|
PureBytes Links
Trading Reference Links
|
Why not test over a 5 years of data, then check your optimisation over the
other 5 years out of sample.
Cheers,
Graham
http://groups.msn.com/ASXShareTrading
http://groups.msn.com/FMSAustralia
-----Original Message-----
From: Chuck Rademacher [mailto:chuck_rademacher@xxxxxxxxxx]
Sent: Thursday, 1 May 2003 3:35 AM
To: amibroker@xxxxxxxxxxxxxxx
Subject: [amibroker] Backtesting Tip
I'm only offering this "tip" since I was finding it very difficult to
achieve what I was trying to do. Now that I've managed to streamline the
process, I thought that I would share my approach. Like most things that I
do, this will not be of interest to everyone here.
Here is my situation:
1. I backtest over 10 to 12 years of data.
2. I backtest my systems against all active and extinct stocks (13,000+).
3. I rely on my filtering to ignore stocks of no interest (low volume, low
price, etc.).
4. I would NEVER have more than one parameter set across all stocks.
The problems arising from this approach are:
1. Excessive amount of time required to test/optimise/backtest entire
universe.
2. AmiBroker limitation of 10,000 stocks residing in cache exacerbates this
problem.
3. Since I generally optimize/backtest over all of my data, I have nothing
left to test out-of-sample.
So, I want to have two randomly selected watchlists. One for designing,
testing and optimising plus one for seeing how results look out-of-sample.
I'm embarassed to tell you how I was making my random lists. But I will
tell you how I am doing it now:
1. I create an exploration with this code in it:
Filter = 1;
AddColumn (Random(),"random");
2. I run this exploration against "all" stocks (excluding composites,
indices, etc.).
3. I then sort the results of the exploration by my "random" column.
4. I then highlight the first half of the stocks (random = 0.00 to 0.49).
5. I then click on "add selected results to watch list" and select a watch
list to use.
6. I then highlight the reamining half of the stocks (random = 0.50 to
0.99).
7. I then click on "add selected results to watch list" and select the
second watch list to use.
8. I then rename my watchlists to something like "first half" and "second
half".
I now have one watchlist for using for research and one for testing
out-of-sample.
No doubt there is a better way to achieve all of this, but it is a 100%
improvement on the way I was doing it.
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 http://docs.yahoo.com/info/terms/
------------------------ Yahoo! Groups Sponsor ---------------------~-->
Make Money Online Auctions! Make $500.00 or We Will Give You Thirty Dollars for Trying!
http://us.click.yahoo.com/KXUxcA/fNtFAA/AG3JAA/GHeqlB/TM
---------------------------------------------------------------------~->
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 http://docs.yahoo.com/info/terms/
|