Higgins, Matthew L.: Testing for Autoregressive Conditional Duration
World Conference Econometric Society, 2000, Seattle

Matthew L. Higgins, Western Michigan University
Testing for Autoregressive Conditional Duration
Session: C-13-22  Wednesday 16 August 2000  by Higgins, Matthew L.
Recently, Engle and Russell (1998) introduced the Autoregressive Conditional Duration (ACD) model. In this paper we derive parametric tests for the presence of ACD in an exponential model. We first present the conventional LM statistic and a parametrically robust version of it. Because durations are nonnegative, the autoregressive parameters in the ACD model must also be nonnegative. The conventional LM test fails to account for the one-sided nature of the alternative hypothesis. To find a test with superior power, we derive the locally-most-mean-powerful-unbiased based score (LBS) test of King and Wu (1990). A Monte Carlo study shows that the finite sample sizes of the robust versions of the LM and LBS tests accurately reflect their nominal sizes when the null data generating process is an exponential or an over-dispersed Weibull distribution. The Monte Carlo experiments, however, reveal that the power of the robust LBS test uniformly dominates the power of the robust LM test, and is therefore the preferred test.


File created by Jurgen Doornik with eswc2000.ox on 2-01-2001