Exploring the Predictive Power of Composite Stock Data in Stock Price Forecasting

Abstract

The development of data-driven prediction models in finance has seen a significant push in the past decades, leading to the exponential rise of machine learning in improving forecasting accuracy. Competitive financial forecasting invites for a thorough assessment of the reliability of various input variables in stock prediction models. This thesis explores the impact of composite stock data in stock price forecasting, particularly focusing on predicting substantial price increases in the short term. It critically examines the comparative power of composite versus singular stock data and sector-based versus S&P 500 data in technical analysis forecasts, by leveraging historical data of ETFs. This research addresses three key research questions through quantitative methods, with the aim of investigating whether combining different types of technical data can result in a more predictive model for stock price forecasting. Utilizing a random forest algorithm, the research also assesses the reliability of short-term forecasting in varying time windows, while acknowledging computational and methodological limitations. The study seeks to contribute to the domain of financial forecasting by delving into technical analysis from a unique angle that incorporates broader-market trends and the concept of stock synchronicity. This investigation provides essential insights for individual investors, financial analysts, and other stakeholders seeking to refine investment strategies by considering less explored variables or alternative perspectives. This approach strives to increase the predictive accuracy of financial models in a cost-effective manner.