|
PureBytes Links
Trading Reference Links
|
<BLOCKQUOTE
>
<FONT face="Times New Roman"
size=2>-----Original Message-----From: tchan95014
[mailto:tchan95014@xxxxxxxxx]Sent: Sunday, April 20, 2003 4:37
PMTo: amibroker@xxxxxxxxxxxxxxxSubject: [amibroker] Re:
Parameter selection, MCS (for Thomas)Hi,
Chuck,Thanks for the confirmation.All my systems generate many
signals during the test history. Thus my question about 1%/99%. The system
itself is not that discriminatory as yours, hence every candidate can
generate many if not hundreds of trades during, say, 10 years
history.During the backtest, many of the potential candidates generate
negative returns over the history. I generally cut them off from the
final portfolio. [ the character shift of individual stock is always
at the back of my mind, but I just can not handle the amount of trades
coming, so I have do something about it. ]I am firstly attracted
to your posts when you mentioned about removng the top performers in
further finding optimal parameter sets, because just started about 2
months ago, I realized there are some superstars in some of my systems, I
suspected that they bend the parameter set so much, it is not
representative any more. In my test on N100, I can repeat the process you
mentioned 2 or 3 times until there is No super performers any more, I then
put every one back to the candidate pool and using the last parameter set
to prune the pool again for the final portfolio, then it is up to the
system as well as my pruning process to pick up or not pick up the
superstars.I force parameters shift, use student t test as well as
XLSim to help me gauge the stability of my system.I heard about
TraderSim but never really look into it, I will now. Using XLSim requires
extensive preparation because it is designed for general use.Thank
you very much for your posts.Thomas--- In
amibroker@xxxxxxxxxxxxxxx, "Chuck Rademacher" <chuck_rademacher@x>
wrote:> My calculations/indicators/functions are such that they take
very few trades> on any one stock. Over ten years of
data, each stock might trade (on> average) about five times.
Many stocks might not trade at all with one> system and very
frequently by another system.> > The only subtle difference
about what I am doing compared to many system> developers is that I
don't pre-screen stocks. I don't have various watch>
lists for different groups of stocks. No watchlist for
sectors. No> watchlists for high dividends.
No watchlists for low P/E ratios. No high> Zacks
ratings. No low Morningstar ratings. I have all of
that> information in my data and each system looks at all of those
things every> day for every stock. If I were to make a
watch list containing low P/E> stocks. Exactly what
timeframe would I use to make that determination.>
Today? Last week? Ten years
ago? Why not put it in the data along> with open,
high, low, close, volume, etc. for each stock for each day and> let
the system do the work?> > I do, however, have watch lists that
contain a purely random selection (out> of the 13,500 stocks) so
that I can try a new system idea without having to> go through all
of the stocks to see if it has merit. I have two
waitlists,> for instance, each containing exactly 50% of the
stocks. Basically, every> other stock. I
have four waitlists that each contain a random selection> of 25% of
the total stock universe.> > What are your
thoughts?> -----Original Message----->
From: tchan95014 [mailto:tchan95014@xxxx]> Sent: Sunday,
April 20, 2003 3:15 AM> To:
amibroker@xxxxxxxxxxxxxxx> Subject: [amibroker] Re:
Parameter selection, MCS (for Chuck)> > > Hi,
Chuck,> > Thanks again.> >
If I read you right, your system is exploiting some
fundamental> movement of the stock universe, since the
pattern does not occur very> often you make it up with
vast amount of candidates (for validity of> your test
results as well as trading opportunities). Sort of like
the> pair trading you talked about. Is this
correct?> > Then, do you have system like, for
example a RSI system that signals> generate on every
stocks in various to times? If yes, how do you> screen for
candidates then? In this case, a system might FIT only
say> 1% of the universe, because the test results show
other 99% over 10> years are net losers.>
> Thanks again for your insight.>
> Thomas> >
> Yahoo! Groups
Sponsor> > > > Send BUG REPORTS to
bugs@xxxx> Send SUGGESTIONS to
suggest@xxxx>
-----------------------------------------> Post
AmiQuote-related messages ONLY to:
amiquote@xxxxxxxxxxxxxxx> (Web page: <A
href="">http://groups.yahoo.com/group/amiquote/messages/)>
--------------------------------------------> Check group
FAQ at:> <A
href="">http://groups.yahoo.com/group/amibroker/files/groupfaq.html>
> Your use of Yahoo! Groups is subject to the Yahoo! Terms
of Service.Send
BUG REPORTS to bugs@xxxxxxxxxxxxxSend SUGGESTIONS to
suggest@xxxxxxxxxxxxx-----------------------------------------Post
AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx (Web page: <A
href="">http://groups.yahoo.com/group/amiquote/messages/)--------------------------------------------Check
group FAQ at: <A
href="">http://groups.yahoo.com/group/amibroker/files/groupfaq.html
Your use of Yahoo! Groups is subject to the <A
href="">Yahoo! Terms of Service.
Yahoo! Groups Sponsor
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.
|