VaR of the Jungle - Don't let the Leopards and Mambas Inside

VaR is a widely used risk measure of the risk of loss on a specific portfolio of financial assets. VaR expressing risk as a single number has been criticized since it moved from trading desk into the public eye. Summarizing it gives false confidence, because it cannot express risk of rare events.
Even VaR supporters agree that it is easy to misunderstand, and dangerous when misunderstood.
Making VaR control the central concern of risk management it is more important to worry about what happens, when losses exceed VaR.


Aaron Brown, risk manager at AQR Capital Management: VaR is a tool of risk management, not risk measurement. Putting in a VaR system always leads to improvements in information systems .... and pointedly, VaR is only as good as its backtest. And other risk tests (I add), dealing with VaR break period distributions, dependance on the Var value ...

Having taken this challenge we developed our VaR module as development environment on top of UnRisk-Q for quant developers and risk managers.
In the UnRisk universe of VaR, calculations can be made

• of portfolios that contain all kinds of financial instruments that can be valued in UnRisk-Q - a wide coverage from the simple bonds, equities, .. to the most sophisticated convertible bonds, range accruals, ..
• across 3 methods - historical, parametric, Montecarlo
• across a vast variety of underlying risk factors - interest rates, equity prices, fx, rates, credit spreads, volatilities, correlations, ...

The comprehensive output includes 

• portfolio and individual VaR based on all risk factors
• portfolio and individual VaR based on risk factors separately
• contribution VaRs - contribution of the instruments along risk factors to the portfolio VaR

Providing such a vast variety of scenarios and outputs, speed-at-accuracy really matters.

• UnRisk-Q and consequently the VaR module are grid-enbled
• developers can choose full valuations, applications of principle component analysis, combined with Taylor series expansion 

If an average valuation of deal types in a portfolio of 100 instruments take 0.3 sec, and you have only a few risk factors and 500 scenarios the output of a VaR cube would take about 13 hours on a single CPU and full valuation.

In a networked pool of 32 core processes and with principle component application and Taylor series expansion this can be done in less than 3 minutes.

Aaron Brown again: VaR is like a fence built around a village in the jungle. The fence is built in a relatively safe place, and remaining dangers inside are cleared out. But people spend, say, 1% of their time outside the fence where dangers are greater and less known.

Our contribution is to safe risk mangers from getting lost in the numerical jungle when building the fence and consequently find leopards, snakes and poisonous plants inside - unexpectedly.

Out VaR developments were challenged by Solventis, a leading risk management consulting firm in Barcelona, Spain.