Continue active refreshing of this index's data?

Continue active refreshing of this index's data?

Blog Posts — April 16, 2020

Risk measurement and modelling in times of crisis

We are experiencing unprecedented volatility in the markets, many experiencing asset losses and volatility spikes in close to or in excess of the global financial crisis of 2008. Risk is at the forefront of the entire investment community. And once again, there are some who are criticising the efficacy of Value at Risk (VaR) models in time of crisis. Some of this criticism is valid, but it is worthwhile outlining the underlying mechanics of VaR – and underline that VaR is a useful tool, but it is critical to understand how it works and that it is only one tool in the risk management arsenal – especially in times of volatility.

Mechanics of VaR models

It is important to remember that VaR is calibrated using measured volatilities and correlations as they occur, so when market volatility is trending higher and where asset correlations are increasing as we’re seeing now, the risk measured by VaR will naturally spike higher. Even for the best hedged portfolios, relative risk can increase as the idiosyncratic risk balloons.

Risk model not working? check the focus

There is a natural tension between having a responsive risk model that reflects the current market paradigm versus having a slower-moving model that helps long-term asset-allocation decisions and does not breach risk limits on a daily basis.

The vast amount of feedback we get from clients is that they prefer to have a responsive and appropriate risk model, pointing towards a shorter-term equity model or fixed income model with DTS and a shorter model half-life. The benefit of your model reacting quicker is two-fold: firstly, it is providing a realistic risk level for the current stressed environment, and secondly, it can give a signal or forced trigger to reduce risk early in a falling market saving potentially larger losses. If your risk model does not appear to be picking up the market moves and you are regularly breaching your reported VaR (P&L) limits, switching to a shorter dated, more responsive model may help.

We saw this happen in previous crises – most recently in 2008. When you are in the middle of a so-called Black Swan event, the risk of another once in a 100 years or 200 years type event (99% or 99.5% VaR) ends up at the very tail of your risk distributions and as volatility increases, naturally increases these tail risk projections.

Mind the Gap

Perversely, when asset values drop significantly combined with large vol increases, pure mathematical VaR models may project you can lose more than the market value of the instrument you hold. Here loss-capping and tools such as Cornish-Fisher VaR can help apply some common-sense to pure statistical models.

Again, relying one single VaR number means your view of risk is incomplete: a one-dimensional view in a multi-dimensional world. You wouldn’t drive a car just looking at a speedometer (and not looking out of the window or in your mirrors) – nor fly a plane just looking at the altimeter.

What then should you be looking at?

More than just a single VaR number, is the simple answer. Or at least looking under the hood with a more forensic lens. An effective risk framework needs to include multiple dimensions of risk – starting with exposures and sensitivities; multiple VaR calibrations; looking at expected loss and tail risk; then ability to stress test portfolios for out-of-sample scenarios and extreme unpredictable events. Of these, the first is arguably the most important – if you are not happy with the sensitivities your fund is exposed to, start fixing your risk there.

Short, medium and longer-term VaR calibrations should be available and the most appropriate one selected for the use case. The current high volatility is likely to nudge you towards the model calibrated using the shorter-history to provide a model most appropriate for the current operating environment.

Linear vs non-linear risk models

Many traditional risk models assume linear, normally-distributed return profiles. That is fine when your assets are purely linear in return profile – e.g. cash equities – price goes up 10%, return goes up 10%. Not all assets however have linear payoff profiles – Puts, Calls, Credit Default Swaps and Swaptions – therefore a linear model may not be appropriate. In stressed times such as this it becomes clear very quickly which assets in portfolios have non-linear payoff profiles. Here, having a full-revaluation risk engine with full revaluation monte-carlo risk models, full revaluation stress testing and scenario analysis is necessary.

Stress test, stress test, stress test

This can’t be stressed enough (sorry). What-if scenarios, regulatory shocks, transitory shocks, factor-based scenario analysis, replaying historical events – the more stress tests, the better. The ability to get the full picture means identifying the outliers — the ‘edge’ events that you can identify and importantly, hedge. And don’t forget about reverse stress testing as well: what is the scenario or event that you do not want to see, what would trigger that event and again, how can you avoid this from happening?

The human touch

In order to get a grasp on your exposures, it won’t do to rely just on VaR or a single risk model. A good risk framework will always need human input and interpretation of these underlying numbers for a complete view of risk in your portfolios.