| IN THE ARCHIVES: May 2010 March 2010 January 2010 November 2008 September 2009 July 2009 More... |
 July 2010 The Value of Market Timing Has Trading Really Helped P5?
That's fortunate for quantitative Portfolio 5 because it has no opportunity to exercise security selection. It has an extremely limited universe of potential holdings: The nine iShare ETFs representing each of the nine style and capitalizations of the Morningstar stylebox. Since each of the ETFs is comprised of all holdings in the respective S&P index, there is absolutely no opportunity for security selection.
The model's allocation does -- or at least can change every three months when it is optimized. There is no requirement that any of the nine stylebox categories ever be represented at and at the extreme, it's even conceivable the portfolio could at some point be concentrated in only one. While that's never happened, the mix does periodically change. It's the ability to change that mix -- both into and out of specific categories -- that enables P5 to employ a certain degree of market timing. Although limited by the rather small universe of potential holdings, the model is able to time wide ranges of the broad domestic equity market. As you can see from Chart 1, P5 has always managed to stay ahead of its benchmark index, the S&P Super Composite 1500. Asset allocation clearly plays a role, but how much is the result of market timing?
A Constant in Change
With thirty-four data points, it's possible to get an idea of the average allocation. In fact, with the heavy reliance on mid caps, there's a certain consistency to the mix. Arguably then, the benefits of market timing are in the differences from this average. It can be measured by comparing the hypothetical return of the average allocation against the actual optimized results. But how should the average allocation be calculated? The easiest solution is to simply sum the allocation across each of the thirty-four allocations for each category and then divide by thirty-four. But remember, there were eight quarters where there were no changes from the prior. This means some allocations were in effect considerably longer than others. Simply relying on the arithmetic average won't really capture this difference. As a result, we calculated the average using both methods. Chart 2 has the results. Surprisingly, there were no major differences. Because of this, performance was quite similar between the two averages (Chart 3). We decided to focus on the time weighted version.
Why Did It Happen?
This is borne out on Charts 4 and 5. Chart 4 shows the risk (measured by standard deviation) and return for both the model and the averages. As you would expect from the similarity in their composition, the two averages plot on top of one another. But P5 comes up short on both measures: Higher risk and lower return. This suggests that although its differences with the average allocations may not have been great, they had an effect -- albeit a negative one. The correlation matrix in Chart 5 offers some clues. Correlation measures how closely two series track one another. The scale runs from 1 to -1 with the former representing perfect correlation (they move identically) and the latter signifying perfect negative correlation (they move in complete inverse lockstep).
Again as you would expect, the correlations for all series are relatively high. All sectors of the market tend to move in the same direction but at different rates. Because both P5 and the average are composed of these elements, their results are similarly correlated. But focus on the last two columns of the table. Aside from its almost perfect correlation with the average, P5's highest correlation was with large cap core and secondly with large cap growth. On the other hand, the average is most closely correlated with mid cap core and mid cap growth. Throughout the period, mid caps handily outperformed large caps.
A Possible Explanation As we now know in hindsight, mid caps outperformed throughout the entire measurement period. Although P5 had a high average weighting there, it also, at times moved a significant allocation to large caps. This first occurred in the five quarters starting in April 2004, then again when the economy faltered in 2007 through the present. While this ultimately dulled overall performance relative to the average allocation, P5 was acting in accordance with its design. Unlike this analysis which has the benefit of perfect 20/20 vision, P5 was designed to act prospectively. Its algorithm is not strictly based on historical performance but rather employs a Black-Letterman approach incorporating estimates and probabilities of future market performance. Again, looking back with complete hindsight, the moves in 2004 and 2007 are reasonable. When equities began their recovery from the deep two-year bear market at the beginning of this decade, small and mid cap stocks led the charge. Large caps that had been so overvalued in the late 1990s continued to lag. By early 2004, the situation was reversed: Mid and small caps were fully (if not over) valued and large caps offered better potential. P5 moved to take advantage of this by putting over 90% of the portfolio in large caps. When the credit crisis erupted in 2007, mid and small cap stocks were more exposed. Not only did the smaller capitalizations face greater danger from tighter credit, they derived less of their revenues from abroad. P5 doesn't have the luxury of moving cash to the sidelines, it must always be fully invested in the domestic styleboxes. In this case, large caps were the safer bet.
To its credit, P5 didn't abandon mid caps in 2007, it simply reduced exposure to 45-50%. It did trim back small caps and bumped large caps to roughly 25%. In recent quarters, it's shifted a greater allocation to small caps, cutting the large cap exposure in half. Again, a reasonable move. (To see the current allocation of P5, click here).
The Future vs. The Past The point is this: P5 was designed to seek benchmark beating performance in the future. That's the purpose of trading and reoptimizing every quarter. As Chart 1 shows, it succeeded. Chart 3 shows that without trading, it could have done even better had we known its average allocation for the period at the beginning of the period, but not only is that impossible, it's not the goal. To return to the original question, no, P5's market timing does not appear to have enhanced performance over a buy-and-hold approach using what would turn out to be its average allocation. But don't take this conclusion further to assume the model would be enhanced by constantly rebalancing back to this average. It would have worked from January 2002 to May 2010, but there's little if any reason to believe it will work in the coming years. If anything, P5's forward-looking algorithm should be viewed favorably for focusing its quarterly allocations around what turned out to be a successful average. It was, after all, the product of those quarterly allocations. Search this site! Just enter you key word or words: Get current quotes or follow your own custom portfolio,
courtesy of E-Line Financials:
|
|||||||||||||||||||||||||||||||
HERE ARE THREE WAYS TO POTENTIALLY add value to a portfolio: Security selection, asset allocation, and market timing. Asset managers tend to tout their stock-picking ability, yet evidence suggests the latter two factors are really the key determinants to long-term performance.
