In this example we begin with three profitable models for the stock Deere (DE), which is known around the world for its brightly painted green tractors.  In the NeuroShell Trader chart below, you’ll note a series of cyclical up and down price movements that lend themselves to model building.

This NeuroShell Trader chart combines three different models into a single trading system that increases profits compared to the individual models.

We built the first model in ChaosHunter, our stand-alone software tool designed to produce readable formulas to model numeric data. We exported the price data from NeuroShell Trader and saved it as a text file for use in ChaosHunter.

We loaded the ChaosHunter built-in template for daily trading models so we didn’t have to worry about the correct model settings in ChaosHunter.  ChaosHunter used genetic algorithm optimization to fuse a selection of model inputs and functions into a formula that is compared to a threshold value to make buy/sell decisions.  In this example, ChaosHunter discovered a formula that generates profitable long and short trading signals in both the training data and the out-of-sample period.  Once we saved the model in ChaosHunter, we used the built-in indicator in NeuroShell Trader to insert the ChaosHunter trading signals into trading rules in the chart.

Next we used NeuroShell Trader’s prediction wizard to create a simple prediction with only three inputs (Average Directional Movement, Stochastic #K, and Williams %R).  We set the maximum number of hidden neurons to four so the model wouldn’t overfit the training data.  The Price Momentum model again showed trading signals that hit peaks and valleys in both sets of price data.

We created a second prediction in NeuroShell Trader using the momentum/acceleration template included in the Prediction Wizard. Once again the trading signals for both data sets were favorable.

Next we combined all three models to increase the probability of generating correct trading signals. (Visit www.ward.net  in the Tips and Techniques section and look for Ensemble Estimators for mathematical proof.)  We set up the Long Entry and Short Entry rules to buy if 2 of the 3 models were producing a trading signal.  The results for all of the models are displayed in the chart above.  You’ll notice that the green equity curve for the combo model exceeded the individual models in general.  This is not always the case when you combine models.

Overall we believe that using combo models might be a technique you want to explore to bring more stability to your trading during these volatile times.