[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[amibroker] Re: Walk-Forward Out of Sample (OOS) Testing



PureBytes Links

Trading Reference Links

Chuck,

By "Stuck in the mud" do you mean you have yet to find a system that 
is viable using this type of approach ?  Work I did some years ago 
seemed to suggest this but the engine I was using wasn't really 
powerful enough to thoroughly test this theory to a satisfactory 
conclusion.

--- In amibroker@xxxxxxxxxxxxxxx, "Chuck Rademacher" 
<chuck_rademacher@x> wrote:
> I'll probably only confuse things, but I'll describe how my walk-
forward
> code works.   I would share the program with you, but it is some 2 
million
> lines of C++ and has nothing to do with AFL, AB, ABTool, etc.   I 
use this
> approach only for futures markets, where I do have different system
> parameters for each commodity.
> 
> I WOULD NEVER DREAM OF USING DIFFERENT PARAMETERS FOR  STOCKS, 
EITHER UNIQUE
> TO EACH STOCK OR UNIQUE TO DIFFERENT TIME PERIODS.
> 
> In any event, I can tell my backtester a start point, an in-sample 
window
> size and an out-of-sample window size.  I also tell it which of 
over 200
> objective functions I want to use when making the decision as to 
which
> parameter set to use for the next out-of-sample data.   An objective
> function might be something as simple as KRatio, or 
Netprofit/Drawdown,
> whatever.    The in-sample window is used for "learning" a 
parameter set and
> then my software measures how well the system does out-of-sample.  
I can
> tell you now that more than 95% of the systems tested in this 
manner are
> rejected at this stage.
> 
> I cannot envisage another way to do walk-forward testing, but then 
I've been
> stuck in the mud using this approach since 1960.
>   -----Original Message-----
>   From: Fred [mailto:fctonetti@x...]
>   Sent: Wednesday, June 18, 2003 2:06 AM
>   To: amibroker@xxxxxxxxxxxxxxx
>   Subject: [amibroker] Re: Walk-Forward Out of Sample (OOS) Testing
> 
> 
>   b,
> 
>   Actually you bring up some interesting points one of which I 
hadn't
>   initially thought of but of course that is why I was asking the
>   questions I was i.e.
> 
>   1.  Are we testing some optimizable system oriented parameters for
>   some fixed user selectable lookback period or window which is 
either
>   based on an anchored first data point or is a moving window as 
time
>   goes along ... or ...
> 
>   2.  Are we testing a static system for some optimizable lookback
>   window ? ... or ...
> 
>   3.  All of the above ? ... or ...
> 
>   4.  Something else ?
> 
>   I suspect it's #1 as I'm not sure optimizing the lookback period 
has
>   merit ... or at least not more than once per full run i.e. we 
might
>   want to see how changing the size of the lookback period or window
>   size affects the results overall as we might want to see how using
>   different lengths of out of smaple data affects the results 
before we
>   reoptimize but I don't think having varying lookback periods or
>   window sizes has merit, do you ?  This "feels" way to curve fit.
> 
>   Fred
> 
>   --- In amibroker@xxxxxxxxxxxxxxx, "Fred" <fctonetti@xxxx> wrote:
>   > It's okay ... I think Mark thinks it's over my head ...
>   >
>   > --- In amibroker@xxxxxxxxxxxxxxx, "b519b" <b519b@xxxx> wrote:
>   > > Fred,
>   > >
>   > > I think you and Mark have misunderstood each other over the 
use
>   of
>   > > the terms "segment" and "period". Like you, I first read 
these as
>   > > synonyms, but after re-reading Mark's post, it appears they 
are
>   > > completely distinct.
>   > >
>   > > Reading them as synonyms gives the following:
>   > > Test 3 time segments. Select the best period (=time segment) 
and
>   > use
>   > > the parameter setting from that segment. As you point out 
such an
>   > > approach would problematic.
>   > >
>   > > Reading period and segment as being distinct gives:
>   > > Test the 3 time segments. Select the best period (=parameter
>   > > setting), ie, select the look-back period in the sample code
>   which
>   > > gives the best profit factor for all 3 of the time segments. 
Use
>   > > this parameter (ie, look back "period") for time segment 4. 
And
>   so
>   > > on.
>   > >
>   > > I think Mark has the second approach in mind. But Mark can
>   correct
>   > > me if I am wrong.
>   > >
>   > > b
> 
> 
>         Yahoo! Groups Sponsor
> 
> 
> 
>   Send BUG REPORTS to bugs@xxxx
>   Send SUGGESTIONS to suggest@xxxx
>   -----------------------------------------
>   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.


------------------------ Yahoo! Groups Sponsor ---------------------~-->
Get A Free Psychic Reading! Your Online Answer To Life's Important Questions.
http://us.click.yahoo.com/Lj3uPC/Me7FAA/ySSFAA/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/