|
Research Projects Supported by HKU's High Performance Computing Facilities |
|
¡@ |
| Researcher: |
| Miss Polly Po-ling Kam, Department of Statistics and Actuarial Science |
| ¡@ |
|
Project Title: |
| Mixture Autoregression with Heavy-tailed Conditional Distribution |
| Project Description: |
| In this project, we try to modify Mixture autoregressive (MAR) model which is introduced by Wong and Li (2000). This new model is called Student t-type MAR (TMAR) model. We developed the method of estimation and computation of standard errors for parameter estimates in the model. We also investigate the model selection criteria for choosing the best model for data fitted. Extensive simulations have been carried out to test the performance of TMAR model. For example, using different sample size, and comparing TMAR model with MAR model. TMAR model is applied to several sets of financial data, like the Hong Kong Hang Seng Index and 3-year Treasury Constant Maturity Rate. |
| Project Duration: |
| September 1, 2001 ¡V August 30, 2003 |
| Project Significance: |
|
MAR model is very useful and flexible in non-linear time series analysis. However, in financial area, strong evidence has indicated that some of the return series have fat tails. Since MAR model allies with the Gaussian assumption, it may underestimate the occurrence of extreme financial events (e.g., stock market crashes). Therefore, the applicability of the MAR model to financial time series might be questionable. Our new proposed model not only retains some nice properties of MAR model, but also handles various kinds of financial data with heavy tailed distributions. Therefore, TMAR model is more suitable than MAR model to model financial time series data. |
| Results Achieved: |
| We have successfully formulated a TMAR model for financial time series. We found that the estimation method has small bias and reasonable standard errors. The performance of estimation can be improved by large sample size. Since the size of data in economics and finance areas has grown rapidly, it is not difficult to obtain large volume of data in real life application. Simulations show that the empirical standard errors match our theoretical standard errors very well. For model selection problem, we found that the criteria does a good job in identifying the model and helps us to fine tune the model. In real life application, TMAR model gives several interesting insight into some extreme financial events. |
| Remarks on the Use of High Performance Computing Cluster: |
| The parameter estimation and calculation of standard errors involve complicated computational steps. HPC Cluster helps us to solve this computational problem. Also, HPC Cluster provides large amount of disk space for our extensive simulation studies. Some of simulations are very time consuming. Using parallel facility in HPC Cluster, it can help us to reduce the time of computation and finish our task in a short time. |
| Email Address: |
| pollykam@graduate.hku.hk |