A key element in understanding the global recession is the movement in house price indices (HPIs). Methodological differences in compiling HPIs plague and can undermine both within-country and cross-country analysis of house price cycles and their determinants. It is a difficult but important area of study. There are empirical questions, such as, whether measurement differences matter and, if so, how and to what extent, and second, how such differences impact on some Fund analytical work including the modeling of house price inflation and the measurement of global house price indices.

In the March 2010 issue of this Research Bulletin, Prakash Loungani summarized research, much of it IMF work, that compared the present housing cycle with previous ones in OECD countries. The article highlighted the broad features of house price cycles and the depth of the current trough; anchoring of house prices; factors that amplify the response of house prices to fundamentals; country coincidence of house price changes; and the effectiveness of monetary policy to keep house prices in check. Understanding the causes and consequences of the housing cycle, and its implications for the broader economy and the appropriate policy response, have become a key focus of attention in recent years of central banks and international institutions (IMF, 2008 and 2011). Yet the underlying series largely used for these analytical studies, house price indices, are particularly prone to measurement differences both between and within countries. More than one national HPI can exist within a country—for example, there are eight national HPIs for the United Kingdom with similar trends and turning points, but the timing and amplitude of the turning points differ significantly (Silver, 2011). Such measurement problems are particularly conspicuous and problematic for IMF country surveillance.

There are country-specific studies as to why national HPIs differ, including Leventis (2008) for the United States and Carless (2011) for the United Kingdom. Measurement issues are not always ignored in studies of house price inflation (Igan and Loungani, 2012, Appendix). However, such studies are the exception. Often HPIs are produced by private sector organizations and metadata on their compilation practices can be quite limited.

HPIs are not easily measured. Transactions on the same property are infrequent and the transactions taking place in any period are for heterogeneous properties. Comparisons of the average price of like-with-like properties on a monthly or quarterly basis require a quality-mix adjustment, the nature and effectiveness of which varies between data source and country. Secondary source data are generally used for HPIs and include appraisal/completion prices from mortgagees or tax offices, transaction prices from land registry records, and asking prices from realtors. The coverage, reliability, and timeliness of such price data depend on the institutional arrangements in a country for selling, financing, taxing, and registering the sale of a residential property.

HPI measurement differences may arise from: (i) the method of enabling constant quality measures for average price changes (repeat sales pricing, hedonic approach, mix-adjustment through stratification, sale price appraisal ratio (SPAR)); (ii) type of price (asking, transaction, appraisal); (iii) use of stocks or flows (transactions) for weights; (iv) use of values or quantities for weights; (v) use of fixed or chained weights; (vi) aggregation procedure; (vii) geographical coverage (capital city, urban, etc.); (viii) coverage by type of housing (single family house, apartment, etc.); and (ix) vintage (new or existing property). Details of research on HPI measurement methods and international standards as to good compilation practices are given in a draft Handbook on Residential Property Price Indices near completion.¹ The Handbook contains particularly detailed accounts of methods of aggregation and quality-mix adjustment, though readers are further referred to Bourassa and others (2006) and Vries and others (2009) on SPAR, Hill (2011) on hedonic regression based quality-mix adjustment, and Mason and Pryce (2011) on the repeat sales approach.

Given the potential for major differences in HPIs due to measurement practices, Silver (2012), in a recent paper, considered: whether measurement mattered and, if so, how and to what extent and, further, how measurement differences impact on some Fund analytical work including models of house price inflation and the measurement of global average HPIs. To explore these issues, a panel data set was compiled that comprised five years of quarterly data (2005:Q1 to 2010:Q1) for 150 HPIs from 24 major countries, along with explanatory variables on each of the methodological and coverage descriptors associated with each HPI.

“Understanding the causes and consequences of the housing cycle, and its implications for the broader economy and the appropriate policy responses, have become a key focus of attention in recent years.”

To determine the effects of measurement on HPIs a regression was estimated of house price inflation (quarterly annual rates) on measurement-related variables and fixed-country effects. We found measurement matters for house price inflation, particularly when it really matters, in a recession. Prior to the recession measurement variables had little explanatory effect on house price inflation. By mid-2009 the regressions with only fixed country effects and measurement variables included—no market structural/financing variables—had substantial explanatory power ${\overline{R}}^2$ at about 0.50 (Silver, 2012, Table 3). The parameter estimates were allowed to vary over time and were shown to have quite distinct patterns during the recession compared with prior time periods.

Estimates of country house price inflation controlled for measurement differences were derived by including in the regression dummy time variables that interact with each of the 23 country dummies. By re-estimating the model without the measurement variables, counterpart unadjusted national HPI change series were also derived. The econometric model of house price inflation in Igan and Loungani (2012) was taken to illustrate the impact of measurement differences on models of house price inflation.

Igan and Loungani (2012) regressed (real) house price inflation on disposable income, affordability, working-age population, equity prices, credit, and the level of short- and long-term interest rates using quarterly data for 22 advanced economies. Implicit in such analysis is the assumption that measurement-related differences in HPIs within and across countries are not of a nature/sufficient magnitude to adversely affect the analysis.

The Igan and Loungani (2012) model was estimated using our measurement-adjusted and unadjusted estimates of house price inflation. Measurement-adjusted HPIs were found to out-perform unadjusted ones in the modelling. Both stock price changes and long-term interest rates had no (statistically significant at a 5 percent level) effect on HPI changes for both the Igan and Loungani model and unadjusted estimates, but did so with the appropriate sign for the measurement-adjusted estimates. Some parameter estimates for measurement-adjusted price changes had larger falls and smaller increases than their unadjusted counterparts. For example, measurement-adjusted and unadjusted house price inflation were estimated to fall by 8.5 and 7.7 percent respectively as (lagged) affordability increased by 1 percent; to increase by 0.40 and 0.52 percent respectively as the change in income per capita increased by 1 percent; and to increase by 0.156 and 0.186 percent respectively as the change in credit increased by 1 percent (Silver, 2012, Table 1).

The adverse effect of using unadjusted HPIs in modelling was mitigated by allowing parameter estimates to vary by country (Silver, 2012, Figure 3). This gives some credence to the Igan and Loungani (2012) model as fairly robust to such measurement differences as long as variable country explanatory effects are specified. It also calls into question simple bivariate analysis of the determinants of house price inflation based on cross-country scatter diagrams.

Measurement-adjusted and unadjusted house price inflation series were also used to compile indices of (GDP-PPP weighted) global house price inflation to determine the distinctive effect of these measurement variables on such global measures.

HPI measurement problems carry over to estimates of global house price inflation, such as Loungani (2012). The evidence is that unadjusted global inflation rates were substantially over-estimated in specific quarters at the start of the recession (Silver, 2012, Figure 4). Given the quite different national country methodologies for and coverage of HPIs, the nature and extent of errors/bias in summary averages is difficult to determine but may be substantial.

A need to improve and harmonize HPIs is recognized by the international organizations responsible for setting standards in economic measurement. The setting of standards on real estate price indicators and the dissemination of these indicators are key elements of Recommendation 19 of the report The Financial Crisis and Information Gaps, endorsed at the meeting of the G-20 Finance Ministers and Central Bank Governors on November 7, 2009. The IMF’s Statistics Department is working as members of the Interagency Group on Economic and Financial Statistics (IAG) and the Inter-Secretariat Working Group on Price Statistics (IWGPS) to set standards for the measurement of HPIs (recently completed) and commercial property price indices (CPPIs). The implementation of such standards for HPIs is more problematic though a notable program is that undertaken by Eurostat to attempt to harmonize HPI measurement across European member states.

The focus on measurement issues for price statistics in this article is part of a continuing IMF research program as outlined by the author in the September 2006 and March 2011 issues of the Research Bulletin. Future research may look at commercial property price indices, for which the underlying transaction data refer to particularly heterogeneous properties whose transaction prices dry up in times of recession, just when it really matters—an altogether harder problem.