R-Shiny Web Application Development for Multilayer Perceptron State Switching Model for Predicting Regimes of Time Series Returns

Forecasting regimes in financial time series is complicated by nonlinearity and small-sample limitations, where conventional Markov Switching Models (MSMs) and regime-switching autoregressive models often underperform. To address this, we developed an interactive R-Shiny Web Application implementing the recently introduced Multilayer Perceptron State Switching Model (MLPSSM). The app integrates neural networks to capture nonlinear intra-regime patterns with Markov switching structures to identify latent regime transitions. Using Nigerian exchange rate returns as a case study, the app demonstrated robust performance across diagnostics, estimation, and forecasting. Residual checks confirmed that the hybrid approach effectively modeled underlying dynamics, while forecasts achieved lower RMSE and MAE than baseline MSMs. The web-based interface further enhances accessibility, enabling both technical and non-specialist users to apply advanced regime-switching methods without coding expertise. The MLPSSM Web App thus bridges machine learning and econometric modeling, offering a practical, reproducible tool for regime prediction in financial markets.