Observer: Diversification may be good, but returns are better

Roger Gibson is an author and speaker on the issues of asset allocation and presented a session at this year’sFinancial Planning Association(FPA) national convention.

His arguments for diversification are convincing and no doubt his numbers are accurate, but I would like to address some of the premises raised.

Gibson presented his case primarily through a hypothetical portfolio covering equal weightings across four asset categories — US stocks, non-US stocks, real estate securities and commodities. He uses this simple model more as an educational tool rather than a practical description of what he implements for clients.

Using data from 1972-2001, he showed how portfolios combining these assets can result not only in lower risk than each of the individual asset classes (as measured by standard deviation) but can also have higher annual compound returns than any of them.

He shows that the equally weighted portfolio of all four asset classes returned 12.98 per cent per annum over the period, above the 12.49 per cent per annum achieved by the best performing of those asset classes — real estate securities.

This is a very powerful conclusion, although one has to consider the circumstances in which this result is achieved and whether this should be expected in other circumstances and time periods, particularly looking forward.

The implication of Gibson’s result is that diversification can not only lower risk but actually achieve extra return. Gibson explains this result as the consequence of compounding the less volatile diversified portfolio returns over time.

However, this view neglects the role of the implicit annual rebalancing between asset classes. Furthermore, we need to question whether the circumstances that allowed this result are likely in the future.

Looking more closely at Gibson’s figures, the four asset diversified portfolio return figures assume an equal weighting (25 per cent) to each asset class’ annual return. This calculation determines the portfolio return for that year. This series of returns then determine the portfolio’s compounded annual return over the full period. This approach is implicitly assuming annual rebalancing, i.e. at the beginning of each year 25 per cent of the value of the portfolio is assumed to be exposed to each asset class.

This can only be achieved in practice by selling better performing assets and buying worse performing assets. This is not a minor matter. At the end of the year 2000 for example, approximately 11 per cent of the portfolio’s combined exposure to US and non-US stocks would need to be bought with these purchases financed by the sale of 11 per cent of the portfolio’s exposure to real estate securities and commodities.

Rebalancing in this case involves buying low (assets that performed badly in the previous year) and selling high (assets that did well in the previous year) a process that should add something to return in a mean reverting world. This type of re-balancing can work particularly well in periods where you have asset classes that tend to alternate from good to bad years, where inter-year correlation between at least some of the asset classes is low but where the long-term returns from all asset classes are broadly similar.

I believe that the ability of the equally weighted portfolio to achieve a better annual compound return than each individual asset class derives largely from this implicit rebalancing, rather than the effect of compounding less volatile portfolio returns.

If an investor were to invest in the same equally weighted portfolio initially but never rebalance, then the portfolio would still be less volatile than the four individual asset classes. Yet it could not possibly have a higher return than all the asset classes, simply because the portfolio’s compounded annual return must be equal to the sum of the various compounded returns weighted by their original 25 per cent exposure.

While Gibson is no advocate of market timing, the equal weighted portfolio is adopting a simple contrarian timing strategy. If such a strategy, which takes a basic calendar year approach, can add value, isn’t it possible that a more disciplined approach involving price and value limits could add more?

Furthermore, it may have been that the period 1972-2001 provided ideal circumstances for this result to occur. For example, the divergence in compound annual return between the four asset classes was just under two per cent per annum, with all asset classes producing returns between 10.55 per cent and 12.49 per cent.

Despite this there was considerable inter-year volatility and low correlation between some assets. Commodities, the worst performing asset class, were the least correlated and most volatile, helping them to add value through rebalancing.

What happens if the situation in the next 10, 20 or 30 years is very different? What if one or two of these assets significantly underperforms the others? In a range of scenarios, the diversified portfolio could significantly underperform the top performing asset classes, despite rebalancing.

This does not argue against diversification (risk will still be significantly lower in the diversified portfolio) but it does suggest that Gibson’s “free lunch” portfolio with higher returns than the individual asset classes is more likely to be the exception rather than the rule. Gibson had the benefit of setting up a hypothetical strategy when much of the data was already in. An investor setting up a portfolio today does not have this luxury.

Gibson shows this effect over a limited period by breaking out the three years to 2001. In that period, the diversified portfolio returned less than half the return of the best performing asset class (6.17 per cent per annum versus 12.83 per cent per annum for commodities) and almost half as bad as the second best performing asset (real estate securities 11.15 per cent per annum).

Of course, the portfolio return was still above the worst performing asset classes (US stocks minus 1.05 per cent per annum and non-US stocks minus 4.9 per cent per annum) and indeed above a portfolio which was not rebalanced for the three years. The point is that this sort of result could occur in the future even over longer periods if one or more asset classes significantly underperform, particularly if such assets are highly correlated.

Although Gibson stressed that the portfolio is hypothetical, one cannot totally ignore issues with the practical implementation of such a strategy. On Gibson’s own numbers, the year 2000 would have resulted in turnover of almost 22 per cent for the portfolio. Clearly such turnover could add significant costs (and reduced tax effectiveness) to a portfolio. Perhaps a more disciplined (but less frequent) rebalancing strategy might be considered.

Then there is the allocation of 25 per cent of the portfolio to commodities. Do commodities really have a place in portfolios? (I believe they can have a significant role at times). But then how is this achieved? Can any ‘set’ asset allocation approach (even with regular rebalancing) adequately cope with the changing risk, correlation and long-term return dynamics of various asset classes that will occur in the future?

Diversification across asset classes is one of the most discussed and accepted principles of investment. Roger Gibson has done an excellent job of encouraging investors and advisers to think about and implement asset allocation diversification.

However, diversification should not be pursued for its own sake — diversification does not work simply because assets move in different return patterns. It still needs most of those assets to produce reasonable returns over time. In this regard it is future not past returns that matter.

Furthermore if one accepts that rebalancing across assets reduces risk and potentially increases returns, then why shouldn’t other contrarian rebalancing strategies that incorporate price and value parameters, rather than a simplistic time element, also be considered.

At the very least, we need to approach historical returns across asset classes and portfolios with some skepticism and to think about how those results might look in different circumstances and time periods.

While the future is always hazy, we have no choice but to look forward. Knowing where you have been is useful but driving using the rear vision mirror is highly dangerous.