Advancing macroeconomic forecasting by integrating big data and machine learning

Abstract

Purpose: to assess the role of integrating big data and machine learning (ML) in improving the accuracy of macroeconomic forecasting and to develop an adaptive multi-indicator forecasting framework.

Method: a quantitative data-driven forecasting approach is employed by integrating historical macroeconomic data, real-time data, and unstructured data. Forecasts are generated using random forests, support vector regression (SVR), neural networks, and macroeconomic random forests (MRF), and model performance is evaluated using MAE, RMSE, and MAPE via rolling-window cross-validation.

Findings: ML-based models consistently outperform traditional econometric approaches. MRF achieves the highest accuracy in forecasting GDP and unemployment, while random forests and SVRs perform better at capturing inflation dynamics and short-term fluctuations. The inclusion of real-time, unstructured data enhances the model’s responsiveness to economic volatility and shocks.

Implications: these findings highlight the potential of ML-based macroeconomic forecasting systems as effective decision-support tools for evidence-based policymaking.

Originality: lies in the development of a multi-indicator ML-based forecasting framework that integrates heterogeneous big data in a single integrated system.