|
TradeStation
is required for the trial of our products.
If you do not have
TradeStation 6
and want to test it realtime
for free during 3 weeks, please click
HERE
SirTrade2000 is collection of
functions, indicators, systems and templates ready for use as Easy
Language codes and user functions.
It basically acts as a neurofuzzy runtime module for TradeStation
and allows to run the neurofuzzy systems built by the Safir-X Assistant.
Works with Tradestation 2000i and TradeStation 8.
The specola version Sirtrade2004
only works with Tradetation 8 and above.
But it has also other extra features
All of them are used as in the examples provided in the User's
manual .
You will even not need a fair understanding of Easy Language
to use them in your systems and studies.
The only thing to do is to modify the copy of the indicator, signal as explained in the
manual
( parts to modify are highlighted in colors in the user manual, and all the
example codes are commented).
The copy of the indicator created from the template matches the SirTrade2000 indicator
explained
Indicators deal with building trading data (TSD) files or evaluating realtime
training performances.
Signals are used to run FZB files.
We have greatly simplified the code: the templates call user functions where all the
tedious easy language and DLL calls are now hidden from the end user.
The templates designed in this manner are
now more readable than before.
-
Possibilities are endless, as we have provided the most flexible and
comprehensive parameters set available to the concept...
-
Output results are stored as TradeStation arrays, as well as the current RMS
error and the Reliability for each output produced.
-
Of course, you may use these values as a control for deciding to train again or
not, to use the given predictor or an other...
-
As all the complex calculations being done by DLL functions
embedded in TradeStation functions, therefore this will
not waste your Easy Language size code.
Return to top
We believe that this feature has never been made available before to the
TradeStation users' community.
It's well known that a neural predictor needs sometimes to be retrained (this is a kind of
sophisticated optimization),
this remark being valid for any trading system if you cannot afford a long and deep
drawdown period, leading to
a lack of confidence in it.
In this case, one dollar lost provides a different emotion intensity than one dollar
earned, yielding to doubt,
interpretation of the system, and generally an increase of losses.
Again, it's a common way to displace the problem when a valid (time stable) solution
cannot be found.
But the problem still remains if you cannot answer to the crucial question:
Return to top
When may I train again my predictor, my system?
There is no definite answer to this question.
Maybe there is no answer at all!
Maybe there is an alternate solution... by changing the question:
Fuzzy logic is a closer interpretation of classical technical analysis
than any other method (except for expert systems,
very difficult to program, and not so easy to modify later on).
Generally speaking, you will observe by yourself that a fuzzy logic trained system yields
to a better
generalization on unseen data.
You may know that the BackPropagation algorithm used to train the NeuroFuzzy rules was
quickly converging
"fairly well", not "perfectly" in a few iterations.
The idea presented here is to frequently train a FIS (Fuzzy Inference
System, as produced by The Assistant)
on recent data, and use the trained FIS during the next bars.
Possibilities are numerous, but we present here a framework from which you may try your
own concepts,
due to the high versatility of this tool.
Please keep in mind that this feature is time consuming when testing on
a huge database, and that a real-time
use needs to load only the minimum of bars necessary to follow the trading system
real-time.
Also notice that we work with PC machine, even powered by the most recent Pentium
processor,
they are still PC and not yet Cray machines .These considerations make realistic the use
of a few real-time
training systems together when using intraday data on the same computer.
Return to top
To conclude with these general concepts, the real-time training option
shows some theoretical advantages:
If you choose to train a FIS on each bar (or every "n" bars to
save time), or under user defined conditions
(noisy data, loss of accuracy) and use it for the next bar, it means that:
-
You will never reply to the question: "When do I train again?",
because you retrain continuously or so.
In this sense, you have not to find the answer because the question is not valid in this
case.
-
A well-known knowledge from system developer and testers is to
"train" on a given period,
and then to test on unseen data.
When both tests succeed, one may consider that the system is a valid one. Maybe.
But how long could it behave "correctly"? The same question arises later or so
(see above considerations).
When training in real-time in respect of the framework explained above, concepts like
"training set",
"testing set", "unseen data set" are inapplicable:
All the database may become unseen data for the real-time training system in
this case.
The basic consideration is that, when you decide to train the FIS, you
make it learn the most recent
past behavior of the market, and that you add this knowledge to the previous one (learned
during the initial training).
This local knowledge is removed (or not if you decide this) during the next real-time
training pass.
This is maybe not yet the Holy Grail, but a very promising way for
research.
We will provide you with the tool and some idea not previously available
(and never with fuzzy logic)
to the public. We have made some efforts to make it feasible and hope to provide a
powerful and understandable tool.
It's up to you now to test.
As powerful as it may be, it does not dispense of any form of intelligence, and at least
logical sense.
For example, training real-time on the 50 previous bars to predict swinghigh or swinglow
points could be dumb,
if we consider that we have less chance to see examples of such points to train the FIS
during these 50 bars.
In the same manner, using the same 50lookback period to predict a price change 10 bars
ahead may be disappointing.
Stay close to reality when deciding what you try to predict.
Return to top
Retraining on any user
defined condition:
The bold curve is the equity of the retrained system
here retraining when ADXis falling
 .
A performance summary comparison is also available
Retraining on the
equity Curve before the trade:
Return to top
Return to top
May improves any existing system by adding fuzzy logic to it !
Available here as a free trial ( download SIR2000D.EXE from the Download page)
Summary
-
NeuroFuzzy Logic at your finger tips within TradeStation. This is the
TradeStation complement program for Safir-X Assistant and Safir-X Workshop.
-
Seamless integration of multiple FIS within the same TradeStation code (Indicator
and/or system).
-
Realtime training during the market: A true approach of really self - adaptive
systems and indicators.
Click here to know MORE ON NEUROFUZZY LOGIC...
Return to top
Last modified:
July 05, 2007
|
