Value At Risk (VaR) is a risk measure widely used by Institutional managers. It is also one of the primary risk metrics required for reporting to many global regulators.
VaR is defined as the maximum loss of a portfolio for a confidence level and time horizon. For example, we can say that a portfolio cannot lose more than USD 1 Million at 99% confidence level for a 10 day time horizon.
Commonly used time horizon periods are 1 day, 10 day and 1 month. And the popular confidence levels include 90%, 95% and 99%.
There are 3 key methods to computing VaR – Parametric VaR, Historical Simulation VaR and Montecarlo VaR.
– Parametric VaR: Simple to implement, Returns based and uses Variance-CoVariance approach.
– Historical Simulation VaR: Popular and uses historical data
– Monte-Carlo VaR: Uses simulation techniques and the most complex of the three methods
Key Challenges to calculating VaR include –
– Identifying and mapping key risk factors to instruments within the portfolio – FX rates, yield curve, asset price, Spread, etc.
– Availability of historical market data for all securities in the portfolio
– Handling data gaps (because of newly issued securities in the portfolio and/or non-availability of market data)
— this can be handled through use of proxy security or by using beta method to calculate returns based on market index
How can we verify the models are indeed predicting the VaR accurately?
As VaR is forward looking measure the accuracy of prediction will be important for it to be a useful measure for all stakeholders including portfolio managers, regulators and/or investors.
Risk teams need to monitor the model’s prediction frequently i.e. daily/weekly.
‘Backtesting VaR’ is a popular technique that can compare forward looking P&L (i.e. VaR) to actual performance results.
When the prediction is out of tolerance, the risk analyst can input comments providing reason for the model’s results (unusual market condition or event maybe one reason). This information capture on exception will be helpful not only from an auditing perspective, but also a good risk management practice.
Based on the results of backtesting the model may need tuning to ensure it continues to be a good predictor of portfolio risks.