Value-at-risk modelling for the Malaysian stock exchange based on Monte Carlo simulation / Zatul Karamah Haji Ahmad Baharul-Ulum

This study puts forward Value-at-Risk (VaR) models based on Monte Carlo Simulation (MCS) that are integrated with several volatility representations to estimate the market risk for seven non-financial sectors traded on the first board of the Malaysian stock exchange which is now known as Bursa Malay...

Full description

Bibliographic Details
Main Author: Haji Ahmad Baharul-Ulum, Zatul Karamah
Format: Thesis
Language:English
Published: 2008
Subjects:
Online Access:http://ir.uitm.edu.my/id/eprint/15518/
http://ir.uitm.edu.my/id/eprint/15518/1/TP_ZATUL%20KARAMAH%20AHMAD%20BAHARUL-ULUM%20BM%2008_5.PDF
Description
Summary:This study puts forward Value-at-Risk (VaR) models based on Monte Carlo Simulation (MCS) that are integrated with several volatility representations to estimate the market risk for seven non-financial sectors traded on the first board of the Malaysian stock exchange which is now known as Bursa Malaysia. In a sample over the years from 1993 until 2004 for construction, consumer product, industrial product, plantation, property, trade and services and mining sectors, the expected maximum losses were quantified for 1-day, 10-days and 25-days at 95% and 99% confidence levels. While the mining sector gave the highest expected value of VaR in all parameter settings, the plantation sector delivered the minimum value of VaR in most circumstances. Next, in a two-year out-of-sample group covering 2005 to 2006, the performance of the RiskMetrics EWMA and GARCH-based models was assessed from three different perspectives; conservatism, accuracy and efficiency. Although mixed results were observed, the study provides some indications of the applicability of some VaR models for the sectors involved besides confirming that data and computational choices affect risk measurement qualities. Under conservatism tests, the GARCH-based models were the most conservative model at both the 95% and 99% levels of confidence of Mean Relative Bias and Root Mean Squared Relative Bias in the single variable cases, while the t-distributed EGARCH delivered good results in multiple variables circumstances. For accuracy performances, tests conducted using Kupiec’s and Christoffersen’s provided evidence that almost every model was found to be accurate for all sets of occurrence. However, using Lopez test, which takes into consideration the magnitude of the impact of exceptions, models with the highest accuracy rate for multiple variables and most of the sectors studied under single variable were the MCS2+EGARCHn and MCSi+GARCHt respectively. In measuring efficiency, restricted to the confidence levels, the t-distributed models were identified as the best representative to track movements in true risk exposure for single variable while MCS2+EGARCHn for multiple variables conditions. As such, this study indicates that a consideration of fat-tails and asymmetries are crucial issues when deciding to estimate VaR in managing financial risk.