In the world where risk-free assets like banking deposits have close to zero or even negative returns, investors are seeking for ways to save and grow their assets.

StocksNeural.net analyzes and predicts stock prices using Deep Learning and provides useful trade recommendations (Buy/Sell signals) for the individual traders and asset management companies.

Predictive models based on Recurrent Neural Networks (RNN) and Convolutional Neural Networks (CNN) are at the heart of our service. We constantly improve them, try new models and new scientific approaches. We believe our model are more accurate than competitors have and our service is much easier to use by either novice or experienced traders. We are communicating with some of the professional quant traders, and working together to make our system better.

Following steps are present in models training:

• Loading of the historical market data from the Quandl and Cryptocompare premium datasets
• Data normalization
• Selection of the optimal Kalman filter parameters using our innovative approach or using causal CNNs for automated feature extraction.
• Smoothing of the source data with Kalman filter using optimal parameters or using causal CNNs for detecting features at different abstraction levels and generalization.
• Optimization of the Recurrent Neural Network or CNN hyperparameters
• Training and validation of the Recurrent Neural Network or CNN
• Model backtesting

Daily pipeline for models includes steps required to load and preprocess new market data, calculate model's accuracy and performance metrics and generate trading recommendations according to forecast made and strategy parameters.