An evaluation of competing methods for constructing house price indexes: The case of Warsaw

Accurate house price indexes allow governments and planners to make more effective decisions with regard to land use policy. We compare different ways of computing house price indexes for Warsaw over the period 2006–2018 using a detailed micro-level dataset of 101,182 transactions. We find that when a hedonic approach is used, the resulting index is reasonably robust to the choice of method. More problematic is the repeat-sales method, which is widely used in the US. We find that repeat-sales price indexes are unreliable, affected by sample-selection bias, and prone to significant revisions when new periods are added to the dataset. Even for the hedonic approach, the choice of method can become important for smaller datasets. In such cases, the hedonic time dummy and rolling time dummy methods tend to perform better than the hedonic repricing, hedonic imputation and average characteristics methods. In comparison with the index produced by the National Bank of Poland (NBP), we find that in 2006–7 prices did not rise as much, while since the end of 2012 prices have risen faster than indicated by the NBP index. We attribute these differences to a combination of the choice of hedonic method (we use time-dummy while the NBP index uses the repricing method) and different datasets (the NBP dataset is smaller, especially in the earlier years of the sample).