Xu, Yongdeng ORCID: https://orcid.org/0000-0001-8275-1585
2013.
Econometrics of high frequency data and nonnegative valued financial point processes.
PhD Thesis,
Cardiff University.
Item availability restricted. |
Preview |
PDF
- Accepted Post-Print Version
Download (14MB) | Preview |
PDF
- Supplemental Material
Restricted to Repository staff only Download (780kB) |
Abstract
Econometrics of high frequency data and nonnegative valued financial point process is addressed in an Autoregressive Conditional Duration (ACD) and Multiplicative Error Model (MEM). The basic idea is to model the nonnegative valued point process in terms of the product of a scale factor and an innovation process with nonnegative support. However, when extending such a model into a multivariate setting, the direct use of multivariate MEM model is restricted since conditional distributions for multivariate nonnegative valued random variables are often not available. A common strategy is to reduce the multivariate setting to a series of univariate problems by assuming: a) weak exogeneity. b) the independence of innovation terms. The objects of this thesis are to examine this strategy and develop a general form vector MEM. Three main Chapters have been developed. We begin with the analysis of weak exogeneity. The independence of innovation terms is considered as a special case of weak exogeneity. The simulation study indicates that a failure of the weak exogeneity assumption implies not only inefficient but also biased estimate of the parameters. We then derive an LM test for weak exogeneity and the empirical results indicate that the weak exogneity of duration is often rejected. Chapter 3 discusses the use of lognormal distribution for financial durations and we propose a lognormal ACD model. The empirical results show that lognormal ACD model is superior to Exponential and Weibull ACD model. It performs similarly to Burr or generalized gamma ACD model. In Chapter 4, we release weak exogeneity assumption and propose general form of vector MEM. Based on the results in Chapter 3, we further propose to use the multivariate lognormal distribution for the distribution of the vector MEM for which maximum likelihood is proved as a suitable estimation strategy. The model is then applied to the trade and quotes data from the New York Stock Exchange (NYSE) for the dynamics of trading duration, volume and price volatility. The empirical findings are generally consistent with market microstructure predictions.
Item Type: | Thesis (PhD) |
---|---|
Status: | Unpublished |
Schools: | Business (Including Economics) |
Subjects: | H Social Sciences > HG Finance |
Date of First Compliant Deposit: | 30 March 2016 |
Last Modified: | 24 Oct 2022 10:48 |
URI: | https://orca.cardiff.ac.uk/id/eprint/45954 |
Actions (repository staff only)
Edit Item |