Conference Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 101 - Time Series Modeling: Mixed Frequency Data, Seasonality, and Model Identification
Type: Topic Contributed
Date/Time: Monday, August 8, 2022 : 8:30 AM to 10:20 AM
Sponsor: Business and Economic Statistics Section
Abstract #320902
Title: A Model Comparison Diagnostic for Differencing Operators Based Upon Multi-Step Ahead Forecast Mean Squared Error Paths
Author(s): Tucker McElroy*
Companies: U.S. Census Bureau
Keywords: Model comparison; Model identification; Seasonality; Unit roots

We study the problem of identifying a differencing operator for a non-stationary time series by considering the multi-step ahead forecast mean squared error path for two different specifications. Each specification is a differencing operator that is deemed sufficient to yield a stationary time series, but there is concern that one (or both) of the operators over-differences. We propose a test statistic of the hypothesis of equal forecast mean squared error (for all forecast horizons) based upon in-sample squared forecast errors. Since we only utilize the differencing operators to generate multi-step ahead forecasts, there is no need to specify a model for the stationary part of the time series; this allows rejections of the null hypothesis to focus our attention on the differencing operators, and avoid confusion with ancillary specifications. Using joint asymptotic normality of the difference of forecast mean squared error paths (expressed as a vector), we devise a Wald statistic, which is easily computed and interpretable. Simulations and data analysis support the new identification statistic.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2022 program