CONSTRUCTION OF R EGIONAL HOUSE PRICE INDEXES – T HE CASE OF S WEDEN Lars-Erik Eriksson (Valueguard) Han-Suck Song (KTH) Jakob Winstrand (Valueguard) Mats Wilhelmsson (KTH and Uppsala University)
Motivation-Objective NasdaqOMX house price index – Insurance/financial products – Complete market Thin markets – Geographical or temporal aggregation? The objective is to construct house price indexes for all parts of Sweden or at least a large part of the economic value on the single-family housing market.
Literature Schwann (1998) – Nearby observation in time Englund et al (1999) – Temporal aggregation – not recommended McMillen (2003) – Nearby observations in space Francke and Vos (2004) – Nearby observations in time and space
Research Procedure Estimate hedonic price equation for each region Perform a cluster analysis – Geographical proximity – Price development – Price level – Price development (2 years) – Combination Estimate hedonic price equation for each cluster Evaluate the performance of the models – R 2, MSE
Data Single-family houses Transaction price, contract date, size, quality, coordinates 100 labor markets (93 with transactions) – Based on potential commuting
No. of transactions per month
Step 1: The Hedonic Price Equation StockholmVästeråsMora Coefficientst-valueCoefficientst-valueCoefficientst-value Living area Room Room Room Room Room Room Room Room Room Quality index Quality index sq Sea front Sea view Semi-detached Detached Building period: Building period: Building period: Building period: Building period: Building period: Urban R2R No of obs No. of obs/month
Temporal Aggregation Month Year
Step 2: Cluster Analysis Cluster method No of clusters Average no of observations No of observations in smallest cluster Average R 2 C11218,0366, C2924,0569, C3924,04912, C41415,4616, C51119,6718, C61021,6464, C1: Price development C2: Price level C3: Price development 2 year C4: Geographical proximity C5: Geographical proximity + Price development C6: All
Price development and geographical proximity
Step 3: Evaluation Constant implicit prices Non-constant implicit prices
Index – Region 1-5
Index – Region 6-10
Summary statistics INDEXR1R2R3R4R5R6R7R8R9R10 average std.dev Coeff. of variation RETURN average0.55%0.45%0.36%0.52%0.36%0.22%0.54%0.52%0.47%0.43% std.dev2.73%2.21%3.44%1.99%2.55%5.36%2.75%2.33%2.52%2.70% Coeff. of variation
Conclusion Thin markets is a problem It is not obvious how solve it Temporal and geographical aggregation has been criticized Especially arbitrary geographical aggregation New method how to aggregate in space based on cluster analysis of regions – Price development and geographical proximity Out-of-sample test Other measures to evaluate the method? Improve the cluster models