Find us on Facebook Find us on LinkedIn Follow us on Twitter Subscribe to our YouTube channel

Research Students

Some current Business Analytics PhD students

Research Student Thesis Topic Thesis Description

Wilson Chen

Estimation and Forecast of Value-at-Risk using Bayesian Methods

Value-at-Risk (VaR) was introduced as a measure of market risk in the Amendment to Basel-I in 1996. Banks have an incentive to obtain accurate VaR forecasts to minimize their capital charge. A model that either over- or under-forecasts VaR will ultimately result in a higher capital charge. The aim of this thesis is to improve the accuracy upon the state-of-the-art forecasting models. To accomplish this goal, this thesis proposes a semi-parametric GARCH-type model that is flexible in the specification of volatility dynamics. The proposed model is more robust to misspecification, and is able to provide more accurate forecasts when the true volatility process departs from the parametric specifications. A key challenge is that the proposed model typically requires the estimation of a large number of parameters. As such, it is important to account for the uncertainty in parameter estimation. In a Bayesian setting, parameter uncertainty can be handled naturally by integrating out the parameters with respect to the posterior distribution. In this thesis, the posterior distribution is estimated using Markov Chain Monte Carlo (MCMC) algorithms. The estimated predictive distribution is then the functional of the estimated posterior distribution.

Professor Richard Gerlach is the principle supervisor and Dr Boris Choy is the associate supervisor.

Chao Wang

Range-Based GARCH Models for Value-at-Risk Estimation with Bayesian Approach

Now we are in a world saturated with data and information, and numerous quantitative methods for financial risk management are proposed and used by many financial research institutions and organizations within recent years. As a commonly used financial risk measurement, Vaule-at-Risk (VaR) summarizes risk through a single number. GARCH-type models are employed to capture the volatility clustering and the leverage effect of financial return series in my research. In order to capture the potential skewness and heavy tails in the conditional return distributions, the GARCH-type VaR forecast models are proposed with the assumption of different distributions, such as mixture of Student-t and mixture of Gaussian. A MCMC algorithm will be designed and implemented for inference and parameter estimation in a Bayesian framework, which is expected to show improved estimation properties compared to the frequentist. The proposed methods and models can be applied to forecast VaR for different market indices, exchange rates, and individual stocks. In addition, the proposed models will be modified and combined with range-based time series. The experimental results are expected to illustrate that the proposed models outperform, or are at least highly competitive with, several popular alternatives.

Professor Richard Gerlach is the principle supervisor and Dr Boris Choy is the associate supervisor.

Christian Contino

Estimation, Inference and Forecasting for Value-at-Risk and Conditional Value-at-Risk using Bayesian GARCH Methods and High Frequency Intra-Day Data

Financial institutions are constantly looking for better ways to both manage and quantify risk following the regulatory disclosure of Value-at-Risk (VaR). An incorrect estimation could jeopardize the returns or risk profile of an institution, leading to loss of confidence from stakeholders. The proposed model introduces high frequency intra-day data into a traditionally daily GARCH specification via mixed data sampling, MIDAS. The goal is to increase accuracy, especially for short term forecasts of both VaR and Conditional VaR. Monte Carlo Markov Chain (MCMC) methods will be employed within a Bayesian framework for inference and parameter estimation. The resulting models should provide application in risk management, derivative pricing and portfolio optimization.

Supervisor: Professor Richard Gerlach

Rasika Yatigammana

Modelling of liquidity risks through the estimation and forecasting of an ACD model

In financial econometrics, vital information regarding the current state of financial market activity is revealed by the dynamics of duration between two consecutive trades or durations. Information that could be gleaned from this series is most useful for risk management, portfolio optimisation, hedging and also for managing liquidity and investigation of market microstructure theories. Liquidity is an important factor for many stakeholders in the financial market and duration is an obvious indicator of financial market liquidity, with its variance implying liquidity risk. This research would focus on Autoregressive Conditional Duration (ACD) models for the estimation and forecasting of liquidity risks primarily through the application of Bayesian methodology. Thus far, limited attention has been paid to reliable estimation and inference on ACD parameters as well as on finite sample and asymptotic properties of estimators. Since it is well established that the financial data exhibit extensive changes in its properties over time, methods for identification and estimation of structural breaks in duration data too would be developed, employing Markov chain Monte Carlo (MCMC) techniques.

Supervisor: Professor Richard Gerlach / Associate Professor Shelton Peiris

Maria Jofre

Financial Fraud Prediction

Financial fraud is a global concern and a critical affair for industries and governments. This type of fraud is becoming a million dollar business that increases every year, representing a significant threat to social stability and national security. In recent years, data-informed quantitative models have been developed to automate and reduce the manual auditing processes. Although the existing techniques have improved the detection rate of financial fraud offences, these are very limited and can be improved in terms of accuracy and efficiency, leading to more targeted and effective examinations. The main objective of this study is to create a new statistical technique to be implemented by public auditors and financial institutions assisting the analysis and prediction of this phenomenon. This technique will be based on the adjustment and improvement of existing quantitative methods including Regression Models, Machine Learning Methods and Bayesian Models.

Supervisor: Professor Richard Gerlach