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

RE: Oddball and the Emperor's New Clothes



PureBytes Links

Trading Reference Links

>the question still remains on the market inefficiency that
>exists that makes 6-7 perform so well but not 5 in the limited sample
>period.

*7* is the number of hourly bars in an S&P day. I think the idea is
to compare today's ADV ROC to yesterday's at the same time of day. The
more you get away from 7, the more you compare apples to oranges since
it takes a while for the ADV numbers to fill in (even in a down day the
ADV's tend to increase as the day goes on).

BW


>From: "Wes Williams" <softexcl@xxxxxxx>
>Reply-To: <softexcl@xxxxxxx>
>To: "Omega List" <omega-list@xxxxxxxxxx>
>Subject: RE: Oddball and the Emperor's New Clothes
>Date: Wed, 20 Feb 2002 07:59:55 -0700
>
> >
> > Hello  Wes,
> >
> > looking  at your figures i can tell you something is wrong.  i can see
> > why  about  1  in 1000 people contact me about poor results.  i do not
> > have  the time to trouble shoot each individual case of what is wrong.
> > it could be a zillion things that have to do with data, settings of ts
> > and  even  what ts you run it on makes a difference, yes.  seeing that
> > you   and  a  few  other  have  blinders  on i sympathize that you are
> > frustrated.  but  don't shoot the piano player dude.  you numbers are
> > way   off,   not saying that's your fault but it's not mine.  have you
> > seen or are you aware of the real time posted trades?
>
>Thank you for your observations, Mark. With all respect, it is the very 
>lack
>of blinders that makes me question, not the system, but its robustness. Nor
>did I ever criticize you, anyone else, or the "piano player." Rather, I
>appreciate your contributions of a core strategy on which to build. Neither
>am I frustrated but I do point out that the results for 7,3,1 perform very
>well but 5,3,1 returned a loss ($10,550) for the period from
>3/27/01-present, not counting slippage and commissions. This is a fact and
>not a personal criticism and I am trying to understand why by asking 
>others.
>Since we all evaluate results without blinders, there is no objection in
>asking that question and dealing with it openly, wouldn't you agree? I have
>only been focused on the specific issue of robustness and I wish to confine
>my contributions on this thread to that topic. Recall in the story that it
>was the masses that had on blinders. I/we can arrive at the conclusion of
>where Trading System blinders are by examining the results with, as I
>suggest, simple robustness testing.
>
>I am not aware of the real-time posted trades. This may help partially
>answer the question. Where does one go for that? It may also be true that
>TS6 has a data problem with $ADV or @SP seeing that you observed that my
>numbers are off from yours. I am also limited by TS6 because I cannot go
>back farther than 3/27 and may need to do this on 2000i. I can email
>privately some thread contributors on this.
>
>To continue pursuing the issue of robustness, can someone do the following
>with OB? Run a parameter optimization using data as far back as you can go
>on only the first variable from 4 to 12 and either post or email the 
>results
>to me. We are looking for a rough bell curve and profitability across all
>variables and not jagged profitability peaks and valleys. If there is not a
>rough bell curve, the question still remains on the market inefficiency 
>that
>exists that makes 6-7 perform so well but not 5 in the limited sample
>period. If there is no explanation, then what would prevent market
>conditions from changing and making 9,3,1 perform optimally and 7,3,1 
>poorly
>in the future? This does not mean that the system is bad but means that
>extra caution is needed because of a weakness in robustness. Perhaps
>periodic reoptimization is a possible solution if the weakness exists but
>that raises other robustness-testing concerns.
>
>Sincerely,
>Wes Williams
>
>
> >
> > WW> I  agree with the principle of having a large in-sample period but
> > WW> optimizing against it all is not good for robustness. Let us apply
> > WW> the  generality  to  the  specific  of  OddBall  with Walk-forward
> > WW> testing and principles of robustness.
> >
> >
> >
> >
> >
> > --
> >
> > Have a Great Day, Mark
> >
> > http://www.markbrown.com
> >
>