Day Trading & Forex Forum

Can a sector rotation strategy improve performance of investment returns? In this article the author looks at sector rotation within the US markets.

Investors can most certainly use a sector rotation strategy to produce returns which outperform the market, or even some hedge funds, despite what was described in a recent article. Many hedge funds would be wise to consider such a strategy. However it's not as easy as the plan described in the article’s abstract. Nor should we expect it to be; the market is not easy-pickings. But recognize what market returns we are talking about: since say 1990, the S&P has had about an 8 pct compounded annual rate of return, while experiencing about a 46 pct maximum drawdown. Those numbers can be beaten, and not just by the pros. As with any game, being an informed participant enhances your success. Without getting too far into proprietary methodologies, let me provide some insight as to what a sector rotator must consider (with a few tips included):

Is the investor going to be fully invested all the time, or does he have an escape? That is, should the first set of sectors be equities vs. treasuries. Let's call that a strategic overlay, or the first on-off switch. Then, is the investment going to be long-only, long with a hedge, or long-short?

Next comes the equities universe. What group are you considering: 10 S&P economy Sectors, the 24 S&P Industry Groups, the larger (ever-changing) number of Industries and Sub-Industries (GICs), high liquidity ETFs of economic sectors, country ETFs, Fidelity funds, Dow Jones 18 European sectors, or small cap funds accepting new money. The list can be endless. We have rotated all of the above, and can attest that the hardest of the lot is the 10 S&P Sectors. We would avoid as much as possible rotating a list of just four, as was done in the referenced article.  If you are going to rotate sectors profitably, you need more focus.

Once you have chosen your universe, you have to pick an out-of-sample period and run cross correlations on the assets. The cross correlations should not be on the levels, but either the changes or preferably the scoring that you will eventually use. Your purpose here is to reduce the likelihood of always being in the same combinations of assets. You wouldn't mind doing so if they were all moving up together, but the opposite would destroy you, and must be avoided. This is the only subjective part of the entire process: you will personally have to decide the level of cross correlation you will accept. You can of course test this also (a huge amount of work), but there will be obvious breakpoints that make sense to the experienced researcher. Thus if you start out with the 67 S&P Industry GICs, cross correlation may reduce that number down to about half. This should be done blind. Also, some of the 67 have only one stock, and you may want to eliminate such a sector.

The choice of ranking or scoring device is the most critical part of the entire process. Most of the industry professionals we know use a "quarterly" rate of change or relative strength calculation. For example, we have been told that R***x uses a moving 63-day rate of change for their sector rotation fund on 67 GICs culled down to 56. They have passable results which outperform the market, and several hundred million in that fund alone. The problem with either moving rate of change or relative strength is that those calculations produce fairly erratic scores. They can be improved by some slight smoothing prior to ranking, or by skipping days (e.g. scoring and ranking every other day), but your task is to find what works best, and there are lots of things that work better than rate of change.   We have done some work with counting as a ranking device, but our experience is that using counting works best only if we choose to be particularly risk-averse in a long-only program. Oh, and don't assume that a ranking/scoring device is just one indicator, as combinations work best.

One ranking device we have found absolutely knocks the cover off the ball, but it is not obvious. Although it is very robust and clearly non-random, we don't yet understand why it works, and are reluctant to “bet the farm” until we do.  You should be equally cautious in a similar situation.

Should the sectors be risk-adjusted? If telecoms and utilities have equal rankings on a given day, do you want to discriminate in favor of the least volatile, say by subtracting half a moving standard deviation? Our results show that doing so on sectors reduces both returns and drawdowns. That's unexpected, as usually reducing drawdowns and increasing returns are handmaidens. However if you are ranking proprietary funds, risk-adjusting outperforms not doing so across the board. That is, penalizing managers for bad behavior really works. This suggests that (a) certain managers really have talent and fat tails, and (b) they can be discovered by some quant work.

Then comes the question as to how many assets out of our population are going to be traded. This can be tested empirically. R***x conducted research which suggested that 3-5 sectors was optimal. Yet R***x uses 8 out of their 67-culled-to-56, probably for marketing reasons (so we've been told). Our results show that whatever number you pick, numbers in that vicinity work well too, so it's robust. We generally recommend using at least 3 assets to reduce volatility. But using more than 10 percent of the population curtails returns. Some of the other professionals we have spoken with have told us that their "second quadrant" outperforms their "first quadrant". Should that happen to you, you need to do more work at finding a better scoring method.