On diagnostic checking of the autoregressive conditional intensity model

Abstract

The autoregressive conditional intensity model proposed by Russell (1998) is a promising option for fitting multivariate high frequency irregularly spaced data. The authors acknowledge the validity of this model by showing the independence of its generalized residuals, a crucial, assumption of the model formulation not readily recognized by researchers. The authors derive the large-sample distribution of the autocorrelations of the generalized residual series and use it to construct a goodness-of-fit test for the model. Empirical results compare the performance of their test with other off-the-shelf tests such as the Ljung-Box test. They illustrate the use of their test with transaction records of the HSBC stock.