In this issue:
I. NeuroShell Trader Wins Readers’ Choice for 2013!
II. Commentary by Marge Sherald, CEO
III. Writers Needed
IV. Help Wanted
V. ChaosHunter and NeuroShell Trader
VI. NeuroShell Trader Add-on Sale
VII. How to Stop This Newsletter
***********************************************************
I. NeuroShell Trader Wins Readers’ Choice Award for 2013!
We’re excited to announce that NeuroShell Trader has won the Technical Analysis of Stocks and Commodities (TASC) annual award for Artificial Intelligence software for the 11th year in a row. Our staff is saying a big THANK YOU to all of our customers who voted for us. Your support gives us that extra boost to continue to develop award winning software in the future.
************************************************************
II. Commentary by Marge Sherald, CEO
1. Thanks for voting in the TASC Readers’ Choice Awards.
I personally want to express my appreciation to our customers. Not only did you take the time to vote for us in the TASC Readers’ Choice Awards, you make it worthwhile coming to work every day. You challenge us by asking our software to do things we never even thought about when we were designing it. The result is a collaboration that has been continuing since 1988 when Ward Systems Group first started. We will endeavor to continue to uphold your loyalty and confidence in our products. I also want to express my high regard for the Ward Systems Group team who develop and support the software. Your technical expertise and creativity cannot be found anywhere else.
If you’ve been thinking of upgrading your version of NeuroShell Trader, we’re offering a 20% discount on upgrades to either the NeuroShell DayTrader Professional or NeuroShell Trader Power User or NeuroShell DayTrader Power User. The sale ends March 4, 2013, and you must order from www.ward.net/60.asp to get the discount.
2. When to retrain neural nets – Adaptive TurboProp2 example
We’re often asked when to retrain neural network models, and the only answer we can give is when the model is no longer working under current market conditions. We have customers who retrain models every day, once a week, and once every six months and they are all making money. There’s a way around a specific answer to this question, however, and that is through the use of the Adaptive TurboProp2 add-on. You can set it up to retrain in a certain number of bars and you can use the NeuroShell Trader’s optimizer to determine how many bars that should be for a particular stock, futures contract, or Forex model.
You can feed an Adaptive TurboProp 2 net the same types of inputs you would use in the prediction wizard in the Trader. There are several different TurboProp 2 indicators to choose from based on the number of inputs you want to use. The indicator lets you specify how many historical bars to use for training the model and how far ahead to predict. As new bars come in, the oldest bars are dropped from the training set and the newer bars are added to the training data. These models always predict out-of-sample (unless you specify that you are predicting zero bars ahead), so you don’t have to hold out the latest data for testing purposes. This is especially important for intraday models where markets change rapidly and you want to use the latest data in the training set. In addition to letting the Trader’s optimizer determine how many bars to include in the training set, it can determine how many hidden neurons to use, and how often to retrain the model.
If you’re using an Adaptive TurboProp 2 indicator on both the long and short side with the same inputs, the optimizer can customize the indicator for either a long or short trade, including both the neural network design and input parameters. If you prefer a symmetrical model, you can link the parameters in the Trading Strategy so the same settings are used on both sides.
We have created an updated example for Adaptive TurboProp 2 that is available from www.ward.net in the Examples section.
3. Amount of training data
Another question we hear frequently is how much training data to use for building models in any of our products. This is not a new answer, but it bears repeating. In fact, I refer to this posting from www.ward.net so often that I now keep a copy on my desktop. The complete write up is called “Selecting in-sample and out-of-sample periods – Commentary by Steve Ward” posted 12/7/2010 in the Changes in Documentation sections of www.ward.net for both ChaosHunter and NeuroShell Trader.
“This brings us to what I will call focus. The traditional wisdom is that the longer the in-sample period the better. This use of a “sledge-hammer to kill the fly” wisdom grew out of the need to include variety and balance in your in-sample data, and the faulty (I believe) assumption that you can find a model that works well forever. But just starting your in-sample data as far back as possible might be counter-productive. For one thing you may get so much variety that the optimizer learns to only detect the largest moves correctly. For another thing, the variety that you capture and learn could be so old as to be useless tomorrow. It might be better to focus on a smaller part of that most recent slow bull market.
But won’t that cause overfitting? It could, but if you are careful and focus on a period of cyclic activity with good inputs/rules, the result might be that the model is better able to detect the onset of smaller activity and reversals. There is less hay to hide the needle. Yes, you might have to retrain or reoptimize as the markets shift.” Steve Ward.
4. Different Objective Functions
I was talking to a user who wanted to use the Maximize Return on Account * Equity Curve Correlation optimization objective, but was concerned because during the historical data used in the model, the price of the instrument changed substantially (think Apple from $200 to $700 over a period of a few years). He said that if a model learned to make a trade with 100 shares at an early part of the data, the return would be much less than during the $700 period, so the trades should be evaluated differently. He wanted me to get the Trader developers to add an objective function that would account for this change in price. It turns out there was already an objective function that did this: Maximize Return on Account * Log Equity Curve Correlation. This objective was designed to work better with position sizing methods such as optimal-F and percent of account. The result in his particular model was a 10% increase in profit. (Sometimes it pays to read the help file!)
**********************************************************************
III. Writers Needed
If anyone writes a description of a technique he or she uses to build a successful trading system in NeuroShell Trader, we’ll publish it in our newsletter if we believe it will be helpful to our customers. Any item that is published will entitle the writer to a free copy of one of the Ward Systems Group add-ons listed in the sale below. Send all submissions to sales@wardsystems.com.
***********************************************************************
IV. Help Wanted
We occasionally get requests for someone to program NeuroShell Trader data pump and trade pump applications from our customers. If anyone is interested in doing this type of work, please send an E-mail to sales@wardsystems.com and we’ll provide your contact details to those customers. Ward Systems Group will only pass on the contact details and will not be involved in any negotiations about price or scope of work.
***********************************************************************
V. ChaosHunter and NeuroShell Trader
On sales calls we are asked about how ChaosHunter and NeuroShell Trader work together. First off, you can obtain data from the Trader that you can use to build your ChaosHunter model. You can either export data from the Trader or have ChaosHunter read a NeuroShell Trader .wsg data file directly (if you have ChaosHunter release 3). Once you have data, the next step is to build a model in ChaosHunter.
ChaosHunter is especially effective if you don’t know which rules to use in a Trading Strategy in the Trader. Unlike a neural network, ChaosHunter produces models that clearly show you the often SIMPLE mathematical relationships between variables.
Once you have a successful ChaosHunter model, you can embed that model in NeuroShell Trader simply by inserting an indicator from the External Program & Library Calls, ChaosHunter indicators. You add the ChaosHunter model as a rule in the Trading Strategy using a relational indicator. A long entry rule could look something like this:
Value of ChaosHunter formula > a threshold value
ChaosHunter produces the threshold value when it builds a model, so you can either use that threshold in the Trader rule or let the Trader’s optimizer figure out a new value. Running a ChaosHunter model in the Trader lets you add stops and limits as well and view the results.
ChaosHunter offers a different approach than the neural networks and optimization techniques used in NeuroShell Trader. ChaosHunter tries to make the simplest models it can, carefully selecting the fewest independent variables and avoiding those tight fits that, without great care, can cause overfitting and poor results out of sample. We have frequently seen ChaosHunter build better fitting, yet less complex models than neural networks which attempt to use all of the inputs you provide.
Want to learn more? Call us on 301 662 7950 or Skype us at wardsystems.
VI. WSG add-on sale
When new customers buy NeuroShell Trader, they often ask which add-ons to purchase. We tell them to wait until they become familiar with the Trader. To help those who are ready to explore, we are offering the following NeuroShell Trader add-ons at a 20% discount until March 4, 2013.
-Fuzzy Pattern Recognizer
-Adaptive Net Indicators
-Advanced Indicator Set 1
-Neural Indicators
-Adaptive TurboProp2
-Cluster Indicators
-Advanced Indicator Set 2
-Turning Points
-Fuzzy Sets
-Pattern Matcher
-Advanced Indicator Set 3
We’re also including the add-ons we sell in conjunction with MESA Software and Richey Enterprises: MESA91 and Cybernetic Analysis, and InterChart Tools 1 and 2.
To buy the add-ons at this special price, order only from the following web link: www.ward.net/60.asp or give us a call at 301 662 7950 or Skype us at wardsystems.
***********************************************************************
VII. One way to stop this newsletter
It is really easy. Just change your E-mail address and don’t tell us.
***********************************************************************