| |
 January 2008 A Long Strange Trip A Look Back at Quantitative Portfolio 4's First 7½ Years
To see why, just think back over the past ten years or so. Ask yourself, "When would have been the absolute worst time to invest?" Odds are, you'll probably pick somewhere in mid-2000. As luck would have it, P4 (and P3 the other quantitative large cap model) were launched on July 1, 2000.
P4 was designed to track and (hopefully) exceed the return of the S&P 500. Again, as luck would have it, the benchmark index peaked in the final week of August 2000 and went into a steady decline until finally hitting bottom a little over two years later, in the final week of September 2002. Guess what P4 did? It tracked and exceeded the index, but it did so on the downside. That's why it's nice to see the ensuing five-year bull market has almost brought the two full circle. Prior to the November sell-off, we had expected P4 to be back in the black by the end of the year. We also thought it would finally be ahead of the S&P 500 then, too. But even with this minor setback, we still anticipate that will happen in the first quarter of 2008. Why the optimism? It's actually a combination of things, including growth's return to favor, a slight tweak back in 2003, and perhaps most importantly, a truly viable model design in the first place.
The Background
We made that same assumption in creating P4, but there were a few twists. First, we didn't assume that the same characteristics would be indicative for all sectors. For example, price-to-book value may be important in the financial sector, but much less so for technology. So rather than look for one set of factors that would work in all sectors (basically the premise of Portfolio 3), we considered each sector separately. We started with eight factors that research had suggested were predictive of future performance:
We then ran a multiple-regression analysis for each sector of the S&P 500, using these factors and monthly returns covering the past ten years. Factors that weren't statistically significant were eliminated resulting in each sector getting its own unique regression equation. These equations could then be used to evaluate stocks for inclusion in the portfolio. (For details of this process and factors, see The Starting Point.) Secondly, we assumed that stocks starting with these desired factors would eventually see them decay over time. As a result, the model would occasionally need to be reformulated and rebalanced. Our recent analysis bears this out. In addition, the goal was to seek return from stock selection, not market timing or sector weighting. As you can clearly see from Chart 2, the S&P's sector weightings have changed dramatically over the past seven and a half years. The most pronounced differences can be found in the tech and telecommunications sectors, both of which are now roughly half of their earlier weights. To accommodate this, P4 is always balanced back to the index sector weights whenever it is reoptimized. Thirdly, we had to decide on a frequency of rebalancing. To avoid any potential for market timing, it needed to be on a regular basis. Given that the statistics used in the regression analysis were long-term in nature, P4 was originally only reformulated on an annual basis, in June of each year. 'Originally' is the key word here. By March 2003 we detected a fall-off in performance in the last six months before reoptimization. This suggested the one-year holding period was too long in light of changing market conditions. When P4 got its first three-year review in July 2003, the holding period was shortened to six months with reoptimizations now occurring in June and December.
To date, that's the only change in the process. Turnover has increased slightly from 49% to 75% (68% in 2007), but it's still lower than the average large cap mutual fund's. Performance has arguably picked up as well. With the exception of 2006 when P4 trailed the S&P 500 (7.6% vs. 13.6%, respectively), the model has bested the index in every other calendar year since the holding period was shortened. This, of course, is why it has been able to make some major inroads into catching the benchmark.
More Than Just Beta P4's beta has always been greater than that of the S&P 500. As you probably know, beta is a measure of the model's sensitivity to market movements. If it simply moved in lockstep with the index, its beta would be around 1.0 meaning its expected return is about 1x that of the market. P4's beta usually hovers between 1.15 - 1.25 giving it an expected return of 115% - 125%x that of the index. Beta cuts both ways: with a beta greater than that of the market, when the S&P goes up, P4 will be expected to rise about 20% more. That could explain why it's been picking up ground during the bull market. But when the market goes down (as it did from 2000 - 2002) P4 should be expected to fall roughly 20% more. That could explain how it got in the hole in the first place. But we would suggest there's more at work here. Consider P3 with an even higher beta averaging between 1.30 - 1.40. Based on that, one would have expected it to have originally fallen harder than P4, but to recover quicker. The first part of that prediction came true: At their lowest points, P3 and P4 were down 75.8% and 55.9%, respectively. Yet through the end of December 2007, P3 is still down more than 30% from inception while P4 is within 3% of positive territory. Without a doubt, P4 has snapped back more strongly than P3, suggesting there's something more than beta at work here. With every reformulation being balanced to the S&P 500's current sector weights, differences don't originate there. Since the model can only hold stocks taken from the index, it's not making up the difference by investing elsewhere like mutual fund managers often do to game the benchmark. No, the differences have to come from the stocks themselves. Growth's recent return to favor may have something to do with it. Again as you probably already know, the S&P 500 is often broken down between "growth" and "value" stocks. Growth stocks are those that offer above average earnings growth potential while value stocks are those that are cheap relative to their intrinsic value or value relative to other stocks in their sector or market. Historically they've taken turns leading the market. P4's regression model was based on returns from the 1990s when growth stocks dominated. As a result, it tends to favor growth stocks over value stocks. In another bit of bad luck, growth stocks fell from favor when the tech bubble burst -- just about the time P4 was released out of sample. That again could help explain why P4 fell so hard (and P3 fell even harder) right out of the gate. It could also help explain P3's stellar performance this year, even relative to P4.
It's quite likely the growth factor has something to do with it, but, as you can see from Chart 1, P4 started to regain ground on the index in 2004, back when value was still in vogue. It still seems that something else must be at work here.
Monthly Data
The correlation coefficient is .920, relatively high for a non-index model. In this regard, P4 has certainly lived up to its goal of tracking the performance of the benchmark. Because the two plots use different scales, it's not immediately apparent but the index returns are much more concentrated around the midpoint. On the other hand, P4 has a higher number of outliers at both extremes. You might expect this from its higher beta. As you would probably guess, P4's monthly standard deviation is higher than that of the index (0.0540 vs. 0.0388, respectively). P4's highest and lowest monthly returns are 10.9% and -16.7%, while the S&P's are 8.6% and -11.0%, respectively. Again, this is consistent with the beta differences. Both series are "negatively skewed" meaning more data points fall in the left-hand tail of their respective distributions. This is slightly more pronounced for P4 (-0.7217 vs. -0.4627 for the index). The average monthly return for the index is 0.10% while it's only 0.02% for P4. However this figure is somewhat distorted by the outliers. The medians are higher as well as reversed: 0.71% and 0.89%, respectively. This is what's to be expected given the relative degrees of negative skewness.
So What Is It? We would suggest there's one other thing working in P4's favor: momentum. As we noted in March of 2003, P4's performance tended to fall off in the latter half of its one-year holding period -- that's why we reduced it to six months. Essentially what was originally happening with the longer holding periods was that the model's stock selection benefits faded over time, leading to a loss of momentum. When the holding period was shortened, the model benefited. Although there really hasn't been a prolonged down market (knock on wood) since the change, we suspect the shorter holding period would help there, too. Part of the reason P4 fell so hard in 2000-2001 was the fact that it was locked into a high percentage of tech stocks which until that point had been market leaders. When they started to fall and investors ran to sell, the model was left with them for the full twelve months. The shorter holding period will help limit the damage the next time a bear market comes along. This isn't just a specific stock issue, either. As you can see from Chart 2, the index's sector weightings have changed dramatically over time. With a longer holding period, P4 was initially stuck with a larger allocation to underperforming sectors than the index itself -- especially in the latter months. Again the shorter holding period will help avoid this. When we last took a look at P4's holding period ( Not only that, the holding period may be correct, too. Evidence So it's quite possible that a six-month holding period is just what's needed when it comes to capturing short-term momentum without being whipsawed by volatile market conditions. This additional momentum may very well be the mystery factor powering P4 back to the index. Regardless, the best thing to take away from all this is the fact that P4 is truly living up to its objective. The goal was for it to track the index and incrementally beat it. It wasn't able to do that initially, but now it is. Additional time out of sample will help determine if it's simply the bull market or the shortened holding period leading to these results. Given P3's inability to gain on the index under the same market conditions, the scales may be tipping toward P4's six-month holding period. Search this site! Just enter you key word or words: Get current quotes or follow your own custom portfolio,
courtesy of E-Line Financials:
|
||||||||||||||||||||||||||||||||||||
T'S TAKEN OVER SEVEN years, but quantitative Portfolio 4 is finally closing in not only on the S&P 500, but positive territory as well. In most instances you probably wouldn't think this was a reason for celebration, but in this case it actually is.
