For an introduction to Neural Networks and Stock Market Prediction please read the first post we have written on the subject: Stock Market Prediction
In this tutorial, we will show you step by step how to create a neural network prediction model and how to setup this model in order to create a good stock market prediction system.
Create a Stock Market Prediction Item
In QuantShare Trading Software, the neural net prediction tool is opened when you click on "Prediction" under "AI" menu name.
The form contains a list of all previously created models as well as some statistics such as the number of inputs, the number of cycles of training and whether the model is trained or not.
To create a neural market prediction model, click on the "Add" button. Select "Neural Network" then click on "Next". The panel that is displayed now allows you to specify learning, validation and testing periods. You must first specify the start and end dates then select the percentage of bars to use for the learning (Green), validation (Bleu) and testing (Grey) samples. The previous settings apply when the sampling method is set to Normal. If you choose a "Random" sampling method then the learning, validation and testing samples will be created randomly from the specified period, which is defined by the Start and End dates. Once this is done, click on "Next" to move to the next screen.
The Most Important Screen
Here we are in the Inputs & Output screen. This screen is certainly the most important one. Here, you can define the output or the time-series that must be predicted and the inputs or the time-series that will be used to predict the output.
Both technical analysis and fundamental analysis indicators can be used as inputs in a neural network system. Some traders also include economic, composite or other trained neural network models as inputs in their neural network system.
Click on the cell under the "Type" field to select the input type. For example, if the "Symbol Field" is selected, then clicking on "Select" will display the six time-series that compose a security (close, open, high, low, volume, open interest). When "Formula" is selected, you can enter a custom vector-based indicator/function as an input time-series (Example: Rsi(14) > 50 or simply Rsi(14)).
For the time-series you would like to predict, the output, you may select the stock close price (close), the one-day rate of return (perf(close, 1)), whether the stock will close higher tomorrow (close > ref(close, 1)) or any custom value.
The next screen presents the model settings/nodes. You can add or remove layers to the model, update the network settings or update nodes' parameters.
If you don't know what all these words mean then you have two solutions. The first one is to study neural networks in details and the second one is to simply ignore all these parameters and click on "Next".
In order to create your stock prediction model, you will need to specify the stock(s) you want to use to train your neural net model. This is what the next screen allows you to specify. It also lets you define a filter function so you can ignore specific bars.
After that, you can define a trading strategy to apply to the prediction model while it is trained and then set stop and best model settings. You should first specify stop conditions or when to stop the training of our stock market prediction model. During the training and validation process, hundreds or thousands of models will be created. You should specify the metric to use in order to select the best model among all these prediction models.
This was a brief tutorial on how to create a neural network model. If you have any question please leave a message on the comments section.