Stresstests recognize extraordinary losses in funds that may arise as a result of unusual changes in the market parameters and their correlations which are not covered by ordinary market risk measurement methods. In addition to the VaR calculation, which is only valid under normal market conditions, by law an adequate stresstest system has to be implemented .
In addition to stress test volatility - an increase in volatility without market movement is simulated - and stress test correlation - an increase in the correlation coefficients without market movement is simulated - RiskMan offers a set of over 100 individual stress scenarios according to the parameters share prices and volatilities, interest curves and interest rate volatilities , exchange rates and volatilities as well as spread over yield of bonds and CDS.
Traditional stress tests provide fixed stress parameters and thus pose the risk of false illusion
security (are there much more dangerous scenarios of similar plausibility?)
transparency (should a portfolio be redesigned if a 1000-year scenario shows a bad loss?)
effectiveness (have the risk factors of the portfolio changed or do not fit all the risk factors to the portfolio?)
Strest tests of the second generation
Stress test methods of the second generation consider all relevant scenarios of each portfolio and systematically determine the worst case scenario over a defined minimum plausibility threshold.
The "Worst Case Finder", which was developed in Austrian National Bank, is integrated part of RiskMan solution (Guidelines for Market Risk - Volume 5, Implementierung von Krisentests, OeNB, see PDF)
Not only the worst case scenario is calculated within a set of plausible scenarios, also the risk factors responsible for the loss are determined. This information can be used to construct hedging positions that reduce the worst case loss.
Due to the integrated multi-processing function even large and complex portfolios can be computed by conventional server systems within minutes.
The Mahalanobis distance is a measure of the distance between points in multi-dimensional vector space where points of identical Mahalanobis-distance will form an ellipse (see diagram; the number of dimensions in vector space is determined by the number of risk factors).
The worst case can be found outside the VaR loss, excluding highly unlikely scenarios. The shape of the ellipse is subject to the correlation of these various parameters.