Evaluating different Value-at-Risk calculation methods: Are the Hong Kong Listed Banks’Market Risk Disclosures consistent with the performance?

Yuen, Lok Hin David (2011) Evaluating different Value-at-Risk calculation methods: Are the Hong Kong Listed Banks’Market Risk Disclosures consistent with the performance? [Dissertation (University of Nottingham only)] (Unpublished)

[img] PDF (MSc Risk Management Dissertation) - Registered users only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (5MB)


This paper studies the quality of Hong Kong listed banks’ market risk Value-at-Risk (VaR) disclosure; of which the Hong Kong stock market contains the diversity of Chinese state-owned banks, international banks with English history, and some Hong Kong local banks. Therefore, focusing on the Hong Kong stock market allows the comparison on the disclosure quality between the three. While the banks usually do not provide the disclosed VaRs with details on their internal estimation model and the actual accuracy of the estimation; this paper aims to (1) replicate the banks’ Value-at-Risk and apply the backtesting on the estimated VaR; and (2) compare the estimated VaR under the selected model with the VaR disclosed in banks’ publicly disclosed information.

In the first part of the research, twelve models are estimated with the variation of their volatility adjustment model, width of data rolling window, and simulation approaches. There are two ways of historical volatility forecast models applied, namely, the Exponentially Weighted Moving Average (EWMA) and the Generalized Autoregressive Conditional Heteroskedasticity (1,1) (or GARCH(1,1)); together with the case of no adjustment, they make up three different specification. 250 and 500 previous trading days are the two elected width of time window; and the Historical Simulation and Monte Carlo Simulation are the two simulation methods that are tested. These variations generate the twelve estimation approaches. These methods are then tested with the Christoffersen’s (1998) three tests of conditional coverage, independence and unconditional coverage. Furthermore, the GMM based duration test suggested by Candelon (2011) is also applied to the VaR data. The results concluded that the EWMA model with a 250 days of rolling window is the best perform model under the Christoffersen’s (1998) tests.

In the second part of the research, the estimated VaR from the best perform model is compared with the disclosed values. Four types of disclosed VaR are included in the comparison, they are the trading book year-end and year average VaRs, and the trading and non-trading book year-end and year average VaRs. The results suggest that the year average forms of VaRs can better reflect the risk and the trading and non-trading book VaR is better than the trading book ones. In the cross country analysis, the group international banks (HSBC and SCB) outperform the two groups with Chinese and Hong Kong banks, which suggests that the international banks have their VaR disclosed with better quality.

Item Type: Dissertation (University of Nottingham only)
Depositing User: EP, Services
Date Deposited: 26 Apr 2012 10:28
Last Modified: 24 Oct 2016 00:36
URI: http://eprints.nottingham.ac.uk/id/eprint/25054

Actions (Archive Staff Only)

Edit View Edit View