AllianceBernstein’s Ashwin Alankar and Mike DePalma consider the merits of measuring expected tail loss.
By any name – Black Swan, three-standard-deviation event or negative tail event – the risk of unexpected heavy losses is a major concern for investors. The question is how best to protect against these low-probability, high-impact market moves.
As the 2008 crisis and all those before it have illustrated, the rules of the game under normal conditions are not the same as they are at times of market stress. The portfolio investors want to hold when asset volatility and correlations are at their long-run averages is not the portfolio they want when volatility is spiking and assets that are usually weakly correlated are all nose-diving together.
That’s why, as history has shown, the old-fashioned static approach to diversification – typified by a permanent 60% in stocks and 40% in bonds – works well enough in ‘well-behaved’ markets but offers little protection in ‘bad-state’ markets.
This is because bad-state correlations are dramatically higher than ‘normal’ correlations. When correlations spike, a ‘diversified’ stock portfolio is not very diversified, nor is any other portfolio that relies on ‘normal’ correlations. In addition, many assets have ‘fat-tailed’ return distributions – in other words, they tend to produce more extreme outcomes on the downside than they do on the upside.
One way to address tail risk is to buy insurance via the options markets. But this costs money – a lot of money. The drag on performance in non-crisis markets can be difficult to bear.
It is not unlike someone who owns a house in an area known to suffer disastrous storms a couple of times a decade. They can pay for insurance every month, or they can opt not to insure, and instead develop an accurate weather warning system that tells them when to set out the sand bags and board up the windows. The question is how to design that warning system.
One idea that has caught on is the notion of balancing risks via risk parity (RP). RP will adjust the weightings in a portfolio such that each asset class contributes equally to overall portfolio volatility. So, at times of crisis, the allocation to more volatile assets like equities would be dialled back. This is a significant improvement to the static approach. And in normal times RP works pretty well. The key words here are ‘normal times’.
Market volatility is not the best indicator to use in an early warning system. For one thing, not all volatility is bad. Think back to the build-up of the Internet bubble or the impact of Abenomics on Japanese equities earlier this year. In both cases, rising equity volatility was associated with upside gains rather than downside risk. So an approach that limits volatility is likely to act as a drag on performance by capping the upside as well as the downside. More important, history shows we live in a ‘fat-tailed’ world characterised by tail risk: a risk that volatility cannot fully capture.
But there’s a better alternative: measuring expected tail loss (ETL). We use a forward-looking ‘implied expected tail loss’ measure distilled from options-market information and use ‘bad-state’ (tail) correlations to assess diversification benefits.
One of the primary uses of options markets is to allow investors to protect their portfolios from large negative price moves. Option prices represent the cost of this insurance: the higher the price, the greater the expectation the asset will suffer a large loss. This is the key intuition behind our implied ETL calculation.
The ‘Goldilocks’ period of 2006-07 illustrates how volatility and ETL differ as measures of risk. In our simulations, during this period, the TRP approach was registering very high levels of equity tail loss risk, whereas volatility was close to historical lows. Options were pricing in the risk of the painful deleveraging process that ultimately contributed to the credit crisis. So, unlike a volatility-based approach that would have been giving the all-clear to take on more equity risk, TRP was signalling a need to reduce equity exposure.
Today, we are facing another correction process as the imbalances created by the US Federal Reserve are re-equilibrating. Amid talks of monetary stimulus tapering off, assets are re-pricing to a new equilibrium. In this environment, tail risk is a very real concern. Premature withdrawal by the Fed could cause a painful downward spiral.
To recap, whereas risk parity focuses on volatility, tail risk parity defines risk as expected tail loss. In practice, the two portfolios would look very similar in normal times, but different in ‘bad state’ scenarios where the risk of market meltdown is higher.
Tail risk parity seeks to reduce tail losses significantly while retaining more upside than risk parity or other mean-variance optimisation techniques. A tail risk parity approach hedges the risk of large losses less expensively than using the options market (historically, we estimate savings of about 75% for an equivalent amount of protection). Tail risk parity offers an attractive solution for investors seeking balanced investment portfolios that can reduce exposures to tail losses cost-effectively.
Ashwin Alankar and Mike DePalma are co-CIOs of tail risk parity at AllianceBernstein. Their full research paper can be found here.