Application of machine learning techniques in forecasting the S&P 500 and NASDAQ indices ETFs

Abstract

The low uptake of Exchange Traded Funds (ETFs) in Kenya has been attributed mainly to the volatility of these financial assets, which is a major challenge that deters investors from achieving their investment objectives. This study aimed to address this issue by predicting the direction of returns for the S&P 500 and NASDAQ index ETFs and determining the most efficient and effective forecasting model. The research questions guiding this study were: Which risk factors affected the direction of returns for the S&P 500 and NASDAQ market index ETFs? Which machine learning model could accurately forecast the direction of returns for these ETFs? How could a web-based platform that integrates machine learning models be utilized by investors and financial analysts to estimate the trends of returns for the S&P 500 and NASDAQ index ETFs? The objective of this study was to develop a reliable and precise model for forecasting the trend of returns, which would enable investors and financial analysts to achieve their investment objectives. This research study examined a dataset of historic ETF prices obtained from the Yahoo Finance website and financial factors to assess the performance accuracy of LSTM models in predicting the direction of returns. The models developed included the LSTM, XGBoost, and hybrid ARIMA-GARCH models. The LSTM models had remarkable results, attaining accuracies of 95% and 96% for the S&P 500 and NASDAQ equity market indices, respectively. The XGBoost models performed satisfactorily, attaining an accuracy of 84% for both indices. However, the hybrid ARIMA-GARCH models performed inadequately, achieving accuracies of 49% and 50% for the S&P 500 and NASDAQ indices, respectively. These results underscore the limitations of conventional statistical methods in handling non-linear patterns and relationships in financial data. The results of the study enriched the existing literature by rigorously examining the ability of selected financial factors to forecast the direction of returns and by exploring and analyzing the relationship between these variables and the direction of returns. The impressive precision and reliability of the LSTM models offered valuable instruments for investment professionals and market experts, enabling them to make more informed and astute decisions in volatile and intricate financial markets. Ultimately, this study sought to refine investment plans, enhance investment policies, and broaden understanding of the characteristics and dynamics of specific financial markets. By offering a reliable and accurate forecasting instrument, this research sought to enable investors and financial analysts to realize their investment objectives and navigate the volatility of ETFs more efficiently.