Once you build a prediction in NeuroShell Trader and let the genetic algorithm fine tune indicator parameters, you can export the refined versions of the indicators and build a complementary model in ChaosHunter. The result may be a more profitable model. If you create several different ChaosHunter models, you can build an entire ensemble trading system that increases the probability of a correct trading signal. (Look at the Example 20 chart in NeuroShell Trader and simply substitute ChaosHunter models for the neural net models.)
When You Don't Know the Rules
In previous newsletters we have talked about using a neural network prediction when you have some indicators (or other data streams) that you follow, but you can't define the rules for when to trade. So you let a neural net combined with a genetic algorithm optimization of indicator parameters do the work for you. We did exactly that and the result is the prediction at the top of the chart.
We let the genetic algorithm pick the number of periods for the Average Directional Movement (ADX) and Williams %R indicators, as well as two copies of the regression slope indicators that were used as inputs to the net.
|
|
|
This chart displays a neural net prediction at the top of the screen and an enhanced ChaosHunter version of the model developed using the same indicators as inputs.
|
In an effort to possibly improve the model, w
e exported the refined versions of the indicators from NeuroShell Trader using the export function on the Tools menu. We saved the data in a .csv file which can be read by ChaosHunter.
If you are unfamiliar with ChaosHunter, it is a stand alone program that can produce readable formulas (RULES) from numeric data using genetic algorithm optimization. The formulas may easily be used as trading rules in NeuroShell Trader using pre-built ChaosHunter indicators.
What Happens in ChaosHunter?
Setting up the model in ChaosHunter requires that you choose input values and math functions as building blocks for the formula. ChaosHunter has the final say as to which indicators and which functions constitute the formula that will meet the optimization goal.
|
|
|
ChaosHunter lets you choose model inputs from your data file. The output is the next bar's open which is used to calculate profit or loss.
|
In the image below we chose to use a Trading Strategy optimization goal for the genetic algorithm rather than a statistical goal such as Maximize R-squared. A trading goal presumes that the fill for a trade occurs on the open of the next bar in order to calculate profit/loss figures. To insure that a prediction from the Trader and the ChaosHunter model would work in sync in a combination system, you would set up the prediction in the Trader for one period into the future.
|
|
|
We chose a Trading Strategy optimization goal function to maximize profit.
|
After training, the ChaosHunter model generated a net profit of $35,155 compared to $33,609 for the neural network prediction, a difference of $1546.
Bringing the ChaosHunter Model in NeuroShell Trader
If you wanted to take this method one step further you could select different functions in ChaosHunter in order to create an entirely different formula. Then you could combine the neural net prediction and the two ChaosHunter models into an ensemble system in NeuroShell Trader that says to buy if two out of the three are true and therefore increasing the probability of a correct signal.
You can easily bring the ChaosHunter models back into NeuroShell Trader by using the ChaosHunter DLL indicators that are included in NeuroShell Trader. That is how we displayed the ChaosHunter Trading Strategy Signal on the chart.
When you choose the ChaosHunter Trading Signal indicator, the indicator wizard directs you to select the appropriate ChaosHunter model file. The indicator then lists the data streams that appear in the saved model and you simply have to match them with data streams that are in the Trader chart. If you display the ChaosHunter model in the same chart where you created the original prediction, the appropriate data streams are already in the chart.
|
|
|
The ChaosSignal indicator links the data streams from the formula to the data on the Trader chart.
|
Once you have added the ChaosHunter Trading Signal indicator to the chart, the trading rules are set up as follows:
Long Entry: ChaosHunter Signal = 1
Long Exit: ChaosHunter Signal = 0
Short Entry: ChaosHunter Signal = -1
Short Exit: ChaosHunter Signal = 0
You can download a copy of example chart and ChaosHunter model file (.md) by going to the Examples section of
www.ward.net
