Modeling and Forecasting of Financial Time Series in Emerging Markets using Multilayer Perceptrons

Abstract

This study develops a data-driven forecasting framework for the Nigerian Stock Exchange All Share Index (NSE ASI) using a Multilayer Perceptron (MLP) neural network. Financial markets, particularly in emerging economies, are characterized by volatility, regime shifts, and nonlinear dependencies that limit the effectiveness of traditional statistical models. To address these challenges, this work applies a deep learning pipeline incorporating rigorous data preprocessing, feature scaling, and supervised learning for univariate time series prediction. The model is trained on daily NSE ASI data and evaluated using standard metrics such as MSE, RMSE, MAPE, and R². Diagnostic analysis includes autocorrelation structure, outlier detection, and SHAP based interpretability to assess feature influence and market anomalies. The MLP model demonstrates strong predictive performance across both stable and turbulent regimes, notably capturing post-COVID market momentum. The results affirm the suitability of neural networks in modeling financial indices in emerging markets and highlight the value of integrating explainable AI into financial forecasting systems.