|
PureBytes Links
Trading Reference Links
|
Dans un courrier daté du 26/08/98 03:09:22 , TJ écrit :
<<
joe krutsinger's (oh-oh man) and pete orphelin's real agenda is to
sell their stuff at ridiculously inflated prices.
they don't give a damn about omega or ts, only joe's system de jour
($5K) and pete's nutty fuzzy thingy ($4K) sales are all that matters.
no ts, no sales, no bucks. it's as simple as that.
so what else is new?
TJ
who says snake oil is dead
>>
What's new Mr TJ ?
First, I'm revamping my web site and will offering free versions of a lot of
TradeStation code with also a lot of explanation, that I had no time to
finish.
"Free" means nothing to pay to test it and use it for your own systems or
indicators .
If you want to but them after, it's your problem, not mine.
Where is the snake oil ?
Among them several DLL's for TS4:
*MEM also known as maximum entropy methos (or spectrum analysis). (Full code
with power spectrum)
*Lomb Periodogram
*FFT Filtering
*Point & Figure DLL with variable inputs (allow variable Pt &Figures)
*Long term overlay profiles DLL (peaks and valleys automatic detection)
*Market Profile
Second, we are polishing the neurofuzzy thing Version 2.0 (alpha).
A free evaluation version will be available too.
Third, and only to break your nerves, here is a copy of the latest review of
the nuttyfuzzy thing published in Futures and Options World Magazine (August
1998 Issue -by Andrew Webb).
And do not claim for the wild advertising and netiquette outrage:
This is the price to pay to recover from the damages produced by your
infamous message on this list.
Do not bother NATO and UN for this unfair publishing.
They have a lot of work to do on this planet before solving for the frog case.
Enjoy or delete!
Regards...
Pierre Orphelin
http://www.sirtrade.com
==============
Warning:
Wild ad starts here:
==============
SUIT US, SIR
Sirtrade International has come up with a
unique, trader-friendly way of combining
the Power of Neural Nets with the
sophistication of Fuzzy Logic,
Andrew Webb Reports.
For the trader hunting for a new trading methodology, artificial
intelligence techniques have a particular allure. The concept of your
artificial expert tirelessly perusing a great mound of data, identifying
subtle relationships that would defeat the perception and stamina of a human
equivalent is undeniably attractive. Unfortunately, though the situation has
much improved in recent years, the snag with these techniques has been that
many of the available packages were rather less than trader friendly. Hours
of pre-processing data and grappling with a user-hostile interface soon
exhausted the limited attention span of the trader and saw the concept (and
software) swiftly consigned to the shelf or bin.
Pierre Orphelin of SirTrade International in Paris appears to have taken
these problems very much to heart when designing the company’s SirTrade
neurofuzzy (a combination of neural net and fuzzy logic) trading application
and its accompanying Assistant for Expert Traders. The two packages interact
to allow the trader to develop neurofuzzy trading systems that can be used
(and even retrained) in real time in Omega’s TradeStation. The only
significant requirements on the user’s part are a good understanding of
technical analysis and more than a passing acquaintance with TradeStation’s
Easy Language and Power Editor.
In view of the significant amount of processing power that the Assistant’s
training routines require, Orphelin recommends that the two programs be
ideally run on separate machines to prevent interference with real time data
flow via the comports. However, I encountered no difficulties running both
applications on a Pentium Pro 200 with 128MB of RAM.
The first step is to produce some suitable data that can be used in
training. This is accomplished in TradeStation using a template indicator
supplied by SirTrade (that also includes a number of popular technical
analysis indicators) and printing the output to an ASCII file. By modifying
the Easy Language code you can include any selection of technical indicators
that you feel are of predictive value. Apart from the numerical output of
the selected indicators for each period, the ASCII data file (referred to as
a TSD file) also contains a target value that would trigger a trading signal
for the indicator and the difference between the closing price of each time
interval and its predecessor.
Once the ASCII file has been produced, one switches to the separate
Assistant program for the training routine. When starting the Assistant, the
first window displayed is the journal file, which records the results of all
activities undertaken during operation. As a tool for debugging any training
problems it is invaluable. Selecting “File: New” from the menu bar and
selecting “Fuzzy Description Files” from the ensuing pop up menu opens the
Command window from which the ASCII file is selected and loaded. The window
also allows the user to specify a wide range of training parameters. These
include some very “real world” applications, such as the ability to generate
additional synthetic price bars and reinforce training under difficult (but
vital) market conditions such as the beginning and end of a trend. You can
also specify how much of the available data is to be used for training and
how much for out of sample testing – the default is a fifty-fifty split.
Having loaded the data file, the next step is to define the fuzzy
description settings for the data that will be used during training. This
consists of defining the min and max numerical values for each indicator
(the software fills these in automatically after scanning the file, but you
can change them if you wish) and specifying the desired number of fuzzy
sets. To ensure effective training, this last should be set to at least the
minimum number of different “states” that the indicator can be in. Orphelin
uses the RSI (Relative Strength Index) as an example that has three states –
overbought, neutral, oversold).
After completing this stage, two new windows appear on screen; a Target
Performance Summary and a Performance Summary. The former contains details
of the hypothetical best results that could be achieved on the data set,
while the latter will display performance results for training and test
runs. The detailed performance analysis available here goes way beyond any
other trading system analysis that I have seen to date.
Having specified such parameters as how many iterations the software should
make through the data and the training method (both back propagation and
stochastic are available), hitting the “Train” command button sets the ball
rolling. There must be some well optimised code underlying the Assistant as
even training large, complex data sets with a large number of iterations
selected didn’t seem to take long. (A great deal faster than some neural
networks I have tested!) After training is complete, an equity curve is
displayed, complete with buy/sell signals and price bars.
However, probably the most potent command in SirTrade is “Explore”, which
examines multiple fuzzy sets to find the best performer for a given set of
indicators. You realise just how much time that this is saving you as you
watch the Journal window’s scroll bar go berserk and multiple equity curves
and performance reports flash before your eyes! Having selected the top
performing FIS (Fuzzy Inference System), it can be further refined with
additional training sessions or saved as an FZB file for use in
TradeStation.
Having saved the file, TradeStation’s Power Editor is again used with
another template file provided with SirTrade to complete the new trading
system. At its simplest this consists of little more than specifying the
path to the FZB file. Load the trading system to a TradeStation real time
chart and away you go!
However, SirTrade offers you a lot more than just this. Using templates
provided, you can specify a trading systems that references two or more FZB
files and Fuzzy Inference Systems. You could, for example, specify and train
FISs for trending and non-trending conditions and then specify a trigger
switch that will automatically flip the trading system between them. If you
want to be really smart, train multiple FISs and trigger trading signals
based on their “voting” consensus.
All this in itself is impressive enough, but Orphelin pushes the technology
to the limit by also allowing real time retraining of FISs from within
TradeStation, so the your system’s expertise can be based on the most recent
market price action. You can specify how many price bars elapse before
retraining takes place (the default is five). Alternatively you can specify
conditions that will initiate retraining. The example provided with SirTrade
uses the ADX indicator falling below a value of 15, which implies a
non-trending market that will need more self adaptation than a trending one.
In brief, this is the most impressive application of artificial intelligence
techniques to trading that I have seen to date. The power and flexibility
are impressive with pretty much every angle covered. Orphelin has clearly
focused on developing a product that will get you quickly up and running
with a viable trading system - not bogged down in side issues and data
pre-processing. Though the performance results I achieved over the test
period were anecdotal rather than rigorous, they were nevertheless extremely
impressive. The methods contained in SirTrade are clearly viable. My only
real criticisms are of the manual and help files, which are in the most
exquisitely tortured English! Unravelling the ruptured syntax does make
learning the ropes more long winded than necessary. That aside, this is an
extremely well planned and effective product.
|