Risk Management
Authorised and regulated by the Financial Services Authority
Extreme Risk
Recent experience highlights the need to consider the likely behaviour of portfolios in turbulent markets as well as in normal markets. This is often coupled with the assumption that widely-used risk measurement techniques failed during the recent turmoil, but a more thoughtful approach yields a more subtle and helpful explanation.
While normal, day-to-day risk management aims to align risk with expected return so that the portfolio meets its investment objectives, extreme risk measures identify exposures that could cause significant damage if a shock were to occur.
uantifying these exposures is the key to designing targeted and cost-efficient hedging strategies, for example using deep out-of-the-money options. Inefficient hedging is a drag on normal performance and mis-estimation of extreme risk can give a false sense of security. It is therefore critical that extreme risk analysis results be relevant and valid.
Markets behave differently in turbulent conditions, and models that are designed to forecast normal risk tend to understate both the likelihood and the severity of shocks.Extreme events are more likely than is predicted by normal risk analysis, largely because nearly all risk analyses draw on historical return information to compute the covariances that are the engine of risk measurement. If the sample data from which this is drawn includes no extreme event, then the risk model will assign a very low probability of one happening. Include a crisis and any risk tool will accord some reasonable probability of it happening again. Add to this the fact that many extreme events are preceded by a period of abnormal calm, and the underestimation is greater still. The most intuitive solution – include longer return histories in the sample data – introduces other, often less tractable problems, such as structural changes in the composition of markets (think of how the wave of privatisations in the early 1990s and the dotcom boom of the late 1990s changed the shape of investment opportunities) and survivorship bias, which render much data irrelevant, confounding the results.
The tendency to understate the severity of shocks is because normal risk models assume that markets are more or less efficient, which they are most of the time. In a crisis this core assumption is violated. Correlations between assets change, sometimes dramatically: “diversification fails just when you need it” as investors sell even cheap assets to raise cash. Another violation of the normal model is correlation in sequential price falls: extreme markets can trend sharply down, as margin lending and dynamic hedging provisions are triggered.
It might be tempting to replace normal risk models with models designed to capture behaviour in turbulent conditions, but this would ignore normal market conditions, which after all are the most frequent. The result would be a mis-allocation of risk and consistent under-performance.
It is not a question of either/ or: effective investment management demands both normal and turbulent risk models.
R-Squared has developed a robust and intuitive technique for estimating the risk that is characteristic of extreme and turbulent markets. Unlike many extreme risk models, it is totally transparent and not overly sensitive to input data history. It also stands out from other extreme risk models in that it is not merely a “scaled up” version of normal portfolio volatility or tracking error. Instead it addresses directly the specific characteristics that distinguish extreme and turbulent markets from normal market conditions.