|
PureBytes Links
Trading Reference Links
|
Hi Dave –
I’ve been following your posts with
great interest, too. Thanks for the kind words.
My responses to yours follow each of your
paragraphs.
Thanks,
Howard
-----Original Message-----
From: Dave Merrill
[mailto:dmerrill@xxxxxxx]
Sent: Friday, October
17, 2003 <span
>5:45 AM<font size=2
face=Tahoma>
To: amibroker@xxxxxxxxxxxxxxx
Subject: RE: [amibroker]
Optimization -- again
<span
>great post
howard, interesting, thanks. couple questions, of course (:-), sorry for the
length.
<span
>- what are your
thoughts on what measures to use to evaluate past trading performance while
optimizing? I figured that percent growth per bar should optimize for simple
profit. it ignores risk, drawdown, etc, so it might not be the wisest metric in
a larger sense, but I'm wondering if it's less effective, even on a purely
profit level, than some other approach.
HB: I’ve just posted a response
to some other inquiries about performance. Look for one title “Objective
Functions”.
<span
>- keeping the in-sample
test period short doesn't mean that conditions won't change during that time,
unless you can somehow align optimization periods to coincide with the points
at which conditions do change. besides, isn't coping with changing market
conditions the main thing a trading strategy needs to be able to do?
<span
>
HB: Right. But the market has
so many dimensions that we cannot see them all, know which are changing, and in
what way. If the conditions change within our in-sample period, that
makes our model fit less well; the out-of-sample performance may be good or bad.
If conditions change within the out-of-sample period, that drops our (simulated
or real) profitability. My feeling is that a single trading strategy
cannot copy with many changes. In order to do so, it will require an
increase in the number of parameters, which increases the likelihood of a
curve-fit solution, which decreases the likelihood of profitable out-of-sample
performance. I prefer to make the in-sample and out-of-sample time
periods shorter until the model fits fairly well throughout the period.
<span
>- I wouldn't go
so far as to say that as soon as we've looked at system performance on a given
piece of out of sample data once, it's no longer useful. you just can't
tweak your model with optimizations on data that's supposedly out of
sample. you're right that you need to be very clear about what's in
sample data and what's out, but that's not the same as saying that data is
"ruined" for further testing once we've seen it at all.
<span
>
HB: You are correct. I’ve
just posted a couple of responses to others who also asked about this. Look
for others titled “Re: Optimization – again”.
<span
>- I understand
that if many people use a specific trading system, each person's profits
subtract from the others' gains. I understand also the general idea that a
profitable trading method represents inefficiencies that over time the market
will reduce. but let's say you just trade your method without publicizing it,
and your trades are a small enough fraction of overall volume that their
pattern doesn't draw much conscious attention. by what mechanism does the
market as a system "realize" what you're doing and compensate, in a way
that your method's profitability disappears? on an intuitive level I see that
that should happen, I just don't get how. for instance, if my trades are based
on some complex formula with many inputs, it seems unlikely that it will get
reverse engineered and intentionally copied, especially if I'm a pretty much
below-the-radar small trader. so how then does my individual trading cause the
market's behavior to shift?
<span
>
HB: I agree. If you have a
system that you are trading alone, and you are trading a very liquid stock or
future, you will have no influence. In stocks, Intel and Microsoft trade
billions of dollars per day. There is no way I can influence that
market. But, if several large players begin using the same system, it
will change the way the market acts.
<span
>- do you have any
thoughts on the idea of constant automatic optimization, mechanically choosing
system parameters every day based on past performance? if past performance is
any guide to future results, I can't see any logical reason why this
shouldn't be the best way to go. if doing this doesn't produce profitable
results, which so far in my experience it mostly doesn't, why doesn't that mean
that past performance *doesn't* predict future results, invalidating the whole
notion of backtesting and trading analysis? I don't really buy that conclusion,
I'm looking for a hole in the logic.
<span
>
HB: If there are not persistent
patterns and trends that we can identify and trade, then we are all wasting our
time!! In my post, I Am advocating automatic optimization. I am
perfectly comfortable if your re-optimization period is one day. The only
thing that counts is the out of sample performance of your system. One of
the difficulties with re-optimizing every day is handling trades that last
longer than one day. You might find the new optimization does not show
you in a trade that the previous one put you in, and vice versa. I
recommend that the out of sample period / re-optimization period be longer than
the typical holding period of a trade. Whatever length of time you
choose, you will need to deal with the “edges” at the start of each
new period. This will require some research on your part. One
approach is to exit trades that disappear, but wait for new signals to make new
entries.
<span
>thanks again for
stopping by here, I hope we'll see you around,
<span
>
HB: AmiBroker is already head and
shoulders above the other system development platforms. I’m expecting
to stay with it as it develops in the areas of trade and portfolio management.
At the risk of getting beaten up a bit, I’ll continue to contribute to
this forum.
<span
>dave
In
my opinion, anything we do in development of trading systems involves a
search for a
pattern than precedes a profitable trading opportunity. Any
time we
examine the results of alternative systems, we are involved in
searching;
and when we select the most promising of those alternatives, we
are
optimizing. Only a system based on truly random entries and exits would
not be the
result some optimization. So the question of "should we
optimize?"
is moot -- we have no choice but to optimize. Consequently, we
should be
aware of our optimization techniques.
<<<<< SNIP >>>>>
Yahoo! Groups Sponsor
ADVERTISEMENT
Send BUG REPORTS to bugs@xxxxxxxxxxxxx
Send SUGGESTIONS to suggest@xxxxxxxxxxxxx
-----------------------------------------
Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx
(Web page: http://groups.yahoo.com/group/amiquote/messages/)
--------------------------------------------
Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html
Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.
|