Forecasting Financial Time Series Using Machine Learning Models

Abstract

Forecasting economic variables is crucial for policymakers, researchers, and financial institutions, since it facilitates informed decision-making and efficient planning. Monetary Aggregates (M3) are one of these variables that is very important for showing liquidity, guiding monetary policy, and measuring economic stability. In the literature, numerous classical and machine learning techniques have been utilized to predict monetary aggregates. This study utilizes four independent methodologies—Autoregressive Integrated Moving Average (ARIMA), Autoregressive Fractionally Integrated Moving Average (ARFIMA), Extreme Learning Machine (ELM), and Multilayer Perceptron (MLP)—to predict M3 using monthly data. We utilize well-known quality indicators like Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) to rate how well each model works. The results show that the MLP model always does better than the other techniques, with the lowest error values in both the training and testing stages. This shows that MLP neural networks are very good at capturing the nonlinear and complicated dynamics of monetary aggregates. The results show that machine learning approaches, especially MLP, could make economic forecasts more accurate and help make financial and monetary policy decisions based on facts.

Keywords: Financial Time Series, Machine Learning, Multilayer Perception (MLP)