| |
 July 2006 Status Quo The Second Triennial Review of Portfolios 3 and 4
When they were initially created, we wanted them to experience a market cycle – typically about three years -- before considering any changes. At the end of the first three years, we made one change to each (more about that in a moment), but they’ve remained unaltered since then.
In the intervening months, we’ve examined different aspects of the models themselves as well as their resulting portfolios. In the process, we came upon several potential changes to be considered at the next triennial review. It’s now time for that review, but there are no changes in the offing. The models will remain unchanged for at least three more years. This is however, a good opportunity to look back at the over the past three years’ analysis to see what’s been working and what hasn’t.
In the Beginning The models are based on data from the ten years prior to their launch, the decade of the 1990s. As you’ll probably recall, that period was marked by a long-running equity bull market in which growth outperformed value. Large cap stocks were market leaders throughout the decade. This had a major impact on P3 and P4. The models were created by running multiple regression analyses on seven fundamental characteristics and annual returns. P3’s algorithm allows us to find what are expected to be the top performing stocks in the S&P 500 regardless of sector or index weighting. The portfolio is comprised of the top 30. P4 also looks for the top potential performers, but within each sector. Instead of having one regression for all stocks like P3, it has a different one for each sector. This is based on the assumption that each fundamental factor has a different impact on various sectors. The goal is to select what are expected to be the top performing stocks in each sector. Initially, once selected, stocks remained in P3 until their ranking fell below 40 in subsequent bimonthly optimizations. Whenever a stock fell below the threshold, the next highest rated stock not already in the portfolio was added. P4 was originally only rebalanced on an annual basis. When the optimization was rerun, the top stocks in each sector were included in the portfolio. Prior holdings were removed whenever they were no longer top scorers.
Turnover Turnaround P4’s optimization frequency was changed form 12 months to 6 months. This was an effort to allow the model to remain more in tune with rapidly changing market conditions. In today’s environment, twelve months is a long time to stay with the same holdings, especially in a trading portfolio. Obviously both changes were anticipated to impact turnover, and in opposite ways. With wider latitude to hold onto stocks that may have slipped from the highest rankings, P3’s turnover was expected to decline. With more frequent rebalancings, P4’s was expected to climb.
In this instance, the estimates were right on target. While noticeable, the differences weren't extreme, but in line with what we had anticipated. We calculated turnover using the same method as mutual funds. Accordingly, annual turnover is found by dividing the lesser of buys or sells by the number of holdings at the beginning of the year. If anything, this approach underestimates turnover because it's based on the lesser of buys or sells. Nevertheless, this is a useful approach since it allowed us to compare results to those reported by mutual funds. From July 1, 2000 through June 30, 2003, P3's turnover was 138%. After the change effective July 2003, turnover fell to 36% from July 1, 2003 through June 30, 2006. Three of the 18 rebalancings in the past three years ended with no changes to the portfolio or turnover of 0%. That never happened in the first three years. P4 went in the other direction. In the first 3-year period, turnover was 49%, but in the second three years it jumped to 89%. Even so, that's still a reasonable number for an totally quantitative portfolio and still below the average equity mutual fund's 100%+. What about performance? Here the results aren't as clear. On the face of it, it would appear that the changes helped. In the 2000-2003 period, P3 was down a cumulative 66.5% while P4 dropped 44.2%. Both were positive in the 2003-2006 period with P3 climbing 38.5% and P4 37.9%. That's a nice turnaround. But it may have more to do with what was going on with the market than what was changed in the models. In the first 3-year period, the S&P 500 was down 33.0% while it was up 27.3% in the past three years. Its drop was less than that of the models in the first period but its rise was also less in the second period. That's exactly what you'd expect when comparing it to high-beta models like P3 and P4: When the market's up, they beat the index but when the market's down, they fare worse than the benchmark. It's quite likely that the improvement in performance was just a coincidence, with the changes happening right as the market was turning. Just as you shouldn't blame the models themselves for the dismal performance from 2000-2002, you shouldn't credit them either for the market-beating numbers posted from 2003-2006. This was more luck than design.
As Expected Looking at the historical characteristics of the models' portfolios and the effects of their algorithms, we found that P3 tended to rank the 500 stocks of the S&P in a normal distribution yet all 30 of its holdings tended to come from the positive tail (see Chart 3) essentially making it a portfolio of outliers. Even so, stocks that were frequently highly rated weren't necessarily more likely to provide additional return if held for the long-term. In fact, there was no marked improvement in long-term performance of stocks with long tenures versus those with the shortest.
What did seem to make a difference was the sector weightings of the various 2-month incarnations of P3. We found that 75% of P3's returns were attributable to sector weightings. This was a major discovery given that P3 simply focuses on the 30 highest rated stocks of the S&P 500 with no attempt at sector weighting. Here again, the superior return over the past three years may be more luck than design. Neither sector weighting nor tenure appeared to play much of a role for P4. Portfolios created with stocks most frequently appearing in the model produced no noticeable improvement in performance when held for the long-term. Altering P4's sector weightings also failed to improve results. In a related matter, the requirement that when rebalanced P4 must match the S&P 500's sector weightings could actually harm returns. By forcing it to hold stocks in each of the 10 S&P sectors as well as requiring the sector weightings match those of the index, the model is unable to focus on those sectors with the highest price momentum at the expense of the laggards. All of its excess return must come from stock-picking, an extremely difficult burden. Although we didn't consider any specific alternatives to P4's algorithm, we did examine potential changes to P3's. The original regression for the model was based on fundamental characteristics of superior performing stocks from the 1990s. Those tended to be growth issues, particularly in the Technology sector. Clearly a lot has changed since then, so perhaps now top performers had considerably different fundamental characteristics. We re-ran the regressions based on data from 2000-2005 only to find that the resulting portfolios showed very little improvement over the historical P3 returns. Truth be told, none of these findings provided any viable reason to change either model. That's why no changes are being made this year. Change for change's sake is never a solid approach. Without compelling reasons or alternatives, maintaining the status quo is the best course of action for the next 3 years.
A Possibility Based on the studies of the past three years, tenure, sector weighting, and diversification are not the issue -- at least not directly. Both already closely track the benchmark index, perhaps even too closely. In light of their above-average betas, it's not a good thing to match the index too closely when the benchmark is declining as it did in the 2000-2003 period. Which brings up the biggest weakness of both models: The lack of downside protection. It's great when they outperform the index in up markets, but its more than equally dreadful when they underperform by the same margin in down markets. Investment analysts call this "upside capture" and "downside capture", respectively. All else being equal, high performing portfolios have a high degree of upside capture but a significantly smaller downside capture. If anything, P3 and P4 actually have a higher downside than upside capture. To see what this implies, just look back at Chart 2 where the losses in the down market are greater than the gains in the up market. This is exactly the opposite of the way it should be. Portfolios with these characteristics will never get back to even, and indeed, P3 and P4 haven't yet made their way back to their July 1, 2000 starting levels. So if there are improvements to be made to either portfolio, greater downside protection would be the place to focus. In the next three years, this will be the main but not sole focus. For now, however, it's still status quo. 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 BEEN SIX YEARS since the launch of quantitative Portfolios 3 and 4 back on July 1, 2000. A lot has changed since then, but the models are essentially the same.
