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

RE: Precision Errors



PureBytes Links

Trading Reference Links

1************-
Of course, if you use 1000 or more maxbarsback settings, such side effects
may appear.
The question is: Who really uses 1000 or more bars is his current system
development?

Value1= average(c, 1000) does make sense to you ?

2********
Regarding to your  precision problem, there is always a workaround.
For example, you may use arrays elements to calculate a very long average.
Store the sum by slice of say 50 increment into an array element, divide by
50, then sum all the arrays results and your long term average will be
correct without losing so much precision ( 50 is an example and can be set
by param or internally to any  slice value, even with an incremental
automated value when the cell precision is up to limit).

This trick may be applied to SumSqr calculation used in correlation, StDev,
Regression.
I am always amazed to see how people are quick to bash without having the
idea of trying to solve with what they have and adding some use of their
brain.

3********
EL allow DLL Calls. If you do not want to use the EL trick abovve for  big
stats functions, you may code a DLL where you will use internal double
precision in it and pass back the results to EL with all the necessary
precision. There are  enough C developers here to build such a thing ( the C
code is even easy to find in a public C library).

Is it necessary to call for a better platform only for that ? Frankly I do
not think so...or it's an endless game.
The next step will be to have valid trading systems as strandard examples in
TS. And why not neurofuzzy logic for free ?
Your better platform quest is not at its end...Just starting.

4-********
The corelation fuction provided by Omega is not the standard formula
 Pearson). Not because of precision, but because it was not added since SWP
inception.
You may probably remember that you can write your own. I have attached one
provided on the late Scilink ( and if someone has some news from Bob
Brickey, please let me know privately, I would like to know how he is
doing).
The Peareson correlation code will be attached to the next message.


5-************
There is something new in TS Pro. See below and ask to TRAD what it should
mean:


 EasyLanguage Preferences - General Tab
==============================
On this tab, you can customize the general settings for EasyLanguage
documents. The following options are available:

 ž      Enable Smart Indent - Select this check box to enable automatic
indenting based on the context of your EasyLanguage instructions.

==============================
ž      Enable enhanced precision checking - Select this check box to apply
the floating precision point checking settings.

Note If you are using analysis techniques created prior to TradeStation
2000i and you experience difficulty with them while this setting is enabled,
you may want to disable this setting to resolve the problem.
==============================

ž      Tabstops box - Enter the number of spaces to indent when the Smart
Indent Enabled check box is selected.

ž      Function Keys - Specify the function key template that you want to
apply. Select the Office Compatible option to use the Microsoft Office
default or select the System Writer option to use the System Writer default.

ž      Advanced - To verify all of the analysis techniques in the
EasyLanguage document, click Verify All.


All the best

Sincerely,

Pierre Orphelin
www.sirtrade.com
TradeStation Technologies representative in France






> -----Message d'origine-----
> De : Neal Falkenberry, CFA [mailto:tnf@xxxxxxxxxxxx]
> Envoye : jeudi 19 juillet 2001 00:38
> A : Bengtsson, Mats; omega-list@xxxxxxxxxx
> Cc : pierre.orphelin@xxxxxxxxxxxxxx
> Objet : Precision Errors
>
>
> Mats,
>
> I just returned from traveling to catch your e-mails on TradeStation's
> precision.  I couldn't agree more.  Try using the coeffR function to
> calculate a correlation.  Correlation, StDev, Regression all require
> squaring large numbers.  This is a disaster with TS's single precision.
> With Currentbar > 1,000 the difference between TS's calculation
> of a 90 bar
> correlation versus Excel or C++ (which agree to 9 decimals) is .32 versus
> .56.  I call that significant.   To add insult, Omega adds code
> to constrain
> the calculation to +/- 1.  Correlation is by definition constrained tp +/-
> 1.  The only reason to constrain it is to hide the fact that you can't
> calculate it correctly.
>
> On an even simpler scale, calculate a simple 1000 bar moving average in TS
> and compare to Excel or C++.   TS can't even do this with precision to ONE
> decimal place.
>
>
> Neal
>
>
>
>
>
>