A deep learning based stock trading model with 2-D CNN trend detection

The success of convolutional neural networks in the field of computer vision has attracted the attention of many researchers from other fields. One of the research areas in which neural networks is actively used is financial forecasting. In this paper, we propose a novel method for predicting stock price movements using CNN. To avoid the high volatility of the market and to maximize the profit, ETFs are used as primary financial assets. We extract commonly used trend indicators and momentum indicators from financial time series data and use these as our features. Adopting a sliding window approach, we generate our images by taking snapshots that are bounded by the window over a daily period. We then perform daily predictions, namely, regression for predicting the ETF prices and classification for predicting the movement of the prices on the next day, which can be modified to estimate weekly or monthly trends. To increase the number of images, we use numerous ETFs. Finally, we evaluate our method by performing paper trading and calculating the final capital. We also compare our method's performance to commonly used classical trading strategies. Our results indicate that we can predict the next day's prices with 72% accuracy and end up with 5:1 of our initial capital, taking realistic values of transaction costs into account.