Deja vu November 2008? Post-August turmoil, global volatility is high and persistent. The Greek sovereign debt crisis, having evolved into the PIGS (Portugal, Italy, Ireland, Greece and Spain), is especially challenging with policy competing against politics. Portfolio managers are grappling with their strategies ahead of the next near-term market shock. Their response so far has varied depending on the models they use in estimating and decomposing their portfolio risk.
For all of the browbeating that modern portfolio theory has taken since 2008, normal distribution value at risk, or VaR, remains a staple of risk reporting and percent contribution to (volatility) risk , or PCTR, remains a key driver in assessing portfolio positions and asset classes. It's accepted that normal VaR metrics are generally reliable during calm, low volatility periods. In the depths of a crisis or market shock, it could be argued that all measures converge with high degrees of risk. Then it's too late and the measure is inconsequential.
That crucial period just prior to any market shock is when risk measures truly stand apart. The widely used simplistic VaR measures not only underestimate the overall level of risk in these pre-shock periods, but they also lack any indicative power of the impending crisis, or worse, provide wrong signals that lead managers down the path of moving into potentially more negative positions. For many institutional investors, the limited value of these risk reports relegates them to “check the box” printouts for the circular file.
Historical VaR, the simplest of VaR measures, offers up risk estimates based on the assumption that past returns are a good predictor. Having the advantage of imposing no distributional assumption, historical VaR simply ranks returns for a selected time window to determine the maximum potential loss. But this simplicity is offset by the human dilemma in choosing the representative period of returns and the lack of insight into the changing correlation between portfolio assets.
Harry Markowitz's mean-variance framework brought volatility as a risk measure to the forefront. The assumption that returns followed the normal distribution made it easy to quantify and decompose total portfolio risk with a straightforward calculation of volatility and correlation. But volatility as a risk measure provided no simple interpretation of likely gains and losses. The introduction of VaR solved this problem. For a long time, Monte Carlo simulated VaR, with its first two normal moments, was seen as a panacea for accurate risk measurement and decomposition. But, as everyone now knows, the implicit assumption of normality ignores the higher moments of asymmetry and fat tails. Since volatility is simply a measure of dispersion around the mean, assessing position risk using PCTR based on the bell curve is both inaccurate and lacking indicative power. A popular approach to improve VaR predictive power is to make the recent data more relevant by applying an exponential weighting scheme, or EWMA. Although this approach adds some predictive power, it fails to address the fundamental shortcomings of not accounting for fat tails and return asymmetry.
The volatile market events of 2008 brought tail risk into the spotlight. Expected tail loss, also known as conditional value at risk, is ideal for measuring tail risk. Focusing only on possible downside events, expected tail loss tells you how much you would expect to lose when you exceed VaR. By isolating tail risk as pure loss, measuring and tracking percent contribution to ETL becomes a powerful tool for both accurately assessing which positions to exit and for advance warning of impending market shocks. But this can only be accomplished when ETL is measured in association with non-normal distributions that account for fat-tails and asymmetry.
So here we are in November 2011, the sovereign debt crisis and stagnant U.S. economy have volatility at new “now accepted” highs, and still the industry standard for measuring risk remains the Monte Carlo normal VaR for the buy side and historical VaR for the sell side.
This summer's Italian sovereign bond crisis is a good case in point. The run-up to Italy finally being downgraded by S&P on Sept. 20 and Moody's on Oct. 4 offers stark differences between risk measures when assessing risk levels prior to Italian bond yields spiking earlier in July.
Focusing on a European government bond portfolio, we measure historical VaR, normal (EWMA) VaR and fat-tailed GARCH VaR using a 500-day window at a 99% confidence level for fixed-income positions across 10 countries. The simulated VaRs have a 30-day horizon. By portfolio weight, the top four countries are Germany (22%), Italy (16%), France (7%) and Spain (7%).