| |
 March 2011 Thinking Ahead Potential Changes for Portfolios 3 and 4
It's not yet time to make any additional changes; we decided to do that only once every three years at the earliest. Inasmuch as the last changes were made at the end of 2009, there's still more than a year and a half to go. Nevertheless, it's never too soon to begin considering what ought to be done next. There's a right way and a wrong way to enhance a quant model. You might just want to look back over the past few years and see which fundamental characteristics commonly led to superior stock performance. Just by breaking down a sample universe (e.g. the S&P 500) you can determine what features were correlated with success and which weren't. Those in the former group would then be candidates for inclusion in the models' algorithm for use in the coming years.
That's pretty simple and straightforward, but it's also one of the "wrong' ways to employ quantitative analysis. By focusing only on the past five or even ten years, you can learn a lot about how stocks performed while learning virtually nothing about the way they will perform. Looking at longer periods isn't much better because it only tells you how stocks behaved on average, rather than what to expect in the next few years. Obviously what's needed is a way to consider predictive factors, those that may be able to isolate stocks poised for better performance in the future regardless of what happened in the past. This means the focus of the search shouldn't be on specific trading patterns of the recent past, but rather on what should work in the future. Clearly the results won't be as precise as looking at historical stock results, but the potential benefit should be much greater. The correct way to do this is to come up with a theory -- more precisely a hypothesis -- based on observations and logic. Rather than relying on the past alone, this should include elements that could be more predictive rather than derivative. The best way to test the hypothesis is to actually put it in action and follow the results. You can "backtest" it, if you want, but the correct procedure would be to use your hypothesis with data from prior to the test period to avoid meaningless data mining. The goal is to find if data available at the time the hypothesis is tested would actually support the theory. So with this in mind, here are two hypothesis, one for P3 and one for P4:
P3: Align the Statistics and the Time Horizon Investors actually "invest" in the companies and stocks they buy. They have considerably longer time horizons -- usually 3-5 years is the minimum -- and are willing to wait for their choices to pay off. Unlike traders who depend on short-term success, they are willing to hold on for smaller, incremental gains over time.
All of the model portfolios at this site -- including the four quant portfolios -- are investment portfolios. By their design they attempt to minimize trading while providing market-beating long-term performance. For the most part (check out their historical graphs), they've done an excellent job. Nevertheless, time horizon may be an issue for P3. While it does have a long-term goal, it has a short-term composition. This comes from the fact that P3 is reoptimized every other month. Conceivably, it's entire thirty stock portfolio could turn over every 45 trading days. That's not really long-term, is it? But what is long-term is the set of factors in its algorithm. We use ten-year market statistics to select the factors that will be used to pick the stocks for P3. In other words, these are the factors we believe will be most predictive of future performance, but when they were selected, there was no time limit on their results. It's very unlikely they'll change much in 45 trading days. As a result, P3 has a short-term holding period and a long-term algorithm. One might think there's a certain disconnect there. In general, P3's short-term holding period has worked to its favor in increasingly volatile markets. The ability to adapt has arguably helped it take greater advantage of market momentum. P4's initial holding period was one year, but was subsequently reduced for this exact reason. As a result, P3's holding period is not the problem. So here's a new hypothesis: P3's algorithm should be based on shorter-term fundamental factors. Rather than ten years, perhaps twelve or even six months should be considered. While there's no need to match the factors and the holding period precisely, 6-12 months would be long enough to show a meaningful correlation for selection while still retaining enough short-term momentum to make a difference in performance. This doesn't mean we wouldn't want to look back ten years in the out of sample period when selecting statistics. What it does mean, however, is that rather than looking at results over the full ten years, we'd instead focus on 6-12 month periods within that ten-year period. This would provide plenty of data points for testing while allowing greater insight into the factors' short-term predictive ability. There's no denying P3 is at heart a momentum-based strategy. By more closely aligning its true time horizon with its quantitative factors, results should be enhanced.
P4: Increase the Opportunity to Add Value As you probably know, diversification like this is a double-edged sword. It can help stem losses when the market leadership abruptly changes because all corners of the market are always covered. However, it can actually serve as a drag on performance when the market is trending and certain sectors are left behind. Even the "best" performing stocks in an underperforming sector will weigh on performance. They are, in essence, the cream of the crap. Then again, P4 does have that six-month holding period. Completely removing the diversification requirement would simply convert it into a longer version of P3. Not only that, some sort of decision rule would have to be added to determine which sectors to represent and which to eliminate when selecting stocks for the portfolio. As an alternative, consider this alternative for P4: Maintain the requirement that all ten market sectors be represented, but modify the requirement that each be market weighted at rebalancing. Instead, allow the two sectors expected to outperform to be overweighed and the two anticipated to trail to be underweighted by the same percentage. As it presently stands, P4's only opportunity to outperform the broad market comes from its stock selection. Because all sectors are represented at market weight, nothing can be done to improve performance other than selecting the best stocks (and avoiding the worst) in each sector. Allowing the two sectors anticipated to outperform for the next six months to be overweighed while underweighting the two expected to lag would offer another means of adding value. Maintaining a presence in all ten sectors still allows the model to benefit from diversification during down markets while over and underweighting sectors at the opposite end of the spectrum allows additional gains without adding an inordinate amount of risk. This is essentially the approach of a long-short fund where the manager is able to take advantage of all his or her research, both to the upside and downside. This allows flexibility in four of the ten sectors with the remainder anticipated to basically perform at market levels -- not an unreasonable assumption.
You, Too, Can Help The best quantitative models evolve over time and that's what's going on here. You can participate, too, if you have any hypotheses that might help. Certainly feel free to submit them and we'll be happy to credit you if they're used. Search this site! Just enter you key word or words: Get current quotes or follow your own custom portfolio,
courtesy of E-Line Financials:
|
|||||||||||||||||||||||||||
UANTITATIVE INVESTING IS NOT a stagnant thing. The investing algorithms used today may reflect our best efforts, but like anything else, they may be improved in the future. As observations mount, additional information imay yield improved analysis. You've actually seen this in action at this site as new data led to changes in Portfolio 3 and Portfolio 4 in 
