Greetings
from NeuroDimension!
The World Leader in Neural Network Software
This issue
of the TradingSolutions newsletter highlights the new TradingSolutions Primer
and features a look at the effects of the training data ordering.
In this issue you’ll find:
What’s New
and News?
Solution
Service Recap
* Training Data Ordering
Note: You are
receiving this newsletter because you requested to stay informed concerning
information about TradingSolutions from NeuroDimension. If you would like to stop
receiving these newsletters, please see the bottom of this newsletter for
removal instructions.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Are you new to technical analysis or neural networks, or
just not sure what TradingSolutions can do for you? Check out the new
TradingSolutions Primer available from our website. This five-page document
provides an insightful overview into each of these topics, with introductions
to technical analysis, neural networks, TradingSolutions, and the Solution
Service.
To view this document and other additional information about
TradingSolutions, see our Additional Information page: http://www.tradingsolutions.com/products/ts/InfoMenu.html
TradingSolutions v2.0 was recently reviewed by Dennis D.
Peterson in the May 2002 issue of Technical Analysis of Stocks and Commodities
(TASC). In his review, Dennis wrote: "If you haven't considered neural
networks as a solution for trading, TradingSolutions is an excellent choice to
see if they might be worth it...All in all, this is an outstanding product."
To view additional quotes and reviews of TradingSolutions,
see our News and Reviews page: http://www.tradingsolutions.com/reviews.html
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
The Solution Service is a free service that provides
daily trading signals and commentary for a select group of stocks. These models
can be used directly as part of your trading strategy or used as examples of
how to generate your own winning signals. To access it, simply press the
“Update Solution Service” button in the toolbar of either the evaluation
version or registered version of TradingSolutions v2.0.
Two new models were introduced in April: Baker
Hughes (NYSE:BHI) and KLA-Tencor (NASDAQ:KLAC). These models
replaced the Medimmune and Xilinx models, which failed to find a foothold in
this year’s markets.
The Baker Hughes model tends to trade fairly
slowly, averaging a new trade every two weeks, with many trades and
out-of-market positions lasting up to a month. However, the trades it does
generate do appear to capture many of the larger cycles in the price. The
KLA-Tencor model tends to trade fairly quickly, with most trades lasting
less than a week.
The recent movement in semiconductor stocks has
produced primarily good results for the Solution Service models. The Novellus,
KLA-Tencor, and Linear Technology models were long at the start of the rally
and correctly went short when the reversal occurred, resulting in six
profitable trades in the last few weeks just between these three stocks. Since
then, these stocks have all re-entered the market long with mixed results.
These will definitely be three models to watch in the coming weeks.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
The neural networks in TradingSolutions learn how a stock
performed in the past in order to predict how it will perform in the present or
future. The historical data used for this learning is typically divided into
three subsets, each with its own purpose.
The training set is used during the training phase
to adjust the weights used in the neural network. This is the only data used by
the neural network that directly affects how the neural network will perform.
Therefore, it is important that the training set include as many different
examples of how the data will perform as possible.
The cross validation set is used during the
training phase to prevent over-specialization. The neural networks in
TradingSolutions are trained using multiple iterations. After each iteration,
the current effectiveness of the neural network on the data in the cross
validation set is tested. The training phase is repeated as long as this
performance continues to improve. As soon as the performance begins to
significantly degrade, the training is stopped and the weights providing the
best results are used.
The accuracy testing set is set aside to manually
verify the ability of the neural network to work with new data. This data is
“out of sample” in that it is not used during the training of the neural
network at all.
Additionally, people sometimes refer to a production set.
The production set typically refers to any data after the testing set. In other
words, the data that is input into the neural network while it is being used on
new data or “while it is in production”.
Default Training Data Ordering
By default, TradingSolutions uses the last five years of
data and divides that data in the order “training, cross validation, testing”.
In other words, the first three years of data are used for training the neural
network weights, the next year of data is used for verifying the adjustments to
those weights are continuing to produce good generalizations, and the last year
of data is set aside for you to verify how the neural network would perform on
new data.
The advantage to this ordering is that the most recent data
is used for cross validation and out-of-sample verification. This recent data
will typically be the most like any new data. Therefore, this effectively
allows you to verify the neural network performance on the next best thing to
new data.
The disadvantage to this ordering is that the oldest data is
being used to adjust the weights. This can have the effect of the data being
least like the current conditions to be used for training the neural network.
Changing Training Data Ordering
The flexibility of TradingSolutions allows you experiment
with settings such as the training set order. For example, you can move the
training set later in the sequence so that the neural network is being trained
on more recent data. The advantage of this would be that the conditions in the
training set should be more like the current conditions. However, the
disadvantage is that testing and/or cross validation would have to be moved to
before the training set, making them less like the current conditions.
To modify the training set order while creating a new
prediction, select “View or modify the settings for training and testing this
prediction.” This setting is on the “Select Options” page of the Predict a
Value Wizard. To modify the training set order of an existing or copied
prediction, select the “Training Data” page of the Modify Field Dialog. With
either method, simply set the “Ordering” setting to the training data order you
would like to try. From this page, you can also adjust the size of the
individual sets to be smaller or larger.
The important thing to remember in adjusting the training
data is that a neural network is only as good as the data it has available for
training. For example, if the majority of the data in the training and cross
validation sets covers a time when the stock was steadily increasing in price,
then the neural network may have trouble predicting what will happen when the
stock price is decreasing – primarily because it has never seen this happen
before.
Because of this, it often worthwhile to adjust the training
set and/or cross validation set to cover as wide a range of price movements as
possible. After you create a default neural network, display a chart of the
price with the prediction. TradingSolutions will automatically indicate where
the training, cross validation, and testing sets are for the prediction so you
can quickly assess whether any of the data sets focuses more on one type of
market than another.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Many TradingSolutions users read Technical Analysis of
Stocks and Commodities (TASC), the Traders’ Magazine. It’s a great way to
trade smarter with detailed information on trading strategies, charting
patterns, indicators, and computerized trading methods. As a free service to
our users, we arrange for the “Traders’ Tips” section of the magazine to
include tips and techniques for implementing many of the indicators found in
the magazine in TradingSolutions.
Each time a tip is included in the magazine, we also include
the implemented functions in the Solution Library on our web site. These functions
can be imported into either the full version or the evaluation version of
TradingSolutions and are completely free. Visit the Solution Library now at: http://www.tradingsolutions.com/download/solution.html
If you aren’t familiar with Technical Analysis of Stocks
and Commodities, we encourage you to visit their web site at: http://www.traders.com
Neural 101
is a free educational web site on trading stock and futures markets with neural
networks. It includes message areas, links, trading indicators, systems and
scans. The message areas include special forums on building trading systems,
using neural networks, and even discussing tips, tricks, and ideas for using
TradingSolutions.
NeuroDimension
encourages the use of this free resource to foster discussions between people
interested in TradingSolutions. You can visit Neural 101 at http://www.neural101.com. The
TradingSolutions message forum can be located using the “Message Forums” link
from the left-hand menu.
Have an event or product you would like to announce? Contact
us at submissions@nd.com.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
We appreciate your feedback! Please send us your comments or
suggestions concerning this newsletter, our web site, or part of the
NeuroDimension product line. Write to us at: feedback@nd.com
Have questions about TradingSolutions? Send your questions
to: info@tradingsolutions.com
Some mail readers may display an attachment at the bottom of
this newsletter. This is typically caused when the mail reader is set to
display text only. The original, formatted version is sometimes included as an
attachment.
This issue and previous issues of this newsletter are
available on the TradingSolutions web site at: http://www.tradingsolutions.com/center/newsletter.html
If you would prefer not to receive these newsletters or
subsequent product updates from NeuroDimension, please reply to this letter
with the subject heading changed to the word REMOVE.
Thank you again for your support of NeuroDimension products!