VaR Methodologies Compared

I recently came across a succinct comparison of the three accepted approaches to computing Value at Risk (VAR). According to Boris Agranovich of GolobalRiskConsult:

VAR or Value at Risk is a summary measure of downside risk expressed in the reference currency. A general definition is: VAR is the maximum expected loss over a given period at a given level of confidence. VaR does not inform on the size of loss that might occur beyond that confidence level.

The method used to calculate VaR may be historical simulation (either based on sensitivities or full revaluation), parametric, or Monte Carlo simulation. All methodologies share both a dependency on historic data, and a set of assumptions about the liquidity of the underlying positions and the continuous nature of underlying markets. In the wake of the current crisis the weaknesses of VAR methodology became apparent and they need to be addressed....

A VAR system alone will not be effective in protecting against market risk. It needs to be used only in combination with limits both on notional amounts and exposures and, in addition, should be reinforced by vigorous stress tests.

I personally maintain that the Monte Carlo (i.e., stochastic) approach is superior to the other methods because the results provide a more "polyvalent" explanation of the financial risk in the subject investment.

Source: GlobalRiskConsult

See also:

VaR Methodologies Compared II