In mechanical trading research, more corroboration is always better than less. Something that produces spectacular results in five years of data can be good; if it also holds up even marginally in the previous 10 years, that’s better. Eye-popping gains in one market may seem like your Holy Grail, but if it’s failing to profit in related markets within the complex, you probably want to discard it anyway.
The more examples of confirmation we get, the happier we should be. You do not have to insist that an average financial idea also works across non-financial sectors but you should be jazzed if it does. We never know the exact degree of pure chance that is factoring into our results. The more extensive and prevalent a tendency, the more reliable it is. “Robustness” is the term that encapsulates this.
The financial arena has been where the action is in recent years, the current surge in traditional commodity prices notwithstanding. Within financials, you find the best trending markets of all, the currencies. You also get the bond complex, one of the best barometers of world events, and of course, the volatile stock indexes. In two decades of specialization in this later complex, it’s become clear how unique the stock-related markets are and, therefore, how they differ from their financial non-index counterparts.
WHY THEY’RE SPECIAL
Unlike the Japanese yen, which can trend for months or even years, the S&P 500, Nasdaq, Dow, etc., tend to be mean-reverting. That is, they tend to come back off overextended high or low levels. Just what makes a market overextended is one of the challenges of research. For now, let’s just accept for argument’s sake that there are enough wide pivots, backing and filling and raiding-of-stops that make these markets hard.
This makes it difficult to develop a one-size-fits-all system that works the same way in all the major financial complexes. It’s not absolutely necessary to do so. It is possible to ride sector-specific characteristics to glory. By definition though, targeted ideas are never going to make the kind of fortification seen from a single idea fit the same way across most of the dissimilar financial environments. Is there any way to bridge these disparate elements?
Let’s start by looking at one of the most renowned and basic mechanical systems of all. It’s known by a few names, among them “N-Day” and its rules are simple: Buy the highest of the last n-day’s worth of highs on a stop and go short at the lowest n-day low.
In its rawest, most unadorned form, the system is perpetual, meaning you’re always long or short — always reversing on entry signals opposite your existing position. A quick study of the methodology would also make it clear that the more a market moves against your position, the closer it is to levels at which you would get flipped. The stops, in other words, are not arbitrary but rather unfold directly out of the market dynamics.
“The best N” reveals that any five-point multiple between 10 and 50 would have been positive during the time frame of the backtest for the Treasuries. Only the lowest (five days) was a loser. Similar results are seen in the euro. The S&P 500 performance, on the other hand, shows more losers than winners with significantly bigger numbers in the former category. It’s not as easy to merely jump aboard a trend in the indexes.
Inevitably, different results are going to top the lists in different markets. Again, random chance is always going to play some part in how numbers fall out in any given study. This brings us to the inexact process of selecting a number that performs well in some cases and adequately in most.
“On and off” shows that 25 is fairly workable across the field. This is obviously intended to be more of a demonstration than an ultimate system, but consider a couple things. First, it’s as stripped-down a version as we can get. We could almost certainly improve our summaries with the addition of qualifiers and filters. Second, that the study originates in the year 2000 means that our results have occurred long after the system was developed and publicized. For all practical purposes, this is real-time performance that shows a bias that has more or less existed throughout the history of markets.
DEFYING THE BIG MO
We’ve obviously chosen a robust number because even the indexes are complying. That isn’t the case across a broad spectrum, however, as shown in the third table in “The best N.” There are index-specific methodologies that work better than pure momentum following; among them, a technique that could pretty much be considered anti-momentum. Here are the rules:
1. If the close is below the eight-day closing average, then buy at the next opening plus 25% of the three-day average range on a stop.
2. If the close is above the eight-day average, then go short at 25% of the three-day average range below the opening.
We are fading closing averages, not going with them. Granted, we’re using small momentum to enter (buy higher than the open, sell lower), but we’re doing so off overextended levels. “Fading the eight-day” shows the performance across our targeted financial array.
This is one of the best ways to exploit stock-related futures. It is not hard to catch typical swings the right way. Granted, you run into problems when you get onto the wrong side of a trending market. The big obstacle with a mean-reversion system is, the more wrong you are, the more your signals insist you’re right. Indeed, you may never find stops that flow out of market action in the natural way that N-Day type momentum systems allow, requiring you to settle for arbitrary places to cry uncle and then deciding where to re-enter on still-persisting signals.
This demonstrates a rather primitive version of swing trading, which can be made serviceable in almost all markets. Again, though, it is the indexes that especially lend themselves to the concept. The unique psychological nature of the sector keeps the action from getting too far afield without significant backing and filling. “Stock index special” shows that whether you applied five or 50 days, or any five- point increment in between, you’d have made huge profits in the Russell. When we apply the same study to the trendier euro however, we get a vastly different picture.
COMPLETING THE PICTURE
We might notice that overall, we’re seeing better fade results using low numbers whereas N-Day tended to work better with higher values. We should therefore be able to combine the two concepts — fade a lower number and follow a higher one. After much optimization, which produced an array of encouraging results, the following emerged: If the close is below the eight-day average close and above the 50-day, then buy the next open plus 25% of the average three-day range on a stop; following the entry day, any time either of the above two conditions ceases to be true, (the close is above the eight-day average or below the 50-day), then exit the trade at the next open minus 50% of the average three-day range on a stop.
Obviously, if you have a setup but no execution, you’re still in the trade and will follow the next day’s new signals at the next opening. As you should expect, the short rules are the mirror opposites. The results are shown in “Piling on.”
As with the basic N-Day system, your risk is finite without being arbitrary; you can get out whenever you travel to the wrong side of the 50-day closing average. Part of the system’s success is that you sometimes exit at favorable levels as well; the eight-day reversal rule tends to lock in profits.
Overall, the combined system is clearly superior. Probably the best gauge of a system’s effectiveness is the return on account (last column). That represents the percentage your account would have increased had your startup amount been equal to the maximum drawdown level. It’s computed by dividing the net profit by the worst drawdown and multiplying the figure by 100. In the top example, the ultimate $225,000 net profit divided by the $29,075 drawdown equals 7.74, which times 100 equals the 774 ROA percentage. A 774% increase since the turn of the century is not too shabby. Even better are the respective 1,402% and 1,212% figures in the other indexes.
The worst drawdown figures with the combo are more reasonable than with either isolated component (Keep in mind that mini indexes would reduce all numbers by 80%.). This is more than a mere demonstration of biases. In all the indexes, the profit-per-trade column vaults dramatically past the standard $100 slippage/commission we should figure to expend per contract. A few other markets cross that threshold as well, and all of them are at least theoretically corroborating the concept.
There certainly are enough trials over the eight years to convince us that we have something statistically valid. The percentage of correct trades is never less than 55%, certainly psychologically fortifying. We should feel similarly confident in that both system halves have held up under their own individual testing.
On top of everything, we get corroboration in some unexpected arenas as well (see “Outside validation”).
A mechanical system won’t work for you if you don’t follow its mandates to the letter. The best chance we have of maintaining such rigid discipline is via exceptional theoretical results that figure to persist into the future. This comes from knowing we tested properly and seeing our single concept perform the same way across a wide data field. Again, non financial performance is not an absolute must for financial strategies. So much the better, though, when we do get it.
Art Collins is co-founder and managing director of Trireme Capital Advisors LLC and a regular contributor to Traderinsight.com. He is the author of “Beating the Financial Futures Market: Combining Small Biases Into Powerful Money Making Strategies.” E-mail him at firstname.lastname@example.org.