Nowcasting Transaction-Based House Price Indices Using Web-Scraped Listings and MIDAS Regression

Timely transaction-based residential property price indices are crucial for effective monetary and macroprudential policy, yet transaction-based data often suffer from significant reporting delays. Online property platforms, by contrast, provide list prices of properties in real-time. This paper examines whether immediately available online list prices can improve timely nowcasts of transaction price movements. Using 16 years of micro-level data from Warsaw and Poznan, we construct quality-adjusted monthly list-price and quarterly transaction-price indices using the hedonic rolling-time-dummy method. We find that list-price indices consistently lead transaction-price indices by one to two months, with the strongest relationship in Warsaw's larger, more liquid market. Building on this lead-lag relationship, we develop a Mixed Data Sampling (MIDAS) regression framework to nowcast quarterly transaction-price growth using monthly list-price data. Our preferred MIDAS specifications reduce one-quarter-ahead root mean square error by approximately 16-23 percent for Warsaw and 5-15 percent for Poznan relative to standard autoregressive benchmarks. The predictive advantage is greatest when incorporating list-price data from the first or second month of the quarter, as third-month data introduce forward-looking noise. Our results show that properly constructed list-price indices can play an important role to provide early housing market signals, potentially enhancing the timeliness of policy responses.