February 25, 2020
IN THIS ISSUE
Adaptive Nets Automatically Update to Market Conditions
There is an add-on for NeuroShell Trader called Adaptive Net Indicators that matches incoming data to similar patterns in the training data. The output of any new pattern will be much like similar known patterns.
Adaptive Nets also have another type of input called the “Actual” value. This value shows the net what you want the output to be like.
Unlike the Prediction Wizard in the Trader, Adaptive Nets retrain quickly on each new bar, so your model is constantly updating to current market conditions.
In addition to the principal indicators, Adaptive Nets include lagged versions that evaluate input values not only from the current bar but values from X number of previous bars.
Classification Models Generate Probabilities
There are separate Adaptive Net indicators for prediction and classification models.
For our example we are going to build a classification model that can generate buy/sell signals based on the similarities between the current set of inputs and like patterns in the training data.
The net’s output will be a number between -1 (strong probability of sell) and 1 (strong probability of buy). Numbers close to zero could be “Hold”. You would decide how close to zero a prediction should be in order to consider the value a sign to hold.
You Tell the Model Which Inputs Are More Important: Contribution Factors
You have more control over the Adaptive Nets indicators compared to the Prediction Wizard. You can tell the model which inputs should carry more weight in making a buy/sell decision, or you can let the optimizer figure it out.
The parameter settings for Adaptive Net Indicators include a list of the inputs followed in order by a Contrib (contribution) factor for each input.
You can boost the impact of any input by increasing the Contrib factor.
Lock the Value
If you customize any of the contribution factors, be sure to lock the value in the Trading Strategy wizard.
If you don’t lock the value, the optimizer will override your ranking value.
Let the Optimizer Choose the Value
If you don’t know how to set the contribution factor, let the optimizer choose the importance for each input. (You might be surprised which inputs are ranked higher by the optimizer!)
The optimizer can also adjust other parameters such as the actual value, the number of periods ahead and the size of the lookback window.
Classification Model Beats Buy and Hold
The chart above shows the results of the Adaptive Net Indicators classification net for DIA. The percent change in price over the time period displayed in the chart was 18.8%, while the Return on Trades for the model was 84.1%.
Click here to download the example chart. You must own the Adaptive Net Indicators add-on to open this chart correctly.
Changes for End-of Day Datafeeds
Quote Media stopped providing free data and Google now only provides data on their webpage. These changes affect both the OnDemand datafeed from Bowfort Technologies and the Historical Quotes Downloader.
The OnDemand datafeed still offers four datafeeds that supply end-of-day data: Yahoo, Quandl, Tingo and FRED.
If you subscribe to OnDemand, you can change the default datafeed. Use Windows Notepad to edit C:\\NeuroShell Trader 7\Servers\OnDemand Server\settings.ini and change default source=quote media to defaultsource=yahoo for example.
When you install OnDemand in NeuroShell Trader, it updates any chart with new data once you open the chart.
Questions? Contact Bowfort Technologies at phone: 281 301 5766 or visit their website at www.bowfort.com
Historical Quotes Downloader
Because of the recent changes, the Historical Quotes Downloader available from www.ward.net will now only supply data from Yahoo.
You need to open the Historical Quotes Downloader and download data every day before you open your chart in NeuroShell Trader.