with Tucker McElroy


We will motivate and describe two new test statistics for deciding if one ARIMA model provides significantly better h-step-ahead forecast than an a second ARIMA model. The two statistics differ in the standard error estimates used to normalize the statistic: one accounts for parameter estimation and the other treats the parameters as fixed. Both offer improvements over statistics of Diebold-Mariano type. Size and power study results will be present, as well as an empirical study using series from two widely used textbooks.

VIEW PAPER