hood to experience a streak of at least a certain number of winners or losers. For
example, the likelihood to experience at least nine winners or losers in a row is just
below 10 percent. The leftmost line shows the likelihood to experience two trad-
ing sequences of the same length within 100 trades. For example, for five winners
or losers in a row, the likelihood to experience two such sequences is below 50 per-
cent, and to experience two sequences of nine winners or losers in a row, the like-
lihood is almost 0 percent.
Note, however, that all this assumes that you’ve done everything correctly,
the logic behind the system makes sense, and all other system statistics are reli-
able. Also, this statistic works both ways: If you just experienced a large number
of winners in a row, it doesn’t mean that the system has gotten much better and
will continue to produce the same type of winning streaks in the future. Quite the
contrary: Depending on the size of the winning streak, the more difficult it will be
to repeat it. It’s all a game of statistics. Nothing else.The Difference Between Trades and Signals
During the initial testing of a system, we really shouldn’t concern ourselves with
the performance of the system, but rather the reliability of the signals it generates.
To do that, we really need to test all the signals generated, using standardized stops
and exits—such as, always exit after a specific profit or loss or after a specific
number of bars in trade.
Normal system testing usually only tests the first signal in a cluster of entry
signals and assumes that the others can be ignored because you would have been in
a trade already. However, whether you would have been in a trade or not depends
on your stop-loss and profit-taking levels and how long the trade should last if none
of these levels is hit. Plenty of other reasons could exist for why the first signal
might have been ignored in real-life trading, such as already being fully invested
because of previous trades, or simply because you missed trading that day.
Therefore, when researching a system, it’s important to look at all the signals
it generates, so that it can be trusted just as much on its second or third signals as
it can on its first signal. For example, assume you missed a signal one day because
of a doctor’s appointment. The next day, you get a new signal in the same stock,
despite the fact that you should have been in this stock already had you been able
to trade the previous day. If you haven’t done your research properly, you have no
idea whether the second signal is likely to be as valid and reliable as the first one.
On a more hypothetical note, there also are plenty of days when the system
just barely manages to produce a signal, or conversely, just barely misses produc-
ing a signal. (Should the near misses perhaps be as valid as the near hits? We won’t
answer that particular question here, but it’s still worth asking.) Look at Figure 3.7,
which shows a chart of all the signals that one of the systems featured in Part 2
generated in Microsoft during the fall of 2001. It shows that in early August 2001,CHAPTER 3 Probability and Percent of Profitable Trades 43