What is CrashMetrics? The normal relationship between market variables is different under stress or ‘crash’ conditions. As such, any statistical measures used to determine how a portfolio value will respond to a crash should be derived solely from the extreme values of the historical data in parameters measurement and crash protection. Yet it is precisely these ‘outliers’ that are often treated with little respect. Furthermore, it is common experience that it is far easier to predict the results of an extreme event, a market crash or a hurricane, than it is to predict the probability of such an event. It would be nice not to rely too heavily on a poor estimate of a probability when we are talking about the scenarios that could result in the collapse of a bank. The CrashMetrics concept can be applied to both trading and hedging of crash events but the simplest and most obvious application of CrashMetrics is the estimation of the value at risk of a portfolio. In creating CrashMetrics, we are aware of the need for an intuitive tool that is easily communicated and accepted in the real world. There will be points of discussion in the model, as there should be in any useful mathematical model approximating the real world, but the framework is such that the model can be readily adapted and extended to the needs of the individual institution. Loosely speaking, Value at Risk, or VaR, is ‘an estimate, with a given degree of confidence, of how much one can lose from one’s portfolio over a given time horizon.’ There have been many suggestions for how to estimate VaR, every bank and academic has their own preferred version. Whatever the precise details of the implementation, the data required for the calculations are typically parameters for the ‘underlyings’ and measures of a portfolio’s current exposure to these underlyings. The parameters include volatilities and correlations of the assets, and, for longer time horizons, drift rates. The exposure of the portfolio is measured by the deltas, and, if necessary, the gammas (including cross derivatives) and the theta of the portfolio. The acknowledged problems with most VaR measures concern the estimation of parameters and of distribution of returns. Rarely are the parameters such as volatility and correlation stable, it is even questionable whether they exist. Moreover the common assumption of Normality of returns is easily shown to be incorrect. Almost all financial time series data shows fatter tails and higher peaks. Even if returns were Normal, the existence of nonlinear products, such as options, in a portfolio distorts the distribution to such an extent that Normality is irrelevant. There have been numerous attempts to modify the simple VaR assumptions to correct for these problems. Unfortunately, none of them are consistently successful. In the technical report which can be downloaded from this site, we outline the new CrashMetrics methodology. We ask what are we trying to achieve when we estimate VaR and can we reduce or even eliminate dependence on any quantities which we cannot accurately measure? If Value at Risk is about normal market conditions then Crashmetrics is the opposite side of the coin, it is about ‘fire sale’ conditions and the farfromorderly liquidation of assets in farfromnormal conditions. CrashMetrics is a dataset and methodology for estimating the exposure of a portfolio to extreme market movements or crashes. Practitioners enjoying the sun in South Africa while learning about CrashMetrics Key points in the CrashMetrics methodology are the following.
Download CrashMetrics dataset (29k) Download Technical Report (154k) CrashMetrics
is a registered trademark.
