Testing the Conditional Mean Function of Autoregressive Conditional Duration Models

In this paper, we suggest and evaluate specification tests to test the validity of the conditional mean function implied by Autoregressive Conditional Duration (ACD) models. We propose Lagrange multiplier tests against sign bias alternatives, various types of conditional moment tests and integrated conditional moment tests which are consistent against all possible alternatives. In a Monte-Carlo study we investigate the finite sample properties of the individual tests. Moreover, the testing framework is applied to a variety of existing and new ACD specifications using financial duration data based on NYSE trading. We show that conditional moment tests have the highest power to detect general types of misspecifications. Moreover, we provide evidence that most ACD specifications are too simple and are clearly rejected. It turns out that flexible parameterizations of the news impact function are necessary to appropriately model financial durations. A semiparametric ACD model proposed in this paper seem to be a valuable alternative to existing approaches.