Diversification is not enough
The diversification of individual risks, which increases expected risk-adjusted returns, is like apple pie, motherhood and the flag. It is so ingrained in the minds of managers and their clients that this aspect of portfolio theory has remained unchallenged during a period of exceptional creativity in the investment industry. It is applications of the theory that have multiplied like never before, powered by massive increases in computing power and by attracting highly numerate graduates into finance where in earlier decades they might have been engineers or teachers.
As in other aspects of finance, the main focus of all this creative activity was risk management, because it offered payoffs in the two valuable forms: the avoidance of ruin and the maximisation of profit. In the motor industry, better brakes, better road holding and the wearing of seat belts have raised average speeds. In finance, the perceived improvements in risk management have led to increased risk taking.
We can now look back with the benefit of hindsight and see that most of the Darwinian evolution of techniques, structures and strategies for transferring, transforming and managing financial risks is doomed not to survive.
The evolutionary dead ends mostly involve debt and this is rightly what the media spotlight is now falling on. Examples are:
shifts in finance theory that mis-specified the effects of replacing equity by debt
securitisation structures that sought to transform the underlying credit risk of a mortgage portfolio
derivatives that transferred credit risk to third parties
‘value at risk’ models used by banks to drive asset volumes and capital adequacy
Whilst debt dominates the dead ends, equity investing has not been immune to failed experimentation in the area of risk management. We characterise these common failures as having to do with that irreproachable principle: diversification.
The failures take two linked forms:
Unrealistic reliance on diversification effects as a form of risk control
Unrealistic estimates of the effects themselves
Portfolio theory focuses on risk-adjusted returns and so the required inputs that determine the diversification effects are expected returns, the standard deviation of uncertain returns and (because we are addressing portfolios not individual investments) the correlations (or co-movement) between the different building blocks in the portfolio.
There will always be mistakes made about the expected returns of individual investments and even historical evidence can be misleading about standard deviations. But the more telling mistakes involve correlations. When the mistakes are in the direction of underestimating the extent to which asset prices move together, there is a natural bias to relying more than is justifiable on diversification as the main means of controlling risk, as in confining uncertain future portfolio values within some defined range.
Reliance on diversification for this specific form of risk control has increased at the same time that investors have become more aware of and also more sensitive to the consequences of volatility. This looks completely inconsistent but it is not the first time an increased need for a solution to a problem has encouraged false confidence in the solutions coming forward.
In institutional investment, the impact of volatility has increased because of regulated accounting changes that trigger wealth-reducing changes in exposures, such as selling low instead of buying low. Specific instances where regulations have radically redefined investor ‘utility’ include occupational pension schemes, life insurance companies and principal trading desks in a bank. In retail investing, we cannot blame regulatory influences. But the general decline in opaque long-term savings contracts in favour of more transparent and tradable investments has made it more likely investors will see, and then respond to, volatility in portfolio values in ways that are also wealth-reducing. The driver here is not regulatory but behavioural. The effect is the same: they buy high and sell low. They also trade too frequently, incur cancellation penalties and pay new setting-up charges for substitute products. Whilst investing too much in particular assets, they may also invest too little in aggregate to achieve the outcomes they desire or need.
Dilution The 2000-2003 bear market, marked by the bursting of a popular investment bubble, crystallised many of these risks in institutional and retail decision making. The obvious conclusion to draw was that the problem lay in excessive exposure to volatile assets that did not provide hedges for their financial needs, and hence a simple solution would be to manage their total exposure to risky assets better, by diluting them with risk free assets.
Dilution is different from diversification because it does not rely excessively on uncertain correlations. Instead of focusing on how the bets are managed, it forces investors to take some or all bets off the table if the consequences of bad outcomes do not appear to warrant the risk.
The effect of taking bets off the table is to narrow the range of future possible returns but because the risk free hedges have lower returns than the likely mean return of an asset with a risk premium attached, such as equities, it will also lower the entire range. There is no free lunch.
Considering that the most likely risk free hedge for most private client needs is index linked gilts, it is particularly ironic that they had performed almost as well as equities for most of the bull run since 2003, let alone outperforming dramatically in the subsequent bear market. After the event, this looks like a free lunch but it was not something that was predictable before the event.
New solutions for risk control Dilution was the solution the institutional marketplace came up with, in the form of liability driven investment (LDI). This focused client by client on the tolerable outcomes. It controlled the levels of exposure to bets, as opposed to hedges, that left risk on the table but only to an extent consistent with the tolerable outcomes. But with the exception (as far as we can tell) of No Monkey Business, this was not the solution the retail investment industry came up with.
Bruised as they were by the false promise of technology investing, private clients were still eager for new product paradigms, particularly if they looked anything like a free lunch. Into this receptive audience the investment industry fed a steady stream of solutions that promised unrealistic risk control: so-called absolute-return funds, hedge funds, funds of hedge funds and multi-asset class investing based on the spectacularly successful endowment funds of Yale and Harvard. They also fed investors’ appetite for structured products with downside guarantees. Though not new, these provided full insurance of downside risk (unlike the disastrous precipice bonds that cut the cost of the upside option, and so increased upside participation, by leaving very high downside risk not just uninsured but actually geared). As a substitute for cash, there may be investors who would deliberately stake pre-tax interest and buy options on risky assets. But as a substitute for risky assets, or for taking risk off the table, the payoffs of a strategy of rolling over a series of call options in practice highly complex and, like all forms of permanent portfolio insurance, self-defeating in theory.
Perverse business incentives
Compared with taking bets off the table, these product or portfolio constructs were much more attractive options for the industry, both because they like selling solutions that look like a free lunch and because they hate to lose assets under management.
The key insight here is that the industry’s portfolio-based fees model, which No Monkey Business rejects, pushes the single most important set of decisions, total risk exposures, into a parallel universe marked by gaming activity: clients hold back assets they do not want a fee attached to and managers try to maximise the base of assets to which they attach their fee, even if it means having to play down the resulting level of portfolio risk. Indeed, to the extent the solutions encourage higher levels of risk taking in aggregate, they may increase assets under management. Moreover, if the products are more profitable than those they replace, they can increase profits that way.
Hedge funds Diversification effects, via correlations, have proved critical for two of these popular solutions for volatility: hedge funds and multi-asset class investing. But the errors in correlation estimates in each case have been subtly different.
The hedge fund argument relied on the assumption that most of the managers’ returns were from alpha, which can be roughly translated as risk-adjusted return that is not explained by exposure to systematic (and non-diversifiable) risk factors known as betas. Beta is the principle source of returns earned by exposure to equity portfolios when the portfolio’s stock-specific risks have been well diversified. The separation of beta and alpha uses the mathematical technique of regression analysis and so is only as good as the specification of the systematic risk sources and the proxies used for the returns to those risk sources, such as an equity index. By definition, the maths make alpha uncorrelated with beta and so it is highly prized by investors who already have lots of beta exposures or who want to get rid of the volatility of market betas and settle for the less volatile but lower returns of pure alpha generation.
When No Monkey Business: what Investors need to know and why was published in 2002, hedge funds were lumped with ‘entrepreneurial’ investments on the basis that these alpha payoffs were completely unpredictable. Since then, the direction taken by academic research is that the returns are in fact more predictable but only because the betas were mis-specified: there are actually a whole set of systematic risk exposures but they are different from those commonly observed in traditional portfolios. Examples are credit risk, illiquidity and volatility but there may be more that may only be identified in the future. There is also now much more conventional beta in the hedge fund universe as a whole, particularly the US equity market.
Multi-asset class investing Yale and Harvard set a lead amongst US endowment funds by embracing ‘alternative’ investments. There are several strands to their thinking unified by a search for lower correlations. Backing hedge funds, their thinking followed the same course described above: finding alpha. Backing private and illiquid investments held as limited partners, they were making investments that stood up on their own merit, relying on business management skill rather than investment management skill, but whose return paths were also bound to be uncorrelated with public markets just because the valuation frequency and recognition of accounting gains and losses are different. This valuation effect on correlations is shared by investors in direct property, incidentally. But these early adopters of multi-asset class investing also turned to new asset classes to increase the diversification effects via genuinely lower correlations, notably forestry and commodities.
When clients asked us if we could provide richer diversification than the core investments used in our equivalent to LDI, ‘Defined Outcome’ goal-based portfolios, we were careful to explain that the whole strategy was not just a bet on manager skill but also a bet on correlations. We explained that these are both highly unstable and poorly predictable. We also warned that logically the benefits of diversification would be smallest precisely when they were most needed: to cushion large losses shared by capital markets in general. We chose to relate this inconvenient truth to the economic cycle and so this became the basis of our multi-asset class portfolio structure, with the proviso that it still needed decisions about exposure levels, or dilution, to control the risk.
In this extract from a 2005 position paper we suggested the following categorisation of entrepreneurial alternatives to well-evidenced assets.
“The dependencies that lead to the clustering to the left are economic cycle risks, as these explain the convergence of correlations in times of distress. The relevant client exposures are usually equities and property but could also be the business risks a client is personally exposed to.
Economic risks associated with distress link property prices and the credit risk premium on bonds. Property strategies for individuals include buy to let residential investments as well as commercial property via life company investment bonds. The housing and equity cycles, though showing some unpredictable leads and lags, have historically been quite highly synchronised in bear markets.
Illiquidity is also likely to affect many of the assets grouped to the left in the same way, so to the extent there are market prices to establish portfolio values they will be set low enough to reflect buyer” higher required liquidity premium. This can also serve to reinforce correlations between assets or vehicles that are normally not highly correlated. Individuals will have a wider choice of vehicle for commercial property and more diversified residential property when real estate investment funds are launched. Experience in the USA suggests these will be much more highly correlated with equity markets than are other property vehicles (except shares of property companies themselves).
Private equity and VCT returns can appear to be largely dependent on idiosyncratic risks rather than the level of public markets but it is the latter that provides the main exit route for funds and so may be critical to the timing and level of cash flows back to the investor. In the case of private equity, when the vehicle is an investment trust higher correlation with public securities is inevitable.
More intuitive dependencies are emerging markets with developed markets and unlisted equity with public markets.
We position high grade government bonds of wealthy creditor nations away from clients’ other assets because of their almost unique function as a deflation bet.
Though commodities are likely to be linked to the economic cycle they are a cost to industry and consumers and so not well correlated with equity returns.
The cluster of assets least correlated with clients’ core assets is dominated by hedge fund strategies. Within our three high level hedge fund categories there will inevitably be some lower level strategies pursued by individual hedge fund managers that are systematic in nature. However, since they mostly can bet in either direction, the assumption of low correlation with systematic returns is generally sound. The least dependent are relative value bets, market neutral and arbitrage strategies – all of which we loosely group under arbitrage as being characterised by betting on differential price movements rather than absolute price movements of single assets (which could be systematic).
Pure currency bets are likely to be made within global macro funds but can also be made via currency funds.”
In 2008 the collapse of the debt-fuelled global liquidity boom in financial and property assets has proved our point precisely. Diversification has not helped, as correlations have converged dramatically. That is what lies behind the 22% fall in Harvard’s endowment fund in just one quarter last year (we do not know Yale’s yet), the 20% typical falls in the hedge fund universe (also in dollars), the unsparing drop in all developed and merging equity markets and falls of about 40% in commodity index strategies.
It is not what happened that means we were right. It was always right because it could happen. But we clearly had a better understanding, however imperfect, of the correlation problem than was evident in the industry at the time.
Conclusion Diversification, as portfolio theory claims, is an integral part of risk management as it leads to a more efficient set of bets, in terms of risk-adjusted expected returns. It is not a means of controlling the range of uncertain outcomes from a portfolio of bets. This can only be assured by controlling the level of bets relative to hedge assets that match liabilities with certainty and are therefore impervious to volatility.
The bets our Defined Outcome clients leave on the table are in what we call ‘evidenced’ equity markets, which means the expected real returns and standard deviations can be quantified using past history, but the predicted range of some mix of hedges and bets still relies on correlation assumptions for the different equity markets. Here too we have been realistic about the diversification effects. Portfolio return projections assume that these can converge in either very good or very bad times, dramatically widening the range of possible returns compared with more ‘normal’ returns. But the solution to the problem of constructing optimal combinations of equity markets places a high value on diversification of risky exposures, just like apple pie and motherhood, so we use a low correlation assumption when optimising the equity portfolio. How this translated into returns in 2008 is the subject of a separate article.