Purpose: We propose a novel scalable regression model that incorporates two-dimensional location dummies based on longitudinal and latitudinal coordinates. This approach easily achieves comparable levels of explanatory power to ML models while avoiding the interpretability challenges inherent in non-parametric ML techniques. Design/methodology/approach: Regression models have long been a fundamental and effective toolkit in economic analysis, facilitating the validation of theoretical predictions and the examination of policy effects. However, their explanatory power is often constrained by linearity constraints or limited information, which can, in turn, distort the estimated marginal effects of explanatory variables. While machine learning (ML) models can serve as an alternative, their improved fit often comes at the cost of increased complexity in interpretation and inference. This paper proposes that incorporating geocoded information into the regression models for property values can significantly enhance explanatory power to a level comparable to ML models while preserving the simplicity of regression approaches. As a case study to validate the proposed method, we examine the impact of a recent deregulation policy in Korea on the market prices of the corresponding properties. Findings: By accounting for location-specific heterogeneity in its full magnitude, our model provides more accurate and reliable estimates for policy evaluation. As a case study, this paper examines the effect of deregulation on apartment redevelopment in Korea, demonstrating how the proposed methodology can be applied to real-world policy analysis. Originality/value: We propose a novel regression model that incorporates grid dummies for the longitudinal and latitudinal coordinates of the properties. This approach easily achieves comparable levels of explanatory power to ML models while avoiding the interpretability challenges inherent in non-parametric ML techniques.
