Answers are grounded in this paper and cite their sources.
This study develops and evaluates models for predicting the realized housing supply from apartment projects permitted in the Seoul Metropolitan Area. Using permit histories from January 2010 to December 2019 as training data and apartment projects permitted in 2020 as a holdout test set, the study compares a simple averaging approach, a survival analysis model, and a Light Gradient Boosting Model (LightGBM) 4-way ensemble model. Whereas the benchmark models rely on static information available at the time of permit approval (forecasting), the proposed machine learning model is designed as a conditional short-term prediction (nowcasting) framework that incorporates dynamically updated administrative information. To isolate the predictive value of administrative supply data, the explanatory variables are limited to supply pipeline- and schedule-related information, including stage-specific backlogs and elapsed-time variables, while excluding market, financial, and spatial variables. The results show that the machine learning-based model, benefiting from this expanded information set, produces lower prediction errors in monthly completion forecasts and more accurately tracks cumulative completions than the benchmark models. These findings suggest that realized housing supply is explained less by aggregate permit volume than by the dynamically evolving backlog and duration structure embedded in the housing supply pipeline. Overall, the study demonstrates that a practically useful conditional prediction framework can be developed using administrative housing supply data alone.
No takes yet. Share an insight, caveat, or question.
Park et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: