Financial risk management has undergone much change and greater regulation in the last twenty years following, and in many ways in response to, the major stock-market crash (“Black Monday”) of October, 1987. Now, another major market incident, the global financial crisis (GFC) in 2008-09, has prompted calls for more and different financial regulation. In order to better control the risk of financial institutions and to protect them against large unexpected losses, the group of G-10 countries agreed in 1988 to sponsor and subsequently form the original Basel Capital Accord.
In the last two decades, however, large unexpected losses have continued to occur with regularity: e.g. in December 1994, Orange County (US) announced a loss of $1.6 billion in its’ investment portfolio; in 1995, Nick Leeson, of Barings Bank (UK), lost $1.4 billion in speculation, primarily on futures contracts; in 1997, the Asian financial crisis began, which started in Thailand with the financial collapse of the Thai baht; among others, and finally the very recent GFC. Financial markets and the products traded on them are continuing to become more complicated and difficult to properly understand and assess by existing risk management tools and regulations. Such methods and rules clearly need to evolve as well.
Value-at-Risk (VaR) was pioneered in 1993, as a part of the “Weatherstone 4:15pm” daily risk assessment report, in the RiskMetrics model at J.P. Morgan. By 1996, amendments to the Basel Accord (Basel Accord II) allowed banks to use an ‘appropriate model’ to calculate their VaR thresholds. Jorion (1997) defines VaR as a measure of the highest expected loss, over a given time interval, under normal market conditions, at a given confidence level: VaR is thus a conditional quantile of the asset return loss distribution. Following Basel II, VaR has become more popular and is widely used in practice for risk management and capital allocation.
The recommended back-testing guideline proposed by the Basel Committee on Banking Supervision (1996) is to evaluate a one percent (1%) VaR model over a 12 month test period (250 trading days). VaR has been criticised for not measuring the magnitude of a loss in case of an extreme event. As such, and following McAleer and da Veiga (2008), we also consider various criteria measuring the loss magnitude given a violation, such as mean and maximum absolute deviation. These measures go beyond assessing violation rates and allow risk management to incorporate loss magnitude. Further, the different measures of model performance allow financial institutions to select different combinations of alternative risk models to forecast VaR using selection or combining strategies to suit their purpose.
Download
Bayesian Forecasting for Financial Risk Management, Pre and Post the Global Financial Crisis
