Machine learning is the ability for computers to learn and improve without being explicitly programmed to do so. The techniques have been around for a while but not applicable due to technological constraints. However, with today’s computational power and the availability of data, one can efficiently create and train powerful models. The aim of this report to build a machine learning model with the object to predict stock prices.
Using a ML technique with the ability to remember sequences of events the model was built and trained on a stock (Nordea). Following the training and optimization, a system was created and trading logic was applied accordingly – in short, buying when the price is predicted to increase and shorting otherwise. Further logic was implemented with the aim to minimize volatility in the returns and hopefully lead to better overall performance with less risk.
On validation data the model returned 10% but for a test period further away from the training the ROI was down to 1 %. The problem with both overfitting and the constant need to update the model might cause limitation in the current version, but with further optimization and automatic, more frequent training, there are much room for improvement. The system still generates a positive average daily return and wins more trades than it loses. The high trading volumes on both sides causes a problem with transaction costs but also leads to a low systematic risk, opening up more possibilities to profit independent on the financial climate.
Analyst: Johan Hammarstedt
