Does Fidelity Special Situations manager Anthony Bolton disprove active management sceptics?
The performance myth persists partly because of the emergence from time to time of ‘legendary’ managers. One such is Anthony Bolton, soon to retire as manager of Fidelity’s Special Situations fund. How does he fare when measured against such tests? He’s good but not good enough to replace a tracker, even to the extent of 20%. He also demonstrates the fatal practical flaw of active management:even for the best managers after the event, relative performance is so inconsistent along the way that most advisers, let alone their clients, cannot stay the course.
Testing for skill
I am coding this for ‘grown-up consumers’ – because the lessons are so important. I therefore need to explain how and why we assess managers’ past performance in a particular way (this is within the wealth management business of No Monkey Business Limited). I shall emphasise the decision-making logic and the statistical principles and keep the algebra hidden. Chris Drew, who developed the application we use and is Keeper of the Algebra for No Monkey Business, has contributed to this item.
The logic starts with the view that ‘asset allocation’ dominates ‘selection’ – as covered in its own area on this site. By asset allocation we mean exposure to asset classes and to markets within an asset class. In this context, the asset class is equities and the market is the UK.
Any actively-managed equity unit trust investing in the UK will pick up two sources of returns: those explained by exposure to the UK market as a whole (known as ‘beta’) and those not explainable by the market and so, by deduction, a reflection of the manager’s skill (known as ‘alpha’). A manager can do well or badly by holding stocks which move with the market but typically by greater or lesser amounts – as in holding more ‘high beta’ or ‘low beta’ stocks. Alpha adjusts for this relative exposure to the market movements and so describes the payoffs to a set of bets, not the bets themselves. Because market beta can be purchased for about one quarter of the cost of an active fund, alpha has to get over a high cost hurdle to be worth buying.
For the sum of all managers, alpha must be negative to the extent of average costs. The average beta is fairly close to 1 (usually below because funds tend to have a permanent cash float) so the cause of this collective outcome is not anything different active managers are doing in terms of the bets they are making: it is just that the payoff before costs is zero. No skill amongst managers as a whole.
This tells us that the search for alpha is a search for exceptions, so we should expect it to be difficult, not easy. It follows that the tests we use to search for exceptions should be rigorously sceptical.
Having determined, as a decision higher up in the hierarchy, what exposure you want to UK equities, your default position is to implement that through trackers. You will only substitute trackers by active funds to the extent that errors in estimating the new set of risks and returns you introduce will not destroy the integrity of the risks and returns on which you based the asset allocation strategy. The lower-order implementation choice must not be able to drive the higher-order strategic plan off the rails. This approach gives you a solid framework for both active-manager selection and active-manager exposure.
Whereas the reasons for separating beta and alpha are clear, the statistical methods, based on regressions, can only be expected to yield approximations of ‘the truth’ about a manager’s bets and payoffs. The presence of ‘statistical error’ gives us a purely objective reason for modifying the observed values resulting from regression analysis to be sure that our recorded alpha is true or ‘credible’. The basis of this adjustment is ‘standard error’ of alpha: the distribution of individual readings around a fitted tendency. You can think of this problem as being akin to the use of a piece of evidence in a court of law when it is not ‘beyond reasonable doubt’ that there is only one way to interpret that evidence. (This is a separate issue from ‘consistency’ of alpha, which is an obvious dimension of manager skill to look for and I will return to it).
Adjusting for statistical error has another advantage. It allows us to include in the same contest funds with different lengths of data history.
Whereas these tests clearly deal with the issue of whether the estimates derived from a regression are really ‘representative’ of the true position, they indirectly recognise that (in this case) we do not know how representative some historical data sample is of the future. The whole point about manager selection, however it is performed, is that we make a prior assumption that skill persists. This is what makes it different from luck.
Anthony Bolton’s stats
The data we are using is 20 years of monthly returns for Fidelity Special Situations accumulation units (so with income reinvested) and the same for the FTSE All Share Index (source: Lipper Hindsight). This is not Anthony Bolton’s entire period at the helm of Special Situations (which began in 1979) but it is a reasonable approximation of the data investors would have used if they were not selecting funds with short histories.
Special Situation’s whole-period beta is 0.92 and does not display any permanent bias to higher or lower risk stocks. Beta is therefore unlikely to be the source of any significant differences in conclusions about Anthony’s skill.
The measured whole-history alpha is 5.3%. This tells us that over 20 years Anthony has comfortably cleared the cost hurdle of his fund (the current total expense ratio is 1.7% pa). A pre-costs alpha of 7% pa is enough to explain his public profile, at the end of his career, of ‘legendary’ investor.
To most observers, point proven. But when we adjust for the standard error of alpha, we find that the ‘credible’ alpha is exactly zero. In other words, eliminating the possibility that the data is not fully representative of the true ‘Anthony Bolton skill level’, we can only say he has made up the cost hurdle. Though not the stuff of legends, it is still a creditable achievement. But it is also not enough to rationalise substituting Fidelity Special Situations for a tracker.
In Fig. 1 we show the three-year rolling unadjusted alpha. The whole-history alpha is shown as a horizontal line with its value of 5.3%. The adjusted alpha, calculated at the 99% confidence level is shown at zero.
Fig. 1 Measured and adjusted alpha
Since we are solving an allocation problem as well as a selection problem, we also calculate an alpha target: the credible alpha needed if we are to substitute a tracker in whatever proportion the active fund’s beta required. For a fund with a beta of 1 the mathematical solution for the active proportion (cutting a long story short) is about 20% when market conditions are ‘normal’, less when the prospective returns are better and more the lower the market return. There is also a standard mathematical solution for adjusting for a higher or lower beta (we have already noted Special Situation’s beta is close to 1).
Its alpha target is 3.0% – the green horizontal in Fig 1. This is the required alpha to substitute just 20% of the tracker exposure. It fails this test because the adjusted alpha of zero is below the alpha target.
Lowering the bar
We have set the bar very high, at 99% confidence, because with a tracker alternative we can afford to be sceptical. Supposing an investor was less sceptical? Lowering the bar enough to let Anthony scrape over it means reducing the required confidence to 65%. That is not beyond reasonable doubt. It’s more like a toss of the coin.
What about consistency?
In the real world you have to live with your choice of active fund: you don’t just lock it away for 10 or 20 years and then look to see how it has done. The alpha you achieve is therefore dependent on your willingness to stay the course. So the active selection problem actually presents itself as one of trying to distinguish, as your own returns cumulate, between a short-lived reversal of fortune for a truly skilful manager and an error you made in estimating the skill. Managers taking big enough bets (or perhaps small bets often enough) to overcome their own cost hurdle have always hitherto shown so much inconsistency in the path of the payoffs through time as to make it impossible to make that critical distinction. Setting aside any theoretical objections, this is the fatal flaw of active management.
You might therefore wonder whether it is worth the effort of trying to convict on the basis of the evidence if you are likely to see new evidence before long that appears to contradict it. On a worse case your conviction at one point that a fund was a ‘star’ may be tested to the extent that the same fund is at another point indistinguishable from a ‘dog’.
Anthony Bolton illustrates this problem all too well. From Fig 1 you can see that his rolling three-year unadjusted alpha has been both significantly positive and deeply negative. This is a fairly typical way to represent the path problem, particularly amongst advisers who know their alpha from their beta and attribute more importance to recent than old information.
Less typically, you might only look at the whole history, because you do not want new data to have any special influence on your judgement, in which case it will only gradually modify your view of the fund. This is shown in Fig 2, starting with enough trailing data to make the exercise realistic – which the industry typically thinks is as little as three years.
Fig. 2 - Whole history Alpha
At the end of the 1980’s Anthony looked like a star yet only a few years later his cumulative alpha was about 2% below his fees – not a dog but certainly no longer a star. His alpha did recover and for about six years would again have looked (on an unadjusted basis) quite compelling but not legendary. However, it is only in the last five years that it has really looked like the stuff of legends.
Supposing you did not know your alpha from your beta? This would certainly be more representative of self-directed investors throughout the period and even of most advisers for most of the period. The path problem then presents itself most simply as a relative price series, cumulating the change in the fund return divided by the change in the All Share Index return.
This is shown in Fig 3. It has more data because we do not have to wait for some early observations. It shows the early outperformance in the late 1980s that was then reversed in the early 1990s. The middle period now presents itself differently: whole-history alpha may still be positive but compared with the index the fund is going nowhere. Considering the expectations excited by the previous outperformance, going nowhere is likely to trigger buyer’s remorse. The rest of the relative return path is similar to cumulative alpha (because we do not have significant beta differences in this case).
Fig. 3. Simple Relative Return
Investors with a different sort of remorse, namely not owning Fidelity Special Situations in the last six years, should not be hard on themselves. Whether you focused on relative return, rolling short-period alpha or cumulative whole-history alpha, it was not ‘predictable’. Those who would have you believe otherwise are either part of the myth machine or else its unwitting victims.
Does anyone pass our tests?
Yes – to the extent of replacing 20% of tracker exposure but not 100%. How many pass? With the bar set high at 99% confidence, just six. Of these, only two have histories longer than five years for the same manager.
Four are income funds. This is an important pointer to successful active management strategies and I will return to it in a later item.