Imputing Seasonal Data in an Advanced Indicator with Forecasts from X-13ARIMA-SEATS

• Activity Number: 300 - Innovations in and Applications of Imputation
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Government Statistics Section
• Author(s): Nicole Czaplicki* and Yarissa Gonzalez
• Companies: U.S. Census Bureau and U.S. Census Bureau
• Keywords: Imputation; RegARIMA; forecasting; Regression; ARIMA modeling; Time Series

Abstract

The Advanced Monthly Retail Trade Survey (MARTS) publishes early sales estimates of retail and food service companies approximately nine working days after the reference month. Retail trade data have strong seasonal patterns and known calendar effects, and the tabulated industry-level estimates are seasonally adjusted. However, the current missing and erroneous data treatment procedures do not fully account for this seasonality, instead relying entirely on respondent-based ratio estimates constructed from current month to prior month data. Such imputation procedures can yield biased estimates, especially when response rates are low and the response mechanism is unlikely to be ignorable. Furthermore, these methods do not utilize additional historic information, at both the individual unit level and at the industry level. In this paper, we demonstrate a novel imputation method for MARTS that utilizes industry level seasonal ARIMA models with calendar effects estimated with X-13ARIMA-SEATS to develop one-step ahead unit level forecast imputed values.