Neural Network Comparison for Stock Prediction

This research project focusses on comparing simple RNN, LSTM and GRU for stock market return prediction. Using data for AAPL and HPQ from an online Kaggle datasource (https://www.kaggle.com/datasets/borismarjanovic/price-volume-data-for-all-us-stocks-etfs/code), I generated numerous EWM features from the open, high, low, close and volume data.

Using these features, I then created multiple versions of training data for the neural networks, each with different lookback and prediction horizons. I did this to not just understand how each of the neural networks perform, but also how they perform over different horizons. These input data are generated in the Feature_Generation notebooks.

Finally, using these different input data, I compared a simple RNN, LSTM and GRU in their ability to predict the return and probability of positive return over the prediction horizon.

This entailed using a different neural network architecture depending on the use case (return or probability). All of the code for this can be found in the Return_ or Prob_ notebooks (labelled accordingly).

The research outputs and conclusions can be found in the PDF provided in this folder.