ch01 JWBK035-Vince February 22, 2007 21 : 43 Char Count= 0
The Random Process and Gambling Theory 37as,“Are bigwinningtradesgenerally followed by biglosingtrades?” “Are
biglosingtradesgenerally followed by little losingtrades?”And so on.
Negative correlationisjust as helpful as positive correlation.For ex-
ample,if there appears to be negative correlation, and the system hasjust
suffered a large loss, we can expect a largewin, and would therefore have
more contracts on than ordinarily.Because of the negative correlation,if
the trade proves to be a loss, the loss will most likely not be large.
Finally,in determiningdependency you should also consider out-of-
sample tests.Thatis, break your data segmentinto two or more parts.If
you see dependencyin the first part, then seeif that dependency also exists
in the second part, and so on.Thiswill help eliminate cases where there
appears to be dependency whenin fact no dependency exists.
Usingthese two tools (the runs test and the linear correlation coef-
ficient) can help answer many of these questions.However, they can an-
swer them onlyif you have a high enough confidence limit and/or a high
enough correlation coefficient (incidentally, the system we used earlierin
this chapter, which had a confidence limitgreater than 95%, had a corre-
lation coefficient of only.0482).Most of the time, these tools are of little
help, since all too often the universe of futures system tradesis dominated
byindependence.
Recall the system mentionedin the discussion of Z scores that showed
dependency to the 95% confidence limit.Based upon this statistic, we were
able toimprove this system by developingrules for passingtrades.Now
hereisaninterestingbut disturbingfact.That system had one optimize-
able parameter.When the system was run with a different value for that
parameter, the dependency vanished!Was this sayingthat the appearance
of dependencyin our cited example was anillusion?Wasit sayingthat
onlyif you keep the value of this parameter within certain bounds can you
have any dependency?If so, thenisn’tit possible that the appearance of
dependency can be deceiving?To an extent this seems to be true.
Unfortunately, as traders, we most often must assume that dependency
does not existin the marketplace for the majority of market systems.That
is, when tradingagiven market system, we will usually be operatinginan
environment where the outcome of the next tradeis not predicated upon
the outcome(s) of the precedingtrade(s).Thisis not to say that thereis
never dependency between trades for some market systems (because for
some market systems dependency does exist), only that we should act as
though dependency does not exist unless thereis very strongevidence to
the contrary.Such would be the caseif the Z score and the linear correla-
tion coefficientindicated dependency, and the dependency held up across
markets and across optimizeable parameter values.If we act as though
thereis dependency when the evidenceis not overwhelming, we may well
just be foolingourselves and cause more self-inflicted harm thangood.