House Price Indexes: Why Measurement Matters

Transcription

1 IMF Statistics Department 1/6/214 House Price Indexes: Why Measurement Matters Mick Silver OeNB Workshop, Vienna, October 9 1, 214: Are house prices endangering financial stability? If so, how can we counteract this? The views expressed herein are those of the author and should not necessarily be attributed to the IMF, its Executive Board, or its management 1

2 IMF Statistics Department 1/6/214 House Price Indices The hard area of irregular transactions on heterogeneous properties Data, methodology and tradeoffs Handbook on Residential Property Prices Indices (RPPIs), 213: p/methodology/hps/rppi_handbook Data sources: Some country illustrations Does measurement matter? Commercial property price indices: the really hard area 2

3 IMF Statistics Department 1/6/214 Data sources, issues and tradeoffs Coverage: Geographical (capital, national, cities) Vintage(existing, new) Type (sfh, apartment, terrace) Cash/loan limit Price: Asking, transaction, appraisal Weight: Stock/transaction Land registry Lender Realtor/Estate agent Buyer Builders (new) Quality-mix adjustment: Hedonic characteristics Repeat sales Mix-adjusted SPAR Private/administrative data: Timeliness Reliability/transparency/reputational risk Longevity Potential for index manipulation 3

4 IMF Statistics Department 1/6/214 Feast UK (E&W): house price indexes, annual quarterly rates ONS Land Registry Halifax Nationwide Also: LSL Acadata HPI (Land registry) and Rightmove (realtor) 2Q4 coming into the trough was -8.7 (ONS) (Land registry) (Halifax) (Nationwide): methodology and data source matter. joint project to consider the development of a single definitive UK House Price Index (HPI 4

5 Mar-2 Dec-2 Sep-3 Jun-4 Mar- Dec- Sep- Jun- Mar- Dec- Sep- Jun-1 Mar-11 Dec-11 Sep-12 Jun-13 IMF Statistics Department 1/6/214 Famine Brazil: Residential real estate indexes Residential Real Estate Collateral Value Index Fipe-Zap House Asking Price Index Banco Central do Brasil: Measures the long term trend. It uses individual loan collateral appraisal data to construct series of median house values. These median house appraisals series are filtered using a HP filter. FipeZap Index of prices of apartments advertised: weighted average of median asking prices by 16 regions and 4 bedroom sizes insufficient count in cell: carry forward previous price. 5

6 IMF Statistics Department 1/6/214 Repeat sales HPIs: United States Repeat sales CoreLogic Federal Housing Finance Agency Case-Shiller Census Bureau is (hedonic) new houses CoreLogic FHFA Census Case Shiller (national) Leventis (2): how repeat sales used: FHFA more muted downweighting than CS: absolute difference from CS in price change 2-2: 2.67 %; value weights (CS); filters; coverage of qualifying loans (FHFA); FHFA extended data and all purchases similar 6

7 IMF Statistics Department 1/6/214 Making your own luck France: Notaires-INSEE second-hand apartment and house price index Notaires-INSEE 1983 apartments in Paris; provinces and national. Notaries draw up deed and collect stamp duty. In 23 estimated 4,6 notary practices. Separate hedonics for apartments and houses by 3 zones comparing transaction prices of fixed bundles of observed characteristics. Hedonic coefficients now updated every 2 years and weights chain-linked. UK: ONS Mix-adjusted HPI 1969: 5% sample of mortgage transactions of a number of building societies. From 1993; building societies to all mortgage lenders; monthly sample 2-3,. 23: 5% sample each lender increased to 1%; mid-23-august 2: about 25, monthly mortgage completions. End-2: about 4, increasing to 5, a month through 2-7, 6 main lenders 2. Six months to May 21: 32 lenders 23, per month. 211: 65-7% of all completions. Pre-23 mix-adjustment potential 3 cells; post-23: 1, cells. 7

8 Does measurement matter and, if so, how? How HPIs might differ Stocks or flows and values or quantities for weights; Index number formula used and nature of weights; Enabling constant quality measures; Coverage: geography, type of housing, and financing; and Valuation stage of prices.

9 Panel data: From 2: to 21:, for 24 countries, 157 series: log changes in quarterly HPIs are given for i=1,,nc series in country c=1,,c over t=1,,t quarters. Fixed country and time effects panel data.

10 IMF Statistics Department 1/6/214 Explanatory (measurement) variables Coverage Age (benchmark: all residences) New: newly-built residences only; Xist: existing residences excl newlybuilt. GeoCoverage (benchmark: national) Capital: major city; Urban: urban areas; BCities: big cities, say population exceeds 1,; Rural: rural areas Type (benchmark: single family houses and apartments) Sfh: single family houses Apt: apartments Methodology Quality-mix adjustment (benchmark: price per unit) Hed: hedonic adjustment; SqM: price per square metre; SPAR: sale price appraisal ratio; MixAdjust: mix adjust (stratify) Repeat: repeat purchase Price (benchmarked on transaction) Ask: Asking price Appr: Appraisal price (tax) Fixed/Changing Weight (benchmark: fixed base) Chain: chained annual Roll: rolling period Unw: unweighted Weight (benchmark: transactions) Wstock : stock of dwellings Weight higher level (benchmark: value) Wquanity: quantity shares Wprice: relative base price Wsqm: relative size (sq. m.) Wpop: population shares Aggregation (benchmark: geometric) Arith: Arithmetic 1

11 Results: for a regression on the panel data R 2 is:.16 to include only measurement variables,.44 to include only 24 country fixed effects,.3 to include 24 country fixed effects & measurement variables,.141 to include just time effects, to include both country and time fixed effects, to also include measurement variables. Measurement variables have little to no explanatory power. [The null hypothesis of redundant variables could not be rejected, LLR= (p-value=.6717) ]

12 Table 1, Regression of HPIs on measurement variables Variable Coefficient Std. Error t-statistic APARTMENT APPRAISAL ASKING CAPITAL HEDONIC MIXADJUST NEW SFH SQMETER UNWEIGHTED WPRICE WROLLING WSTOCK XSTING R-squared.11 Adjusted R-squared.3 S.E. of regression.4 Log likelihood Sample: 2 to 21; 155 cross-sections; 3,156 obs. Fixed country effects not shown for brevity.

13 Table 2, Fit of measurement variables in moving window regression: time varying RbarSq including: Time; Country; Country; Measurement. Measurement Measurement Measurement Coverage Methodology Q Q Q Q Q Q Q Q Q Q *

14 Table 3, Illustrative regression results for 2: to 2Q2 Variable Coefficient Std. Error t-statistic Prob. C WSTOCK APARTMENT APARTMENT--2Q APPRAISAL APPRAISAL--2Q ASKING ASKING--2Q CAPITAL CAPITAL--2Q HEDONIC HEDONIC--2Q MIXADJUST MIXADJUST--2Q NEW NEW--2Q SFH SFH--2Q SQMETER SQMETER--2Q UNWEIGHTED UNWEIGHTED--2Q WPRICE WPRICE--2Q WROLLING WROLLING--2Q XSTING XSTING--2Q Q2--C.431 R-squared.592 Adjusted R-squared.5 S.E. of regression.33 Log likelihood 619.6

15 Figures 1 6, Varying estimated parameters Existing properties Mix-adjustment Hedonic regressions Sq. meters Unweighted Appraisal prices

16 Does it matter in modeling? Igan and Loungani (21) Illustrative model applied as they did (specification, dynamics, estimator) for both our measurement-adjusted and unadjusted HPIs. Rationale in Igan and Loungani.

17 Table 4, Pooled regression results for house price indexes Dependent variable Affordability, lagged Income per capita, change Working-age pop, change Igan and Loungani (21) -.17*** (.158).431*** (.84).999*** (.197) House price index, log quarter-on-quarter change: Measurementadjusted estimates -.291* (.1772).392*** (.1516).735*.3941 Unadjusted estimates (.121).519*** (.17).494** (.2354) Excluding: Affordabilitylagged Measurementadjusted estimates -.5** (.37).395* * (.411) Unadjusted estimates -.7*** (.271).52*** (.19).53** (.2438) Stock prices, change.44* (.26) -.17** (.6) -. (.1) -.16*** (.1) -.4 (.7) Credit, change.19*** (.3).165*** (.268).191*** (.253).156** (.31).186*** (.273) Short-term interest rate -.** (.4) -.1** (.46) -.** (.25) -.1 (.) -.*** (.25) Long-term interest rate -. (.4).1***.. (.).*** (.).2 (.) Affordability, lag, squared -.19* (.12) -.14 (.121) -. (.5) Construction costs, change.129*** (.366).32* (.1671).312* (.17).285* (.172).295* (.1738) Constant -.243*** (.54) ** (.6384) -.838** (.4232) -.553** (.247) -.54*** (.1796)

18 Country-specific parameter estimates for stock prices Austria Australia Belgium Switzerland Canada Denmark Spain Finland France Greece Ireland Japan Netherlands Norway Sweden United Kingdom United States Measurement-adjusted Unadjusted

19 1/6/214 Commercial property price indices Highly heterogeneous and very few transactions Appraisal data poor Advantages to aggregating within regression framework. Quality adjustment: hedonic/repeat sales Confidence intervals Inclusion of other variables conditioning More efficient estimators; use counts data; How to aggregate in regression framework Spatial autoregressive panel data two way fixed effects estimator: get rid of omitted variable bias and use for weights 19

20 1/6/214 Data Panel data of transaction-based US CPPI quarterly series from 2:Q4 to 212:Q4 by 34 metro areas for each of apartments and core commercial properties. Each metro area CPPI estimated using repeat sales method. Data provided by Real Capital Analytics (RCA) acknowledge help. 2

21 IMF Statistics Department 1/6/214 OLS and WLS estimates US apartment property price inflation: q-on-q rates US core commercial property price inflation: q-on-q rates Pooled WLS Pooled OLS Pooled WLS Pooled OLS -8-1 Moody's/RCA Inflation -8-1 Moody's/RCA Inflation

22 IMF Statistics Department 1/6/214 Spatial autoregressive model Y W Y Z μ Z γ V n, t n n, t n t n, t μ n.. (3) where Wn is a n n row-standardized spatial physical proximity weight matrix and the estimated spatial autoregressive parameter (Anselin, 2, Elhorst, 21 and Debarsy and Ertuur, 21; (Rambaldi and Rao, 211-SEM for RPPIs),). The matrix of partial derivatives of Y nt, with respect to a change in a dummy time variable, is: Y 1 γ I W B nt, t n n t Z....(4) The spatial direct effects are not γt but are given for each area n by the diagonal elements of B t. 22

23 IMF Statistics Department 1/6/214 Fixed and varying (chained) weights Apartments Core commercial SAR Direct fixed weights SAR Total fixed weights SAR Direct varying weights SAR Total varying weights OLS SAR Direct fixed weights SAR Total fixed weights SAR Direct varying weights SAR Total varying weights OLS