Time-Varying Autoregressive Models for Economic Forecasting

This research introduces a time-varying autoregressive (TVAR) model, developed to improve the precision of forecasting in economic time series data. The model advances the conventional TVAR framework by incorporating potential mean adjustments and utilizing the Kalman Filter within a Maximum Likelihood Estimation (MLE) framework, with further optimization through the NelderMead method. Applied to the real gross national product (GNP) data of the United States (U.S.), the model effectively captures dynamic patterns and structural changes that traditional models often overlook. The model’s performance is rigorously compared with the widely used Markov switching autoregressive (MSAR) model, demonstrating superior results in both training and testing forecasts. The TVAR model consistently achieves lower error metrics, underscoring its robustness and flexibility in capturing dynamic economic trends and providing reliable forecasts. This research emphasizes the TVAR models potential for broader applications in economic policy analysis, strategic planning, and decision-making processes, particularly in understanding and predicting economic growth. The models adaptability and precision make it a valuable tool for economists and policymakers aiming to navigate complex economic fluctuations with greater confidence as well as accuracy.