|
Activity Number:
|
610
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 6, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Business and Economic Statistics Section
|
| Abstract - #304260 |
|
Title:
|
Multivariate Mixture Transition Distribution Model for Financial Transaction Data
|
|
Author(s):
|
Musen Wen*+ and Keh-Shin Lii
|
|
Companies:
|
University of California, Riverside and University of California, Riverside
|
|
Address:
|
Department of Statistics, Riverside, CA, 92521,
|
|
Keywords:
|
Mixture Transition Distribution ; Copula ; EM algorithm ; ultra-high frequency data ; tick-by-tick data
|
|
Abstract:
|
We propose a framework of Multivariate Mixture Transition Distribution model to model marked point processes in general and financial tick data in particular. In the proposed class of models, the conditional distribution of the current observation is a mixture of conditional distributions given each one of the last p observations, where each multivariate distribution in the mixture is constructed via a certain class of copulas. We show that the MMTD model can be decomposed to include the univariate MTD model and its hybrid models. An EM algorithm is developed to estimate the model parameters and shown to work well by simulation studies. A specific construction of the MMTD model is then applied to model three randomly selected IBM stock tick data sets from year 2007. We show that the MMTD model outperforms traditional linear time series models and the BMTD model in terms of prediction.
|
