The conventional "strategic" approach to asset allocation does not work, never did work and, most importantly, never will work.
In other words, this approach offers little prospect of being able to reliably converge on the expected return, no matter how long the long run.
The conventional wisdom
The common element of all approaches to strategic asset allocation is that they require the maintenance of a static asset mix whose long-run expected return is sufficiently high to ensure the actuarially required return can be reliably achieved over the long run. The actual portfolio selection process is generally some variation on the following:
1. Identify a reasonably diverse asset universe, typically including U.S. stocks, bonds and bills, and increasingly, but still sparingly, U.S. and foreign real estate, foreign stocks and bonds, venture capital, etc.
2. Forecast the returns and risks of these assets, generally by relying heavily on their long-run historical returns and risks.
3. Impose minimums and maximums on each asset and/or construct a composite benchmark and impose tracking error restrictions in order to ensure palatable efficient portfolios.
4. Run an optimization to generate the efficient set and identify the optimal asset mix, i.e., the portfolio that achieves the required return, or some increment above it, with minimum risk.
5. Implement each asset class with passive or active management vehicles.
6. Rebalance back to the optimal mix periodically to correct for market drift.
A good example of this static mix approach can be found in Trotter, which summarizes a study that was commissioned by the Council on Foundations for the purpose of improving foundation asset management. (See Donald W. Trotter, "Mixing It Up," Foundation News, July/August 1990, and Lester Salamon, "Managing Foundation Assets," Council on Foundations, 1989.)
In this case, input returns are based on long-run observed returns, modified somewhat by various means to improve (it is hoped) their predictive quality. A classic mean-variance optimization is then conducted to identify the efficient asset mixes. The chances for success are then discussed in the context of achieving, for a given efficient portfolio, the predicted dispersion of returns in any given 10-year period. Clearly, the expectation is that this static mix strategy - employing a single asset mix to achieve a specified target return - will be reliable in the long run. Such an expectation seems optimistic. Interestingly, if all of the assumptions in the DeMarche Associates Inc. study are accepted at face value, the chance of achieving the required return after 10 years is only barely greater than 50%, and then only for the most aggressive recommended portfolio. (See DeMarche Associates Inc., "Pay-out Policies and Investment Planning for Foundations," Council on Foundations 1990.)
The conceptual problem:
It can't work
The problems with the conventional wisdom are twofold: both conceptual and empirical.
Conceptually, success requires the accurate estimation of the long-run equilibrium returns for the various asset classes. Implicit in this requirement are two very strong assumptions: that long-run equilibrium returns exist, and that they can be accurately estimated.
There is, however, no basis in theory for asserting the existence of a fundamental long-run equilibrium return. There are certainly long-run equilibrium conditions predicted by theory, and theory certainly does predict the existence of equilibrium returns, but neither of these predictions implies nor requires the existence of long-run equilibrium returns.
For example, "higher return with higher risk" is a long-run equilibrium condition predicted by theory, but this implies nothing about whether those returns are the result of one long-run equilibrium or a series of many short-run equilibria.
It seems more likely, and far less heroic, to assume, given a dynamic market economy, that each asset's equilibrium return is ever changing over time in exactly the same way the market clearing price for bread is ever changing over time. For either the securities market or the supermarket, the clearing price (or equilibrium return) always exists, but it is always changing. If so, then, no matter how accurate the estimation process, an asset's equilibrium return must be re-estimated frequently.
But let's assume for the moment that long-run equilibrium returns do exist. There still is little evidence that the these returns can be estimated accurately. Parameter estimation (forecasting) is a dismal science. Any given estimate, whatever the methodology employed, is likely to be inaccurate (See William S. Gray, "The Anatomy of a Stock Market Forecast," Journal of Portfolio Management, Fall 1989.)
The best one can and should hope for is that any given estimate is at least unbiased, i.e., that it is as likely an overestimate as an underestimate of the true equilibrium value. If so, then, no matter how long the long run, an asset's equilibrium return must be re-estimated frequently.
Given these concerns, strategic asset allocation ensures only the selection of a long-run static mix of assets based on inaccurate estimates of old equilibrium returns. If so, then one would expect the realized returns using this method would bear little or no relation to the expected returns, no matter how long the time horizon. That is, in fact, exactly what is observed.
The empirical problem:
It hasn't worked
This brings us to the empirical problem with strategic asset allocation: There is little evidence to suggest that it has, or could have, worked. The literature is devoid of careful, ex post, out-of-sample, analyses of the reliability of strategic asset allocation. (Remember, success is not defined here in terms of beating the benchmark or the competition. Instead, success is defined as reliably converging on some specified expected return.) An example of what such an analysis might look like is presented below. While this example may seem overly specific and arbitrary, the results should nevertheless be quite troubling to anyone relying on strategic asset allocation to reliably achieve a required return.
The test hypothesis is that the range of annual portfolio returns observed over the prior decade is a reliable indicator of the location of the annual return over the next decade. Specifically, reliability is defined as obtaining a real return, ex post, that falls somewhere within the very broad ex ante 90% confidence interval. For the sake of exposition, we assume a normal portfolio of 60% S&P 500, 30% Treasury bonds, and 10% Treasury bills (See Mr. Salamon's 1989 study). Despite many variations on the above configuration, most portfolio strategists end up doing something similar (See graph on this page and table on page 39).
For example, the expected real return for the 1940s is 7.65% 6 7.21%. This is the observed annual real return in the 1930s. The observed annual real return in the 1940s is 1.74. The Z statistic measures the distance between the expected and observed returns in units of standard error; the larger Z, the less reliable the result.
The graph shows that in four of the five decades, the observed return falls outside the 90% confidence interval, i.e., the return falls more than 1.61 standard errors from the expectation. If reliable, we would expect the failure rate, i.e., the probability of falling outside the 90% confidence interval, to be only 10%. How likely is it that four of five, or 80%, of the observations would fail if the failure rate were really only 10%? Using the binomial distribution, given n=5, x=4 and p=.10, then p(x)=.0004, i.e., there is only one chance in 2500 that the failure rate is only 10%. Therefore, given our definitions above, it is extremely unlikely the prior decade provides a reliable measure of the next decade's return outcome.
It seems unlikely such a poor result could be overcome by simply extending the look-back period and/or the look-forward (or ex post) period, especially given the conceptual problems discussed above. These results should at least cause one to consider that perhaps the problem lies not with the length of the time frame, but rather with the basic long-run orientation. Perhaps we should concentrate on reaching the long-run via a series of short-runs? In other words, perhaps the problem is not with the use of historical data per se, but rather with the way it is used. Rather than using the long-run past to predict the long-run future, why not use the near past to predict the immediate future?
Some would argue that reliability (converging on your expected return) is not the point; the point is to get good returns. But even on this basis, the results are not encouraging for the future. The real return observed in three of the last five decades ('40s, '60s and '70s) would not have achieved the minimum required return of most of today's ERISA plans given the current inflation assumptions and current actuarial discount rates. Foundations, needing at least 5% real return to survive, would have been even worse off. Clearly, whatever your objective, the conventional wisdom offers little prospect for success.
An alternative method
The minimum requirement for an alternative strategy is that it must at least overcome the conceptual problems of strategic asset allocation by satisfying two conditions:
It must not imply nor require the existence of long-run equilibrium returns, and
It must not require accurate forecasts.
As previously discussed, both of these conditions are met by any strategy that frequently re-estimates an asset's equilibrium return. Such a strategy is one that maintains the portfolio's short-run expected return equal to the long-run target return. For example, assume a strategy in which the next year's returns are forecast every year. If so, then each year, with changing one-year forecasts, the portfolio must be rebalanced - assets bought and sold - in order to reset the portfolio's forecast annual return back to the long-run target. The effect of this "myopic portfolio revision" strategy, in the limit, is to create a portfolio whose asset mix is ever-changing but whose short-run expected return (however imperfectly estimated) is held constant and equal to the long-run target return.
The following matrix compares the characteristics of a long-run static mix strategy (strategic asset allocation, or SAA) and the alternative myopic portfolio revision, or MPR, strategy (see table).
Because of the frequent re-estimation required by a short-run orientation, the myopic portfolio revision strategy offers the reassuring prospect of success (i.e., convergence on target) without having to rely on accurate equilibrium return estimates (only unbiased ones), or the existence of long-run equilibria (only short-run equilibria). The conceptual basis for this reliability is explained in detail in my study, "A Short-Run Target Return Strategy for Achieving Long-Run Target Returns: Fifty Years of Evidence," in the Journal of Portfolio Management, Summer 1991.
It can, however, be summarized as follows:
Whatever your parameter estimation technique, as long as each asset's short-run return estimate is unbiased, and therefore is as likely an overestimate as an underestimate of the true, current equilibrium return, then each short-run period's observed portfolio return will as likely fall above as below the portfolio's expected return. If so, then, since each period's expected return is equal to the long-run target return, the cumulative average of each period's observed return will converge on the target over time. Turning to the simple but apt coin flip analogy, while we can't call each flip of the coin reliably, we can rely on the fact that after many flips we will approach 50% "heads."
While a conceptual basis for the efficacy of this alternative method is by far the most important, there is, contrary to the conventional wisdom, at least some supporting empirical evidence as well (Journal of Portfolio Management, Spring 1990, Summer 1991 and Summer 1992).
These studies suggest, using a naive parameter estimation method also explored by Nils H. Hakansson and Robert R. Grauer (Journal of Finance, July 1987) that:
1) There is a close correspondence between expected target returns and observed returns;
2) A wide range of target returns are achievable with relatively modest asset universes; and
3) The methodology exhibits considerable robustness.
This article has briefly discussed some fundamental flaws in the conventional approach to asset allocation, namely its reliance on heroic assumptions for success, and the lack of supporting empirical evidence. An alternative "myopic portfolio revision" strategy is then presented, for which supporting empirical evidence does exist, that offers the prospect of success (i.e., convergence on the expected return) assuming only the existence of short-run equilibrium returns and unbiased (if not accurate) return forecasts.
These assumptions are far less stringent than those required for the success of conventional strategies.
Edward Carr Franks is senior vice president at Trust Co. of the West, Los Angeles.