Economic Forecasting of New Zealand with Electronic Card Transaction Data Using Machine Learning Models

Electronic card transactions encompass various economic activities like retail sales, e-commerce, and business services, offering a comprehensive view of consumer spending essential for forecasting overall economic activity. Policymakers and investors are increasingly interested in reliable high-frequency indicators to track real-time economic conditions. This study evaluates the economic forecasting performance of several machine learning (ML) models using New Zealand data from January 2023 to May 2024. ML models can quickly process large data volumes, enabling near-real-time forecasting of economic variables, facilitating timely decisions for businesses and policymakers. Our results reveal that linear regression (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}=0.99$</tex>), Lasso (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}=0.99$</tex>), decision tree regressors (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}=0.99$</tex>), and OLS regressor (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}=0.98$</tex>) outperformed random forest regressor (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}=0.30$</tex>), KNN (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}=0.06$</tex>), and XGBoost (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}=0.13$</tex>). Overall, using electronic card transaction data with ML models enhances the accuracy, timeliness, and granularity of economic forecasts, providing valuable insights for stakeholders.