It is about this time of the year that I start thinking about the strategic asset allocation (SAA) of our Balanced Fund - actually, I think about it all the time, but I start formally addressing it in the middle of each year. My philosophy is that the SAA (or the benchmark mix, as it is often called) should be properly reviewed every year. That does not necessarily mean that it gets changed every year, but it needs to be formally reviewed.
Of course, it should also be reviewed in the light of changing environments as well - the deteriorating environment in 2008 was a good example. Fortunately, last year the realization of a worsening environment coincided with the timing of our formal review, so I was able to cover that off as well.
Anyone who reviews the asset allocation more frequently than annually (with the exception of extraordinary changes in the market environment) is bound to be engaging in what is called tactical asset allocation. TAA is basically guessing which asset classes are going to outperform others in the short term. Given the difficulty of calculating market turning points (how many got the peak of equity markets right in 2007, how many predicted the global property volatility in 2008 and when will the equity bottom occur, or has it done so already?), I have yet to meet anyone who can do this successfully on a consistent basis. Sure there are some who claim to be able to do it, but it is questionable whether any success they have is due to luck rather than any skill. No-one seems to be able to consistently do it successfully over a multi-year period.
SAAs are supposed to be set for the long term. I agree with that philosophy, so why do I bother reviewing it every year? Well, the long term is a somewhat indeterminate time frame. I believe that the key is to get the general framework right and then any adjustments made at the annual review are relatively small. Another reason for reviewing it is that asset classes tend to trend and mean revert over various multi-year cycles and there are opportunities to take longer-term advantage of these trend directions. Only reviewing an asset allocation once every three years, say, leaves a lot of time between reviews to miss out on trend changes.
Some call this annual adjustment process dynamic asset allocation, rather than strategic asset allocation, and accuse it of being sort of midway between TAA and SAA. There is some truth to this, but I think this is overcomplicating the whole process. Basically, you have a long term structural position that would barely alter over, say, 10 years and then each year tinker around the edges to reflect the prospects for each asset class over the next three to five years.
Just how important is SAA to a fund's overall performance? Well, it is often quoted that it is responsible for 90% of a fund's return. This is actually a misquote from the seminal 1986 paper on portfolio attribution by Brinson, Hood & Beebower, where they stated that across many US pension funds, the SAA was, on average, responsible for 90% of the variation in the fund (that is, the volatility). There was another paper written in 2000 by Roger Ibbotson and Paul Kaplan titled "Does Asset Allocation Policy Explain 40, 90, or 100 Percent of Performance?" The conclusion was that it determined 40% of the variability between funds, 90% of a fund's variability over time and slightly more than 100% of a fund's total return.
In the case of the Tyndall Balanced Fund, the gross return since 1 October 2002 (the inception date of the fund in the Tyndall Super Fund range) has returned 6.95% p.a. to 31 March 2009, whereas the SAA (if invested passively) would have returned 6.54% p.a. Hence, in this instance, it has provided 94% of the fund's return. These numbers show just how important it is to get the SAA right.
Which leads nicely to the question: How do you get the SAA right? How does one go about achieving the optimal strategic asset allocation? This is both far harder and far easier than you think. Let me explain.
There are lots of wonderful theories about how to allocate assets. Many are based on what is called mean-variance portfolio optimisation. This basically involves supplying each asset class with an expected average return, an average variance (or more commonly, standard deviation) and a correlation with other asset classes. Then, an overall portfolio is "optimised", usually via a computer model or spreadsheet-based program that runs a large number of allocations to determine which overall allocation has the highest return for a given level of risk. Then, for each level of risk from lowest possible (usually 100% cash) to highest possible (usually 100% aggressive equities or the like), you will get a portfolio for which no other provides a higher level of return at that risk level. Sounds straight forward? No, and there are many caveats here.
One is how do you go about determining the expected return, risk and correlation numbers for each asset class? Most of the ways of doing this use some element of past history as a guide to the future, when we all know that it is not a terrific predictor. The theory is the longer the history used, the more "information" that is contained in it and hence the better guide it will be. The problem is that some asset classes have far shorter histories than others. Also, what really does 100 years of history tell us about what the next three to five years will bring?
Correlations and standard deviations are far more likely to have better predictions from past results than do returns, but even these are not stable and change over time. Correlations in particular, which are seen as relatively stable long term, tend to increase alarmingly towards one during times of market turbulence.
The big problem with this whole mean variance approach is that is assumes asset class returns follow a nice normal (or bell-shaped curve) distribution. This has been repeatedly been proved to be false. In particular, the tails of the actual distribution are fatter than those of a normal distribution. What this means is that results end up being far more extreme than you would expect - witness global equity returns over the past twelve years and tell me how many of these were predictable, or "normal".
Believing that past performance is no guide to the future then, some use a process of producing a purely forward-looking return estimate, by taking the risk-free rate (usually the OCR or a bond/cash rate) and adding various premia to it (such as an equity risk premium). This is probably a more sensible approach, although the problem of how to determine the risk premia remains - again, some fall back into past history for that and you still need to calculate risk and correlation estimates.
Other approaches to the assumptions include using more moments than just the return and variance of a distribution, or focusing on turbulence distributions - what happens to an asset class during market shocks. The New Zealand Super Fund released its SAA paper a few years ago that looked at a global wealth portfolio (basically all of the assets in the world) and interpolated from that a set of return and risk assumptions that were internally consistent. It was relatively complicated and very impressive, although to be fair it didn't predict the financial crisis any better than any other method.
Which brings me to the easy way of allocating assets. In the recent biography of the world's greatest allocator, Warren Buffett, he outlined his approach to determining how to allocate his capital. He does not use a computer or spreadsheet, but simply looks at the prospects for each area of allocation, using a fair bit of experience of how they are likely to perform in the environment and then, using his words, takes a "whack" at what the number will be.
And he seems to have done better than anyone else for a long period of time - hence my skepticism of using complicated mathematical models to determine optimal allocations. No matter how sophisticated your model, there is no guarantee that you can do any better than with any other method. The best way may probably be to use a bit of simple forecasting based on intelligent estimation and experience and an amount of quantitative analysis to provide some support to your rationale.
However, whether you use very simple estimation or the most complicated mathematical models, there are still some basic criteria you need when forming a SAA. One is diversification, of course, because it becomes hard to justify a "balanced" fund when it is heavily slanted towards a particular asset class or just a couple of classes, no matter how strong your conviction that those classes will outperform. "Balanced" funds are invested in to provide that diversification between asset classes. An asset class should not be added, though, unless it can offer some level of gain to the overall portfolio - this could be either a return enhancement or a risk reduction (or both). In this respect, I am very comfortable incorporating non-standard asset classes (those other than bonds, equities and cash) into a balanced fund.
The main criterion is to achieve your objectives. This basically means the desired level of risk or return that the portfolio is required to achieve. This could also relate to a certain level of income or capital gains, or a hedge against some other parameter, such as inflation. If it is an open-ended balanced fund, then performance relative to peers becomes an issue, meaning that you don't want to stray too far from peers' SAAs (although you want to be different enough to outperform them). Loss avoidance or mitigation is also desired, as no one likes losing their capital.
Finally, fees and taxes should always be taken into account, as it is no use coming up with a great SAA from a gross perspective, but which becomes less efficient than another one after all the deductions are taken into account. SAAs are the most important part of any balanced fund, or indeed asset allocation exercise. Whatever method is used to determine them, there needs to be a good dose of common sense in there.
Download the May 2009 Tyndal Comment.