|
PureBytes Links
Trading Reference Links
|
Volker:
Sometimes our intuitions fool us. I understand your doubts, but,
here, intuitions tells us one thing, while methodology arguments
tell us the opposite. Once you think through your methodology,
you may note that it IS legitimate for results to change whenever
we change the underlying data. In fact, the results SHOULD
change.
The classic explanation for change in results is given by Pierre
DUHEM in 1905 "La Theorie Physique, Son Object et Sa Structure".
(The English translation is "The Aim and Structure of Physical
Theory", New York, Athenium, 1962. There are also a prior
translation from 1954, published by University of Chicago if my
memory serves me well.)
The most famous of modern arguments dealing with this issue is
given by Imre LAKATOS, in "Criticism and the Growth of
Knowledge", Cambridge Univ. Press; 1970 (with several reprints).
See the essay by LAKATOS 'Falsification and Methodology of
Scientific Research Programmes'. The essay is brilliant!
Back to rolling data:
Roll by calendar data, Roll by Open Interest, Roll by Volume are
roll TRIGGERS.
A trigger generates an EVENT to do something.
Roll by Ratio Adjusted is NOT an event, it is an algorithm.
Other algorithms are also possible including, Gann, Perpetual, no
gap adjustment, etc.
Additionally, you can roll forward, or backwards, again, by date
Trigger, OI, Trigger, with Point Adjustment, Ratio, adjustment,
Perpetual, etc.
Bottom line: about 100 different kinds of continuous contracts
can be built. Each one is valid, but some are more useful for
certain styles of trading than others. For the style of trading
that I do, point based back adjusted roll, triggered by Open
Interest is conceptually suitable. For my trading, Ratio
Adjusted algorithm would work about as well, but I know of only 2
vendors who support Ratio Adjusting Algorithms. Therefore, to
have more vendors support my trading style, I adjust on a point
basis.
Do you know any vendors that support Ratio Adjusting (other than
Pinnacle and CSI)?
If so, kindly post.
Thanks,
Leslie
VK wrote:
>
> Hi.
>
> I am always a bit suspicious if results change because of the roll over
> method. Especially if they go from positive into negative. I came to the
> conclusion that RatioAdjusted data seems to be the most realistic data
> to use. What do you think?
>
> Any way, the best thing would be to be able to use the REAL contracts.
>
> Regards.
>
> Volker Knapp
> Wealth-Lab Inc.
> www.wealth-lab.de
> www.wealth-lab.com
>
> -----Ursprüngliche Nachricht-----
> Von: Leslie Walko [mailto:l.walko@xxxxxxxxxxx]
> Gesendet: Montag, 27. Januar 2003 03:00
> An: Trey Johnson
> Cc: omega-list@xxxxxxxxxx; realtraders@xxxxxxxxxxxxxxx
> Betreff: Re: Roll over methods
>
> Trey:
>
> Here is the Spread Utility. (ELS enclosed.)
>
> I check the data when setting up a new machine, switching data
> vendors, after installing a new version of the downloading /
> contract building software, when changing rolling methods, etc.
>
> Differences in data can be huge depending on rolling method.
> For a while, I rolled on Open Interest, 1st trigger, but CSI had
> problems with Open Interest. I switched to rolling by fixed
> date, but that threw off my system and generated losses. Since
> CSI fixed their bugs, I retested everything and back to rolling
> on Open Interest.
>
> If you roll on fixed calendar dates, you might want to use the
> dates the Pinnacle has posted on their web site. Their choices
> correspond very very closely to when Open Interest rolls would
> take place. Good choice.
>
> You can also roll on Volume, but need to use at least three
> consecutive triggers because volume is less stable than OI.
>
> Last but not least, here is beginning and end of a comparison
> between two Five Year Note contracts, both rolled on the same
> machine, both rolled using CSI Unfair Advantage 2.5.4 but one
> rolled using fixed date, the other rolled using OI. Note the
> difference.
>
> Keep in mind, both roll methods are valid. You just need to pick
> one and stay with it.
>
> Beginning of SAMPLE:
> File 1 FV20@xxxxxxx File 2 FV21340B.TXT
> Date Data 1 Data 2 Point Diff $ Diff
>
> 900102 70.7969 71.1406 -0.34375 -343.75
> 900103 70.6094 70.9531 -0.34375 -343.75
> 900104 70.6406 70.9844 -0.34375 -343.75
> 900105 70.625 70.9688 -0.34375 -343.75
> 900108 70.5781 70.9219 -0.34375 -343.75
>
> End of SAMPLE output:
> 1021224 112.547 112.547 0 0 ___OK___
> 1021226 112.656 112.656 0 0 ___OK___
> 1021227 113.203 113.203 0 0 ___OK___
> 1021230 113.406 113.406 0 0 ___OK___
> 1021231 113.25 113.25 0 0 ___OK___
>
>
> Start Date 900101 End Date 1030101
> The spread changed 64 times during the study.
>
> Maximum bar to bar change in spead occurred on 990226 valued at
> 109.375 dollars. Data 1 was not equal to Data 2 for 2929
> of 3268
> bars -- namely 89.6 % of bars.
> Cheers, Leslie
>
> Trey Johnson wrote:
> >
> > Thanks Leslie. If you don't mind, please post the utility. I also
> noticed a
> > difference when I tried different roll methods with the same vendor.
> > However, they weren't as drastic as the difference between the two
> vendors.
> > Which method of rolling do you use, calendar based or liquidity based?
> How
> > did you decide? Thanks for the warning about different computers, I'm
> > getting a new one in a couple of months.
> > Best Regards,
> > Trey
> >
> > -----Original Message-----
> > From: Leslie Walko [mailto:l.walko@xxxxxxxxxxx]
> > Sent: Friday, January 24, 2003 12:22 PM
> > To: Trey Johnson
> > Cc: omega-list@xxxxxxxxxx; realtraders@xxxxxxxxxxxxxxx
> > Subject: Re: Roll over methods
> >
> > Trey:
> >
> > Welcome to the mystery of rolling.
> > Even if you use the SAME vendor and use the same raw contracts to
> > base your rolls on, different roll methods supported by the same
> > vendor will produce huge differences in results. Then, to make
> > it worse, the SAME vendor, SAME roll method, and SAME software
> > can also produce major differences if the data is run on two
> > separate computers.
> >
> > Confused?
> > When I tested the same vendor, same rolling method, same
> > settings, but different computers produced over 3,000 dollars
> > worth of spread in the Heating Oil contact between 1990 to
> > present. Now imagine what happens to your trading results when
> > the underlying data is so different!
> >
> > I wrote a utility to measure and log the differences. In fact, I
> > need to do some checking and revise the utility. If desired, I
> > can post the updated utility tomorrow.
> >
> > Your immediate problem:
> > If you change vendors, or rolling methods, or computers:
> > (1) test your data for differences using a utility;
> > (2) re-test your trading system ;
> > (3) you may need to reoptimize the trading system;
> > (4) you may need to redesign and revise the logic of your trading
> > system.
> >
> > Why the retesing and re-design?
> > Different vendors use different definitions of Open, Close, High,
> > Low. Each vendor sets their own method of rounding (or not
> > rounding) to nearest tick, or offers ability to be set by the
> > user. A vendor may carry quotes at one half tick, one quarter
> > tick, or may not be able to carry one quarter ticks, etc. (For
> > example, CSI is unable to quote 2-year Notes in the exchange
> > specified format of 1/32nd and 1/4 of 1/32nd -- equal to 1/128th
> > .)
> >
> > Net result: No system will ever perform the same with 2 different
> > vendor's data or rolling method. Each combination is a unique
> > system.
> >
> > Leslie
> >
> > Trey Johnson wrote:
> > >
> > > Hello All,
> > > This has been driving me crazy and I just can't seem to
> figure it
> > out. I
> > > recently changed data vendors. The new data rolls on specific
> calendar
> > days
> > > for each commodity when building continuous back adjusted contracts
> on
> > daily
> > > data. The previous data rolled based upon liquidity, when volume and
> open
> > > interest of the next month exceeds the current. I tested four
> different
> > > systems on the Yen and Dollar Index. In each case, the average trade
> and
> > > profit factor of each system was superior on the liquidity roll
> contracts.
> > > Then I tried three different systems on Natural Gas, again, better
> > > performance on the liquidity rolls. Even current open positions were
> more
> > > profitable for the liquidity based rolls than the calendar based.
> When I
> > > asked the new provider why they chose specific days, the response
> was, we
> > > made that decision ten years ago and I don't remember why. Well,
> that's
> > > certainly not the answer I expected. Has anyone else noticed this?
> Does
> > > anyone have any possible explanations as to the discrepancies in
> > > performance? Do I need to take something else in consideration,
> other than
> > > just the back testing results, in deciding which data to use? To me,
> it
> > > seems rather logical to roll based upon liquidity.
> > >
> > > Confused,
> > > Trey
> >
> > --
> > Regards,
> > Leslie Walko
> > 610-688-2442
> > --
> > "Life is a tragedy for those who feel, a comedy for those who
> > think"
> > Horace Walpole, 4th earl of Orford, in a letter dated about
> 1770
>
> --
> Regards,
> Leslie Walko
> 610-688-2442
> --
> "Life is a tragedy for those who feel, a comedy for those who
> think"
> Horace Walpole, 4th earl of Orford, in a letter dated about 1770
--
Regards,
Leslie Walko
610-688-2442
--
"Life is a tragedy for those who feel, a comedy for those who
think"
Horace Walpole, 4th earl of Orford, in a letter dated about 1770
|