Breaking down asset class into their respective common factor risks seems to yield real analytical insights. Martin Steward asks what investors should do in practice with those insights
Pension funds are awkward. They want it all. They want to generate capital growth – but in a variety of market conditions. They want to minimise volatility and avoid significant drawdowns during market corrections – but they also want to take risks that are rewarded over the long term. Modern portfolio theory was supposed to square these circles, or at least help investors find the perfect balance between risk and return. And the major recommendation implied by modern portfolio theory? Diversify your risks.
Unfortunately, the major recommendation investors took from modern portfolio theory was to diversify their assets. It turns out that’s not quite the same thing. That is why those investors who spent 10 years swapping their domestic equity allocation for diverse global markets, and some of their government bonds for corporate bonds, their investment grade for high-yield – not to mention adding ‘alternative’ investments from statistical arbitrage hedge funds through toll bridges and airports to Nordic forests – were so disappointed by the pain inflicted on their shiny new ‘diversified’ portfolios in 2007-08.
The keenest disappointment was focused on the hedge funds, probably because they were often sold as an ‘asset class’ designed precisely to deliver diversification. But the response of institutional investors was interesting: rather than reject all of the underlying strategies wholesale, they took a closer look at what had happened and found that some of those strategies had indeed delivered diversification during the financial crisis, that some emphatically had not, and that some that had appeared to be diversifying during ‘normal’ times turned out to be the least diversifying during the crash.
“Where we have seen positive developments is a better understanding of the distribution of what is in portfolios,” says Divyesh Hindocha, a senior partner with Mercer. “Especially with hedge fund exposures, investors are looking through to understand what risks they are delivering and what function they are performing in the portfolio. That is especially important when an investment has a useful function but has a cost associated with it – there is almost always a trade-off, and it’s important to understand that trade-off and remember why you bought the asset in the first place.”
One result has been the spectacular collapse in assets allocated to broadly diversified funds of hedge funds (even as hedge fund assets have reached new all-time highs) and the re-shaping of the fund-of-funds business model in general. More profoundly, the fact that investors decided not to dump their hedge funds but to try to understand their underlying risks more completely seems to have led to a greater recognition that a similar problem applies in other assets – that those assets are made up of bundles of different risks, not just one.
“As the world – and pension fund investments – has become more complex, investors have found themselves owning assets whose sources of returns are varied, and varying, as opposed to unique and stable,” as Towers Watson’s CIO Chris Mansi puts it. “If you’re in a world with equities and government bonds, you have interest rates, a bit of government credit risk and equity risk and you could argue that that’s it – and that you don’t need to make things any more complex. But if you have credit, private equity, options, convertible bonds, hedge funds and other diversifying assets the need to supplement the asset allocation approach with some cognisance of a more specific set of risk factors become more obvious. We’ve been working towards this for much longer than the last four or five years, partly because it becomes something of a necessity once you start to add more alternative assets.”
Indeed, it was out of the insights gained from analysing hedge fund risks to detect ‘style drift’ that one of the more recent reports into the question of risk factors underlying traditional asset classes grew, according to one of its authors, APG’s head of client risk management Pieter van Foreest.
‘Investment risk – an approach aimed at controlling risks in pension fund investment policies’, published by Dutch trade union FNV Bondgenoten and co-authored by its pensions adviser Jose Suarez Menendez in collaboration with KAS Bank and Ortec, suggests that asset allocation should not aim for a fixed asset-class mix but rather fixed mix of risk factors – such as interest rate risk, inflation risk, country risk, credit risk, equity risk and currency risk – where the asset-class mix is allowed to vary over time.
“It occurred to me that one could apply some of those insights into hedge fund-style drift to the full portfolio,” says Van Foreest. “The project started with using other people’s tools to investigate whether there was a kind of ‘style drift’ happening within a broad portfolio. Asset classes are not ‘plain-vanilla’, they are exposed to not one but multiple factors, so at the beginning of the project I decided to think about the problem as if asset classes were structured products of risk sources.”
Like Mansi, Van Foreest focuses on the obvious hybridity of credit to illustrate his point. We know that US corporate bonds deliver a positive return to European investors over time, on average, but does that return comes from taking the risk associated with company credit risk, or with holding the US dollar, or with locking in a fixed interest rate for a certain amount of time.
“I need to know this – first, because I have the interest rate risk embedded in my liabilities, and second because if the return is mostly to corporate-credit risk then an instrument like a CDS enables me to get exposure to that without the interest-rate risk,” says Van Foreest. “If I want to, I can strip out the useful risk from an asset class that is really a structured product.”
This sums up the major advantages of analysing your portfolio according to these kinds of underlying risks – credit, duration, currency – rather than asset classes: it enables you to identify those risks and assess which ones are rewarded and which ones are not, clearly; but it also gives you the opportunity to re-structure your investments for purer exposure to the risks that you want and need.
“Once you begin thinking in this way, you are better able to build a properly diversified portfolio that does not rely solely on equity risk or economic growth for its returns – as the traditional 60/40 portfolio does,” says Rita Gamelou, portfolio manager on BlackRock’s Market Advantage Strategies fund, a multi-asset strategy that allocates to six distinct risk factors. “And we totally separate the components when we allocate, in the form of a physical bond minus a synthetic, for example.”
So how many of these common factors are there, and what are they?
The key word here is ‘common’. In a very real sense, every security in the world carries its own idiosyncratic risk – but it would be impractical to try and build portfolios based upon those many thousands of varying and often incomplete data series. The solution has been to model the risk of groups of these securities by identifying the small number of risk factors they share in common that explain most of the variation in their performance.
Markowitz mean-variance optimisation is the most radically simplifying model in that it reduces all risks to a single factor – volatility. Its advantage is that the volatility of any asset, asset class or risk factor can be computed easily, so it is a useful explanatory factor for any investment you could make. Its disadvantage is that, by over-simplifying matters, it understates the dynamic nature of all those other common factor risks.
Thinking in terms of asset classes is another simplifying model, of course. It has the advantage of acknowledging the differences between equities and bonds (and indeed between healthcare equity and mining equity) but, in doing so, it loses the advantage of recognising common risks that cut across asset classes that is inherent in mean-variance optimisation.
Risk-factor analysis is part of this family of simplifying models. That is important to note, because it indicates that we should be looking to identify, not hundreds or dozens of common factor risks to take account of the full complexity of market interrelations, but probably a similar number to the number of asset classes we can identify – perhaps even fewer.
“It makes more sense to think of this as simplifying a very, very complex problem than making a simple approach more complex,” as Van Foreest puts it. “In the Markowitz model there is just one risk – everything has a standard deviation and that is the only measure. But we don’t live in a Markowitz world. At the other end of the spectrum, every asset has a certain idiosyncratic return distribution. The idea behind the FNV project was to find a smaller set of explanatory variables that correlate within factors but not across them – and then, indeed, you find yourself stopping at perhaps five or seven factors.”
At BlackRock, the Market Advantage multi-asset strategy stops at six after rejecting some of the newer, behavioural finance-based systematic risk factors like value and momentum which pay an empirical but not an ‘intuitive’ risk premium, and focusing on fundamental economic risks that cannot be arbitraged away. They are pretty representative of what other practitioners have come up with: real interest rates; inflation; credit; liquidity; politics and regulation; and economic growth.
Some would expand these – Mansi at Towers Watson might add the insurance premium, for example, others would argue for those behavioural factors – but when pressed, most practitioners say that if you want to be very conservative with diversification you should probably assume even fewer common factors.
“There are probably very few genuinely unique risk factors,” says Ewout Van Schaick, head of multi-asset strategies at ING Investment Management, for example. “We suspect that there may in fact only be two dominant ones.”
He would argue that investors would only want to be long in economic growth, liquidity, credit, political and inflation risks if they were confident about the economy and market sentiment. He brings these all together as “common market risk”. If they were fearful, they would only want to be long duration – the “safe haven”. In asset-class terms, you have bonds and cash on the one side and everything else on the other.
Philip Hodges, a director in the multi-asset strategies research group at BlackRock, agrees to some extent. The principal components analysis his team did to identify its risk factors actually showed that a lot of the time six factors are not required to explain the price movements of assets.
“Three will do it,” he suggests. “Growth, inflation and interest rates will explain most asset movements in most market environments. But it is very helpful to have an additional liquidity factor – in 2008, the majority of the spike in correlations was explained by liquidity and not the other three, whereas in normal times liquidity risk doesn’t seem to be anything to worry about.”
This is an insight into why the simplifying aspect of this model is valuable. Asset classes are not only bundles of different risks, but bundles of different risks that become stronger or weaker drivers of asset-class performance in different market environments. In the 1970s, the main driver of high-yield bond performance was inflation expectations and interest rates; in the 10 years up to 2007, it was the corporate credit risk; during 2007-08 it was illiquidity. In the first period, correlation with government bonds was high; in the second period they correlated with equities – and these were quite slow-moving developments linked to the economic and business cycle. In the third period, high yield bonds correlated with everything that was remotely illiquid in comparison to cash – and the speed of that correlation was brutal.
We have moved on from bewilderment as to why correlations in a range of apparently diversifying asset classes ‘went towards 1.0’ to an understanding that the inherent or incipient liquidity risk in all asset classes (which we get paid for taking in most asset classes and which we pay for as a premium to hold cash) had burst forth to dominate performance (positively in cash and negatively in everything else).
What do we gain from understanding this?
“It doesn’t change the nature of the problem, which is that we cannot predict the future,” suggests Van Foreest. “But it can help us to know that our diversification is more robust.”
In other words, once we know that there is significant liquidity risk inherent in an investment (or duration risk, inflation risk, economic growth risk, and so on), we can begin to model more clearly how it will correlate with other investments during periods when that risk comes to dominate performance, and construct our portfolio accordingly.
But as the liquidity example indicates, this is not only about knowing that the risk is there but understanding the shape of the distribution of returns to that risk.
“If you allow for the fact that some of these factors are less stable, and more reliant on the assumptions you make, more left-tailed, more subject to liquidity freezes, you will understand that you have to correct or address some of the results you’d expect from a pure optimisation,” as Mansi puts it. “A big number popping out for credit, which has low ‘normal’ volatility but fat left tails, for example, is something that you would know has to be corrected.”
In our simplified high-yield example, the interest rate risk fluctuates gradually over time. The credit risk has fatter tails in its return distribution: a credit keeps paying its coupons until the day that it doesn’t, but price performance will begin to reflect any growing risk of default before it happens. And the liquidity risk has the fattest tail of the three: for long periods it might barely be detectable in the pricing of the high-yield asset class, before suddenly overwhelming all of the other risk factors combined. But the insight is even deeper than this. When liquidity risk suddenly dominates the performance of an asset class like high-yield bonds, the result is that the growing size of the illiquidity premium being demanded pushes up the total yield – and hence the cost of capital of the issuers.
The result is that a common market factor, liquidity, can damage the access to credit of all companies issuing bonds regardless of their idiosyncratic creditworthiness: as a result of a spike in liquidity risk, we also get a spike in the common market factor of credit risk.
This is what the statistically-inclined call ‘tail dependency’, and it helps explain how correlations can rise, not only between asset classes that share common factors, but also how they can rise between asset classes that do not share common factors, but which share common factors that are tail-dependent.
“If you set up a model for the whole portfolio you have to think about the dependencies,” as Wolfgang Mader, head of asset allocation strategies at Allianz Global Investors’ Risklab, puts it. “Are the relationships linear? Or is there more dependence in the tails of distributions? Are there some factors that become most important when we are already in the tails of the distributions of some other factors? That’s what we are trying to build into models as realistically as possible.”
This is why a simple risk-factor model is necessary but not sufficient to improve upon a model that only takes account of asset classes, agrees Mansi: “Crucially, once we have arrived at that, we need to think about the kinds of scenarios under which those factors might correlate.”
In a way, this is the justification for practitioners reducing their six factors to three or two – it is just a way of recognising the tail dependency of many of the (usually independent) factors.
The practitioner might object that all this stuff about common risk factors has failed to address the fundamental problem that we set out to deal with: the fact that unexpected correlations occur between asset classes. If unexpected correlations occur between risk factors as well, what have we gained from this shift in the focus of our analysis?
The answer is that these correlations are no longer unexpected. Without identifying common factor risks, we can know that an asset class has a fat-tailed distribution of returns but still be unsure exactly what causes the tails in its distribution – and therefore when those tail events will occur and what correlations they will cause with other asset classes. But once we understand that, say, the tail event in a corporate bond is likely to be caused by a tail event in either credit risk or liquidity risk, we can begin to analyse the extent to which those risks exhibit tail dependency with each other, but also the extent to which other asset classes are exposed to those risks and therefore liable to correlate when those risks are experiencing the tails of their distributions. In other words, the tail events will always come as a surprise, but the tail dependency no longer will – with all that that implies for improvements in portfolio optimisation.
For some investors this will lead to revolutions in asset allocation. For others, the extent to which derivatives are required to hedge unwanted risks and leverage others will be a constraint. Many will implement practical solutions based on these insights as far as they can within those constraints. But all, it could be argued, can find some way to use them to make their diversification more robust than it was in 2008, and remains today.