Using administrative data sources to develop real estate

price statistics: The case of Portugal

Rui Evangelista, Statistics Portugal

European conference on quality in official statistics

Vienna, 4 June 2014

Outline

• Introduction

• Description of administrative data sources

• Methodology

• Results

• Conclusions and final remarks

Introduction• Recent economic and financial crisis reinforced

the need for more and better statistics on the housing market

• Statistical offices started to develop strategies to meet new (…old…) user’s needs:– Eurostat’s statistical pilot program on Owner-

occupied Housing (started in 2002)– Statistics Portugal joined the program in 2008:

“double data source approach”

Introduction

• Legal framework: – Regulation No 93/2013: regular provision of the

HPI to Eurostat– Regulation No 1176/2011: HPI in the scoreboard

of indicators for the early detection of macroeconomic imbalances

Administrative data• Two sources:

1 - Bank appraisals• Value of appraised dwellings (mortgage credit

processes)• Before any transaction actually takes place• National coverage (almost complete universe of banks

conceding mortgage credit)• 326 thousand observations, an average of 16.3

thousand per quarter (1Q2009-4Q2013)

Administrative data

• Two sources (cont.):

2- Fiscal administrative data• Transaction value: Municipal Tax on Real Estate

Transfer (IMT)• Characteristics of the dwellings: Local Property Tax

(IMI)• 450 thousand observations, an average of 22.5

thousand per quarter (1Q2009-4Q2013)

Administrative data

0

5000

10000

15000

20000

25000

30000

35000

40000

Bank appraisals available for HPI compilation

Dwelling trans. for residential purposes (IMT)

Administrative data

• Points to highlight (from the chart): • The number of Bank Appraisals only outscores those of

transactions in the first quarters (mortgage credit more abundant)• During the 2Q2009-4Q2012 period, bank appraisals

drop considerable (and generally at a faster rate than transactions) • From 4Q2011 onwards, bank appraisals numbers

represent less than half the number of transactions

Methodology

• Appraisals-based HPI: – Compiled using a stratification approach– The strata are defined using the following basic design:

• Location of appraised dwelling: as defined by the 7 NUTS II regions for Portugal

• Dimension of appraised dwelling: 2 categories based on the number of rooms

• Type of dwelling: house or apartment; and• Occupancy status of dwelling: “new” and “existing” dwellings

• 56 strata (elementary indexes, geometric mean formula)

Methodology

• Transactions-based index:– Fiscal administrative data– Hedonic price index • Adjacent time dummy approach

Methodology

, , ; , ; ,1

ln( )K

i q k i q k Q i q Q i qK

P a X D

1, ; 1,..., ( ) q Q Q i n q

,

,

ln( ) , is the log price of the dwelling transaction in quarter ;

, stands for the value of the characteristic of the transacted dwelling in quarter ;

, is the temporal indicator o

i q

i q

Q

p ith q

X kth ith q

D

, ;

,

f quarter , which is defined as:

1, if , and 1, and 1,..., ( ),

0, if otherwise.

, is the parameter associated to the temporal indicator of quarter ; and

, corresponds to

i q Q

Q

i q

Q

q Qq Q Q i n q D

Q

an error term.

where,

1 exp .100QQI

Methodology

• The parameters of the hedonic equations are estimated by ordinary least squares (OLS) for the following strata: – existing apartments– existing houses– new apartments– new houses

• Special attention was given to location, area and age effects

• Robust statistics, tests of individual and joint significance of parameters are used in the specification and estimation process

Methodology

• House sales indicator– Based on IMT data; restricted to reflect transactions of

residential properties only– Agricultural land, commercial and non-arms length

transactions (i.e., inherited dwellings) were excluded from the scope of the indicator

– As in the transactions-based HPI, transactions of parts of dwellings were excluded from the calculations of the indicator

– Results for: apartments/houses and new/existing splits; and by NUTS II region

Results

• Comparison between: – Bank appraisals HPI: HPI_BankA– “Hedonic” transactions-based HPI: HPI_Hed– “Stratified” transactions-based HPI: HPI_Strat– “Unadjusted” (four basic strata) HPI: HPI_Raw

– Asking-price HPI: HPI_Ci – Base 100 = 1Q2009

– Number of house sales: N_trans

0

5000

10000

15000

20000

25000

30000

35000

40000

45000

50000

55000

60000

0,800

0,850

0,900

0,950

1,000

1,050

HPI_Strat HPI_BankA HPI_Hed HPI_Ci N_trans

-15,0

-10,0

-5,0

0,0

5,0

10,0

HPI_Strat HPI_BanA HPI_Hed HPI_Ci HPI_Raw

HPI_Strat HPI_BanA HPI_Hed HPI_Ci

Mean -3.9% -3.0% -3.2% -1.0% Stdev 4.67 p.p. 3.46 p.p. 3.53 p.p. 2.17 p.p.

Results

• Few issues to point out:– Despite the drop in bank appraisals counts,

appraisals-based HPI seems to mimic its transactions-based counterpart reasonably well:• Same turning point in 2Q2012• Strong correlation between the two indicators

Results

• Few issues to point out (cont.):– Asking prices indicator (HPI_Ci) seems to lag behind

both appraisals- and transactions-based HPIs: • Contemporary correlation between HPI_Ci and HPI_BankA

and HPI_Hed are 0.68 and 0.44, respectively• The figures increase to 0.85 and 0.64 when the HPI_Ci of

quarter Q+1 is compared with HPI_BankA and HPI_Hed of quarter Q

• HPI_Ci is less volatile, more “resistant” to price drops• Should come as no surprise: representative of prices at the

start of the buying and selling process, which tend to be, when the market is depressed, higher than real transaction prices

Results

• Few issues to point out (cont.):– Difference between HPI_Strat and HPI_Hed is bigger when

the market hits its lowest point– Stratification approach seems not to fully account the

change in the quality mix of transacted dwellings– A stratification scheme with less strata (only 4; HPI_Raw)

shows even sharper price decreases in the 1Q2011-2Q2012 period

– Results suggests that at least part of the price decreases shown by the HPI_Strat indicator should be (at least partly) attributed to the fact that cheaper and worse quality dwellings are driving average prices down

Results

• Few issues to point out (cont.):– Number of sales indicator is synchronized with the

behavior shown of appraisals- and transactions-based HPIs

Conclusions and final remarks• Results suggest that:

– Asking-based indicators may lag behind transactions- and appraisals-based HPIs

– Bank appraisals may be a reasonable source to develop a HPI (“second-best approach”; need for more research)

• Overall, it is possible to develop good-quality real estate statistics based on administrative data sources

• In the case of Portugal, a change from bank appraisals to fiscal administrative data would represent a jump in the quality of provided official statistics:– Methodological soundness : e.g., use of transaction values (instead of a

proxy)– Accuracy and reliability: use of more appropriate methods to tackle quality

change (“pure” price change would be better measured)