A Composite Lognormal-Pareto Autoregressive Conditional Duration Model With Its Application

With the development of information technology,the analysis of financial high-frequency data is more and more important.Financial high frequency data contains a lot of information about the microstructure of the market,so the analysis of financial high-frequency data is quite important to understand the financial microstructure.At present,most of the researches on microstructure theory are qualitative research,and to what extent these theories are practical,we need empirical analysis to test them.ACD model is an important method to deal with the positively valued time series of financial high frequency data,and the choice of the conditional distribution for innovations has pose challenges.In general,the trade duration has high density around zero and often exhibit asymmetric shape with long right tail.Therefore,the appropriate innovations distribution should be able to capture the inherent characteristics of the data,particu-larly the heavy tailedness.At this point,the composite Lognormal-Pareto distribution in this paper performs better than the existing exponential distribution,Weibull distri-bution,and the gamma distribution.This paper is not from the perspective of shape pa-rameter or mixing the same distribution considering innovation distributions,but from"paste" two different distribution of composite distribution into consideration which provides a new ideal to innovation distributions of the ACD model.The data in this paper is got from the transaction data of Starbucks stock part of between july 25,2011 and july 29,2011.Then,the performance of the Lognormal-Pareto distribution that is LP-ACD is compared with the performance of EACD,WACD and GACD.