Korean Real Estate Price Forecasting through Volatility-Aware Time-series Reconstruction

Real estate transaction data in Korea are inherently event-driven, resulting in irregular observation intervals and extended periods without transactions, which make it difficult to construct continuous time-series representations. To address this issue, this study proposes a volatility-aware reconstruction method that transforms fragmented transaction records into continuous apartment-level time-series. Transaction data are reorganized into monthly sequences, and unobserved intervals are reconstructed by jointly considering local temporal continuity and region-level price variations. Experimental results demonstrate that the proposed reconstruction-based approach consistently outperforms conventional strategies that discard incomplete observations, achieving superior predictive performance. Furthermore, incorporating regional market volatility leads to additional performance gains.