then released. Most financial maikets were affected simultaneously, as were the energj maikets. In the case of the currency markets, many traders profited from the first shock, which occurred on a Sunday, then gave it all bad; on the following Wednesday.

12 1 1 C=61350 BOO 0=61.430 H=61510 L=61 190

Gorbachev is abducled on Surrfay, August ie. Market plunges on open igust 19, men recovers stSl.

of August 19, men onWednesdAuf when crisis ends.

Source Chart created with TrodeSftrtion* by Omega Research, Inc.

FIGURE 22-2 A more common shock.The election of conservatives in Britain in January 1992 was a su rise, but the effect was relatively local.

Qomi -640 Hieioo biBOieo.

l/l 1 election s1<ock " ill" January8.18 2

I ll

1 itir¹:t>ati6ncirei<

Source: Chart created w

CO* bf Omega Research, inc.

The process of identifjing price shocks can be speeded up with a small amount of automation. A price shock will be defined as a day with a frue range greater than the average frue range by a multiplier, called the shock fador This would be coded as

shocK = 0;

truerange = hiqh - low;

if higti - close[l]> truerange then truerange = high - closefl]; if close[l] - low > truerange then truerange = close[l] - low; volatility = @everage(truerange[l],vperiod); if truerange > volati1ity*shockfector then shocK = truerange;

Having coded the identification of a price shock, we can write a record to a file or print log to create a list of all price shocks, the date they occurred, and the size of the initial shock. It might also be usefiil to count the number of periods until prices retum to a relatively normal level, it may be unrealiatic to assume that the volatility resulting from a price shock will completely dissppear; therefore, the point at which some degree of normal might begin could be when volatility declines to one-half the level of the initial price shock. This would be found using the code

if truerange < shock/2 then stiock = Q;

With this simple method of identifjing both the beginning and end of a price shock, we can study the effects of price shods on testing performance. A simple test was performed in which an optimized frend sjstem was inspected for periods of price shocks, where a shock was defined as a price move of three times normal volatility The profits and losses from these periods was removed from the net performance over a 10-year period, with the results shown in Table 22-1. The reduction in profits was more than 50°owhen price shocks were eliminated. This means that the selection of parameters heavily favored profits that were generated by large, probably unpredictable, price jumps.

FIGURE 22-3 The invasion of Kuwait by Iraq. News of massing armies and froop movements near the Kuwait border helped the maiket anticipate the August 7 990, price jump in oil. At the beginning, many traders were on the riejit side of the maiket, although the move was not very dramatic.

02G0¹ C=4429Q .Offl 0-44320 hU4.4IO LM4156

Nervous market shows volatile changes

38.000 36 000 34000 32000 30000 28-000 26.000 24.000 22.000 20 18,000 18,000

Source: Chart created with TnnJeStooon* by Omega Research. Inc

How can we use these results if we cannot see the price shock until after its hsppened? The only way is to correcl the optimization process so that the parameter selection is not based on which calculation period is best tuned to get the most profit from these events. Df you remove the profits or losses during the initial price shock from the historic tesl results, as shown in Table 22-1, the optimization would then show the best performers, net of price shocks. Because price shocks are unpredictable, selecting from a shock-adjusted result should yield a better frading sjstem. Some traders might also prefer to use the identification of a price shock period to change the current frading sfrategj to one of crisis management.

GAMBLING TECHMQUE-THE THEORY OF RUNS

The application of Gambling Theory to high-leveraged or short-term trading satisfies two important conditions. First, it presumes no statistical advantage in the occurrence of profits

TABLE 22-1 Results of Removing Price Shocks from 10-Year Historic Performance

and losses but concerns itself with the probable patterns. Each trade could be treated as an occurrence of red or black-up or down price moves or profits. Second, Gambling Theory stresses money management under adverse conditions. A successful professional gambler and an active trader must both be highly disciplined and conserve capital. This section will look at a gamblers approach to money management and ride using the Theory of Runs.

If the assumption is made that each trade is unrelated to the previous trade or that each successive price has an equal chance of going up or down, the situation closely resembles roulette. In Monte Carlo, the roulette wheel has 37 compartments: 18 black, 18 red, and I white, assuring a loss of 2.7°o in the same way that transaction costs are a handicsp to trading. Continuously betting on only red or black will break even, less 2.7° o, over long betting periods The only way to change the odds is by money management-varjing the size of the bets. Although the variables in trading are more complex, we will look at this first.

The most well-known method for winning in a gambling situation is based on the probability of successive wins, the Theory of Runs. On each spin of the wheel, the likelihood of the same color (red or black) reoccurring is 50° o, or 1/2 (ignoring the white slot). To define a run of 3 reds, it is necessary to show a black on each side, that is,

Black-Red-Red-Red-Black

otherwise, ihe run may not be complete. If each color had a 50°o chance of occurring, the probability of a run

of3is

one run of 3 can be expected for every 32 spins of the wheel (called coups). Extending that to runs of n consecutive reds gives (V2)"". For 256 coups, which is both a power of two and the approximate number of trading dajs in a year (if you consider daily closing prices rather than intraday trades), there are the following possibilities for runs of red: