House Price Indices: Does Measurement Matter?
The significance of a housing bust
October 2009 Report to the G-20 Finance Ministers and Central Bank Governors on
the Financial Crisis and Information Gaps argues that data on
dwellings and their associated price changes are critical ingredients for
financial stability policy analysis. The six major banking crises in advanced
countries since the mid–1970s were all associated with a housing bust (Reinhart
and Rogoff, 2009). Claessens,
Kose, and Terrones
(2008, page 25) also find that “..recessions
associated with house price busts are on average over a quarter longer than
those without busts. Moreover, output declines (and corresponding cumulative
losses) are typically much larger in recessions with busts, 2.2 (3.7) percent
versus 1.5 (2.3) percent in those without busts. These sizeable differences
also extend to the other macroeconomic variables, including consumption,
investment and the unemployment rate.”
an understanding of deviations from equilibrium prices in housing markets requires
reliable and, for international comparisons, consistently-measured, residential
property price indices (RPPIs)—hereafter the terms RPPIs and house prices
indexes (HPIs) are used interchangeably. RPPIs may also be used in the
measurement of the cost of owner-occupied housing in consumer price indexes and
inconsistencies in RPPI measurement practices may compromise the integrity and
comparability of such measures (Diewert, 2004). RPPIs
are particularly prone to methodological differences, which can undermine both
within-country and cross-country analysis. It is a difficult but important
area. There is an important empirical question as to whether measurement
background, three case studies are considered in this paper to illustrate the
problem—RPPIs for the United Kingdom
the United States
, and the Russian Federation
potential for mismeasurement
price indices depend for their data on the way in which houses are bought and
sold and the type of records kept. Different agencies have different records
and can and do compile their own measures. They may be based on the asking
prices by realtors, the appraisal prices of mortgagees/tax authorities, or
registered transaction prices. A house sold in one month will not have a
corresponding sales price in each subsequent month, so there is a need to
control for changes in the quality-mix of houses sold: higher proportions of
more-expensive houses sold in a month should not manifest themselves as price
increases. Methods of adjusting for
quality mix include the use of hedonic regressions, repeat sales data only, and
mix-adjustments by weighting detailed strata. There is a need to weight price
changes over different regions and over types of “houses” (apartments,
detached, townhouses etc.) and the weights may be the relative number or the
relative value of either transactions or the stock of houses. And house price
index measurement can differ in many other respects including their coverage
and the manner in which the quality-mix adjustment and weighted aggregation is
implemented—Fenwick (2006), Hill (2011) and Eurostat
care is required when comparing the house price indices of different countries
for which the nature of source data and methods employed may be quite
different. But what of measuring house prices changes for an individual
country? Surely that should be straightforward. Economists should have a single
reliable measure upon which to base their work. The proliferation of house
price indices in the UK
illustrate that this is not the case.
Figure 2 shows the more volatile quarter-on-quarter
changes, though for 63 percent of the periods all series changes in the same
direction and for 85 percent of the periods only one showed a different
direction of change. Yet the magnitudes of the difference are often
substantial, especially for the Rightmove. For
example, Rightmove showed a 2009 Q2/Q1 quarterly fall of 1.1 percent when all
other indices showed increases averaging 3.5 percent. The coincidence of the
trends and turning points suggests either some commonality in measurement or,
in these respects, measurement differences don’t matter.
Land Registry and AcadHPI
are both based on the same data and it can be seen from Figure 2
that they do track each other more closely. They
both record the price “on completion,” that is, the price returned to the Land
Registry as part of the legal process of registering the completed sale. They are
comprehensive in their coverage of transactions, at
least for England and Wales. There is
a need to control for the changing mix in the quality of houses sold. More
expensive houses may be sold one month leading us to think average prices have
increased when this may not be the case. The repeat sales methodology employed
by Land Registry constrains its coverage to properties transacted more than
once, so that an average of the price changes of the same houses is compiled. The
prices of like are, broadly speaking, compared with like, at the cost of using
a more limited sample and a selection bias. AcadHPI
uses the Land Registry’s entire transaction database. Its mechanism of adjusting
for the differential quality mix is to weight strata categorized
by property type and location. The
weights are transactions-based relative quantities between
January 2000 and December 2003. Land Registry is implicitly weighted by the
relative number of repeat transactions in the sample.
Nationwide and Halifax
include prices for properties for which they are the mortgagee. The DCLG index
covers all transactions bought with a mortgage issued by one of about 50
lenders (about 55 percent of mortgage transactions—more than Nationwide and Halifax) reporting to the
Regulated Mortgage Survey of the Council of Mortgage Lenders. All three indices
cover transactions in the United
Kingdom but, unlike the Land Registry-based
indices, exclude cash sales—about 25 percent of all purchase. While the
lender-based source data have some similarities, there is much in their
construction that differs.
Nationwide and Halifax are based on the asking price
when a mortgage is first approved—when the property is under offer, that is
later than when first advertised but prior to completion. Not all approved
applications will go through to completion.
DCLG is based on transaction prices “on completion.” DCLG is a
average of individual stratum indices with weights updated annually based on a
three year moving average. Nationwide and Halifax are complied
as stock quantity-weighted averages of the strata. Weights for Nationwide are updated every two years based on four year
moving averages of data, while Halifax uses constant weights from
1983. All three indices use hedonic regressions to minimize the
effect of changes in the quality-mix on price measurement, though the
specifications of the regressions differ. DCLG strata are both valued and defined
by quality (price-determining) characteristics of properties estimated from
hedonic regressions. As an example,
DCLG include in their regression variables relating to location (local
authority district or London borough), property type (PT), type of neighborhood
(using the ACORN classification), local authority cluster membership, defined
by the Office for National Statistics, number of habitable rooms, old or new
(New), first-time buyer or former owner-occupier (FT), plus interaction terms
for ACORN and PT, ACORN and FT, and PT and New. Each combination of the
variables in the regression forms a stratum defined by the combinations of
characteristics of the property, about 1,000 property types/strata. Halifax and
Nationwide define “typical” properties by fixed characteristic sets and value
them over time using the estimated coefficients of hedonic regression
In spite of these quite substantial differences, the
indices can be seen to track the broad phenomena of the trend and turning
points in residential property prices. Interestingly, there is no more
commonality between mortgage-based indices than those based on Land Registry
advertises properties for sale online throughout the United Kingdom covering
asking prices of the 90 percent of estate agents stated to advertise on their
site. Properties that do not sell are also included. The index is compiled from
the asking prices of properties, the prices at the very beginning of the buying
and selling process. Weights (and mix-adjustment) are according to the stock of
properties in terms of geographical distribution and property type. Rightmove is
distinct in its use of asking price and has the least commonality with other
points are apparent:
Similar trends and turning
points are tracked by all indices in spite of their data and methodological
differences, yet substantial differences remain especially at some peaks and
troughs, and more so for quarter-on-quarter comparisons.
The use of quarterly annual changes in Figure 1 does much to
smooth the discrepancies endemic in the quarter-on-quarter indices in Figure 2.
- Indices using similar source
data seem to move more closely together, even though the coverage of the data
and methodology may vary considerably. The correspondence between the Land
Registry and AcadHPi is the most striking, being
based on the same data but having very different coverage, weighting, and
methodologies for controlling for quality mix.
But these points are based on a single
country’s experience. Consider a further example, the United States.
States has two main indices for residential property prices, the Federal
Housing Finance Agency (FHFA)
“purchases only” house price index and the S&P/Case-Shiller
National Home Price Index. The FHFA produces an “all transactions” HPI that includes refinance appraisals that are
not sales that comprise nearly 90 percent (about 35 million of the 40 million
repeat transactions). FHFA itself notes evidence that prices based on
appraisals submitted for refinancing tend to lag market trends and have an
appraisal bias. The “purchases-only” HPI excludes refinancing transactions. Leventis (2008) estimates that removing appraisals accounts
for 1.54 percentage points of FHFAs 4.27 percent average difference over Case-Shiller for the four-quarter price change estimates over
2006Q3-2007Q3for the ten original MAs. Both
FHFA and Case-Shiller use the same repeat-sales
methodology to control for quality changes in the mix of houses sold. They have
the same coverage of type of properties; that is, they include transactions on
one-family houses and exclude 2- to 4-family houses, condominiums and
cooperatives, and weight changes in regional price indices over 9 US census
The FHFA, Case-Shiller and
Census Bureau indices do not incorporate Condominiums. However, in
November 2008, Standard & Poor's launched indices designed to track
condominium prices in five major metropolitan areas—Boston, Chicago, Los
Angeles, New York and San Francisco. The National Association of Realtors
provides median values (by quarter) for a larger number of cities for
condominium prices, but these are not quality adjusted.
Figure 3 shows the Census
Bureau’s index to be quite different from FHFA and Case-Shiller,
though this is to be expected since its coverage is of new houses only.
What is striking from Figure 3
is the different timing of the downturn in house prices: Annual changes in
Case-Shiller turn negative in 2006 Q4 in Figure 3,
but FHFA turns negative a full year later in 2007 Q4. Figure 4 shows quarterly
changes in Case–Shiller to turn negative in 2006 Q3,
while FHFA dips into a negative change in 2006 Q4 to subsequently have positive
changes for the next two quarters to then turn negative in 2007 Q3.
The difference in the magnitude
of the 2008/09 downturn is also striking. Figure 3 shows for 2008 Q4 and 2009
Q1 Case-Shiller registering annual falls of over 18
percent compared with falls of around 7–8 percent for FHFA; similar
discrepancies are apparent from Figure 4.
Even the nature of the
differences cannot be relied upon. From Figure 3, up to 2006 Q2 Case-Shiller exceeded FHFA, this being reversed between 2006 Q3
to 2009 Q4, and reversed again from 2010 Q1.
- Figure 4 shows highly volatile
quarter-on-quarter changes. Yet in spite of methodological differences,
the peaks and troughs of Case-Shiller and FHFA roughly coincide—peaking in Q2—though
their amplitudes differ.
Differences in the indices are to be expected. While
both indices use repeat sales methodology and cover the same type of houses, the
coverage, weighting, and implementation of the repeat sales methodology differ.On coverage, the Case-Shiller National Home Price Index is based on publicly available
transaction sales prices from local recording offices while the FHFA index is based
on data on conventional, conforming mortgage transactions obtained from Fannie
Mae and Freddie Mac. The “conforming” loan limit for mortgages is a capped and
FHFA data are biased against houses purchased with relatively “high” or “low”
The Case-Shiller National HPI does not have valuation
data from 13 states while FHFA’s index uses data from all states (363 metropolitan
weighting, the FHFA HPIs is a
geometrically-weighted average of price changes of the nine census division;
the weights are the relative number
of one-family housing units. Case-Shiller is an
arithmetically-weighted average of price changes; the weights are the relative
dollar value of one-family housing
units. For example, the Case-Shiller index places a 22 percent weight on the Pacific
division in contrast to the 14 percent weight of the FHFA HPI, due to the relatively
higher house prices in California. The weights used for aggregating both indices are
estimated using US Census data, updated every ten years, that is, in 1990, 2000,
and 2010, though linear interpolations are used by FHFA to chain-weight the
indices retrospectively once the subsequent benchmark census results are
available.On the implementation of
the repeat sales methodology, sales pairs with longer time intervals are given less
weight than sales pairs with shorter intervals The down-weighting for
lengthy intervals used for Case-Shiller National HPI is
more modest than that used by FHFA.Studies undertaken by FHFA economists on
why the two indices differ find the most important reasons are the non-coverage
by FHFA of “low and moderate-priced” sales,
somewhat offset by the non-coverage of “high-priced” sales, and differences in
down-weighting long-intervals and filters used exclude non-arms-length sales.
Difference in RPPIs might well be expected
for new versus existing properties. The evidence of Figures 3 for the US bears
this out with, prior to 2005, the Case-Shiller
showing inflation for existing dwellings as about double that shown by the
Census Bureau’s index for new dwelling. The relative positions were reversed
from 2006 with inflation for new property prices below that of existing properties, and substantially so: for the
end of 2008 and early 2009, the Case-Shiller index
showed property price inflation to be three
times lower than the Census Bureau’s estimate for new dwellings.
We can of course look to further evidence
from other countries for the effects of differences in estimates of boom and
busts using differing indices of house prices. There exists for Russia a
national price index for both new and existing dwellings. Figure 5 shows that
while the differences can be sizable in absolute terms, about 15 percentage
points in 2006 Q3, the two series broadly track each other.
Information on coverage and methodology for
these indexes is less readily available, at least to the author, than for the US and UK, but appear to be
comparable. Both are produced by the national statistical office, cover
apartments in urban areas, and are measured as average prices (rubles) of
properties subject to market transactions. Qualitative and quantitative
characteristics including location are stated as being taken into account at
the time of valuation, but it is not immediately apparent how this is done.
Both the average prices for new and existing dwellings are compiled for the
Russian Federation from average prices of Federal districts, weighted by some
combination of population and number of apartments commissioned.
While in spite of major coverage and
measurement differences, it seems that RPPIs can track each other both in terms
of trend and turning points, such differences can also give rise to sizable
discrepancies. There is a natural question of what should be done given an
unsatisfactory situation of competing measures with different outcomes.
nothing. Since many indices are produced by private
organizations this may be the only practical alternative. House price indices
sink or swim on their own merits. It can be argued that different indices have
different virtues and it is for users to choose between/reconcile them: Rightmove, for example, may be based on asking prices, but
it is an earlier indicator than indices based on transaction prices. Yet there
are complexities in the way house price indices are calculated and supporting
documentation, especially from some private organizations, is flimsy. Lay users may not be in a position to attempt
any reasonable understanding of the differences between the measures.
Identify an official series. A recent report by the UK National Statistician recommended: “A single definitive house price index and
accompanying statistics should be produced by the official statistics producer
community.” and “..a regular official statistics report should be developed
presenting and analysing official house price
measures and their relationship to other non-official sources and wider housing
market indicators.” UK GSS
(2010, Recommendations 1, paragraph 1.8, and 2, paragraph 1.11, respectively).
A regular report analyzing why any differences exist would indeed be useful, though
the unambiguous determination of a definitive index may be problematic.
possible ray of sunshine is the proposed publication in late 2011 of
international guidelines on residential property price index methodology: the draft Handbook on Residential
Property Price Indexes (Eurostat, 2011), though see also initiatives for
house price indices for the euro area and European Union (Eurostat (2010a and
2010b). The handbook provides an
excellent account of methodological issues in compiling a RPPI and associated
recommendations. Yet, while a naturally welcome and valuable step, publication
of standards does not of course assure their implementation. The coverage, types of price data, quality-adjustment
procedure, nature and frequency of weighting are for a large part dictated by
the nature of source data available.
A complicating issue is that house price
indices by private organizations such as realtors and lenders serve to
advertise their business. The available information on their methodology is
generally not up to the standard of a statistical office, and for some users,
there will always be skepticism as to conflicts of interest, whether justified
or not. Even if the source data/methods of
the private agencies do not meet the standards of the international guidelines,
private organizations are unlikely to abandon their indices.
Many of the
residential property price indexes used in this study have been drawn from the
Bank for International Settlements’ (BIS) database of property price indexes
available at: http://www.bis.org/statistics/pp.htm.
The codes cited below alongside “BIS” refer to this database. Use of the
database requires a citation of the appropriate national source as noted at: Link
and given below along with the websites used.
country series have been supplemented by further residential property price indexes,
not always published, from the national sources indicated below.
BIS: Q:RU:9:1:1:1:1:0 and Q:RU:9:1:2:1:1:0; Indices
of Prices in Primary/Secondary Market of Dwellings; original source: Federal
State Statistics Service: Link.
United Kingdom: BIS:
Q:GB:3:1:0:2:0:0; Halifax House Price Index; original source and further
series: Halifax Research: Link
(historical house price data). BIS:
Q:GB:0:1:2:1:0:0; Communities and Local Government House Price Index; original
source and further series: Department of Communities and Local Government,
available at: Link.
Also available from UK (Office for) National Statistics at: Link.
(© Crown copyright 2008 Land Registry). Acadametrics;
LSL Property Services/Acadametrics House Price Index;
Land Registry; House Price Index; source: Link.
Nationwide; Nationwide House Price Index; source: Link.
Rightmove; House Price Index; source: Link.
United States: BIS: Q:US:0:2:2:1:0:0;
US Census Bureau; Constant Quality (Laspeyres) Price
Index of New One-Family Houses Sold; original source: Link. Federal Housing
Finance Agency (FHFA); FHFA “Purchases-Only” House price index; source: Link. Standard &
Poor’s; S&P/Case-Shiller National Home Price
Index; source: Link.
Carless, Emily (2011), “Reviewing Residential Property Price Indices in the UK.” paper
presented at the Workshop on Residential Property Price Indices, Statistics
Netherlands, The Hague, 10-11 February 201. Available at:
Stijn, M. Ayhan Kose, and Marco E. Terrones, 2008,
“What Happens During Recessions, Crunches
and Busts?” IMF Working Paper 08/274 (Washington: International Monetary
Sumit (2007) “House Price
Indices of the UK – Methods Explained,” UK Office for National Statistics, Economic & Labour
Market Review, 1,1, January. Link.
Diewert, W. Erwin (2004), “The Treatment of Owner
Occupied Housing and Other Durables in a Consumer Price Index.” In W.E. Diewert, J. Greenless and C. Hulten (eds.), Price Index Concepts and Measurement, NBER
Studies in Income and Wealth, University of Chicago Press.
Eurostat, (2010a), European Commission, Directorate G:
Business Statistics, Unit G-6: Price Statistics;
Purchasing Power Parities, Research Paper, Experimental
House Price Indices for the Euro Area and the European Union, December.
Available at: Link.
(2010b), European Commission, Draft Technical Manual on Owner-Occupied Housing
for the Harmonised Index of Consumer Prices, v.1.9, February. Available at:
European Commission, Directorate G: Business Statistics, Unit G-6: Price Statistics; Purchasing Power
Parities, draft Handbook on Residential
Property Price Indices. Available
Fenwick, David (2006), “Statistics
on Real Estate Prices: The Need for a Strategic Approach.” Paper presented at the OECD-IMF Workshop on Real Estate Price
Indexes, November 6-7, OECD: Paris. Available at:
(2011), “Hedonic Price Indexes for Housing,” OECD Statistics Directorate Working Paper 35, STD/DOC(2011)1/REV1,
Leventis, Andrew 2008, “Revisiting the Differences
between the OFHEO and S&P/Case-Shiller House
Price Indexes: New Explanations,” Office of Federal
Housing Enterprise Oversight, January, Available at: www.ofheo.gov/media/research/OFHEOSPCS12008.pdf .
Mason, Phil and Gwilym Pryce
(2011), “Controlling for Transactions Bias in Regional House Price Indices,”
forthcoming, Housing Studies.
Meissner, Chris and Stephen Satchell
(2010), “A Comparison of the Case-Shiller House Price Index Methodology with the Academetrics House Price index Methodology,” Acadametrics Limited, December. Available at:
Reinhart, Carmen M. and Kenneth S. Rogoff, 2009, “This Time is Different, Eight Centuries of
Financial Folly,” Princeton University
Press, Princeton and Oxford.
Shiller, R.J., (1991). “Arithmetic Repeat Sales Price
Estimators,” Journal Housing Economics, 1,
U.K. Government Statistical Service (GSS), (2010), National Statistician's Review of House Price Statistics,
London: GSS, December. Available at: Link.
Wood, Robert (2005), “A Comparison of UK Residential House Price Indices.”
In Real Estate Indicators and Financial
Stability, Bank of International
Settlements Papers 21, 212–217. Proceedings of a joint Conference Organized
by the BIS and IMF in Washington DC 27-28 October, 2003.
 Claessens, Kose, and Terrones (2008, page 25)
 There remains a
sample selection bias if the indices are used to represent price changes of the
stock of houses (Mason and Pryce, 2011).
 Due to delays in processing Land Registry (LR) data, the AcadHPI results are not termed “final” until a significant
volume of LR data is available which is normally after three months have
forecast results makes use of Halifax,
Nationwide, and DCLG indices. One month after any given month, LR provides
average house prices based upon circa 70% of the eventual total transactions,
which are used to replace the AcadHPI “forecast”
result with an AcadHPI "updated" result. A
further month later, LR provides prices based upon circa 90% of the eventual
total transactions which are used to replace the first with a second AcadHPI "updated" result. Three months after any
given month, LR provides prices based upon circa 95% of the total transactions
for the month. Taking the current month as month T, in month T + 4 the AcadHPI results are regarded as sufficiently updated to be
described as the AcadHPI “final” index (Meissner and Satchell, 2010, page