 July 2003 Time for a Change
We also knew the models needed some time to work before seriously considering any changes. To this end, the initial trial period was set at three years, the historical duration of a typical market cycle. The initial portfolios were put in place on July 1, 2000 so June 30, 2003 brings an end to the test period. So now it's time to look back, review past performance, and make some changes based on what we've learned over the past three years. The archived Work in Progress pages document the most important observations. A Look BackPortfolios 3 and 4 are a quantitative attempt to track and hopefully exceed the performance of the S&P 500. The models are "quantitative" in the sense that they rely completely on regression models designed to select those issues with the greatest potential. No fundamental or subjective analysis is involved.Since the S&P 500 is the benchmark, all potential holdings must be drawn from the index. While it may be simpler to beat the 500 by holding stocks not in the index, it also begs the question as to the appropriateness of the benchmark to the model. In other words, if the model focused on small cap stocks and outperformed the S&P 500 -- a large cap index, is the S&P 500 really the proper benchmark for the model? We'd argue it isn't; a small cap benchmark would be more appropriate. Portfolio 3 was created by running a regression analysis on eleven different fundamental factors for all the stocks in the S&P 500. This model makes no distinction between sectors, industries, or market caps, it simply seeks those stocks that score the highest in its regression formula.
Portfolio 4 takes a slightly different approach by running the same regression analysis separately for each of the ten S&P sectors. The underlying belief here is that the specific factors that lead to outperformance differ from sector to sector. Whereas P3 is a model where one size fits all stocks in the S&P 500, P4 is more tailored to each sector. The models differ in other ways as well. Given it's emphasis on sectors, P4 is sector weighted to match the S&P 500 each time it's reconstituted. P3 has no such requirement and as a result, is often concentrated in only a few, top-performing sectors. P3 is realigned more often, too. It's been revisited on the 15th of every even numbered month whereas P4 is only reconstituted one a year on June 15th. Finally, as a rule, P3 always holds thirty stocks while P4 has as many as necessary to match sector weights with the S&P 500. Although it doesn't have to, P4 has always had more holdings that P3. (A detailed description of the portfolios' creation and characteristics can be found in The Starting Point while a summary is in the Sidebar below.) Looking back at three years of performance, it's easy to see that P4 has cumulatively done better than P3. But that hasn't always been the case. In 2003, P3 is solidly ahead of both P4 and the benchmark, S&P 500. The three-year history of both portfolios and the S&P 500 is shown on the nearby chart. Over the past three years, we've used the Work in Progress page to document observations about what worked, what didn't work, and why. Now we're in a position to put those findings to work. What Hasn't ChangedOf course, everything isn't changing. In fact, some critical aspects will remain unchanged.First and foremost are the algorithms for stocks selection in each model. The models worked fairly well despite adverse market conditions. It's quite possible that performance can be enhanced by changing other, less quantitative factors. By leaving the algorithms unchanged, the models will also maintain their growth bias. This tilt is a result of ten years of data from the decade of the 1990s that was the basis in creating the initial regression formulas. Growth was in favor at that time and this carried over into the models. Value outperformed since 2000, but now growth may once again be poised to take the lead. We really haven't seen a complete market or style cycle since the models have been under review, but we may have that opportunity if growth comes back into favor. Given the growth tilt, we've questioned (Down in the Data Mine, Apples to Apples, and More Than Just Return) using the S&P 500 as the most The number of portfolio holdings isn't changing, either. P3 was initially set up to hold 30 stocks. This number was chosen to allow sufficient diversification while still focusing on the highest rated issues. Sure, it's arbitrary, but what number (short of 500) wouldn't be? At this point there's no compelling reason to change it. There's no fixed number of holdings for P4. Instead, issues were initially selected based on share price. Each sector was populated in such a way as to closely mimic the weightings of the S&P 500. Initially there were 52 holdings. This process was repeated in each subsequent rebalancing. The 2001-2002 version held 51 stocks, and the 2002-2003 had 55. The current version has 52. Given P4's sector-regression approach, this weighting scheme still seems appropriate. Some changes occurred without any intervention. In reviewing P4's sector regression models, we noted (But Does It Make Sense?) the Transport's made very little fundamental sense: [T]he coefficients are off the wall...[T]hose associated with return on invested capital and long-term future earnings growth rate are not only negative, but large. This is clearly counter-intuitive. Add to that the negative coefficient for the PEG ratio, and you probably have a spurious regression relation. Standard and Poor's took care of this problem for us. In July 2001, they restructured their major market indexes, eliminating the Transportation sector. Stocks from this sector were combined with Capital Goods to create today's Industrials sector. Unlike the one for the Transports, the regression equation for the Industrials actually makes sense. Also changing without intervention were P3's "core holdings". Back in March 2002 (Return Engagement) we noted that two years into the test period, P3 still had six charter holdings, Amgen, Medimmune, Mercury Interactive, Network Appliance, PayChex, and Yahoo. We wondered if this signified there was anything special about these stocks. Apparently it wasn't. As of June 2003, five of the six were still in the portfolio, but two had left and returned. With only three charter members -- about 5.5% -- surviving after just three years, the model probably isn't telling us anything special about them. Portfolio 3As you've probably already gathered, P3 has been a high turnover model. Over the three years from July 1, 2000-June 30, 2003, annual turnover has averaged 157%, high by almost any standard.To a certain extent, this is no big surprise. The model focuses on the top 30 stocks of the S&P 500 as determined every two months. It's quite understandable these would change fairly frequently, especially in volatile markets. But how much turnover is too much turnover? More importantly, is the turnover really predictive or is it just turnover for turnover's sake? We addressed this issue in September 2001, concluding [O]ne could argue that P3 reflects recent performing sectors rather than those that will outperform. This is momentum bias. In a trending market, momentum based models can be helpful in stock selection. But in volatile or directionless markets, they will generally trail the overall market since previous winners will not repeat.The market certainly was volatile over the last three years, so maybe that's right. The fact that so many stocks left the model only to return within several months (as documented in Return Engagement) also suggests the effects of short-term momentum -- or perhaps a problem with the holding period. For almost any investor -- and most traders -- two months is an extremely short holding period. If it were lengthened, perhaps fewer stocks would be ousted only to return two or four months later.
On the other hand, the problem might lie in the way stocks are removed from the model. As we noted in March 2002, "The fact that the same stocks come, go, and come back again may simply mean the cutoff to be removed is too low." The model originally held the top 30 ranked stocks, and removed them when their ranking fell below 40. If allowed to fall further before being removed, turnover would be reduced. Indeed, two of P3's "core" holdings that were eventually removed only to return within six months never fell below 62 (Amgen (62) and Mercury Interactive (46)). Turnover in and of itself isn't bad and to a certain extent was expected in this portfolio. Requiring stocks to remain in the "top 40" was arbitrary. The fact that stocks come and go only to come back again suggests the level was set too low. As a result, stocks will now only be removed from P3 when their value falls below 100. In essence, stocks will remain in the portfolio as long as they stay in the top quintile (20%) of the S&P 500. Yes, this is arbitrary, too, but the top 20% still seems quite reasonable. Changing this limit rather than the two-month time frame should still allow the portfolio to benefit from short-term momentum while allowing it to maintain its core holdings. At least that's what we hope to accomplish. Portfolio 4Looking back over the past three years, P4 has kept up admirably with the S&P 500. This is despite the fact that growth was out of favor until early 2003. It suggests few, if any, changes are needed.There is one potential problem we ran across from two different angles. Just as the stock selection process for P3 may be influenced by recent momentum, the same may hold for P4. This is potentially more of a problem for P4 since it's only reconstituted less frequently than P3. In January 2001 we examined P4's selection process for momentum bias. What we found was that P4 did outperform in our backtest, but then the backtest was conducted from 1990-2000, a period when growth was generally in favor. In an upwardly trending growth market, momentum bias was a plus. In March 2003 we approached this same question from a different direction. This time for each one-year period from inception, we compared P4's first six month's performance to that of its second six months. If the stock selection algorithm was influenced by near-term momentum, the portfolio's performance should have been better in the first six months of each period. Indeed it was as we found, "all of P4's performance deficit with the S&P 500 is attributable to its second-half underperformance." Since we only had two and a half year's worth of data, we tested the hypothesis further by looking at quarters. Presumably performance would be best in the first quarter and worst in the last. Again P4 acted exactly as you would anticipate: ...the first three months had a higher correlation with the index than the last. In fact, correlations fell in each quarter although not in a linear fashion.All of this suggests that one-year is too long of a holding period. It's quite likely momentum does play a role in P4's stock selection algorithm so the holding period needs to be shortened to minimize potential longer-term negative effects. Back in March 2003, we concluded: P4 probably needs semi-annual revision. The correlation with the index and relative performance are high enough over the first six months of each year to suggest this is a more appropriate holding period.That's still sounds right, and that's what we'll test. Starting with the portfolio created in June 2003, P4 will be reconstituted every June and December. Watch the WorkAs before, we'll monitor portfolio performance in Work in Progress. We'll continue to test these as well as other aspects of the portfolio's algorithms. Also, as before, the test period will be three years.Weekly performance as well as that of benchmark indexes can be found on the Home Page. Let's see what happens. Search this site! Just enter you key word or words:
Get current quotes or follow your own custom portfolio,
courtesy of E-Line Financials:
|
||||||||||||||||||||||||||||
ACK IN EARLY 2000 WHEN WE created our quant portfolios, we knew they'd be a work in progress. The odds are really against hitting on the "perfect" model on the first attempt.
