In this case, whatever you think is best depends on what you expect and how
you would like the system to perform in the future. After a few bad trades, it could
be comforting to know that the positive outliers historically have had a tendency
to be larger than the negative ones; on the other hand, having to rely on a set of
not-yet-seen outlier trades to find your way out of a drawdown seems awfully
close to desperate gambling if you ask me.
If we don’t know for sure whether the distribution is approximately normal or
not, we cannot use our normal distribution measurements, which, for example, say
that 68 percent of all observations will be within one standard deviation of the mean.
Instead, we have to use Chebychef’s theorem, which says that for a kgreater than or
equal to 1 (k1), at least (1 – 1 / k^2 ) observations will fall within kstandard devi-
ations of their mean value. Note that with this method, we cannot give an exact value.
For example, if we have a normal distribution, we know that 95 percent of all
observations will fall within two standard deviations; but if we don’t know what
the distribution looks like, we can only say that at least 75 percent (1 1 / 2^2 ) will
fall within two standard deviations. Furthermore, without knowing the exact dis-
tribution, we cannot say if they will be equally distributed around the mean. We
have to estimate this from our kurtosis and skew measurements. Similarly, to
encapsulate at least 50 percent, 67 percent, and 90 percent of all observations, the
standard deviations must be 1.41, 1.74, and 3.16, respectively.
When we’re building a trading system, we do not want the distribution of
trades to be normally distributed, but instead to fall into a few very distinctive bins
or categories, such as a certain sized profit and a certain sized loss. Thus, we know
exactly what we can expect from one trade to the next. Figure 2.7 shows one such
distribution of trades. In fact, if you ask me, a normal distribution of trades with-
in a specific market or time period indicates bad research and sloppy trading.
Remember that the normal distribution is a function of a random variable, and we
don’t want the outcome of our trades to be entirely left to chance, do we?TIME IN MARKET
Many traders and analysts pay very little or no attention to the number of trades a
system is likely to generate. However, this is very important information that will
give you a first clue to whether the system is suitable for you. The questions you
need to ask yourself are, “Does this system trade often enough and does it keep me
in the market enough to satisfy my need for action?” These are seemingly silly
questions at first glance, but the truth is that a specific system will not suit every-
body, no matter how profitable it is. If it doesn’t fit your personality or style of
trading, you will not feel comfortable trading it.
Even more important, however, is how much time the system is expected to
stay in the market, because time spent in the market also equals risk assumed.
Therefore, the less time you can spend in the market to reach a certain profit, theCHAPTER 2 Calculating Profit 31