number of profitable markets continued higher to 76 percent, the average profit
per trade increased to $3,577, the profit factor to 2.23, and according to the risk
factor, we’re now making a whopping 72 cents for every dollar risked.
At first glance, this looks very good indeed, but as already pointed out earli-
er, when a system produces results this good, they might be too good to be true,
and these results are no exception to that rule. Let’s start by comparing the aver-
age profit per trade with the standard deviation for all average profits. Note that
the standard deviation, at $6,094, is almost twice as large as the average profit of
$3,577, which results in a one-standard-deviation interval stretching from
$2,516 to $9,672. Thus, this system still produces a lot of trading sequences with
an average profit per trade below zero, and the chances for the true, always
unknown, average profit to actually be negative are quite high. This annoying fact
is also confirmed by the profit factor that has its lower one-standard-deviation
boundary at 0.81 and the risk factor with its lower one-standard-deviation bound-
ary indicating a loss of 17 cents per dollar risked. Another unwanted feature of this
system is that the average trade length increased from around six days to 35 days,
which no longer can be considered short-term and may be too long when using the
system as a filter for other systems.
The high standard deviations probably ruined the original version of this sys-
tem. Testing it on only three markets, unknowingly and unwillingly, I managed to
produce a system with a decent hypothetical average profit per trade, but also a
very high standard deviation, thus making it very likely that the system would lose
plenty of money over long periods, which was exactly what happened during the
period following the original testing period.
Granted, the upper one-standard-deviation boundary for the average profit
indicates also a high likelihood that the system will perform better than indicated
by the original research, but good periods are easy to deal with. Of the really bad
periods, we only need one to end our careers as traders (and find a cheaper place
to live). The trick is to keep the likelihood of this happening as small as possible;
for that we also need to sacrifice some upside potential.
Before we move on to the short side, I would like to make one more adjust-
ment, and that is to equalize the lookback periods for the entry and exit formations
that now are set to two and four days, respectively. I would like to do this because
otherwise this system simply has too many adjustable and optimizable variables,
and as we will see later, when we talk about stops and exits, finding robust solu-
tions for more than three variables is very difficult. It also is important that we
keep the entry as simple and logical as possible, with as few rules as possible.
Therefore, if it turns out we’re better off without these variables, or if the results
are inconclusive, the variables will be scratched. If not, they will both be set to the
same value.
As it turned out, the results were rather inconclusive, so both variables will
be scratched. The results of this version of the system can be seen in Table 11.5.CHAPTER 11 Relative Strength Bands 129