS) and multivariate signal extraction with latent component models. We present both new findings and provide discussion of practical implications. We explore univariate models applying SEATS methodology, a model-based signal extraction paradigm. Additionally, we pursue multivariate methodology that allow M3 aggregate series to be viewed jointly by the lower level composition series. For example, we investigate the added benefit to jointly performing signal extraction on inventories and shipments for automobile, dairy product, farm machinery, and petroleum refinery manufacturing. We also examine joint signal extraction on unfilled orders and shipments for iron and steel mill manufacturing.