This paper applies a Bayesian Variable Selection (BVS) framework to forecast the year-on-year growth rate of China’s newly built housing price index (HPI YoY). Using a broad set of macroeconomic and financial predictors, we implement a hierarchical BVS model with rolling-window estimation and direct multi-horizon forecasting. Out-of-sample performance is evaluated against autoregressive (AR) models, the random walk (RW), and a machine learning benchmark, the Random Forest (RF). The results show that BVS consistently outperforms AR and RW across most horizons in both point and density forecasts, and dominates RF within two years ($h=1$–$18$), while RF is only slightly better at very long horizons ($h=24,30,36$). Horizon-specific predictors provide further insights: lagged HPI and market sentiment (RECI) drive short-run dynamics; RECI, inflation (CPI), and housing credit conditions (HPF, IHLL) matter at medium to long horizons; and demographic fundamentals (PNGR) dominate at long and ultra-long horizons. The forecast results point to a subdued housing market over the next two years, consistent with a structural break around April 2022 and the lasting impact of demographic shifts. Overall, the study demonstrates that BVS not only improves forecasting accuracy but also enhances interpretability, making it a valuable tool for academic research and housing market policy design.