Machine Learning for Economic Modeling

Abstract

Accurate estimate of public expenditures is needed for budgetary planning and government decision making. Recent advances in machine learning offers the opportunity for modeling such problems. The paper introduces a novel modeling approach using a machine learning tool to forecast public expenditures and compare and contrast the effectiveness of this approach to traditional modeling alternatives. This research uses historical quarterly data from 1960-2016 to model public expenditures. Various accuracy measures (MAD, MAPE, and RSME) show that the machine learning model is the best alternative formulation and offers 97% forecasting accuracy. This model allows government decision makers to assess alternative policies with specific budgetary impacts. Furthermore, the study also shows that population aging is an important predictor of public expenditures; suggesting that demographic monitoring is indispensable for efficient fiscal planning and management in South Africa.