|
PureBytes Links
Trading Reference Links
|
A SYSTEM TOO GOOD TO BE TRUE?
> By Jack Schwager
>
> Imagine a trading system that often generates sell signals near relative
> highs and buy signals near relative lows, and also has more winning
> trades than losing trades. Interested? Chart 1 illustrates the SRD
> system (I'll explain the name momentarily), which exhibits these
> characteristics. During the period shown, the system provided four
> signals in the corn market, all of which were profitable, with an
> average per trade gain of $1,031. Chart 2 is even more impressive. All
> five of the signals generated for the British pound during the period
> depicted were profitable, with a whopping average gain per trade of
> $4,662.
>
> Of course, you are probably curious to know the details of the system
> that generated these series of impressive trade signals. However, I'm
> sure you will agree that it is unreasonable to expect me to divulge this
> information for free in a magazine article. Certainly, it would make
> more sense to market this system at several hundred dollars a copy, or
> more, providing only a very general description in this article.
>
> Well, I'm actually going to reveal the specifics of the SRD system in
> this article. What possible motivation could I have for disclosing this
> information for free? That will soon become clear. The great trading
> signals illustrated in Charts 1 and 2 were generated by an actual
> system. The rules of this system are as follows:
>
> Buy Signal-Whenever the market closes below the lowest
> low of the past 90 days, go long.
> Sell Signal-Whenever the market closes above the highest
> high of the past 90 days, go short.
>
> It's that simple. And yes, this very basic system generated the
> excellent signals indicated in our examples. However, before you rush
> out to trade this system, let me provide you with a few more
> illustrations of the SRD system in action.
>
> Chart 3 depicts the same system applied to the Japanese yen market. In
> this instance, 3 out of 5 trades were losers. However, the real problem
> is that one of the losing trades was for a staggering $39,062 per
> contract, resulting in an average per trade loss of $7,017 over the five
> trades combined. In Chart 4, for the period illustrated, the SRD system
> started off promisingly enough with four straight wins in the coffee
> market, with an average per trade gain of $2,250. The next two trades,
> however, were mammoth losers, with an average loss of $25,047 per trade!
> For the six trades combined, the average per trade result was a loss of
> $9,849 per contract. As one final example, Chart 5 shows the SRD system
> applied to the gold market. Here the system generated four consecutive
> winning trades, followed by a fifth trade, which lost more than three
> times the gain of the first four trades combined.
>
> So what happened to our great system? The truth is that although the SRD
> system will work great in wide-swinging trading range markets, it will
> get massacred in strongly trending markets. The problem is that although
> you can easily identify trading range and trending markets in the past,
> it is virtually impossible to predict whether a market will be in a
> trading range or trend in the future. In fact, generally speaking, you
> are likely to do far better by fading the SRD system-that is, taking
> trades in the opposite direction of the signals-than trading the system
> as proposed. The system only seems to do spectacularly well because the
> examples were carefully selected to make the system look good. (By the
> way, the SRD name of our system stands for Super-Razzle-Dazzle.)
>
> Well chosen examples are pervasive, appearing in books, articles,
> advertisements, direct mail, and seminar presentations. Accepting well
> chosen examples at face value is one of the most common and critical
> blunders made by traders. Our mock presentation of a "super" trading
> system was intended to highlight three critical points:
>
> 1. Any, repeat any, system can be made to look great using a few
> well chosen examples. In fact, I believe it is virtually impossible to
> devise a system that can't be made to look great on some segment of
> market price history.
> 2. Therefore, never, repeat never, judge a system based on a few
> examples.
> 3. When it comes to trading systems, what works well in one case,
> may kill you in the next.
>
> As another illustration of the well chosen example, let me tell you a
> true story. Back in 1983, when I had only been working on trading
> systems for a couple of years, I read an article in a trade magazine
> that presented the following very simple trading system:
> 1. If the six-day moving average is higher than the
> previous day's corresponding value, cover short and go long.
> 2. If the six-day moving average is lower than the previous
> day's corresponding value, cover long and go short.
> The article used the Swiss franc in 1980 as an illustration. Without
> going into the details, suffice it to say that applying this system to
> the Swiss franc in 1980 would have resulted in a profit of $17,235 per
> contract (assuming an average round-turn transaction cost of $80). Even
> allowing for a conservative fund allocation of $6,000 per contract, this
> implied an annual gain of 287 percent! Not bad for a system that can be
> summarized in two sentences. It is easy to see how traders, presented
> with such an example, might eagerly abandon their other trading
> approaches for this apparent money machine.
>
> I couldn't believe such a simple system could do so well. So I decided
> to test the system over a broader period-1976 to mid-1983 (the time of
> my test)-and a wide group of markets. Beginning with the Swiss franc, I
> found that the total profit during this period was $20,473. In other
> words, excluding 1980, the system made only $3,238 during the remaining
> 6½ years. Thus, assuming that you allocated $6,000 to trade this
> approach, the average annual percent return for those years was a meager
> 8 percent-quite a comedown from 287 percent in 1980.
>
> But wait. It gets worse. Much worse. When I applied the system to a
> group of 25 markets from 1976 through mid-1983, the system lost money in
> 19 of the 25 markets. In 13 of the markets-more than half of the total
> survey-the loss exceeded $22,500 or $3,000 per year, per contract! In
> five markets, the loss exceeded $45,000, equivalent to $6,000 per year,
> per contract! Also, it should be noted that, even in the markets where
> the system was profitable, its performance was well below gains
> exhibited for these markets during the same period by most other
> trend-following systems.
>
> There was no question about it. This was truly a bad system. Yet, if you
> looked only at the well-chosen example, you might think you had stumbled
> upon the trading system Jesse Livermore used in his good years. Talk
> about a gap between perception and reality.
>
> This system witnessed such large, broadly based losses that you may well
> wonder why fading the signals of such a system might not provide an
> attractive trading strategy. The reason is that most of the losses are
> the result of the system being so sensitive that it generates large
> transaction costs. (Transaction costs include commission costs plus
> slippage.) This sensitivity of the system occasionally is beneficial, as
> was the case for the Swiss franc in 1980. However, on balance, it is the
> system's major weakness. Losses due to transaction costs would not be
> realized as gains by fading the system. Moreover, doing the opposite of
> all signals would generate equivalent transaction costs. Thus, once
> transaction costs are incorporated, the apparent attractiveness of a
> contrarian approach to using the system evaporates.
>
> It should be emphasized that the foregoing is not merely intended as a
> cautionary tale to purchasers of trading systems and readers of articles
> on trading systems. The points made are equally relevant to traders who
> design their own trading systems. Traders need to be wary of the
> potential of well chosen examples cropping up in their own system
> testing. Don't jump to conclusions based on a few test cases-they may
> simply be chance occurrences of well chosen examples. Always do
> sufficient testing before assuming the efficacy of a trading system.
>
> The moral is simple: Don't draw any conclusions about a system (or
> indicator) on the basis of isolated examples. The only way you can
> determine if a system has any value is by testing it (without benefit of
> hindsight) over an extended time period for a broad range of markets.
>
> Kindest regards,
>
> Janette Perez
>
> -----------------------------------------------------------
|