High volatility stress tests a strategic asset allocation. The dispersion of returns across managers will likely be wider and the dispersion of returns across the peer group against which the plan sponsor is judged will also likely be wider. Other events associated with high-risk environments greatly increase the chance of a weak point in an allocation springing a performance leak. The past two or three years have exposed several of these weak points.
The measurement and prediction of asset return volatility is one of the most important areas of empirical finance. Almost all market participants would agree with the fact that stylised returns have fat tails – meaning that there are more big moves than might be expected using the normal distribution – and, a lot of the time, markets go sideways for all major asset classes.
A growing body of research supports the idea that asset price returns are generated by regimes. From currencies to fixed income to equities, good evidence has been found for the idea that there are occasional sharp changes in the level of risk and return that an asset is likely to generate. In a bull market, investors always see the silver lining, in a bear market, only the cloud.
This is interesting because it turns out that changing regimes can generate the classic fat-tailed and peaked-in-the-middle (or leptokurtic) pattern of asset returns.
Like most asset price returns, the actual distribution of the Russell 2000 Index has fat tails and is peaked in the middle. In other words, extreme events are mixed in with long periods of tedium.
Regimes don’t come and go randomly. Even in this simple case with two distributions, the high-volatility regime needs to be far less persistent than the low- volatility regime for the end result to look like the observed pattern.
Only if we are in the low-volatility regime most of the time and then occasionally in the high-volatility, will the pattern build up to a fat-tailed distribution with a peak in the middle.
So, in addition to identifying the levels of volatility in the different cases, it is also vital to try and generate insight into how persistent each regime can be expected to be. As an illustration, the table shows the results of a regime analysis for 1 June 1991 to 30 June 2001 for the Russell 1000, Russell 2000 and EAFE Indices (all in US dollar terms). This regime analysis simultaneously estimates the return and risk in each regime and how persistent each regime is.
The risk column is the annualised volatility in each regime. The return column is the annualised mean return in each regime and the persistence column shows the likelihood of each regime lasting more than a week.
The persistence levels confirm the intuition that, thankfully, the high-risk state isn’t as long-lasting as the others.
If this way of thinking about volatility is right, then there are a lot of implications. First of all, conventional approaches to the estimation of market risk which measure volatility as some form or other of trailing average based on the past will be doubly misleading. Such measures will tend to overestimate risk in the low- or medium-volatility regimes and radically underestimate it in the high-volatility state.
Imagine that markets are relatively tranquil, meaning that the prevailing regime is low or medium. The trailing average of history will do a good job as long as this regime stays in place. However, if an event happens that triggers a switch to the high-volatility regime then the trailing history volatility measure may well be worse than useless, based as it is on data that is not relevant to the current higher-volatility regime. If the market spends a reasonable amount of time in the high-volatility regime, then the trailing volatility measure will eventually catch up – perhaps just as the regime changes back to a more tranquil one. At this point, the trailing measure will likely overestimate risk, again for a considerable period of time until the trailing window includes a sizeable chunk of the more tranquil environment.
This is relevant for all risk measures built this way. For value-at-risk measures, which are specifically used to measure the probability of loss over very short-term horizons, the regime-based critique is a damning one. In essence, it means that the capital set aside to cover loss based on this measure will be too high most of the time and woefully inadequate in the high-risk regime.
Strategic asset allocation analyses often include a stress test or optimisation analysis based on a long run of historical data. In a regime-switching world when the prevailing regime is low or medium, this form of analysis probably overestimates risk because the long run contains some higher-volatility times. By the same token, in a higher-volatility time, this form of analysis understates risk. It is, of course, true that any estimator based on an average will be underestimated part of the time. What is really important is the degree of the underestimation. The regime-switching view suggests that because volatility is radically higher in the extreme environment, the underestimation of risk based on long averages is also extreme.
This is something of great importance for defined benefit plans right now. Many market participants believe that the recent period of very high equity returns is not just temporarily on hold. There is a strong case to be made that returns will likely be lacklustre for the next few years, even after the bear market comes to an end. Thus, the last thing a pension fund needs is to be too conservative in allocation or have the underlying managers be too conservative because of risk measurement systems assuming that the extreme conditions of the past few years should be projected forward into the post-bubble era.
If your strategic asset allocation is based on long-run analysis and estimation of risk, then it is likely to be a very poor guide to what will happen to your portfolio in the high-volatility regime. The first practical suggestion then is to apply regime-switching analysis to your overall strategic allocation. This will allow those in charge of the strategic direction of the fund to get a better feel for how the fund might perform in different environments. As with many things, forewarned is to some extent forearmed. A regime-based analysis can allow a fund to determine more realistically what they can live with in terms of volatility and can serve as an educational tool for boards. If markets really do work this way, then even a very well-constructed allocation will occasionally result in some wild swings in portfolio values.
These swings will be very difficult to avoid because the only way to realistically do this is to accurately forecast when the regime is likely to shift. Then the plan would need to make shifts in the plan’s strategic allocation quickly enough to make a difference in the outcome. Both of these are heroic assumptions to make.
Alternatively, a risk-averse plan could put together an allocation based solely on the likely distribution of returns in the high-volatility regime. This would certainly give better control over the investment outcomes in this high-volatility environment, but only at the expense of a much-too-cautious approach in the other environments. If the portfolio is managed so that it will not suffer wildly from excessive swings in a high-volatility regime, then in more normal times, it will likely be way too conservative.
Two very important conditions that contribute to this effect are increased correlation of manager returns and increased concentration in benchmarks.
Manager diversification has, in the past, been a fairly constant mantra from the consulting community. However, high-volatility regimes can be quite a challenge to this approach for a couple of reasons. To analyse these reasons it is necessary to distinguish between the correlation of the positions taken by managers and the correlation of the returns of the assets that the managers are choosing between.
Heightened correlation across and within asset classes is an important aspect of high-volatility market environments. Fear is even more contagious than greed, and since down markets lead to losses, which lead to demands for liquidity, which lead to sales of assets, there are very practical reasons for correlations to rise in tough market conditions. This rise in correlations, however, magnifies at the portfolio level the impact of the increase in volatility. Thus, even if the positions taken by managers have the same degree of correlation as on average through more tranquil times, the correlation of returns from those managers will likely increase in more volatile times.
However, there are good investment reasons to believe that the correlation of manager positions also tends to rise when markets approach extreme conditions. When valuations are stretched in a given set of asset markets, the pool of assets that will likely pass valuation screens is naturally going to tend to shrink. Thus, value managers who stay entirely true to their style will probably end up with positions that are more highly correlated with other similarly pure managers. How many technology stocks would a value manager have been holding in mid-1999?
Even when the number of opportunities isn’t directly affected, extremes of valuation can make certain positions almost irresistible regardless of approach. At the start of 1999, an overwhelming majority of global fixed income managers were underweight Japanese government bonds (JGBs) because it was felt, quite incorrectly as it turned out, that Japanese yields just couldn’t go any lower. JGBs were the best-performing bond market in 1999 and the relative return streams of global fixed income managers were therefore much more highly correlated, and more negative, than usual.
This raises the very real possibility of a ‘double whammy’: increased correlations in the assets in which managers invest and increased correlations in the positions they end up taking. These changes, occurring at a time of high volatility, will magnify the increase in tracking error that might naturally be expected to occur.
As more managers are added, the aggregate tracking error will fall if they are all managing to a 2% target. With four active managers, the aggregate tracking error is going to be only 50% of the target. A fund with this structure would likely end up with something close to a passive or enhanced index return but at an active fee level and with the administrative burden of dealing with four managers.
Assuming this is not a desired outcome, what most funds would presumably do is give the individual managers a higher tracking-error target so that in aggregate, the tracking error comes out at around 2%. Assuming that the excess return streams are uncorrelated, this would mean that each manager would need to be given a tracking error target of 4% resulting in an aggregate 2% outcome.
However, if the excess return streams become more heavily correlated, then the aggregate tracking error will increase. In the event that all the managers ended up with roughly the same positions, or positions that substantially overlapped, then the aggregate tracking error – even assuming no change in the volatility of the underlying assets – would approach 4%. If correlations of assets and positions are sometimes associated with periods of high volatility, it is easy to see how the high-volatility environment can really test a strategic allocation and the assumptions on which it rests.
Another practical implication of this work is that plan sponsors should look carefully at the manager structure that is employed. Some degree of manager diversification is important but there may be simpler and better ways of getting at the excess returns a four- or five-manager structure delivers for (say) the US equity portion of the portfolio, without incurring the risk of greatly increased tracking errors in tough market environments.
In the US, in particular, equity management is dominated by the twin distinctions of capitalisation and value/growth (cap–v/g). Proponents of this approach argue that a fund can ensure that the correlations between managers are kept low while retaining the benefits of manager diversification.
However, a potential drawback with this approach is that, again, particularly at market extremes, the benchmarks can become questionable investment choices which a rational investor probably would not pick as a starting point for a portfolio.
The debate over the merits of style benchmarks is growing in volume and there are many good arguments on both sides that are being glossed over here. The main point that needs to be drawn out when thinking about volatile market environments is that these environments are often associated with high levels of concentration in benchmarks. This concentration again puts pressure on the assumptions underlying the strategic allocation. The behaviour of value and growth indices in the past few years does raise a serious question about allocating to managers on this basis. When markets reach extremes, this can have unintended negative consequences that greatly increase the risk of a serious performance shortfall because of the potentially large increase in volatility driven by the benchmark concentration and the likely increase in correlation between manager returns.
High-volatility times shine a light on some of the assumptions typically made in a strategic allocation. The period from 1999 to 2001 has been a protracted period of high volatility associated with extremes of valuation. Multiple-manager approaches and style benchmarks have had their weaknesses exposed by the combination of benchmark concentration and high volatility.
Interestingly and not surprisingly, this has been associated with growing interest in the ‘whole stock’ portfolio concept developed by Ennis Knupp#. They too argue that style benchmarks are not useful and also don’t recommend multiple-manager set-ups. Thus their basic conclusion is that a fund should have a large allocation to passive management and a small allocation to a few active managers with broad benchmarks not constrained by style. By avoiding multiple managers and, crucially, by staying away from style benchmarks, two of the biggest contributors to serious performance risk in high-volatility regimes are eliminated.
However, the downside is that if, as in the example used in the paper, 80% of the portfolio is invested passively, then the active managers have to generate significant alpha if the fund is to outperform in a meaningful fashion.
Anthony Foley is principal and managing director of the Advanced Research Center of State Street Global Advisors in Boston