PureBytes Links
Trading Reference Links
|
I have developed an indicator which flags a possible change in trend
(CIT)
based on a volatility breakout of some patterns. I noticed that what end
up
being false breakouts often terminated at the same breakout filter i.e.
the pattern plus a breakout filter.
For example, a filter of 2 cents in wheat (hypothetical value) might
lead to
lots of false CITs whereas 2.5 cents does not.
What I am saying is that the same value repeats through the history of
the
market in question. In this case you will see a breakout of the patterns
by
2 cents then the old trend resumes.
In my observations this value is never 1 tick.
The next step is to quickly find this value for each market or stock as
to
do it manually would severely limit the number of items that could be
researched.
In general it would probably be better to use an Average true range type
filter rather than a fixed value.
For a CIT from short to long:
maxlist(atr(4),atr(20)) * longmult where longmult is an input variable
to 2
decimal places.
For a CIT from long to short:
maxlist(atr(4),atr(20)) * shortmult where shortmult is an input variable
to
2 decimal places.
My premise is that the market does not "know" about my or anyone else's
patterns, therefore for breakout filter testing purposes we can just
test a
simple system that uses something like h > h[1] +
maxlist(atr(4),atr(20)) *
longmult.
I imagine I can use optimisation to find the value that gives the best
results.
Keep in mind this is not a standalone entry system but is to be used as
part
of a pattern based trend filter.
Given the goal of the test, can anyone suggest a sensible stop for such
a
system ? For example, if periods since entry = 1 then exit. The next
paragraph addresses a possible problem with this type of exit.
One other thing to consider is that even after a valid CIT breakout the
market often retraces so that x periods later whilst there has been a
CIT
the market is back trading around the breakout level then the new trend
resumes with force. However if the periods since entry stop is 1 then
each
new thrust will trigger a new trade and a new exit 1 period later
therefore
increasing the sample size.
Given the goal of the test, what would be the best basis for the
optimization i.e. % winners, ?????????????
In summary although I think I can use the optimization feature in
Supercharts ( ver 4 ) to achieve my goal but I am not sure of the right
questions to be asking and therefore the coding required.
I would appreciate your ideas on how best to address the problem.
Chris Jackson
|