- 2010年12月6日(月) 14:30-16:00
- 統計数理研究所 セミナー室5 (D313-314)
- Selection Between Non-Nested Models Through Multi-Step-Ahead Forecasting
- David F. Findley (U.S. Census Bureau)
- We develop and show applications of two new test statistics for deciding if one ARIMA model provides significantly better h-step-ahead forecasts than another, as measured by the difference of approximations to their mean square forecast error.
The two statistics differ in the variance estimates used for normalization. Both variance estimates are consistent even when the models considered are incorrect. Our main variance estimate is further distinguished by accounting for parameter estimation.
The simpler variance estimate ignores parameter estimation effects.
These effects are asymptotically negligible only when h = 1.
The h=1 result also applies to the log-likelihood ratio tests and to AIC differences. These statistics can be calculated for any pair of scalar ARIMA models with the same differencing operator, and their broad consistency property offers improvement to what are known as tests of Diebold and Mariano (1995) type. They can also be calculated for vector autoregressive models. This is joint work with Tucker McElroy of the U.S. Census Bureau.