Application of Nonlinear Autoregressive with Exogenous Input (NARX) neural network in macroeconomic forecasting, national goal setting and global competitiveness assessment

TL;DR

This paper investigates how NARX neural networks perform in macroeconomic forecasting, national goal setting, and global competitiveness assessment across different countries, analyzing the impact of various exogenous input configurations.

Contribution

It constructs specific NARX neural networks tailored for macroeconomic and competitiveness applications and evaluates their performance with diverse exogenous input sets across multiple countries.

Findings

NARX neural networks' forecasting accuracy varies with input set size and relevance.

Training on data from similar economies improves prediction accuracy.

NARX models can potentially replace human analysis in macroeconomic tasks.

Abstract

This paper selects the NARX neural network as the method through literature review, and constructs specific NARX neural networks under application scenarios involving macroeconomic forecasting, national goal setting and global competitiveness assessment. Through case studies on China, US and Eurozone, this study explores how those limited & partial exogenous inputs or abundant & comprehensive exogenous inputs, a small set of most relevant exogenous inputs or a large set of exogenous inputs covering all major aspects of the macro economy, whole area related exogenous inputs or both whole area and subdivision area related exogenous inputs specifically affect the forecasting performance of NARX neural networks for specific macroeconomic indicators or indices. Through the case study on Russia this paper explores how the limited & most relevant exogenous inputs set or the abundant & comprehensive exogenous inputs set specifically influences the prediction performance of those specific NARX neural networks for national goal setting. Finally, comparative studies on the application of NARX neural networks for the forecasts of Global Competitiveness Indices (GCIs) of various economies are conducted, in order to explore whether the specific NARX neural network trained on the basis of the GCI related data of some economies can make sufficiently accurate predictions about GCIs of other economies, and whether the specific NARX neural network trained on the basis of the data of some type of economies can give more accurate predictions about GCIs of the same type of economies than those of different type of economies. Based on all of the above successful application, this paper provides policy recommendations on applying fully trained NARX neural networks that are assessed as qualified to assist or even replace the deductive and inductive abilities of the human brain in a variety of appropriate tasks.