Macroeconomic forecasting with echo state networks

Abstract

Forecasting macroeconomic indicators plays a crucial role in economic planning and policy formulation. With the increasing availability of large datasets, there has been a surge in interest towards employing sophisticated forecasting models. This paper explores the performance of Echo State Networks (ESN) in forecasting Gross Domestic Product (GDP) growth, both one-period ahead and multi-step ahead. In addition to ESN, traditional models such as Autoregressive model with lag 1 and Vector Autoregressive models are included for comparison. The Model Confidence Set procedure is adopted to assess the forecasting performance across these models. Through empirical analysis using US Macroeconomic data, the study reveals that ESN exhibits notable forecasting performance, demonstrating its potential as a valuable tool in macroeconomic forecasting.