Are taxes too high? A machine-learning approach to Laffer curve estimation

Abstract

This paper estimates Laffer curves for personal income tax, corporate income tax, and value-added tax across OECD countries. While the Laffer curve is widely used for assessing the revenue effects of taxation, existing empirical estimates typically rely on restrictive functional forms and are vulnerable to misspecification, when the true relationship between tax rates and revenues is unknown. In response to this limitation, this paper develops a model that allows data-driven flexibility and enforces the defining properties of the Laffer curve. The parameters governing the curvature and turning points of the curve depend on a rich set of structural and institutional characteristics while LASSO regularisation mitigates overfitting. The results reveal substantial cross-country heterogeneity in revenue-maximising tax rates among OECD countries and suggest there is limited scope for further revenue mobilisation through higher income tax rates in several countries, while highlighting a comparatively greater fiscal space in consumption taxation.