This repository contains the pdf of my Dissertation, along with the Google Colab Notebook holding the Python code for its reproduction. I will be gradually refining the code from the project over the next few months.
This project was undertaken as my postgraduate dissertation. Under the supervision of Dr. John O’Hara, Director of the Center for Computational Finance, I used Python to implement and compare the performances of four machine learning models on forecasting the daily returns of the S&P 500 index.
Forecasting the S&P 500 Index: A Comparison of Long Short-Term NeuralNetworks, Support Vector Machines, Random Forests, and Feed-Forward Neural Networks
Industrial and academic finance, alike, have become inextricably linked to the application of machine learning models and the study of financial time-series. Consequently, the race to develop a model that would allow the greatest advantages to be gained over the market has become increasingly competitive. Nevertheless, and despite being driven by the 20th and 21st Century innovations in computer science and engineering, a truly robust and all-around model has eluded researchers. Instead, the financial literature is rife with papers examining the application of evergreen and novel models to specific scenarios, with little room for cross-study comparison of results, except in the case of a handful of open-source papers available on Git-hub. In this paper, we, i) survey the sparse developments in the literature on financial forecasting, ii) attempt to derive conclusions from it, and iii) implement three of the most commonly deployed models to forecast the daily directional movements of the S&P 500 Index (GSPC). Namely, we apply a Random Forest (RF) with XG-Boost, a Long Short-Term Memory Neural Network (LSTM), and Support Vector Classifier (SVC) with rbf-kernel, compare them according to four metrics, and judge them against a bench-mark Feed-Forward Neural Network (FFNN).
Keywords: Support Vector Machines, LSTM, Random Forest, Stock Price Prediction, S&P 500 Index.
