|
PureBytes Links
Trading Reference Links
|
Tim McCaughey wrote:
> I was wondering if some of the more experienced system designers might be
> able to share some of their experience with a novice.
>
> I have been testing some entries for a trend following system that I have
> been working on. I have been using the testing method that is described in
> the Chuck LeBeau book. That is simply exiting the market after 5, 10, 15, 20
> days.
>
> I have been using 5 years of data, daily bars, across 6 markets
> (Oil,Gold,eurodollars,Yen,T-Bonds,Wheat) using a large number of different
> entries.
>
> What I have found is that most of then come out with % profitable of Between
> 56-59%. This figure represents the average % profitable for all of the
> markets for all of the time frames(average of 6 mkts * 4 times frames = 24
> results per entry). I have been getting 5-15% variation between markets and
> time frames for each entry but the average of all these is remarkably
> stable.
>
> Considering the different techniques I have been employing I have found the
> low variability between entry methods very interesting. Is this because I am
> unintentionally testing the same idea expressed a different way or is this a
> common experience.
>
Tim,
It's hard to pinpoint the reason(s) for small variation in results w/o more info
on the correlation between markets you're testing, etc. But, given the
information in your post, testing various (potentially correlated) entries and
fixing the exits, my first guess is that your entry signals are (as you say) are
simply the same idea expressed differently. You might try testing one market
using the various entries you've devised, and see if the entry dates are
similar.
But my main comment is on the subject of isolating the quality of an entry by
using exits of varying time frames (you mention 5-20 days) and then using the
results from this type of testing to make a judgement as to the validity of the
entry signal. I've read the very excellent posts by Dave DeLuca and Omega Man on
this subject but I feel both miss the mark to some extent. I'd like to extend
the thread and interject one idea for comments/criticisms/etc.
First of all, I like this method for isolating entries provided the proper basis
for distinguishing a profitable entry is used.
>From your post, it looks like you simply look at the system stats for each entry
given the random exits (5-20days) and then make a profitability determination. I
don't think this is correct and this is why.
For an entry to be "valid", it must be able to identify (for a long) a period
when prices will tend to go up and (for a short) a period when prices will tend
to drop. If found, the entry has a "positive expectation of profit" (as Dave
DeLuca points out in his response to Omega Man on this thread). If a particular
market had no trend then the expectation of profit would be zero and your
method, I believe, would be valid.
However, when markets get into long term trends (take INTC or MSFT for example)
most any entry technique will show an overall profit and be successful. But the
success is simply the result of the fact that the majority of the price action
is up. "Don't confuse brains for a bull market" is an oft-quoted saying that
embodies this trait.
To account for this background price action, I'd first determine a stock's
average return over various time periods (1, 2, 5, 10, and 20 days for example)
and then extend Dave's question somewhat to ask, "what is the conditional
probability that this entry makes money over various time frames given the fact
that the background price action has been such-and-such over those same time
frames". I answer this by comparing the returns 1, 2, 5, 10, and 20 days
subsequent to the signal to the average returns over the same periods.
If you find that your entry technique can identify periods (definitely more than
one) where prices rise more than the average (for long entries) and/or fall
faster than average (for short entries) I think you can say you've isolated a
good entry and at that point, you can move on to developing your exit.
Great thread!
Regards,
Bill Vedder
> thanks in advance
>
> regards
>
> Tim.
|