Applications Of Machine Learning And ARIMA In Emerging Market: Evidence From Predictive Inefficiency In The Nigerian Stock Exchange

Background: The challenge of forecasting stock prices in the capital markets of developing countries, such as Nigeria, persists. Traditional statistical models, such as the Autoregressive Integrated Moving Average (ARIMA) which have mostly been used for this purpose, are no longer as effective as they are limited in their ability to handly non-linear time-series data. This study investigates the predictive powers of the ARIMA model against selected machine learning algortihms such as Random Forest (RF), Support Vector Regression (SVR), and Long Short-Term Memory (LSTM) networks, for some selected stocks from the Nigerian Exchange Group (NGX). Materials and Methods: The study employed an ex-post facto research design, and utilized historical stock price data of five major companies on the Nigeria Exchange Group (NGX) from 2015 to 2022. Stationarity tests were conducted using Augmented Dickey Fuller (ADF) and thereafter, the ARIMA model was specified using Autocorrelation function (ACF) and Partial Autocorrelation function (PACF) analysis. The Machine Learning models were evaluated using hyperparameter tuning and time-series cross-validation. The Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) were utilized for the assessment of the forecasting accuracy of the ARIMA model and the Machine Learning models. Results: Random Forest had the lowest MAPE, MAE and RMSE for four out of the five stocks indicating that it was the most adequate model for the forecasting of stock prices in Nigeria. Support Vector Regression was second for 4 of the 5 stocks and also had the lowest MAPE, MAE and RMSE for one of the stocks. Random Forest and Support Vector Regression both outperformed ARIMA for all 5 stocks, while ARIMA outperformed Long ShortTerm Memory model., this could be due to overfitting of the model. Conclusion: In summary, the findings of the study show that Random Forest models and Support Vector Regression models are better predictors of the stock market prices in Nigeria. Their superior forecasting capacity challenges the weak-form Efficient Market Hypothesis, which in turns suggests the presence of exploitable nonlinear patterns in the Nigeria Exchange Group. Recommendation: The study recommends the integration of Machine Learning Algorithms such as Random Forest, into equity analysis and trading systems to improve the forecast accuracy and portfolio returns in the Nigerian Market.