In this issue:
I. ChaosHunter in science and business by Steve Ward, CEO
II. Other commentary by Steve Ward, CEO
III. One way to stop this newsletter
***********************************************************
I. ChaosHunter in science and business by Steve Ward, CEO
For many years now we have been recommending the NeuroShell Predictor to create numeric outputs to produce information like predicting sales, the future prices of raw materials, consumer response, the time it takes for pill coatings to dissolve in the stomach, survival of species in different environments, response to diseases, fish catch, oil exploration, and a host of other relationships that organizations need or would like to know.
Neural nets have their advantages and disadvantages. Some of ours are quite fast, and they can effectively assimilate a fairly large number of independent variables, say 50 or 60 depending upon the quality of your training data. They can give nice tight fits for the training data, and if you set up your data effectively the models will be robust.
Unfortunately, many novice users frankly have great difficulty setting up their data effectively, because it is daunting. In addition, the equations inside neural nets tell you virtually nothing about the data relationships, aside from a rough estimate of variable contribution, which is usually misinterpreted anyway.
So now I am recommending that most future business and scientific studies be conducted with ChaosHunter, www.chaoshunter.com, with some exceptions which I will cover in a minute.
The main reason is that ChaosHunter produces models that clearly show you the often SIMPLE mathematical relationships between variables. You can show them to your boss and publish them in your company newsletter or journals. You can send them to your colleagues or embed them in computer programs or the WEB with no real need for any Runtime Server or DLLs. Stick them in a spreadsheet.
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. To be sure though, I have frequently seen ChaosHunter build better fitting, yet less complex models. ChaosHunter builds neural nets and chaotic models too, but I do not recommend them for business and science.
To be clear, ChaosHunter users should set up their data effectively too. But ChaosHunter has better facilities for evaluating and setting in-sample and out-of-sample relationships, as well as evaluating runner-up models. More such facilities may be coming.
Now for the exceptions. Continue to use NeuroShell Predictor if you cannot wait a few hours or (for bigger problems) a few days for training. (Never mind that if you have multi-core computers or several fast ones on a network. ChaosHunter does multiprocessing). Continue to use Predictor if you feel all of your independent variables are important, must be used, and you have more than say 15 or 20 of them. Finally, do not replace the entire AI Trilogy with ChaosHunter, because it is still the only thing to find optimal values of variables of prebuilt nets, i.e., you are involved in the deep concepts of what we call the AI Trilogy problem: http://www.wardsystems.com/learnmore.asp
In view of my new recommendations, we are going to give any owner of NeuroShell Predictor a discount of $395 towards the purchase of ChaosHunter, including AI Trilogy owners. This promotion will end at the end of April 2012.
***********************************************************
II. Other commentary by Steve Ward, CEO
1. ChaosHunter for TradeStation users. Last month we reported on our new ChaosHunter converter that converts ChaosHunter artificial intelligence models into EasyLanguage code. So it is now easier than ever for TradeStation devotees to build sophisticated “Ward Systems Group” trading models, and do all of your trading fully inside your TradeStation account. ChaosHunter is no longer involved. The converter is free to anyone who is running the latest release of ChaosHunter; you can get it from ward.net.
So I thought I’d show you a couple of simple examples of ChaosHunter code and the EasyLanguage equivalents that are to be pasted into a TS strategy:
============ CH Code 1 ================
Predicted = (30.83793 / (-29.93429)) – (71.66399 + ((n2((-29.93429),
48.96743, ChaosVar, 33.38189) * ROC%(Low,13)) + (RelStr(High,Low) /
Sprd%(Open,Close))))
At next bar Open:
Buy Long if Predicted >= 68.74312
Sell Short if Predicted <= -27.44999 ============ Equivalent TS Code 1 ============ var: ChaosVar(0); var: Predicted(0); If CurrentBar = 1 Then ChaosVar = 1.23; //—- Raw formula output —- Predicted = -1.030187 – (71.66399 + ((n2(-29.93429, 48.96743, ChaosVar, 33.38189) * Proc(Low,13)) + divide(RelStr(High,Low), SprdP(Open,Close)))); ChaosVar = Predicted; //—– Trading signals —– //—— True-reversal strategy —– If Predicted >= 68.74312 Then Buy next bar at open;
If Predicted <= -27.44999 Then SellShort next bar at open; ============ CH Code 2 ================ Predicted = 74.48262 / %chg(Close,15) At next bar Open: Buy Long if Predicted >= -21.44052
Sell Long if Predicted < -21.44052
Sell Short if Predicted <= -75.48004 Cover Short if Predicted > -75.48004
============ Equivalent TS Code 2 ============
var: Predicted(0);
//—- Raw formula output —-
Predicted = divide(74.48262, Pchg(Close,15));
//—– Trading signals —–
//—— Non-reversal strategy, both Long and Short positions —–
Switch (MarketPosition)
Begin
Case 0: //currently in Neutral
If Predicted >= -21.44052 Then Buy next bar at open;
If Predicted <= -75.48004 Then SellShort next bar at open;
Case 1: //currently in Long
If Predicted < -21.44052 Then Sell next bar at open; Case -1: //currently in Short If Predicted > -75.48004 Then BuyToCover next bar at open;
End;
TradeStation and EasyLanguage are registered trademarks of TradeStation Securities.
***********************************************************
III. One way to stop this newsletter
It is really easy. Just change your email address and don’t tell us.
***********************************************************