INFERENCE IN A NEARLY INTEGRATED AUTOREGRESSIVE MODEL WITH NONNORMAL INNOVATIONS

Robust tests and estimators based on nonnormal quasi-likelihood functi ons are developed for autoregressive models with near unit root. Asymp totic power functions and power envelopes are derived for point-optima l tests of a unit root when the likelihood is correctly specified. The shapes of these power functions are found to be sensitive to the exte nt of nonnormality in the innovations. Power loss resulting from using least-squares unit-root tests in the presence of thick-tailed innovat ions appears to be greater than in stationary models. (C) 1997 Elsevie r Science S.A.