A Modification to Khandakar and Hyndman's ARIMA Model Selection Algorithm Using an Empirical Information Criterion

• Activity Number: 129
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2008 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economics Statistics Section
• Author(s): Brian C. Monsell*+
• Companies: U.S. Census Bureau
• Keywords: Automatic ARIMA model selection ; out-of-sample forecast error ; outlier regressors ; calendar regressors ; history diagnostics

Abstract

Khandakar and Hyndman (2007) propose an automatic ARIMA model selection algorithm based on the Linear Empirical Information Criterion (LEIC), which penalizes the maximized log likelihood by a linear function of the number of parameters in the model. A penalty term is derived using an out-of-sample forecast performance criterion evaluated for a set of related time series. A modification to this algorithm will be examined that (a) includes selection of trading day, holiday and outlier regressors identified initially using a default model, and (b) incorporates 12-step ahead out-of-sample forecast performance. This modified algorithm will be applied to a set of Census Bureau series examined in McDonald-Johnson, Hood, Monsell and Li (2007). The model choices made by the algorithm will be compared to those made by Version 0.3 of X-12-ARIMA.