FIGURE 22-6 Actual futures movement compared with random movement.

able direction and set new trend positions after I or even 2 dajs of price reversal. Either variation might improve the retum ratio after a random occurrence of runs.

For the anti-Martingales approach, double the position after each profitable day, take profits after runs of 3, and begin again after a reversal day. By trading with the trend, profit sequences may be increased to 4 dajs; it may also be expected that, on average, the price change will be smaller on dajs that move counter to the trend.

Sjstem Trading

standard trend-followmg sjstem can expect 30°o to 40°o successtlil trades (called its reliability) of varjmg profitability, which can be defined in terms of a profit/loss ratio, in considering the Martingales or antiMartingales methods, it should be noted that the lower probability of profitable trades would make the wait for a long run of profits less ath-adive. if a sjstem has a profile of 35°o successful trades and a profit/loss ratio of 3 : 1, then trading equal amounts on each signal results in the following, assuming $750 each profit and $250 each loss:

Total h-ades 100 x $25 (commission) = $-2,500

Profitable trades 35 x $750 = $26,250

Losing hades 65 x $250 = $-16,250

Net ($75 per hade) $7,500

Taking these same figures and appljing Martingales to a uniform sequence of two losses and one profit gives $275 as the first loss ($250 plus a $25 commission), 5550 for the second loss, then $2,175 for the profit, netting $1,375 for three trades, which would rdum $229 per contract traded

The disadvantage of appljing Martingales to the irregular returns of high-leverage maikets is that a long run of losing trades requires a laige amount of capital. If it is necessary to keep a laige reserve in the event of a long run of losses, why not simply trade equal amounts of a larger position on each signal, without using Martingales? A sound trading strategj should differ from a pure gambling situation because a specific sjstem win win in the long term, while you will lose at roulette. The results show that frading the same laiger position on each signal will retum a higher profit than an application of Martingales, which requires the same available capital; however, a less aggressive betting method that assumes favorable odds can improve the retum on investment.

In another shidy of appljing Martingales, an actual trading sjstem based on Wilders RSI was used to create buy and sell signals. During the test, a total of 277 contracts were held at one point, which is far beyond the realm of the normal speculator. That means an investment of 277 times the initial maigin, plus any accumulated loss, just to recover one unit of your initial maigin! It should not be surprising that the risk-toreward ratio of this betting method was found to be unacceptably laige.

Complex Martingales

The simple Martingales discussed do not present an attractive way of managing risk,, however, a slightly more complex form is far more practical and has very different results, instead of doubling the bet each time there is a loss as in the case of simple Martingales, increase the bet by a smaller amount, for exanple, 40" o. This slows down the process and avoids the rid; of ruin, or table limits. You may even choose to divide one bet into more than

1 Roger Altaian, "Tactical Trading I eviEite.L" TecAnncal Analysis .-& Cinmodities iceftember 1993)

James William Fu-»ison, "Martees," ! Analysis -i - & Cimiiodities (Febru.*y 199iii, and "lever.-e lJ.imiigales," TecAnncal Analysis -i - « & mn,odities . lJ.*cl, 19911.

one, so that a situation in which you would trade 8 contracts may be traded in two trades of 4 each. You must simply remember to stop the sequence and begin again when your profits have exceeded your losses. In the following , each loss is followed by increasing the bet by an equal amount. When a profit occurs, the bet size is reduced by one unit and the sequence ends when there is a net profit. Profits and losses are equal to the bet size.

EXAMPLE I Equal-Size Bets witli Equal Payoff

In Example 1, it was necessaijto win twice to realize a profit the size of the original invesbnent; however, that profit could be achieved with 40° winning trades and 60° losing trades.

Appljing this strategj to a more realistic trading profile, consider a sjstem with an average loss of $1,000, an average profit of $2,000, and a win to loss ratio of 1 to 3. This means that there are 2 losses and 1 profit in every 3 trades, a net sjstem result of zero, equivalent to a gambling situation. Example 2 shows another sequence of trades. Even though there are 3 losses to start, a single win nets a profit for the sequence.

EXAMPLE 2 Equal-Size Bets with aTradin System Profile

When appljing a more complex form to reverse Martingales, it is best to taiget sequences of profitable runs that are likely to occur with reasonable frequency Instead of betting the entire accumulated profit on the next trade, bet a fixed percentage of those profits. Alternatively, if you are looking for profitable runs of 4, remove the amount of the initial

bet from the total bet after the sjstem has returned two profits in sequence. This tends to conserve capital but slows down the process

SELECnv-E TRADING

It may be unrealistic to think that you can ahvajs follow a technical, fully automatic sjstem without manual intervention. Overriding sjstem signals may occur for reasons of sound risk management during a crisis period when you feel that the sjstem is not programmed to deal with the current volatility and that a conservative action is the best business policy.

The beginning sjstem frader often has concems that the impersonal buy and sell orders will not react properly to the underljing forces that are driving prices. This can

cause the trader to override the sjstem signals. For example, if the program is geared to take profits after a move of 2" o, the trader might judge that the current price move is going to continue; therefore, he will hold the trade rather than take profits. The application of this special intuitive filter generally favors holding a position through ups and downs until it shows a profit, rather than appljing consistently good rid; management, judgment tends to interpret all news in favor of the position being held until it shows a profit. A frader who will survive must leam not to take a loss personally, to keep losses small, to establish and stay with a frading plan, and to filter out unsupported opinions. Trading is a business, and business principles must apply.

Technical sjstems can be taken to exfremes in an effort to automatically select only profitable trades. Two approaches to this are most common. The use of more than one trend indicator, such as two moving averages, is one method of isolating the best trend. It is intended to find a smaller piece of a trend that is more reliable or one thai performs better than a single, longer frend. In principle, it is no different from using three, four, or more trends of different periods to isolate the one point of a price move that ahvajs generates profits. As you become more and more selective, you will find that your opportunities have disappeared, you have used your capital poorly, and the performance profile has actually deteriorated. Using multiple frends, just as any sjstem with many rules, is another attempt at perfection, but only succeeds at overfitting the problem.

There are many sjstems that can be bought that promise profits. Trying to inco orate a black-box approach, a method that does not disclose its exact rules, to improve the performance of your own program is not a good idea. Without knowing everjthing about the other sjstem, there may be inadvertent duplication of a faster trend, a 3-day cycle, or some other feature. You may think that an analjsis of the correlation of results between the black-box sjstem and your own method will show if the methods work well together, but there are still too many uncertainties. The unknown sjstem may have been overfitted and will not show its real risk, or it may have a particular weakness to a maiket pattern that has not occurred during your testing period. You must understand every method that you trade. Its your rid;.

The best way to create a successfiil frading sjstem is to look for unique elements, each with a sound basis, and combine them one at a time, testing each to be sure they successfully stand on their own. The selection of the speed of a moving average will be more important than either the smoothing technique or the number of moving averages to be