Share price action analysis using natural language processing

Abstract

Stock price prediction has been a widely researched topic, primarily through technical and fundamental analysis. While technical analysis relies on historical stock data and mathematical indicators, its effectiveness diminishes in illiquid stock markets such as the Colombo Stock Exchange (CSE) due to low trading volumes and irregular price movements. Fundamental analysis, on the other hand, focuses on intrinsic company value but does not fully capture short-term market reactions to external events. This research explores an alternative approach by applying Natural Language Processing (NLP) techniques to conduct an event study analysis. The study examines how news articles influence stock price movements in the CSE by transforming textual data into numerical representations using Large Language Model (LLM)-based embeddings. The extracted feature vectors are then analysed using machine learning algorithms to identify correlations between news representation and stock price fluctuations. By leveraging NLP-based vectorization and predictive modelling, this research provides new insights into price action analysis in illiquid stock markets, where traditional prediction methods often fail. The findings contribute to the field of financial analytics by demonstrating the feasibility of using textual data to enhance stock price forecasting in under-researched market conditions.