Handbook on Residential Property Price Indices Chapter 5

Handbook on Residential Property Price Indices Chapter 5: Methods Jan de Haan UNECE/ILO Meeting, 10 -12 May 2010

House price measurement: three problems Exact matching of properties over time is usually not possible because their quality is changing (we are not comparing like with like). Low incidence of re-sales. The mix of properties changes over time, and a quality mix or compositional change problem arises. Index compilers often face a lack of data, particularly on property characteristics. Those problems have implications for the choice of measurement method. 2

Four main approaches Methods for compiling house price indexes come in four main varieties: central tendency measures: mean or median (Section 5. 2) hedonic regression methods (Section 5. 3) repeat sales methods (Section 5. 4) appraisal-based (SPAR) methods (Section 5. 5) Each of those methods can be combined with stratification, which itself is a form of ‘mix adjustment’ 3

Mean or median indexes Track some measure of central tendency from the distribution of house prices, usually the median, over time. Advantages - easy to explain - reproducible Disadvantages - noisy estimates due to compositional change - does not adjust for quality changes of individual houses - sample selection bias problem (if number of sales per type of houses does not properly reflect number of houses in stock) 4

Stratified mean or median indexes Stratification: separating the total sample into subgroups or strata. The aggregate index is a weighted average of stratum-specific mean or median indexes (using either stock weights or expenditure weights) Trade-off More detailed stratification (according to size, type of house, amenities and possibly price) increases homogeneity and reduces the quality-mix problem, but too detailed stratification leads to a considerable amount of sampling variability Disadvantages - some mix changes will remain - no adjustment for quality changes of the houses 5

Hedonic regression methods Heterogeneous goods can be described by their (performance) characteristics, i. e. a good is essentially a bundle of characteristics. Housing context: characteristics of both the structure and the land (location) of the properties Demand supply for the properties implicitly determine the characteristics’ marginal contributions to the property prices. Hedonic regression techniques can be used to estimate the (implicit) characteristics prices. 6

Hedonic regression methods (2) Issues - choice of model specification: linear or log-linear - choice of explanatory variables (in practice some potentially important variables will be unavailable or hard to measure directly) - multicollinearity (produces ‘unstable’ coefficients) - treatment of outliers - treatment of land/location and land-structures split (separate chapter in the handbook) Two options for constructing hedonic indexes - imputation approach: separate equations for each time period - time dummy variable approach: pooling data into one overall equation for multiple periods (restrictive assumption of ‘fixed’ parameters) 7

Hedonic regression methods (3) Advantages - method adjusts for both sample mix changes and quality changes if the list of property characteristics is sufficiently detailed - method can be modified to decompose the property price index into land structures components (though there additional practical problems) Disadvantages - data intensive: requires data on all relevant characteristics - it may be difficult to control sufficiently for location - different choices lead to varying estimates of aggregate price change (need for adequate metadata) - some technicalities may not be easy to explain to users 8

Repeat sales methods Essentially a matched model approach: it utilizes information on ‘identical’ properties which have been sold more than once. Because ‘matched houses’ are used, there is no change in the quality mix to control for. The method also automatically controls for micro location (address). Low incidence of re-sales: pooled regression on repeat sales data (time dummies ‘explain’ the price change) Best understood as a modification of the time dummy hedonic approach, estimated on a sample of repeat sales 9

Repeat sales methods (2) Advantages - needs no characteristics of the properties other than address - repeat sales regressions are easy to run and the indexes easy to construct - no imputations involved Disadvantages - inefficiency: the method does not use all of the available sales prices - no adjustment for quality changes of the properties - sample selection bias: repeat sales sample may not reflect all sales or the housing stock - detailed stratification impossible due to a lack of repeat sales observations - ‘revision’ of previous estimates when new data becomes available 10

Appraisal-based methods Such methods combine sales prices and assessed values or appraisals (which are often collected for taxation purposes). It is possible to uses appraisals in a repeat sales context by using them as the first or second price measure in a repeat sales regression. Better alternative: If (official government) assessments are available for all properties, pertaining to a particular reference date, then the matched model methodology can be applied in a standard index number formula framework. The ‘missing’ selling prices are imputed by estimated values, i. e. the appraisals. Price relatives are sale price appraisal (SPAR) ratio’s. 11

Appraisal-based methods (2) Advantages - no information on property characteristics needed - compared with repeat sales methods: uses much more data; sample selection bias is likely to be smaller; SPAR method does not suffer from revisions Disadvantages - the methods do not deal adequately with quality changes - they are dependent on the quality of the (base period) assessment information (the exact way the valuations are carried out may not be clear and has an unknown impact on the results) - if property characteristics are completely lacking, the methods can only be used to estimate one overall index 12

Further work Chapter 5 includes - details on the four main approaches - index number formulae - references to the literature Work to be done - coordination with Chapter 4 (conceptual framework) and Chapter (? ) on land-structures split using hedonics - incorporation of comments - additional chapter on model pricing? 13