| |
 November 2005 Too Long, Too Short, or Just Right? An Evaluation of P4's Holding Period
The S&P 500, the model's benchmark, is also in the red. Cumulatively, P4 has done a little worse (slightly less than 3% below the index) for its first five years, but it's still closely followed to the benchmark. Maybe those market factors really are playing an important role. The demise of the 1990's equity bubble has certainly taken its toll on P4. The model had the misfortune of being launched on July 1, 2000, just as stocks were entering a three-year bear market. P4 spent virtually no time in the black, dropping precipitously with the overall market. But results have been much better since early 2003. This, of course, is the point when earnings and share prices began to turn up. Perhaps most encouraging for P4 is the fact that it not only began to post gains, it also began to regain lost ground versus the index.
As a new and untested model, we wanted to give P4 adequate time to prove itself. The initial plan was to wait three years -- a traditional market cycle -- before considering any changes. When that time was up -- right about the point when results improved -- we made one alteration. So, was that change the reason for the improvement or was it just improving market conditions? Once again, that question's not easily answered. Goldilocks vs. Lady LuckThe idea behind P4 is fairly simple. The goal is to find the stocks offering superior annual performance in each of the ten S&P 500 sectors. To do this, we ran a regression analysis for each sector based on seven fundamental equity characteristics and one-year return. Data was taken from the period January 1990 through December 1999.Since the regression was based on annual returns, a 12-month holding period seemed logical. Not only did it match the return input, it also had the potential to increase return by minimizing turnover and the associated costs. (For a more detailed explanation of P4's construction, please see The Starting Point.)
It all seemed so logical, but there was one thing we didn't anticipate: An abrupt shift in market conditions. The model's sector and security diversification provide stability in a changing environment, but much less so when there is a major shift in underlying conditions. As luck would have it, one of those major shifts occurred just a few months after P4 debuted. The model's official inception was July 1, 2000. Its initial holdings were selected using data through mid-June -- just about the time the great 1990s bull market hit the wall. Over the next twelve months, the S&P 500 fell 16%, with growth stocks -- the previous decade's leaders -- leading the decline. Many investors bailed out to preserve their capital, but P4 didn't have that option. With its growth stock portfolio locked in, it lost 29%. Over the next twelve months, P4 lost an additional 25%. As a small consolation, the difference with the benchmark declined to 6% as the S&P 500 dropped 19%. From July 2002 to June 2003, the model actually outperformed the index (+5% vs. -2%, respectively), but the damage was already done. When it came time for the 3-year review, the question arose: Was P4 hamstrung by too great a holding period? Could it have avoided (or at least diminished) its losses if it had been reoptimized more frequently? To be sure, there's little benefit in turnover for turnover's sake. In fact, data suggests that what turnover there has been in P4 hasn't necessarily been helpful. Nevertheless, P4's first two years were truly trial by fire, a stress test of a completely untested model. Low turnover may indeed be the best course under normal market conditions, but July 2000 - June 2003 wasn't normal market conditions. Limiting P4's flexibility may have unintentionally cut its return. With this in mind, the only change in P4's 3-year review was a new, shorter holding period. While we didn't want the model to be whipsawed by frequent trading, twelve months appeared to be too long. What was needed was one that wasn't too long or too short, but just right. Without a whole lot of science behind the decision, the holding period was reduced to six months. Effective in 2003, P4 was reoptimized each December as well as each June. Results immediately improved. But was this because we had finally hit upon the correct holding period or because stocks had again turned to the upside? Who's entitled to the credit, Goldilocks or Lady Luck? You Can Go Home AgainIt is actually possible to see what would have happened in those first three years with a 6-month rather than 12-month holding period. The model would have still been reoptimized each June so all that's needed are three additional optimizations from December 2000, 2001, and 2002. Results can then be combined and compared with P4's actual results.We went back to mid-December data for each of the three years and ran P4's algorithm. The resulting portfolios were then balanced to the then current S&P 500 sector weightings. With no additional turnover until the actual reoptimization the following June, Baseline easily provided the hypothetical returns.
This exercise was truly a blast from the past. The hypothetical portfolios included some notorious stocks of yesteryear including Global Crossing, WorldCom, and Enron. Of course there were good ghosts, too, including the likes of PeopleSoft, Veritas Software, Nextel, and Concord EFS. With so many changes in even the potential universe of holdings, maybe a shorter holding period is justified. The real test, however, lies in the returns. Chart 1 shows the actual six-month returns of P4 and the S&P 500 as well as what P4 would have returned if optimized in December 2000, 2001, and 2002. The latter are designated "Px". The results aren't completely consistent. From December 15, 2000 through June 15, 2001, all three were down, but P4 and the S&P 500 fared significantly better than the hypothetical portfolio. The exact opposite occurred in the other two periods where Px was the winner. One thing is abundantly clear: A shorter time between re-optimizations wouldn't have helped P4 when stocks began their decline in 2000. Instead, it would have increased the loss by just under 75%. Is there an explanation for this? Perhaps there is. Recall that as a consequence of its 1990s background, P4's regression model favors growth stocks with positive price momentum. By December 15, 2000, stocks had already been declining for almost nine months. Many of P4's constituents had already suffered serious losses. At that point, the re-optimized model replaced them with other that had managed to maintain more of their forward momentum -- just in time for them to sell off, as well. You can see this by comparing the individual stocks that were in P4 with those that would have been in it (ie. those in Px). P4 contained many of the most overvalued stocks, those that were the first to fall when stocks turned down. These included Enron, Yahoo, and WorldCom. Px replaced them with stocks that subsequently fell such as Calpine, JDS Uniphase, and Qwest Communications.
On the other hand, P4's search for growth stocks with positive price momentum would have helped it when stocks stabilized and began to rally. Although it would have fallen from December 14, 2001 through June 14, 2002, it wouldn't have dropped as far as the actual portfolio or the S&P 500. When stocks again began to climb in 2003, Px did even better relative to P4 and the index. Cumulatively, the additional rebalancing would have actually diminished overall returns. This is illustrated on Chart 2 where we've combined P4's actual returns from July-December with the hypothetical December-June results. This again is illustrated by "Px". By the end of the five years, Px is considerably below both P4 and the benchmark index. Although it's hard to see from Chart 2, the gap was actually closing as time went by. In essence, Px was never able to make up for ground lost in the December 2000 - June 2001 period. In fact, if the cumulative measurement period had run from June 14, 2001 - June 15, 2005, Px would have actually been up 21%, putting it 12% ahead of P4 and 18% in front of the index. What Goes Up
There are actually three things to take away from this:
|
|||||||||||||||||||||||||||||||||||||
|
|||||||||||
| Homepage | Return to Top |
UANTITATIVE PORTFOLIO 4 HAS been around for over five years now. Unfortunately, most of that time has been spent underwater. Is this because the model is inherently flawed or is it due to other external factors? It's hard to tell.
