Filling the Gaps with MICE: Addressing Missing Data in Real Estate Price Indices
Filling the Gaps with MICE: Addressing Missing Data in Real Estate Price Indices
Authors: Miriam Steurer, Sabrina S. Spiegel
Abstract: Missing data are a common feature of micro-level transaction data used to construct hedonic real estate price indices. Missingness typically occurs in the descriptive characteristics required for quality adjustment rather than in transaction prices. Since these characteristics are central to hedonic quality adjustment, complete-case analysis can skew measured price dynamics through sample-selection and composition effects. This paper proposes multiple imputation as a way to handle missing characteristic values in index construction. The aim is not to recover individual missing values, but to restore incomplete observations and reduce variability in the estimation sample. We employ multiple imputation by chained equations (MICE) as a flexible imputation framework. Since conventional aggregation rules for multiple imputation, Rubin’s rules, do not align with the multiplicative chaining structure of price indices, we introduce an alternative aggregation method based on pooled growth rates. Empirical evidence from two applications, a large dataset of Vienna apartment transactions and a smaller, more heterogeneous Austrian office market, shows that index estimates are relatively robust to missing data in large, homogeneous settings. In contrast, in thinner and more heterogeneous markets, imputation can materially affect index dynamics. In both settings, flexible MICE specifications with rich predictor sets perform better than simpler imputation methods.
Seminar Notes
Venue
NBER CRIW Pre-Conference 2025
Objective
To introduce the MICE-RF algorithm for imputing missing data in real estate transactions
Importance
Real estate prices are an important economic indicator
Background
EU countries are required to compile hedonic residential property indices. New requirement to compile them for commercial properties starting in 2027
Data & Key Variables
Complete transaction database for Viennese apartments 2015-2023
Austrian commercial real estate data 2015-2024.
Missing around 50% of observations
Methodology
Multiple Imputation by Changed Equations combined with the Random Forest (MICE-RF)
Introduce artificial missingness and check to confirm that MICE-RF corrects biases
Results
MICE-RF preforms well relative to complete case analysis in Viennese apartment data with introduced missingness.
Imputation using MICE-RF significantly alters the price index for Austrian office transaction data.
With missingness, it looks like prices are rising, but with the imputation they are stable or declining.
Key Table/Figure
