Addressing Missing Data in Real Estate Price Indices Using MICE

This National Bureau of Economic Research working paper addresses the challenge of missing data in micro-level transaction records used for constructing hedonic real estate price indices. Missing values typically occur in descriptive characteristics essential for quality adjustment, potentially skewing price dynamics through sample-selection biases if handled via complete-case analysis. The authors propose using Multiple Imputation by Chained Equations (MICE) to restore incomplete observations and reduce estimation variability. Recognizing that standard aggregation rules do not fit the multiplicative chaining structure of price indices, they introduce a novel method based on pooled growth rates. Empirical tests on Vienna apartment transactions and the Austrian office market reveal that while large, homogeneous markets are robust to missing data, thinner and heterogeneous markets see significant impacts from imputation. The study concludes that flexible MICE specifications with rich predictor sets outperform simpler methods, offering a more accurate approach to index construction in complex real estate environments.