APPENDIX CHANGES TO THE WEIGHTS AND STRATIFICATION METHOD USED IN HPI
INTRODUCTION
1 The purpose of this appendix is to describe changes to the weights and stratification method used in the construction of the Established House Price Index (HPI) which take effect in this issue of 6416.0.
2 The ABS undertakes periodic reviews of its price indexes to ensure that they continue to meet users' needs and reflect current economic conditions. The main objective of a review is to update the quantities of goods and services that underpin the weights of the index, but it also provides an opportunity to reassess the structure and compilation methodology of the index.
3 During 2007 and 2008, the ABS undertook a review of the HPI. As a result, the housing stock weights have been updated using quantity data from the 2006 Census of Population and Housing and the method of stratification used to compile the index has been refined.
STRATIFICATION IN THE HPI
4 The HPI uses the change in the prices of detached houses sold each quarter to represent quarterly changes in the value of the stock of detached houses. The stock of houses is a heterogenous set of dwellings, as is the subset of houses sold every quarter. Not only will the houses sold in any quarter vary in terms of characteristics such as location, size and price, but from one quarter to the next there will be a change, in terms of these characteristics, in the composition of the set of houses sold. Stratification is used to control for this 'compositional' effect by grouping (or 'clustering') houses according to a set of price determining characteristics.
5 A major review of the HPI was undertaken in 2004, with the changes implemented in the September quarter 2005 issue of 6416.0 and backcast to March quarter 2002. The main outcomes of the review were the introduction of stock weights (see the section on New Weights, below, for more information on this method of weighting) and improvements to the timeliness of the index and the method of stratification. Improvements to timeliness were achieved by changing the timing of transactions from settlement date to exchange date, and the use of the more up-to-date data from lending institutions to supplement valuers' general data in the two most recent quarters. The 2004 review found that given the absence of the detailed data required for hedonic and repeat sales approaches, the most feasible option for controlling compositional effects remained the stratification approach.
6 A stratification was developed based on attributes that can be broadly defined as the structural, locational and 'neighbourhood' characteristics of suburbs. An analysis determined that four structural variables, four locational variables and one neighbourhood variable were the most relevant in determining the similarity of suburbs for stratification purposes. The structural variables were determined from 2001 Census of Population and Housing data and described the percentage of dwellings in a suburb with particular characteristics, such as number of bedrooms. The locational variables were determined from geographic data and described average distance to facilities, such as the nearest shops, by suburb. The neighbourhood variable was represented by the ABS Socio-Economic Index for Areas (SEIFA(footnote 1) ), which is a measure, derived from Census data, summarising different aspects of the socio-economic conditions of people living in an area. The number of non-SEIFA variables were reduced into two principal components, one each for the structural variables and the locational variables. A process of cluster analysis was then undertaken using these two principal components and SEIFA as variables to select the optimal number of clusters. As there was an aim at the time to publish the HPI at lower levels than the city, this analysis was applied with a constraint to ensure that only suburbs within the same statistical subdivision(footnote 2) (SSD) could be grouped together and clusters could not cross SSD boundaries. A detailed description of the changes to the index resulting from the 2004 review is provided in Information Paper: Renovating the Established House Price Index (cat. no. 6417.0).
7 As part of ongoing improvements to the quality of the indexes, in 2007 and 2008 a stratification review was undertaken. Its purpose was to take account of economic, social and demographic changes since the 2001 Census; to identify improved methods of managing compositional change; and to examine the current SSD constraint, to determine whether its relaxation might enable clusters to comprise broader groupings of similar suburbs. The review also considered work by the Reserve Bank of Australia which suggested that a stratification based solely on long-term median price would produce a robust measure of the movement in house prices (Prasad and Richards, Measuring Housing Price Growth: Using Stratification to Improve Median-based Measures, RDP2006-04). With the SSD constraint removed an analysis with updated data was undertaken of alternative stratification methods using combinations of the variables used in the 2004 review, plus a further variable: that of long-term median price. The outcome of this analysis is the selection of a stratification method based on SEIFA and long-term median price which balances the homogeneity of suburbs within the same cluster with sufficient sales observations to construct reliable measurements of price movement. For more information on the analysis, refer to the ABS Research Paper: Refining the Stratification for the Established House Price Index (cat. no. 1352.0.55.093).
IMPACT OF CHANGES TO THE STRATIFICATION METHOD
8 The refined stratification method has resulted in fewer clusters. They are more homogenous and there should be a smaller number that suffer from volatile movements in median price. This should result in an improved measurement of the quarter-to-quarter change in established house prices. The following table shows the number of clusters now used for each city (Series 2), compared to previous series (Series 1):
Number of clusters |
| |
| Series 2 (from June quarter 2008)(a) | Series 1 (from March quarter 2002) | |
| |
Sydney | 22 | 55 | |
Melbourne | 20 | 39 | |
Brisbane | 20 | 51 | |
Adelaide | 11 | 27 | |
Perth | 10 | 14 | |
Hobart | 5 | 8 | |
Darwin | 6 | 5 | |
Canberra | 7 | 14 | |
| |
(a) Refer to paragraphs 12 and 13, below, for details on the introduction of the new series. |
NEW WEIGHTS
9 The HPI is compiled using weights relating to the stock of established houses. The weights are expressed in terms of stock values (originally derived from the 2001 Census of Population and Housing). An initial value of the established housing stock in each cluster was estimated by aggregating suburb counts to clusters and valuing them at March quarter 2002 mean prices. It is important to understand that it is not the stock values that are held constant from period to period. What is held constant is the number of houses underpinning these values. The ratio of the observed median prices of the clusters for the current and previous quarters (known as the price relative) is used to move forward these stock values for each cluster in each city. Algebraically, this produces the same outcome as weighting together prices for each cluster in each quarter using quantities as the weights but it is much easier to implement operationally.
10 Over time the number of houses in a city will change, therefore to keep the index relevant it is necessary to update the quantities which underpin the housing stock weights. In this reweight, house counts from the 2006 Census of Population and Housing have been used to derive new values to replace those calculated with data from the 2001 Census. The quantities have been valued at March quarter 2008 mean prices, thus March quarter 2008 is the period which links the new series to the old.
11 Cities are defined by the geographical classification level of the statistical division
^{2} (SD) as used in the Census. Consequently, for instance, the weight of Brisbane (and the price information collected) is based only on the Brisbane SD and does not include houses in nearby SDs which might cover, for example, suburbs of the Gold Coast. Weights for a city can change over time due to changes in the number of dwellings and/or in the price of dwellings relative to other cities. The following table shows the weights for the new series compared to the previous series:
Percentage contribution to eight capital cities(a) |
| |
| Series 2 (at March quarter 2008)(b) | Series 1 (at March quarter 2002) | |
| % | % | |
| |
Sydney | 33.5 | 43.5 | |
Melbourne | 27.3 | 27.5 | |
Brisbane | 13.2 | 9.8 | |
Adelaide | 7.6 | 6.3 | |
Perth | 14.2 | 9.2 | |
Hobart | 1.2 | 0.9 | |
Darwin | 0.6 | 0.5 | |
Canberra | 2.5 | 2.3 | |
8 capital cities | 100.0 | 100.0 | |
| |
(a) Percentages may not add due to rounding. |
(b) Refer to paragraphs 12 and 13, below, for details on the introduction of the new series. |
IMPLEMENTING THE NEW WEIGHTS AND CLUSTERS
12 The new HPI series commences in June quarter 2008, which in this issue is the most recent quarter of the benchmark series. The benchmark series consists of index numbers produced with only valuers' general data and are not subject to revision (see paragraphs 12 and 13 of the Explanatory Notes for further detail).
13 The new price index series with updated weights and structures is joined to the existing index to form a continuous series via a process known as chain linking. At the link period, which in this instance is March quarter 2008, new housing stock weights and structures are introduced in parallel to the old basis and median prices are calculated using both the new and old clusters. The published index number for the link quarter is produced on the old basis, however index numbers from this quarter are derived by moving forward the new link period values with price relatives of the new clusters.
IMPACT ON REVISIONS
14 In the HPI, estimates for the two most recent quarters in each issue (the leading indicator series) are preliminary and subject to revision. In this issue, the revisions to the June and September quarters 2008 not only reflect changes in the composition of banks and valuers' general data in the set of prices used to derive medians (the usual reason for revisions), but they also reflect changes to the weights of the index, and changes arising from the use of a new stratification method.
IMPACT ON TABLES 7 AND 8
15 There are some revisions to the unstratified medians and numbers of house transfers, published in Tables 7 and 8 respectively, arising from changes to compilation necessitated by the new clustering approach.
FURTHER INFORMATION
16 For further information on the changes described above, contact Mark Dubner, Assistant Director, House Price Index Section, on Sydney (02) 9268 4448.
17 The ABS publication,
A Guide to House Price Indexes, Australia, 2006 (cat. no. 6464.0), contains further detail on the concepts, sources and methods used in the HPI. An updated edition of this publication covering the above changes, will be released during 2009.
1 For more information on SEIFA refer to Information Paper:
Census of Population and Housing - Socio-Economic Indexes for Areas, Australia (cat. no. 2039.0).
<back
2 The Australian Standard Geographic Classification (ASGC) is a set of hierarchical geographic structures. The main structure consists of spatial units in each of the following hierarchical levels: Australia; States/Territories; Statistical Divisions (SDs); Statistical Subdivisions (SSDs); Statistical Local Areas and Census Collection Districts. The HPI weights and structure have been updated using the 2006 edition of the ASGC (which was used in the 2006 Census of Population and Housing).
<back