COMPARATIVE ANALYSIS OF GEOMETRIC BROWNIAN MOTION, ARTIFICIAL NEURAL NETWORK AND NAIVE BAYESIAN TECHNIQUES USING NIGERIA STOCK EXCHANGE DATA

Abstract

This paper presents a comparative analysis of three market price forecast models, namely Geometric Brownian Motion, Artificial Neural Network, and Naive Bayesian techniques, using data from the Nigeria Stock Exchange. The exploratory data analysis results indicate slight variations in the mean and median of the log stock price over a 5-year period. The data shows a relatively small spread from the mean and approximately symmetric distribution, as indicated by a skewness value close to zero. The normality test confirms that the log stock price data follows a normal distribution. The forecast using Artificial Neural Network (ANN) shows a minimal change in future stock price, suggesting low returns and moderate risk in the Nigerian stock market. The graphical representation of the ANN model demonstrates a constant path with little variation. Similarly, the Naive Bayesian technique provides a similar forecast to the ANN model, indicating limited profit potential. The Geometric Brownian Motion model also forecasts little variation in future stock prices, with 2023 showing slightly higher values. The accuracy of the forecast models is evaluated using the Mean Absolute Percentage Error (MAPE). The results indicate that the ANN model has an error of 4.60%, the Naive Bayesian model has an error of 9.29%, and the Geometric Brownian Motion model has an error of 12.67%. These findings suggest that the ANN model performs better in terms of accuracy compared to the other two models.