| |
 May 2010 Turnover Unleashed Turnover has Risen Dramatically in P3
There's no denying research has been dedicated to models fitting this description. But there are plenty of other quantitative trading strategies that don't rely on minute by minute trading or require several statistical PhDs to implement them. The quantitative models at this site would certainly fall into the latter category. However like all misconceptions, the fear of quantitative investing strategies contains a certain level of truth. Turnover has always been a major nemesis of any mathematically based trading system. The basic problem is easy to see: When an algorithm determines it's time to trade, it trades regardless of the frequency or associated costs. Individual investors are much more sensitive to investing costs, as well they should be. As rap on quantitative investing goes, it doesn't matter if the model generates gross returns if they're more than eaten away by high transaction costs. The advent of discount brokers helped ease this concern, yet turnover still must be considered when evaluating quantitative models. A certain amount of trading is necessary in any investing strategy (except perhaps, buy-and-hold, but then is that really investing or just a different version of saving?), yet when does it become, too much? That's a question all would-be quantitative investors must face, and it's become increasingly important for our model known as P3.
Fighting Turnover P3 seeks to find and hold the top thirty stocks in the S&P 500. It uses an algorithm to locate those stocks that have the highest levels of key fundamental factors which over the past five years have led to (or at least been associated with) outstanding large cap equity performance. P3 is reoptimized every two months to take advantage of changing market conditions or in less technical terms, play the momentum. A lot can change in two months and as you'd assume, stocks that scored in the top thirty two months ago may not do as well this month. This is particularly true given the universe of potential holdings is the five hundred stocks of the namesake S&P 500. Put another way, holdings must remain in the sixth percentile or above in order to stay in the portfolio. That's a pretty tall order.
If we had been pointy-headed academics holed-up in an ivory tower, we would have just let it go at that. But hoping that P3 would be more than a theoretical construct and something you could actually invest in, we were concerned about the negative effects of frequent trading. In an effort to reduce it, the model was launched with a trading rule: Stocks that scored 40 or better could remain in the portfolio even though they were not in the top 30 at the time of reoptimization. Not only did this mean veterans ranked 31-40 could stay in, it also meant that an equal number of those ranked higher would remain out. It didn't seem like much of a stretch given the greatest difference would have been ten spots out of 500. P3 used this trading strategy for its first four years. When it and its quantitative sister P4 were created in June 2000, it was the intention to let them run at least three years before making any substantive changes. This would provide a longer to term to evaluate them, perhaps even a market cycle. It didn't quite work out that way -- after the first six months or so, the Tech bubble blew up and the rest of the three years was spent in the depths of a deep bear market. Regardless, allowing holdings to fall to 40 probably did cut down on the turnover.
Then again, it only helped so much. Turnover averaged 25 percent at each reoptimization over the first three years. That may not sound high, but don't forget the model was being reoptimized six times a year. That means the real annual turnover was 150 percent. While this may be a little on the high side for a large cap portfolio, there are plenty of equity mutual funds where it's even higher. One thing that was bothersome, was the fact that many stocks left in one rebalancing only to return the next (see Return Engagement). While we thought the cushion of ten places would help prevent such whipsawing, it really didn't have much of an effect in practice. As a result, when the three years was up and it was time to reconsider the structure of the quantitative modes, P3's trading rule was changed to allow holdings to remain in the portfolio until their rating fell below 100 rather than 40 (see Time for a Change). That may sound like a major concession, but it really isn't when put in the context of 500 stocks. Now rather than remaining in the top six percentiles, holdings could remain as long as they were in the top twenty percentiles (also known as top two deciles). Most investment policies allow new holdings to be purchased from the top 25 percentiles (top quartile) of the investment universe, so if anything, this was still a little more stringent. Just two months after making this change, we noted that turnover had noticeably declined. Although that was only one data point, we speculated it was a sign of things to come. As it turns out, it was. Average turnover at reoptimization declined to 8.6 percent over the next six years. Annual turnover of 51.6 percent was slightly over one-third as high as in the first three years. To put this in dollar terms, this would be roughly thirty-two trades (16 buys and 16 sells) per year. If done through a discount broker charging $10 per trade, this would be $320 per year, not an overwhelming amount. Turnover was now under control.
New Algorithm, New Issues So it really wasn't surprising when P3's turnover jumped to 53 percent in December when the new algorithm was installed. The new algorithm was quite different from the original (see the details here), resulting in higher than normal turnover. Again, this is to be expected in such a transition. But it didn't end there. When the model was reoptimized in February, turnover dropped by half to 23.3 percent. While that was considerably lower than during the transition, it was still well above the previous average and right in line with the higher turnover from the original trading rule. At least the trend was in the right direction. That abruptly changed in April. One might have thought that would be far enough removed from the old algorithm that it would have settled down. Quite the contrary, April's figure was 66.7 percent -- twenty of the thirty stocks turned over. In just three rebalancings, turnover was just under 150 percent, on target for 300 percent for the full year. Suddenly that discount brokerage $10 per trade total jumped to $1800 annually. Keep in mind this is occurring with the higher 2003 trading rule, not the original lower one. This was neither good nor expected. It wouldn't be so noteworthy if it was just turnover which can usually be rectified with a different trading rule. Instead, there are other concerns involved as well. For example, of the seven stocks that entered the model in February, only one survived the April rebalancing. That's not turnover, that's churn. When 20 percent of the holdings are in the model only two months -- in a relatively quite market -- it's hard to see how that's adding any value over the short term, not even to mention the long. Sectors are churning as well. Healthcare went from 3.5 percent in December, to 27.2 percent in February, back to 4.5% in April. Even the passage of Obamacare can't explain that one.
Finally, in 2005 we found that turnover (at least until that point) had worked against portfolio performance. To test it, we constructed three portfolios of stocks that had been in P3 based on their tenure. We then compared the performance of these unaltered portfolios over the then 5-year life of P3. In this test case, P3 was the only sample with turnover and -- you guessed it -- it dramatically underperformed the other three buy-and-hold versions. Not only that, the three buy-and-hold portfolios also outperformed the S&P 500. The results from this suggested P3's algorithm was quite good at picking good performers, but its turnover prevented them from providing much benefit. But there's one other interesting piece of information: Over its lifetime, P3's annual return has had a mild positive correlation (.2353) with its turnover. While this is a relatively weak correlation, it does suggest P3's return benefits when turnover is higher. This effect is even more pronounced on Chart 1 which compares annual turnover (red line, left scale) to annualized return (green line, right scale). Taken in isolation, this would suggest P3's turnover is indeed, helping it extract the benefits of market momentum. However, there's one other important factor to consider -- the overall market condition during this comparison. In the early years (2000 - 2003) return climbed the fastest when turnover was falling rapidly (2002). In the later period (2009 - 2010) the market moved up strongly along with turnover. But we know why turnover moved up -- it was because of the transition to the new algorithm, not because the model stepped up turnover to capture more momentum. In this case, looks (and statistics) can be deceiving.
What to Expect One thing, however, is certain. No additional changes will be made to P3 until three more (well, two-and-a-half now) years pass. That's the way it was designed and the way it will stay. The trading rule (stocks remain in the portfolio until they score below the top 100) will remain, too. While broadening it would slow turnover, going much beyond its present cutoff waters it down too much. While it's too early to come to any definitive conclusions about this latest incarnation of P3, it's not too early to pick out those features worth watching. Turnover is certainly one to be tracked closely over the next three years. In the meantime, anyone applying P3's algorithm for in their investment portfolio will certainly have to deal with considerably higher transaction costs. Indeed, through the first four and a half months of its use, they already have. While unfortunate, this in unavoidable at this point. Which brings us back to the major rap on quantitative models: They're too transaction oriented and expensive to employ in the real world. P3 has certainly stepped up the transactions and along with them, the associated costs. On the other hand, its performance has hung in and it's still got the potential to provide some stellar performance. If that were to occur, turnover and its associated costs would be less of an issue. Quantitative or not, performance has a way of crowding out any other concerns. Search this site! Just enter you key word or words: Get current quotes or follow your own custom portfolio,
courtesy of E-Line Financials:
|
||||||||||||||||||||||||||||||
ANY INVESTORS HAVE A POOR impression of quantitative investing. The common misperception is that quants are academics who dream up astonishingly complex investing systems that can only exist in the vacuum of academia. Not only are they complex, they are difficult if not impossible to implement in the real world. If nothing else, the turnover and constant rapid trading are beyond the reach of the normal human investor. Even if they were successful, transaction costs alone doom such strategies to underperform traditional alternatives.
