1. Introduction

This paper deals with the residential property price indicators for the euro area and euro area countries jointly collected and compiled by the ECB and the EU National Central Banks (NCBs) from various national sources. Additionally, available dwelling price statistics for non-euro area EU countries have been collected since early-2005. Furthermore, from 2001 on, the ECB has compiled an aggregate for the euro area by weighting together changes in prices for houses and flats available for the euro area countries. Since then, the statistical features of both the indicators for the countries and the euro area aggregate have improved, but still remain below the standards of other economic statistics and price indicators for the euro area.

Section 2 outlines the relevance of indicators on changes in residential property prices for the ECB. This is followed by a presentation of the national sources used and the statistical features of the price statistics currently compiled for the euro area. Section 4 outlines the compilation of an aggregated euro area indicator for residential property prices. Finally, future developments are outlined in section 5.

2. The role of residential property prices for ECB analyses

The buying or selling of a dwelling is typically the largest transaction a private household enters into. Changes in residential property prices are therefore likely to influence substantially the budget plans and saving decisions of the potential buyers and sellers. Changes in prices of houses and flats will also have an impact on the wealth of owners of dwellings given that it is the largest asset in their portfolio. According to estimates recently compiled by the ECB, euro area household housing wealth (net of capital depreciation) increased by an annual average rate of more than 7% in the period from 1980 to 2003, and the growth of house prices explains a significant part of this increase. Furthermore, housing wealth accounts for almost 2/3 of total (gross) household wealth. 2

Houses and flats are also purchased as an investment, which creates returns in the form of rental payments in the case of the dwelling being let. In this case, changes in house prices may impact on rents. This is for the time being the main channel through which house price inflation may impact on the rates of change of the Harmonised Index of Consumer Prices (HICP) - the ECB’s measure to define price stability in the euro area. The impact materialises via rents and not directly via house prices, since the HICP covers rental payments by private households (weight in the euro area HICP: 6.3%), but excludes – for the time being – expenditures by owner-occupiers for purchasing or using their own house or flat. Housing price developments may also have an effect on residential construction investment. Finally, housing prices can provide important insights for financial stability analysis, since sharp increases and declines in prices can have a detrimental impact on financial sector health and soundness, by affecting credit quality and the value of collateral.

Given these important uses of changes in dwelling prices, price indicators of good statistical quality are required. Since the analysis of the ECB is usually focussed on euro area wide developments, an aggregated residential property price indicator for the euro area is of particular importance. Due to the divergent development of dwelling prices across euro area countries, results for individual euro area countries are also essential. In addition, a distinction at national and at the euro area level dividing between price developments in urban areas (and/or capital cities) and non-urban areas may be very informative. For the prices for different housing types, the breakdown between new and existing dwellings is the mostly used distinction.

As euro area statistics are compiled from national results, a meaningful euro area aggregate calls for a sufficient degree of comparability between the national data. In order to be useful in the monitoring of price developments, a quarterly frequency of the results is desirable. The degree to which residential property price indicators are able to eliminate the effect of quality differences between different dwellings compared over time is crucial for analysing price changes.

For the time being, the collection and compilation of indicators for changes in residential property prices primarily aim at providing insights into changes in transaction prices. Besides that, statistics on the value of the housing stock can also contribute an important piece of information, e.g. for analysing wealth effects. Additionally, price level data would allow the identification of differences between regions and countries at a certain point in time. Finally, a comprehensive analysis of the real estate market should also refer to units used for trade and commerce.

Price data are not the only statistics required for a comprehensive analysis of the housing market. The share of rented and owner-occupied houses and flats, the number and value of transactions, statistics on building permits, housing starts and completions provide important insights into the structure and the dynamics of the market and their driving factors from the supply and the demand side. Therefore, the ECB has also built up data sets on these statistics, calling them “structural indicators on housing”.

3. ECB dataset for euro area and selected EU countries: Statistical properties

3.1 Available data

The data set on residential property prices collected and compiled by the ECB and the NCBs of the European System of Central Banks (ESCB) comprises indicators for all euro area countries and for eleven non-euro area EU countries. For the time being, only for Cyprus and Slovenia are data on changes in house prices are lacking in the European Union. The available country results are shown in charts 1 and 2. An overview table of the development of residential property prices in euro area countries is shown in the annex.

Chart 1: Residential property prices for euro area countries, annual percentage changes

Sources: National data and ECB calculations. Due to confidentiality reasons data for Italy are not shown.

Chart 2: Residential property prices for non-euro area EU countries, annual percentage changes

Sources: National data and ECB calculations. For Denmark, annual percentage changes are available from 1972 on. Data for Poland are not shown due to experimental character of the underlying indicator.

These data have been collected by NCBs and the ECB from various sources. As alternative indicators from different sources exist in several countries, NCBs and the ECB have typically chosen the indicator that corresponds best to the “target definitions” and other criteria (frequency, timeliness, quality adjustment) that have been agreed upon by the European System of Central Banks (see section 3.10). Nevertheless, in order to understand and interpret these non-harmonised data it is essential to take into account their very heterogeneous statistical features and quality. These differences, which concern presentational aspects like frequency and timeliness and, especially important for the interpretation of the results, the coverage and method of the calculations, are presented in the following sections.

3.2 Data sources

3.3 Periodicity

Monthly statistics are available from Ireland, the Netherlands and the UK, quarterly data from Belgium, the Czech Republic, Denmark, Estonia, Greece, Spain, France, Lithuania, Austria, Finland and Sweden. Data from Italy are reported in semi-annual frequency, whereas for Germany, Luxemburg, Poland and Slovakia only annual information is available. Data on changes in dwelling prices published at a lower than quarterly frequency usually do not satisfy the ECBs requirements for short-term economic analysis. However, it has to be taken into account that in particular for smaller countries and statistics with a smaller coverage, higher periodicity of the data implies fewer covered transactions and this may reduce the representativity of higher frequency results and increase their volatility.

3.4 Period covered

For euro area countries, the original data have - with a few exceptions - a sufficient series length for economic analysis, i.e. 10 years or more in terms of index levels. For non-euro-area EU member states, the longest time series are available for Denmark, starting in 1971. Indicators from Sweden and the UK both cover 15 years. For the eastern European EU member states, time series for Hungary, Latvia, Poland and Slovakia are short, i.e. cover less than five years. However, it should be noted that long time series covering several decades often requires the linking of data sources that may differ in definition and coverage. Therefore, international comparisons of non-harmonised national sources have to take into account that the quality of the data usually lower for periods before the 1990s. 3

3.5 Timeliness

The timeliness of the available data differs considerably. The earliest available figures are the monthly data from Ireland, the Netherlands and the UK, for which data become available between one month and two months after the reporting period. Most other data are provided in a time span of two to eight months after the reporting period. For Luxembourg, the annual price indicator becomes only available about 1.5 years after the end of the reporting year. The timeliness of the data varies also among alternative country indicators. It should be noted that timeliness should not be considered separately, but in connection with the frequency of the data.

3.6 Coverage

Most of the data sources refer to a certain segment of the housing market. Most important are limitations in the geographical coverage and the coverage by dwelling type, since price changes in these sub-markets are in most cases not representative for the whole housing market. The quarterly price index for France and the monthly indicators for the Netherlands, for example, cover only existing dwellings.

In most cases, the intrinsic characteristics of the data source imply that only a segment of the market is covered, since the databases are collections of the purchases financed by a certain mortgage bank or cover the transactions of real estate agencies in which they are involved. Other limitations in coverage are due to the sampling design, i.e. the definition of the objects and the geographical areas for which prices are observed over time.

For the interpretation of the results it is decisive whether the covered share of housing purchases is representative for all market transactions. Geographical limitations to the capital city are unlikely to be representative for the country wide price development. The exclusion of new dwellings might also have a systematic effect on the results. To which extent the limitations to certain quality categories or the restriction to purchased financed by mortgage banks affects the results is not obvious and may vary over time.

3.7 Prices

Transaction prices are typically available to mortgage banks, notaries, land registry offices and tax authorities. Some of the other sources are based on price data of different types: In Spain, e.g. the valuation of dwellings serves as a basis for calculating property values (“open-market appraised housing”). For Germany, the data source used by the NCB comprises typical values quantified by real estate experts who refer to price data of various types, including also non-transaction prices. The Austrian price index makes use of offer prices, provided on an internet platform of real estate agencies. For the non-euro area EU member states, the index of Malta is an example of an indicator which is based on offer prices, i.e. asking prices advertised in newspapers. Given that the transaction price of houses and flats is usually fixed in a bargaining process, the price changes of offer prices might not always properly reflect the dynamics of transaction prices.

3.8 Quality adjustment and sampling

Measuring the “pure” price change means to quantify the price development over time for goods or services keeping their product characteristics constant (“matched models”). Whenever a product’s quality changes over time, the impact of this change on the price development has to be excluded. However, it is not possible to follow this conceptual idea for dwelling price statistics for two main reasons: In the first place, no two dwellings are identical and even standardised terraced houses or apartments differ in important details like location and equipment. Secondly, even if a sample of houses and flats could be designed, so that the physical attributes are almost constant over time, too few of them may be sold in a certain period, thereby making the calculation of reliable high frequency data difficult. Additionally, the location and the size of the land might vary across sampled houses.

The simplest way to form a dwelling price indicator is to calculate average prices, i.e. the arithmetic mean or the median of recorded price data. However, such indicators do not take into account that the price-determining characteristics of the houses and flats entering the sample at different points in time are usually different. Therefore, the reliability of an average-price measure depends on the way it is controlled for changes in the composition of the sample. A detailed specification of the dwelling’s physical attributes and the location sets tight limits to changes in the composition and may therefore allow the identification of pure price changes, whereas average prices of broader defined categories may be influenced by non-price factors and their changes. However, the tighter the specification of the categories is, the less transactions per category can normally be recorded. Hence, there is a trade-off between the accuracy of measuring pure price changes on the one hand, and the representativity of the underlying information on the other. In practice, prices are collected for typical houses and flats usually defined on the basis of more general criteria like “good quality”, “good location” or “medium size”. Others refer to certain, more broadly specified dwelling types like single-family houses, terraced houses or medium sized flats. Combining quality and location assessments and house types can help to come closer to a true price index.

Most of the EU countries have applied the average-price concept for compiling the indicators. However, the criteria used for defining the categories and the typical units can vary greatly across indicators. Information about the physical attributes of a house or flat and its location can also be used for applying a hedonic regression in order to calculate a true quality-adjusted price index. Generally, hedonic regression analysis of house prices usually requires a broad and detailed set of data about housing characteristics. Usually, hedonic house price indices are calculated for geographical regions, taking into account the importance of regional differences. The NSIs in France and Finland calculate the overall index for the whole country by weighting together the regional results. Alternatively, the hedonic price index for the total market can be calculated independently from the geographical breakdowns. This type of hedonic price indices is available for Ireland and the UK.

For the Netherlands, the data of the Land Registry Office are used to compile a so-called “repeat sales”-index. By comparing purchase prices for the same dwelling over different points in time, a “repeat sales”-index aims at controlling for differences in the physical attributes and the location. However, the overall condition of a house or flat might have deteriorated between to sales; the location might have become more attractive, e.g. due to a better connection to public transport. Additionally, since only prices for dwellings are taken into account which are sold more than once, this might not be representative for the whole housing market.

3.9 Aggregation

Aggregating price data by simply averaging the dwelling prices collected in a certain period forms a unit-value index. Various types of such unit-value indices exist in practice, calculated for well-defined and homogeneous segments of the housing markets, but also for total indicators of the whole country, e.g. for Luxembourg and the Netherlands. Although no explicit weighting structure is applied for unit-value indices, an implicit weighting materialises by the number of transactions, valuations, offers etc. for which prices have been sampled the concurrent period. Besides the lack of quality adjustment, another drawback of a simple average indicator is, that the variation of its implicit weighting structure can create a certain amount of volatility in the time series, since the number of transactions may vary substantially over time.

For aggregating indicators broken down by regions, by house types or by quality categories different weights can be used. Applying weights based on housing stock data usually implies a high degree of stability, so that the indicator’s variation over time is almost entirely driven by the changes in prices. If reliable information about the housing stock is not available, it is common practice to use population weights as a proxy. However, for certain purposes, weights which reflect the structure of transactions might be more useful than a stock-related weighting scheme, e.g. in order to fully reflect the dynamics of the current market conditions. A way to limit the volatility stemming from concurrent transaction value weights is to apply a fixed weighting structure, which reflects, e.g., the average number of purchases over several years. However, this might imply that in periods in which only a few houses are purchased in a certain segment of the market, the price changes might still get a high weight in the overall index resulting from former periods’ high transaction values or volumes.

The dwelling price indices for Germany are compiled by using population weights for aggregating average prices for the cities. The indicators for Greece and Italy use the size of dwellings in the housing stock for weighting. For Spain, price data are weighted together on the basis of the number of valuations, whereas the French NSI aggregates its sub-indices for geographical areas by applying the share of the sales value in the index base period.

Generally, the question whether, for certain purposes, a time-varying weighting structure might be preferable to a weighting scheme held constant over several years has not yet been intensively discussed for dwelling price indicators. This might be a consequence of the fact that housing-stock related data are usually fairly stable over time, given the much larger number of existing houses and flats compared to newly built units. Hence, the application of fixed weights might be considered the most straightforward way of compiling an index with stock-related weights. Regarding weights which are supposed to reflect the structure of transactions, their potential volatility is usually considered a non-desirable property. Additionally, in practice, the lack of up-to date information on housing stock and transaction values usually does not allow a high frequency of updating the weights used for residential property price indices.

3.10 Specification and compilation of headline indicators for euro area countries

Since for some of the euro area countries more than one source exists, the ECB specified headline indicators for each of the euro area countries. The following criteria have been applied for selection the country’s headline indicator to be used for compiling the euro area aggregate:

1. Coverage: should be broad, if possible, similar with other euro area countries.