If a 20-day and 30-day moving average each showed a maximum drawdown of $10,000, but the 25-day average only lost $5,000 you should assume that the 25-day results could have had the same drawdown as the nearby tests. By some quidi of timing one test was able to avoid a loss while the normal test was not. It would be ridcj to assume sudi good fortune in trading

If you have a record of major price shocks, such as a chart published by the exchanges showing key events, you can verifi how many of these events produced profits or losses in your final sjstem. If you have profited from more than 50% then you should reassess your rid; based on losses rather than profits from some of those trades. Also check to see that there is an even distribution with regard to the size of the price shocks You may have selected parameters that took small losses on minor shod and large profits on major ones.

If you do not freat price shocks as a serious risk, you may overleverage and undercapitalize your trading account. This is the most common cause of catasfrophic loss. Any price shock that causes a windfell profit could have, just as easily, produced a devastating loss. The example that follows, Anatomy of an Optimization," shows how" expectations and reality are often quite different Further examples can be found in Chapter 22 ("Practical Considerations").

ANATOm OF AN OPTIMIZATION-

Using the Iraqi invasion of Kuwait on August 6, 1990, and the U.S. retaliation on January 17, 1991, we optimized a simple moving average strategj for crude oil. The test period began Januarj 2, 1990, but 100 dajs were needed to initialize the calculations; therefore, the first frade was entered May 24, 1990 (see Tables 21-14, 21-15, and 21-16). In both Test 1 and Test 2, the optimization selects moving averages that produced the highest profits. By acc ting these choices, the trader would have held a short position before the invasion of Kuwait, and a long position before the U.S. retaliation. In reality, there would have been a large loss rather than an exceptional profit. When a price shock is an important part of the data being tested, the best sjstem is the one that took the most profits out of that move, even at the cost of other frades

"PerrvKaiibuan, "Price .- I eevaluatiug Ei.+ eturnEiifectations," Futures b.bi.-trv (june/july 1995i

TABLE 21.14 Tesi: I: Optimizing Crude Oil. January 2. 1990, through August 3. 1990*

* Iciting stopped ifie day before the invasaon of Kuwait Test selects I S-day moving avera.

Holding short on August 6.1990. when Irq invade* Kuwait

in Test 2, although there were laige profits from the first price shodc the trader would have still held the wrong position when the second shock occurred. This method of testing hides the real trading ride focuses on profits thai cannot be predicted, and does it all at the expense of consistent trading profits.

DATA MINING AND OVEROPTIMIZATION

Advances in computing power and testing software have made it very easj to test trading strategies using historic data. It is only logical that you should prove that your ideas would have worked in the past. It is also sensible that you understand the amount of risk that you must take to gain the results. For more sophisticated analjsis, a relationship may be found between fundamental factors, volatility, and rid;. For fundamental analjsis, the expected price of oil is greatly dependent upon the crude inventories., for many stock issues, the per capita disposable income is the key element in predicting revenues; and, the anticipated volatility of soybeans is often found in the size of the carrjover sto(ts. Although past prices

30-». I

ss.o

Tratfe Decail fram Testing

Jul Buy-

Oct I9,90 Sell 40. IO

Oct 30, eo Buy I .23

Oct3I,eo Sell -*l.92

Nov 19. 90 Buy 38 39

rsi.w 21, -SO Sell 37.30

141 », 26. 90 Buy

rcw 27-, -SO Sell •O.S3

and their relationships to economic data caimot be ignored, they can also be misleading, or even erroneous, if used carelessly.

The most important factor in the success of a trading program is a sound premise; that is, you must find a real-life relationship between the price and the way participants react. For example, when the mondarj authority of any country raises interest rates, the price of equities tends to fall. Or, when there is bad news for the economy, investors shift from equities to guaranteed interest rates, driving the stock martlet lower and the bond martlet higher. Once these relationships have been identified, it is perfectly sensible to look