|
PureBytes Links
Trading Reference Links
|
Chande himself tested synthetic data extensively.
Here's what he did. He tested a system on a continuous contract Swiss francand
then generated 8 series of synthetic data from this real continuous contract
with the following conclusions:
1-
Performance can vary significantly for
different synthetic data series for the same market (stock) which comes as no
surprise. He advises to generate 5 - 10 series for the same market and average
the results. The average results are a very good approximation of the results on
the original data set
2-
One can use the standard deviation and other
statistical parameters to estimate future drawdowns, risc of ruin calculations
and expected
future performance. The same 'boundary conditions'
as to real data remain, i.e. it are merely estimates.
He also describes one limitation of scrambled data.
The new patterns are synthetic and as such cannot represent real market
psychology and real supply and demand. Many test (like the one described above)
suggest that these limitations can be overcome by using average
system performance on synthetic data.
Regards
Leo
<BLOCKQUOTE
>
----- Original Message -----
<DIV
>From:
Al Venosa
To: <A
href=""
title=amibroker@xxxxxxxxxxxxxxx>amibroker@xxxxxxxxxxxxxxx
Sent: Thursday, June 20, 2002 6:40
PM
Subject: Re: [amibroker] Trends, random
series
This is a fun discussion.
Bill, you said "But once in a while is not good enough, as a result the
general applicability of Chande's approach is questionable, although as I
noted it might be OK for non-pattern based indicators (e.g., MACD, etc.)."
Perhaps you are right about that. I don't know. I would ask, however, how
you define "once in awhile." You can be a successful trader with a systemthat
generates only 20% winners if the system being traded is a positive expectancy
system (i.e., if the avg win/avg loss ratio is high enough). I'm convinced
that money management is the key (Holy Grail, if you will) to success in
trading. If you trade a system that gives you 80% losers, but those losses are
tiny compared to the occasional huge winner, you can do quite well over the
long run. It's going to take lots of intestinal fortitude to experience 80%
losers, but I'm sure there are those out there who do.
Look, all I'm saying is that the scrambler is just another tool that can be
used by us in Amibroker. That's all. I'm not advocating trading random price
patterns. I'm just saying that it might be useful to test your system on
synthetic data. After I've done it, I'll report back and let you guys know
what happened. Like I said, I'm willing to eat Markov monkey if I'm wrong.
|