Predicting Nigerian Stock Market Returns Based on Daily Business News Headlines

Abstract

This study predicts daily returns of the Nigerian Stock Exchange (NSE) using Nigerian daily news headline. The Vanguard newspaper is used as the source of information, to collate 11 years (2February 2, 2012 to September 29, 2023) of daily news headlines and data on the daily returns All-Shares Index (ASI) of the Nigerian Stock Exchange NSE was collared from the website https://ng.investing.com/indices/nse-all-share-historical-data. Text mining techniques such as Term Frequency-Inverse Document Frequency (TF-IDF) are applied to pre-determine important words and sentences and their influences on daily market returns. N-gram sentences are used to build bigrams and trigrams, which helps us determine their positive and negative returns. The result shows that when words such arabia, crypto, renewal, paris, and tradermoni appear in news headlines, there is negative returns in the stock market, but when words such as lawmakers, multinationals, ebonyi, constraints and double appear in the news headline, there is positive return. For bi-gram sentences when sentences such as price dip, earning rise, tax compliance, and econbank partners appear, there is often negative return, on the other hand when sentences such as trustfunds pensions, inter agency, business insurance, index rise and forex supply appear in the news headline, there is often positive return. And for tri-gram sentences, when sentences such as profit taking NSE, government private sector, cross boarder trade, cargo tracking note, and capital market sec, there is often negative return in NSE, on the other hand, when sentence such as poor purchasing power, naira watch cbn, mtn google empower, google empower smes and external reserve hit appear in the news headlines, there is often a positive returns in the stock market. Three machine-learning models were used to build the predictive models. The models were logistic regression with a prediction accuracy of 0.52, Support Vector Machine (SVM) with an accuracy of 0.51, and K-nearest Neighbour (KNN) with an accuracy of 0.99, indicating higher prediction evidence of news headlines by the KNN model for the NSE index over the alternative models. We limited fitting the models using only unigrams and left fitting models using bigrams and trigram sentences for future research.