We have built many NeuroShell Trader example charts that used the optimizer to find the most effective trading rules, neural net inputs, and parameters for successful trading systems. You can make the optimizer do even more work for you by letting it find the most effective bar sizes or data streams for your models. The set up for this technique is as easy as setting up ranges for either neural net inputs or trading rules.
We'll show you how with a neural network example based on a 1 hour chart for SPY, the S&P 500 ETF. In order to let the optimizer choose between two different bar sizes, we added 15 minute bars for open, high, low, close, and volume to our chart.
For inputs to the net we chose ADX, CCI, RSI, and Stoch %K from the Price Momentum category and Money Flow Index from the Volume category. When the indicators were displayed on the inputs screen of the Prediction Wizard, we added the 15 minute versions of the price and volume data to the 1 hour versions in each of the indicators. We did this by selecting each indicator parameter, clicking on the "Modify Parameter" button, and choosing both the High from the hourly base chart and the 15 Minute High. We did this for each of the indicator parameters that included a data stream. This enables the optimizer to choose which bar size to use to calculate the indicator.
|
|
|
The data stream parameters for the ADX indicator include both hourly and 15 minute bars. The same is true for all of the indicators that were included as inputs to the prediction.
|
Note that in addition to picking bar sizes, you can let the optimizer choose another data stream. For example, the RSI defaults to the close for its calculations, but you could add the high and low data streams and let the optimizer choose.
|
|
|
Results show that the optimizer mixed different time bars to calculate the ADX indicator, used 15 minute bars for the CCI and MFI indicators, and used hourly bars for Stoch%K.
|
|
|
|
The finished chart found trading signals that corresponded to peaks and valleys in the data.
|
Technique Also Works for Trading Rules
You can use the same method to customize trading rules. Simply add another bar size or data stream to the trading rules and let the optimizer decide which ones create more effective models.
|
|
|
This RSI trading rule lets the optimizer choose between hourly and 15 minute versions of the close.
|
A few minutes spent setting up optimization ranges for either neural net inputs or trading rules can enable the optimizer to find the best data streams for improved trading models.
The neural net example chart from this article is available for download from
www.ward.net in the examples section.
