Supply and demand dynamics in the housing market in Turkey: An analysis using Spatial Panel ARDL and machine learning approaches

This study aims to analyze the supply and demand dynamics of the housing sector in Turkey within the framework of spatial interactions and nonlinear relationships. Although the housing market holds strategic importance for economic growth, financial stability, and social welfare, traditional models based on linear structures and single-region assumptions remain insufficient to explain the complex nature of the market. To overcome these limitations, this study adopts an integrated analytical framework that combines spatial econometric methods and machine learning techniques. Using a monthly panel dataset covering the period from January 2013 to January 2025 at the NUTS-2 regional level in Turkey, the study first examines spatial dependence structures and cointegration relationships. Subsequently, housing supply and demand are analyzed separately through a Spatial Panel ARDL model, allowing for the distinction between short-run and long-run dynamics. The empirical findings reveal that spatial spillover effects are statistically significant on both the supply and demand sides of the housing market, and that ignoring regional interactions leads to model misspecification and bias. To complement the econometric analysis, forecasting models for housing supply and demand are developed using the XGBoost algorithm within a machine learning framework. Model interpretability is ensured through the SHAP approach, which demonstrates that the marginal effects of explanatory variables vary across regions and over time. The design incorporating spatial information indicates that the inclusion of spatial features significantly improves predictive performance. The results suggest that uniform and demand-oriented policy approaches have limited effectiveness in the Turkish housing market. Region-specific credit policies, land-use and zoning regulations aimed at increasing supply elasticity, and the development of spatially informed early warning systems emerge as key policy implications for ensuring market stability. Overall, this dissertation provides a data-driven, integrated, and policy-relevant methodological framework for analyzing the housing market in Turkey.