Tracking Owners’ Sentiments: Subjective Home Values,Expectations and House Price Dynamics

Anthony Lepinteura and Sofie R. Waltl∗b,c

aUniversity of LuxembourgbLuxembourg Institute of Socio-Economic Research (LISER)

cVienna University of Economics and Business (WU)

March 1, 2019

Abstract

We study subjective house price data collected in wealth surveys. We apply suit-ably adjusted hedonic and repeat-sales methodologies to construct subjective house priceindices measuring changes in owner-occupied housing. We benchmark the resulting in-dices to objective indices compiled from transaction data and find strikingly similar pricepatterns. We hence conclude that subjective data are well suited for studying housingmarket dynamics. We estimate a subjective hedonic house price equation that links owner-reported willingness-to-accept prices to the owner’s expectations on future price changes.We find that optimistic owners report larger prices than pessimistic owners and henceprovide support for the individual association between house prices and expectations onthe supply side of the housing market. Furthermore, we find evidence for loss aversion,which leads to lower adjustments among pessimists as compared to optimists.

Keywords: Subjective Data; House Price Indices; Housing Markets; Expectations; Willing-ness to Accept; SHIW; HFCS

JEL codes: C43; D9; G4; R31

Notes and Acknowledgements: This is a very preliminary draft as of March 1, 2019. We would like to thankFrancisco Amaral, Ronan C. Lyons, Giorgia Menta and Martin Eiglsperger for providing us with additional dataand information, as well as Francesco Fallucchi and participants in the 2019 EUROSTAT International Con-ference on Real Estate Statistics for valuable comments.This article uses data from the Household Finance and Consumption Survey provided by the European CentralBank as well as the Survey on Household Income and Wealth provided by the Banca d’Italia. The results pub-lished, and the related observations and analyses may not correspond to results or analyses of the data producers.

∗Corresponding author: Sofie R. Waltl; Maison des Sciences Humaines; 11, Porte des Sciences; 4366 Esch-sur-Alzette/Belval; Luxembourg; sofie.waltl@liser.lu; www.sofiewaltl.com

“If molecules could talk, wouldphysicists refuse to listen?”

Alan Blinder1

1 Introduction

We study subjective data on owner-reported house prices together with expected future pricedevelopments. Often, self-reported data is treated just as if it was objective data. For instance,wealth distributions measured from survey data claim to describe reality. While the subjectivityof the data is usually simply ignored, those articles that mention subjectivity rather see it as aflaw.For instance, the accuracy of self-reported home values has been analyzed and Agarwal (2007)concludes over this literature that: “[t]here is general agreement [. . . ] that homeowners signifi-cantly mis-estimate their house value.” He reports substantial average absolute mis-estimationfound by previous studies (mainly focusing on the US) ranging between 14% and 25%. Al-though evidence is not fully consistent (for instance, Kain and Quigley, 1972, find a negativebias in owners’ estimates), there is a tendency that homeowners overestimate the price andrental potential of their dwelling.Likewise, Heston and Nakamura (2009) document that equivalent market rents (i.e., hypothet-ical rental prices homeowners think they could earn when renting out their owner-occupieddwelling) reported by owner-occupiers are on average higher than market rents in the US.Often, this literature classifies divergence from a factual, appraised or estimated market value asestimation error. However, Goodman Jr and Ittner (1992), who find a bias in owner estimatesof house values of plus six per cent, point out that for purposes like household consumptionand savings behaviour owners’ perceived home value may indeed be the appropriate measuresof housing wealth.In this article, we do not follow a normative approach. We believe that objective and subjectivedata are complements and the preferred use of either one depends on the question asked.We use wealth surveys as data source for subjective house price data, which constitute a richsource of information on owner-estimated house prices. As a home tends to be the singlemost important asset in most households’ portfolios, an estimate on the current value of thehousehold’s main residence is included in all such surveys. We focus on the household’s mainresidence as the information owners have about this property (including information on theneighbourhood, the quality of the house, etc.) is likely larger than for other properties. Also,large shares of the population own a home – roughly 70% of Europeans (EU-28) were owner-occupiers in 20162 – whereas ownership of additional properties is less common.There is hence ample information on housing values available in surveys, which is, however, notwidely used for studying housing markets. Work that has been done on owner-estimated values,almost exclusively focuses on the US (see for instance Chan et al., 2016; Choi and Painter, 2018;Gallin et al., 2018). This article takes a first step to fill this gap.Our contribution is threefold: first, we construct quality-adjusted subjective residential propertyprice indices (S-RPPIs) for several European countries by applying different index construction

1See Senik (2014).2See EUROSTAT (EU-SILC): http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=ilc_

lvho02&lang=en, accessed on January 24, 2019.

2

techniques (see also de Haan and Diewert, 2013). We show that both main approaches, hedonicor repeat-sales, can be adapted to be applied to subjected house price data.Second, we empirically benchmark subjective indices to objective indices (O-RPPIs) to as-sess whether the aggregate information in self-reported data is comparable to the informationretrieved from market data. We study the Italian Survey on Household Income and Wealth(SHIW) in-depth and show that collectively home-owners track price changes very well evenover long periods of time. Additionally, we analyze data from the Household Finance andConsumption Survey (HFCS), which provides harmonized survey results for currently 18 par-ticipating European countries. We show that even the limited information available in theHFCS (currently only two waves are available) yields indices that are strikingly similar toobjective series.These findings allow us to confidently use subjective house price data for further analyses andwe link owners’ self-reported prices to beliefs and expectations. Hence, our third contributionis a subjective hedonic equation that measures the importance of expected future price growthon the owner-reported value of the house. We find evidence that optimistic individuals tendto report significantly higher values than pessimists. This finding is in-line with the user-costformula that justifies high equilibrium prices in the presence of high expected future capitalgains (see Poterba, 1984; Himmelberg et al., 2005).While such a connection is not surprising, we establish an explicit link between owner-reportedvalues – a measure of the willingness-to-accept price (WTA) – and owners’ beliefs, and hencedescribe a pure supply side effect complementing Piazzesi and Schneider (2009) that look atoptimism on the demand side. We find that beliefs about general developments of house pricesare less important than the expectations on the change in the value of the owner’s home.Additionally, we find evidence for loss aversion: being optimistic leads to relatively largerchanges in the WTA than being pessimistic, i.e., if one beliefs that the value of one’s home risesby a certain amount, the upward adjustment of WTAs is larger as compared to the downwardadjustment in case of an equally-sized expected drop in value. Overall, optimism appears tobe roughly twice as important as pessimism.The remainder of this article is composed as follows: section 2 elaborates on the use andbenefits of subjective data and how they complement objective measures. Thereafter, section 3describes how standard index construction techniques can be applied to subjective data. Insection 4, we empirically compare different types of subjective indices to objective counterparts.The following section 5 discusses the connection of house prices and expectations. We estimatea subjective hedonic equation to link owners’ stated home prices to expectations about futureprice changes. Finally, section 6 concludes.

2 Subjective versus Objective Data: Complements not Substitutes

While the use of subjective data is widespread in other social science disciplines, such as psy-chology or sociology it remains relatively unusual in mainstream economics. This comes fromthe suspicion that what individuals say does not reveal their preferences while choices do, asstated by the revealed preferences theory.However, economists use more and more subjective data. One of the most popular measures ofhealth is self-assessed health. The notion of ‘health’ is itself complex and depends on severaldimensions. One may study objective measures of health such as the Body Mass Index, thenumber of visits to the hospital over a given time period or the daily consumption of cigarettesto create an index of health. However, the creation of such an index would raise the concernregarding the selection and the weighting of all the dimensions of a “good health.” The most

3

pervasive and democratic measure remains probably self-assessed health.Subjective data are also used in labour economics. Clark (2001) models separations and quitsusing a British panel, and shows that job satisfaction data are powerful predictors of bothseparations and quits in a bivariate analysis. Job satisfaction being a function of objectivejob characteristics, one may argue that job satisfaction is a simple substitute for job quality.Nevertheless, even controlling for a large set of job characteristics (such as wages, workinghours or occupations), job satisfaction remain a strong predictor of job quits and separation.This study supports the idea that subjective data do not only reflect objective components,they also carry a useful source of information linked to perceptions that are likely to affectdecision-making. By seeing which aspects of a job are most likely to induce quits, Clark (2001)concludes that job satisfaction, security and the pay are the most important job characteristics.Following a similar approach, D’Ambrosio et al. (2018) show that the larger the gap betweenown income and the income considered as ‘fair’ increases the probability to quit.Retrospective voting theory recently benefited from the use of subjective data. This litera-ture states that voters evaluate diagnostic information, such as unemployment, inflation andpersonal financial circumstance, to re-elect satisfying incumbents or to punish the one who arebelieved as inefficient. Liberini et al. (2017) finds that life satisfaction predicts political supportfor the incumbent, even controlling for a large set of objective characteristics. Using 40 yearsof Eurobarometer data covering 15 countries, Ward (2015) even shows that life satisfaction is astronger predictor of incumbent vote share than economic growth, inflation and unemployment.But concerns remain about the validity of subjective data: what are self-assessed health, jobsatisfaction or life satisfaction really measuring? One way to assess the validity of subjectivemeasures is to perform tests of convergent validity. The convergent validity refers to the degreeto which two measures that should theoretically be related are in fact related. The literaturein health economics adopted this approach and demonstrates that self-assessed health is highlycorrelated with objective measures of health. Doiron et al. (2015) shows that self-assessedhealth predicts future health such as hospitalizations and prescription drugs. They also reportthat the predictive power of self-assessed health is particularly high in case of chronic andserious diseases. Some similar tests of convergence have also been performed for measures ofsatisfaction. Urry et al. (2004) show positive correlations between measures of satisfaction,brain activities and the production of cortisol.Subjective data are neither idiosyncratic noise nor imperfect substitutes for objective data.They are rather complements.An analyses of the additional information inherent in subjective data on house prices is sofar missing in the literature. One goal of this article is to extend the literature on subjectivemeasurement with this regard.

3 Subjective Housing Price Indices

In a wealth survey, the owner is asked to anticipate the outcome of a bargaining process: sheis not asked to state the price for which she is willing to sell the property nor an estimate ofthe “fundamental price” but really what she thinks she could earn when the dwelling was puton sale at the day of the interview. This is anything but easy but provides us with ampleinformation on beliefs.3

We use these subjectively reported home values to construct residential property price indices3When a person plans to buy or sell property, she is likely to seek advise from real estate experts. It is,

however, more than unlikely that such costly actions are taken when preparing for a survey interview.

4

(RPPI). Just like in the case of an objective residential property price index (O-RPPI), for con-structing a subjective index (S-RPPI) one needs to perform some kind for quality-adjustmentto guarantee that like is compared with like. This is important as a large degree in the vari-ation of house prices cross-sectionally and over time is due to differences in locational andphysical characteristics of properties. Both major quality-adjustment approaches, repeat-salesand hedonics, are suitable for survey data.Hedonic approaches assume that the total price of a good can be decomposed into a sum ofshadow prices associated with the good’s characteristics. In a house price setting, the relevantcharacteristics describe both the major features of the structure and the quality of the dwelling’slocation.Hence, a sufficiently comprehensive list of physical and locational characteristics needs to beavailable to filter out price changes over time net of differences in characteristics. While surveysusually collect the most important physical characteristic (such as the age of the structure andthe living surface), they tend to report only very basic information on location due to dataprotection concerns.Hedonic indices are criticized for the high data demand in terms of dwelling characteristics, anda potential subjective specification of the hedonic equation. Neither is there consensus on thefunctional form in which characteristics should enter the equation nor on the ideal selection ofthe same.4 While the list of potentially important characteristics is endless, many importantfeatures are rarely available for transaction data and even more are simply not measurable in anobjective way. For instance, the exact layout of the rooms is an important price-determiningfactor, but it is unclear how such information could enter a hedonic equation. Usually, theselection of characteristics is predominately driven by data availability.In general, all types of hedonic index construction techniques can be applied to subjective datajust as to transaction data (see Hill, 2013, for a survey on techniques). Here, we focus on themost parsimonious approach, the hedonic time-dummy index.A single hedonic model is estimated that regresses the logged price of the dwelling on a matrix ofphysical X and locational characteristics L with associated shadow prices β and λ. Additionally,a matrix of time dummy vectors D enters the model. In the classical time dummy approach,time dummies refer to the period of sale. In the subjective version, the time dummies indicatethe period of valuation. The associated vector of parameters δ filters out pure price changesover time net of differences in dwelling characteristics:

log(P ) = Dδ +Xβ + Lλ+ ε. (1)

Finally, ε denotes a vector of independent and identically distributed error terms.The hedonic index is obtained via

{exp(δ1), . . . , exp(δT )}.

Usually, the index is normalized to a suitable base period.A repeat-sales index (see Bailey et al., 1963; Case and Shiller, 1987, 1989) exploits price infor-mation from repeatedly sold dwellings. Only price pairs of the same dwelling are considered andaggregated to an overall index. As the price of the same dwellings is compared over time, analmost perfect quality-adjustment is obtained. The standard repeat-sales index is criticized for

4A growing body of literature thus suggests modelling the functional form of characteristics non-parametrically. See Hill and Scholz (2017) and Waltl (2016) for such approaches for location and Pace (1998)for physical characteristics.

5

ignoring changes in the age of the structure, measurement errors from unobserved renovations,the exclusion of first sales, and a potential lemons’ bias, as low-quality homes tend to sell morefrequently (see Wallace and Meese, 1997).Some wealth surveys include a panel component, which enables linking self-reported prices overtime. These linked prices can be used to compile a repeat-sales type of index. In fact, the majorpoints of critique of the repeat-sales technique are irrelevant for panel data: there is no reasonto believe that surveys over-represent owners of low-quality homes and hence a lemons’ bias isunlikely to be present.5

First sales are not per definition excluded. Although the price of a dwelling will only be includedinto the analyses as soon as the household was contacted again in a subsequent survey wave.Surveys, however, take place much more frequently than re-sells.Due to short periods of time between survey waves (usually wealth surveys take place every twoto three years), the aging of the dwelling is less important. Additionally, some surveys containinformation on major renovation, which can be made use of.In the absent of a panel structure, the original acquisition price (which is in fact quite objective)can be used together with the estimated current market price to identify price pairs entering arepeat-sales model.To construct a standard repeat-sales index, the logged ratio of prices related to subsequentsales is regressed on dummy variables indicating the period of first and repeated sale, i.e.,

log(Pi,t

Pi,s

)=

T∑t=0

τ tDti + εt

i, (2)

where Pi,t and Pi,s denote the prices of dwelling i at times t and s respectively, and Dti denotes

dummy variables with entry 1 for the period of re-sale, -1 for the period of the first sale and 0otherwise. Finally, εt

i denotes an independent and identically distributed error term.The estimated coefficients τ t are used to construct the price index: the change in prices betweenthe base period 1 and period t is given as

exp(τ t),

which forms the index.To construct a repeat-observations index (RO) from survey data, price pairs may either stemfrom repeatedly owner-estimated values collected in a panel survey or by linking the originalacquisition price to the current owner-estimated price. The latter approach can be applied tocross-sectional surveys.

4 Subjective Versus Objective Indices: An Empirical Comparison

4.1 Indices for Italy: the SHIW

In this article we mainly rely on the Italian SHIW. The SHIW contains both, a rather longlist of dwelling characteristics and a large number of panel households. The start of the surveydates back to 1965, but data is publicly available from 1977 onward only. The survey is carriedout on a roughly biennial basis. While the survey has been extended over time, information onowner-estimated house prices has always been part of the questionnaire.

5However, wealth surveys are known to miss the wealthiest of the wealthy households (Vermeulen, 2016),which also tend to be the owners of the most valuable properties in a country (for differences in average housingwealth across the distribution, see Waltl, 2019). This affects, however, only a very small part of the population.

6

Cannari and Faiella (2008) compare average per-square meter prices implied by the SHIWto averages obtained from market data. They find a quite good correspondence for regionalaverages.We construct a subjective hedonic index6 as well as a subjective repeat-observations indexrelying on panel households. Since we have information on every second year only, we linearlyinterpolate the resulting index number between survey years to obtain an index at annualfrequency.We compare the resulting indices to two objective counterparts estimated from transactiondata. The first index is compiled by the Banca d’Italia since 1990. The quarterly index isquality-adjusted and representative for the entire country. The second index was compiled byCannari et al. (2016) and starts in 1924. The annual, quality-adjusted index also refers to theentire country. See also Table 3 for a summary of all objective indices used in this article.

Figure 1: Subjective versus Objective Indices.

Notes: The subjective SHIW repeat-observations (RO) panel index makes use of all owner-occupierpanel households between 1977 and 2016. The subjective SHIW hedonic index makes use of all owner-occupier households between 1986 and 2016. The subjective HFCS RO Acquisition index relies on theoriginal acquisition price and the current self-reported price available in the two HFCS waves. Theobjective benchmarks are transaction-based indices representative for the entire country. Details aredocumented in Table 3. All indices are normalized to (the average in) 2000.

Figure 1 depicts nominal objective and subjective indices. All four indices report the samegeneral trend in prices. In particular, the subjective RO index and the objective indices matchalmost perfectly indicating that the repeat-sales methodology, which relies on more than 50,000price pairs is very well suited to construct a subjective index.

6We include the logged living surface, the number of bathrooms (one versus more than one), age as a smoothfunction, the rating of the dwelling, the size of the municipality, the location within the municipality as wellas time and region dummies. (We use the same specification as in section 5.) The information on the ratingand the number of bathrooms is not available in the early years. We include an additional category to thesevariables indicating missingness. Additionally, the definition of location with the municipality changed over theyears. We merge identical categories and include an extra category pooling non-matching categories.

7

The hedonic index, which suffers from quite some shortcomings due to missing characteristicsin several years (see footnote 6), reports a very similar price pattern. However, the size of theincrease before the price peak in 2008 is estimated to be much higher than reported by thesubjective RO or the objective indices.Additionally, the left panel of Figure 1 also depicts a RO index based on pairs of acquisition andcurrent self-reported prices. Replicating the indices in subsection 4.2, we use only two surveywaves for this index. Given the very limited information provided to the index, the similarityto the other indices is striking. The disturbing wiggliness results from the low number of overallobservations used to estimate this index. Every additional HFCS survey wave will improve thisindex as more price pairs will enter the estimation process.

Figure 2: Annualized Changes in Housing Prices.

Notes: The figure compares annualized changes in the subjective repeat-observations index and theobjective annual index (IT 2) by (Cannari et al., 2016). Additionally, the median over all individualprice changes per year is reported.

An even better comparison between objective and subjective indices is achieved by looking atannualized increments, which are not affected by the choice of normalization period. Figure 2plots annualized changes in aggregate housing prices implied by the annual objective index (IT2) and the subjective RO panel index. Additionally, the median over all individual incrementsis shown.Annualized changes match almost perfectly. In fact, the Pearson correlation between objectiveand subjective increments equals 80.2!From this analysis, we conclude that a well-designed and well-informed subjective index repli-

8

cates aggregate results derived from objective market data. It is important to note that wefind good correspondence both during periods of rising and falling prices, which differs fromfindings by Chan et al. (2016) for the US, who document that – particularly when prices turnfrom increasing to decreasing – owners’ assessments lack behind the market trend. Our findingssuggest that subjective data can indeed be used to study housing markets. The large amountof additional information available in surveys enables a more detailed analysis of house pricedynamics, but also beliefs and expectations.

4.2 Indices for Europe: the HFCS

The HFCS is a relatively new survey co-ordinated by the European Central Bank. Results ofthe first and second waves were released in 2013 and 2016, respectively. In countries, where awealth survey existed before the start of the HFCS, these surveys were usually integrated intothe HFCS framework and extended/adjusted to achieve the desired degree of harmonization.This was the case for the Italian SHIW. In other countries, 2013 was the first time that resultsfrom a wealth survey were ever released.In particular, the HFCS consists of a core questionnaire, that is ex-ante harmonized acrossall currently 18 participating countries. The results of the harmonized questionnaire are wellcomparable across countries. Due to the large geographical scope of the HFCS, we analyze howthis survey can be used to construct subjective house price indices.Although some countries have a panel component (such as Italy), the HFCS is in generala cross-sectional survey. Also, the number of dwelling characteristics available to constructhedonic indices is very limited, and in particular no locational information is available. Hence,we can only construct subjective repeat-observation indices matching the acquisition price (andyear) with the owner-reported current value. As in the Italian case, we exclusively focus on thehousehold’s main residence, and the resulting indices thus reflect changes in owner-occupiedhousing prices.We pool data from the first and second wave of the HFCS, which increases the number of pricepairs substantially. We treat the year, where the HFCS fieldwork took place, as the period ofthe repeated ’sale’. Indices for countries that did not participate in the first wave are based onsecond wave data only.7 We include only reported values and leave out imputed ones.8

The ECB also compiles an aggregate RPPI for the euro area (EA). We use the index thatreports changes in house prices for EA-17, i.e., the euro area as composed in 2011 including 17countries.9 The EA-RPPI is a weighted average of individual, country-specific RPPIs. Weightsare determined by the contribution of a country’s GDP to the overall EA-17 GDP.To create a subjective counterpart, we calculate a weighted average over all S-RPPIs belongingto the EA-17. The weights are held constant over the period and refer to the 2013 GDP atmarket prices. As the HFCS lacks data to compile a S-RPPI for Finland (no original acquisitionprice is available), this country is excluded. Finland’s share of euro area GDP amounts toroughly 2% only, which is why the impact of this exclusion is expected to be minor.

7Indices start in the earliest year for which there are at least five acquisition prices. This guarantees thatearly index numbers are not driven by too few observations.

8Values are imputed when interviewees refuse to provide an answer to a particular question. The HFCSimputation process aims for consistency of the joint distribution of all HFCS variables, i.e., missing values areimputed to guarantee that the originally observed joint distribution is conserved.

9These countries are: Austria (AT), Belgium (BE), Finland (FI), France (FR), Germany (DE), Ireland (IE),Italy (IT), Luxembourg (LU), The Netherlands (NL), Portugal (PT), Spain (ES), Greece (GR), Slovenia (SI),Cyprus (CY), Malta (MT), Slovakia (SK) and Estonia (EE).

9

Figure 3: Subjective versus Objective Indices in Europe.

10

Notes: The figures compare S-RPPIs estimatedfrom pooled HFCS data to objective, transactions-based RPPIs (O-RPPIs). All indices are nominaland normalized with respect to the year 2012 or,in case of quarterly data, the average in 2012. Theeuro area figure also shows the coverage of the indexmeasured as share of EA GDP (ranging between10.4% and 97.9%). Figures are provided for coun-tries, where objective indices are at least as longas subjective indices. Figure 8 in Appendix B pro-vides results for more countries. Details about theobjective indices are provided in Table 3.

Figure 3 shows S-RPPIs together with transactions-based, objective RPPIs for the six countries,where long-enough objective indices are available: Belgium, France, Germany, Ireland, Italyand Spain.10 The bottom left panel plots the results for the euro area. The bold line indicatesthe part of the S-RPPI index, where data from all EA-17 (excl. Finland) are available. Theindex is also calculated when S-RPPIs are not available for all countries. The coverage of theindex in terms of per cent of euro area GDP ranges between 10.4% and 97.9% and is indicatedby the horizontal lines at the bottom of the plot.Note that the objective indices are far from harmonized in terms of geographical coverage, typesof dwellings and transactions covered, data sources used and index construction methodologyapplied (see Table 3). The subjective indices presented here are, in contrast, fully harmonizedand show price developments for owner-occupied residential dwellings. The indices do notperfectly represent the stock of all owner-occupied dwellings due to item non-response in theHFCS and pooling over two HFCS waves. However, they come close to that and it is fair to saythat the indices present the (subjectively measured) price development of the housing stock asof 2009 – 2014.11

Subjective and objective indices, in general, follow a very similar trend, which supports theconclusion from the SHIW, that – at least collectively – people have a quite accurate perceptionof general price developments in housing markets. Whereas there are some countries where thecomparison yields worse results (see Figure 8), the aggregate result for the euro area is striking.Results are remarkable given that the indices make use of only two survey waves and produceresults covering decades!For many countries there is only a very short objective index available, which complicatescomparability. However, in particular for these countries, the subjective indices provide newinsights into the long-term trends in housing price inflation and help to put current develop-ments into perspective. The more HFCS waves become available, the longer and more reliablethe subjective indices get as more and more price pairs enter the index, i.e., with new data theindex grows into the past.

10Figure 8 in Appendix B shows results for other countries. The appendix also provides methodological detailsregarding the objective indices.

11The HFCS fieldwork took place in the years 2009 – 2011 (wave 1) and 2012-2015 (wave 2).

11

5 Tracking Sentiments: The Role of Expectations

5.1 Theoretical Considerations on Expectations

The user cost of housing is defined as the total cost of owning a house for one period (Poterba,1984; Himmelberg et al., 2005). From a simple no-arbitrage condition, it follows that in equi-librium the user cost must equal the rent. A stylized version of the user cost formula can bewritten as:

P · u = P · (r + δ + ω + γ − π − g) = R,

where u is the per-dollar user cost, r the interest rate, δ depreciation, ω running and averagetransaction costs (including taxes), γ a risk premium, π expected inflation and g expected realcapital gains (or losses) on housing. An additional term is included when owners can tax-deductmortgage interest payments. P denotes housing prices and R rents, respectively.From the user cost formula, it follows that price-to-rent ratios are related to expected capitalgains. Hence, price-to-rent ratios constitute an indirect measure of expectations, which is amajor reason why such indicators are closely followed for financial stability purposes (see alsoHill and Syed, 2016).From this formula, one concludes that ceteris paribus high expected future capital gains justifyhigh current prices as compared to it current yield, the rent. Vice versa, expected decreases inthe future imply lower current prices.In the extreme case, excessive expectations may lead to a housing bubble. Stiglitz (1990)defines an asset bubble as follows: “[I]f the reason that the price is high today is only becauseinvestors believe that the selling price is high tomorrow – when fundamental factors do notseem to justify such a price – then a bubble exists.” Similarly, Himmelberg et al. (2005) write,“We think of a housing bubble as being driven by homebuyers who are willing to pay inflatedprices for houses today because they expect unrealistically high housing appreciation in thefuture.”Both definitions of housing bubbles refer to the demand side of the housing market and describehow excessive expectations drive up prices. For a bubble to evolve, however, both sides of themarket need to share the same beliefs: when owners are unwilling to sell at a low price due tohigh expectations, buyers need to be willing to bit up the price, which they only do when theyshare owners’ beliefs on future price growth.Standard economic theory claims that for rational market participants (at least in the absence ofincome constraints) the willingness to pay (WTP) equals the willingness to accept and coincideat the fair market value (Willig, 1976). However, experiments consistently find WTP-WTAgaps potentially driven by an endowment effect (Thaler, 1980; Kahneman et al., 1990).A prerequisite for a bubble to emerge is hence that WTAs are high due to high expectations onfuture price growth. If these beliefs among sellers are shared by buyers, houses are transactedat high prices. Irrational exuberance as described by Shiller (2015) may be the result.This mechanism is shown below: sales take place when expectations of owners and buyersmatch, or when expectations of owners are lower than those of buyers as indicated by the greenfilling. The latter situation is a transitory state as low prices will be exploited by buyers andhence lead to increased expectations of owners. No sales take place when owners have highexpectations but this is not shared by buyers.

12

In this setting, an rise in house prices can only result from the state where owners and buyershave high expectations – the bottom-right state.The described mechanism relies on two fundamental relationships: first, a positive associationbetween expectations and WTAs among owners, and a positive association of expectations onWTPs among buyers.Piazzesi and Schneider (2009) find an increasing share of optimists during the built-up of theUS housing bubble. Optimism is measured on the demand side: the survey they use askswhether it is a good idea to buy a house given the current market conditions.Our contribution relates to the supply side: we measure the influence of expectations on WTAsamong owners. The positive relationship between expectations and WTAs is fundamental andis expected to work in both directions: high expectations coinciding with high WTAs, and lowexpectations coinciding with low WTAs. We do not expect this relationship to change withmacroeconomic conditions, i.e., we expect it to be present during booming and gloomy marketconditions. However, we do expect that the share of optimists in the population changes withmacroeconomic conditions.For our analysis, we rely on owner-reported values, which, as argued before, reflect a WTA. Welink individual WTAs with individual expectations on future price developments. For owner-occupied housing, the owner can indeed only take the role of a seller and not of a buyer.

5.2 An Empirical Analysis for Italy

We test the relationship between WTAs and expectations for Italy. Italy has a quite high owner-occupancy rate, which equaled 72.3% in 2016.12 Owners’ beliefs are hence of predominantimportance.We use the SHIW as data source. Albeit all waves from 2010 onward include a question onexpected future house prices, the phrasing of the question, the proposed answers and the type ofhouse prices referred to differs in every single survey wave: there are strongly framed questionsformulated in terms of price decreases or increases and questions phrased in neutral language;there are rather simple questions asking whether prices are likely to go up, stay the same orfall, and very complex versions asking for a full distribution of beliefs; questions sometimesrefer to house prices in general, to the price in the household’s neighbourhood or to the priceof the household’s home directly. We exploit these different ways to elicit expectations to testthe robustness of our results.

12See EUROSTAT (EU-SILC): http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=ilc_lvho02&lang=en accessed on January 24, 2019.

13

Table 1: Expectations and Self-Reported Housing Prices.

2016 2014 2012 2010†

Optimist 0.073∗∗∗ 0.103∗∗∗ 0.143∗∗∗ 0.052∗

Intercept 8.824∗∗∗ 8.855∗∗∗ 8.643∗∗∗ 8.023∗∗∗

log(surface) 0.733∗∗∗ 0.755∗∗∗ 0.764∗∗∗ 0.801∗∗∗Number of bathrooms ≥ 2 0.189∗∗∗ 0.188∗∗∗ 0.186∗∗∗ 0.226∗∗∗age -0.001∗∗∗ -0.001∗∗∗ 0.000 0.000age2 0.000∗∗∗ 0.000∗∗∗ 0.000∗ 0.000∗Rating of the property:

Luxury (baseline)Highly desirable -0.304∗∗∗ -0.351∗∗∗ -0.269∗∗∗Mid-Range -0.580∗∗∗ -0.572∗∗∗ -0.460∗∗∗Modest -0.717∗∗∗ -0.776∗∗∗ -0.655∗∗∗Low-Income -0.862∗∗∗ -0.885∗∗∗ -0.793∗∗∗Very Low-Income -0.969∗∗∗ -0.983∗∗∗ -0.863∗∗∗

Location of the dwelling:Isolated area, countryside (baseline)Town outskirts 0.061∗ -0.012 -0.017 0.079o

Between outskirts and town centre 0.070∗ -0.004 -0.019 0.117∗Town centre 0.121∗∗∗ 0.042 0.038 0.204∗∗∗Hamlet 0.057o -0.021 -0.076∗ 0.065

Size of the municipality (no. of inhabitants):up to 20,000 (baseline)from 20,000 to 40,000 0.217∗∗∗ 0.189∗∗∗ 0.222∗∗∗ 0.107∗∗∗from 40,000 to 500,000 0.232∗∗∗ 0.235∗∗∗ 0.280∗∗∗ 0.207∗∗∗more than 500,000 0.431∗∗∗ 0.489∗∗∗ 0.614∗∗∗ 0.511∗∗∗

Administrative region dummies are included.

Adj. R2 (in %) 58.06 60.82 58.78 54.12n 5,337 5,805 5,799 1,866

Notes: Each model regresses the logged self-reported house price on a dummy optimist, which indicatesindividuals with high expectations on future house price growth. Additionally, the subjective hedonicregressions include physical and locational dwelling characteristics.† In the 2010 wave, many respondents answered the question on expectations with “Don’t know”leading to a low number of valid observations. This option was not explicitly given in the followingyears. In 2010, the interviewer was not asked to rate the household main residence.Significance codes: ‘***’ if the p-value is lower than 0.001, ‘**’ if the p-value is lower than ‘0.01’, ‘*’if the p-value is lower than 0.05, ‘o’ if the p-value is lower than 0.1 and ‘ ’ otherwise.

We link owners’ stated expectations with their reported value of their home. We start from ahedonic valuation approach: the (logged) house price equals the sum of physical and locationalhouse characteristics multiplied by associated shadow prices. Additionally to these fundamen-tal factors explaining variation in current house prices, we expect self-reported prices to behigher when reported from an individual with high expectations toward future price growth.Consequently, we include an indicator into the hedonic equation distinguishing optimistic frompessimistic or neutral individuals. The standard hedonic equation (1) becomes a subjectivehedonic equation:

log(P ) = α · optimist +Xβ + Lλ+ ε. (3)

14

The dummy optimist equals one when referring to individuals with high expectations. Weexpect a positive parameter α.

Table 2: Self-reported Housing Prices: Optimists versus Pessimists.

2016 2014 2012 2010

Optimist(O1)†,‡ P(∆ ≥ 2%) > 50% 0.073∗∗∗ 0.120∗∗∗(O2)†,‡ P(∆ ≥ 2%) 0.134∗∗∗ 0.165∗∗∗(O3)† P(∆ ≥ 5%) 0.181∗∗∗(O4)‡ P(∆ ≥ 10%) 0.145∗(O5) E(∆) > 0 0.103∗∗∗(O6)‡ P(∆ > 0) > 50% 0.168∗∗∗(O7)‡ P(∆ > 0) 0.201∗∗∗(O8)§ P(∆ > 0) > 50% / P(∆ ≥ 2%) > 50% 0.143∗∗∗(O9)‡ P(∆ ≥ 5% | P(∆ > 0) > 50) 0.071(O10) P(∆ < 0) < 50% 0.052∗(O11) P(∆ < 0) = 0 -0.016

Pessimist(P1)†,‡ P(∆ ≤ −2%) > 50% -0.064∗∗∗ -0.080∗∗∗(P2)†,‡ P(∆ ≤ −2%) -0.076∗∗∗ -0.087∗∗∗(P3)† P(∆ ≤ −5%) -0.069∗∗∗(P4)‡ P(∆ ≤ −10%) -0.156∗∗∗ -0.001(P5) E(∆) < 0 -0.042∗∗(P6)‡ P(∆ > 0) < 50% -0.102∗∗∗(P7)§ P(∆ > 0) < 50% / P(∆ ≤ −2%) > 50% -0.088∗∗∗(P8)‡ P(∆ ≤ −5% | P(∆ > 0) < 50) 0.013(P9)¶ P(∆ < 0) -0.104∗(P10) P(∆ < 0) > 50% -0.073∗(P11) P(∆ < 0) > 50% ∩ P(∆ < −10%) > 50% -0.056

Notes: The table reports the estimated coefficient to the subjective component in the hedonic equation. Allmodels include the same physical and locational control variables as reported in Table 1. The models differ in theway optimistic or pessimistic attitudes toward future changes in house prices are measured. The specificationin bold highlights the one reported in Table 1. Across survey waves, the question on expectations refers todifferent house prices: the household’s main residence (2016 and 2014), a house in the neighbourhood (2012),house prices in general (2010). ∆ denotes the anticipated change in house prices over the next 12 months.In the 2012 and 2016 wave, survey experiments were run within the SHIW. For details see Appendix A.† For 2016: pooled model across both treatment groups.‡ For 2012: separate model for one treatment group.§ For 2012: pooled model across both treatment groups.¶ Zero probabilities are excluded.Significance codes: ‘***’ if the p-value is lower than 0.001, ‘**’ if the p-value is lower than ‘0.01’, ‘*’ if thep-value is lower than 0.05, ‘o’ if the p-value is lower than 0.1 and ‘ ’ otherwise.

Next to the subjective hedonic equation (3), we also look at specifications, where optimist isreplaced by a dummy referring to pessimists (where we would expect a negative associatedparameter) or by the probability mass allocated to certain price development scenarios. Weexpect negative coefficients for scenarios of falling prices and positive coefficients for scenariosof increasing prices.

15

Table 1 reports the central results. Optimists report significantly larger WTAs than individualswith neutral or negative expectations.The estimated parameters for physical and locational characteristics reported in Table 1 fulfillour expectations: prices increase with additional living area, additional bathrooms and therating of the dwelling. Prices are higher in larger municipalities.13 The location of the dwellingwithin a municipality is not significant in all years. However, when significant, the signs andorder of magnitudes of the various categories are as expected.Age is a delicate characteristic due to its complicated functional form. Two potentially opposingeffects drive the impact of the age of the structure on prices: depreciation reduces the valueas dwellings age, while structures built in certain time periods may be more valuable due tovintage effects. Hence, age is often included as a quadratic polynomial. We also re-estimate ourmodels including age as a smooth function with an a priori unspecified functional form, f(age).The functional form is purely data driven and estimated via penalized least squares (see Wood,2006, for details on Generalized Additive Models). We stick to the parametric estimation as thesigns and magnitudes of the effects most interesting here remain virtually unchanged.

Figure 4: The Full Distribution of Beliefs.

Notes: The figures show the joint distribution of expectations in future house price changes for 2012and 2016.

The results reported in Table 1 relate to different concepts of optimism due to changes inthe survey questionnaire. The definitions are reported in Table 2. This tables also comparesestimated coefficients for various definitions of optimism and pessimism (see also Appendix A).Results are only marginally affected when varying the definition and concept of optimism used,which strongly increases confidence in our findings.While in 2012, 2014 and 2016 the estimated coefficients are highly significant, we find less clearresults for 2010. The 2010 wave is the only one, where the question on expectations refers tohouse prices in general. There is no association established to the owner’s own home. The

13According to the 2001 census, 52% of all dwellings were located in municipalities with less than 20,000inhabitants, 13% in municipalities with 20,000 to 40,000, 23% in municipalities with 40,000 to 500,000 and only11% in municipalities with more than 500,000 inhabitants. See Cannari and Faiella (2008).

16

other survey waves refer to either house prices in the owner’s neighbourhood (2012) or theowner’s home directly (2014, 2016). This finding indicates that the link we establish indeedrelates expectations about one’s own property and its stated value. Expectations about thedevelopment of the more abstract general house price does not seem to be so relevant.We also observe from Table 2 that the magnitude of symmetric effects tends to be larger foroptimists than for pessimists, e.g., when comparing (O1) to (P1). Even when owners believethat prices are likely to go down, the downward adjustment of the WTA is smaller than theupward adjustment when they believe that prices will go up by the same amount. A possibleexplanation is loss aversion.In 2012 and 2016, respondents are asked to specify the full distribution of their beliefs, whichallows us to study loss aversion in more detail.Respondents are asked to allocate probabilities to five different scenarios covering all possiblefuture price developments. The options offered to survey participants differ in the two surveywaves: in 2012,

I ∈ {I1 = (−∞,−10), I2 = (−10,−2), I3 = (−2, 2), I4 = (2, 10), I5 = (10,∞)}

and in 2016,

I ∈ {I1 = (−∞,−5), I2 = (−5,−2), I3 = (−2, 2), I4 = (2, 5), I5 = (5,∞)}.

Figure 4 shows the empirical overall distributions. In both years, larger probability mass isallocated to falling prices (2012: 36.7%, 2016: 38.2%) than to increasing prices (2012: 13.8%,2016: 17.3%). Stable prices receive the largest weight in both years (2012: 49.5%, 2016: 44.5%).

Figure 5: The Impact of Expectations on Owner-Reported Prices.

Notes: The figure depicts the estimated coefficient in the subjective hedonic equation associated withthe probability mass allocated to different scenarios of future house price developments. The shadedareas represent 95% confidence intervals.

17

We estimate separate subjective hedonic models, where we include the probability mass allo-cated to a specific interval as explanatory variable, i.e.,

log(P ) = α · P(∆ ∈ I) +Xβ + Lλ+ ε.

Figure 5 shows the result: a respondent allocating a high probability mass to the event ofincreasing prices reports a higher WTA today. The larger the probability mass the larger theeffect. Vice versa, the more probability mass an individual allocates to the event of decreasingprices, the lower is the self-reported price today. The probability mass allocated to the scenarioof stable prices does not have an effect on self-reported prices.

Figure 6: Optimists versus Pessimists: Loss Aversion.

Notes: The figures compare the magnitudes of estimated coefficients between optimists and pessimists.The left panel compares the magnitudes for the opposing intervals respondents could choose from, e.g.,the magnitude of the effect of P(∆ ∈ I1) and P(∆ ∈ I5). The bars indicate 90% confidence intervals.The right panel pools the probabilities of all negative and positive scenarios, respectively, i.e., wecompare the estimated coefficient for P(∆ ≤ −2%) versus P(∆ ≥ 2%). Results are pooled over bothyears. The bars indicate 95% confidence intervals.

Moreover, we find generally lower coefficients for negative scenarios as compared to positivescenarios, which is a hint for loss aversion. These differences are, however, not consistentlysignificant as shown in Figure 6. Half of the differences are insignificant at a 10% level. To get adefinite answer, we pool the two survey waves and all optimistic (and pessimistic, respectively)scenarios. The pooled model additionally includes a year fixed effect.14 The difference inmagnitudes is significant even at a 5% level.To study the shape of the potential influence of loss aversion on WTAs, we estimate anothermodification of (3). The first model includes the probability mass allocated to optimisticscenarios and the second model the probability mass allocated to pessimistic scenarios,

log(P ) =s1(P(∆ ≥ 2%)) +Xβ + Lλ+ ε, and (4)log(P ) =s2(P(∆ ≤ −2%)) +Xβ + Lλ+ ε. (5)

14Since we can only use the data from one treatment group in 2012 for this comparison, we include allobservations from 2016 with a weight of 0.5. The weighting does, however, not impact the conclusions.

18

We do not specify the functional forms s1(·) and s2(·), but let again the data determine it.15

From the model formulated for optimism (4) and pessimism (5), we extract the functional formof how expectations impact WTAs. The benefit of this modelling strategy lies in the fact thatwe do not have to estimate separate models for each interval I and further do not impose anystructure on the effect.To compare the estimated functional forms between the two models, we compute a relativedeviation curve. Let X = x and L = l be an arbitrary but fixed set of physical and locationalcharacteristics. To ease notation, we write po = P(∆ ≥ 2%) and pp = P(∆ ≤ −2%). Assumingorthogonality between dwelling characteristics and the probability mass allocated to positiveor negative scenarios, i.e., (X,L) ⊥ pO and (X,L) ⊥ pP , yields

log(P )|X = x, L = l, po

log(P )|X = x, L = l, po = 0 = log(P )|po

log(P )|po = 0

and an analogous formula for pp. We use a plug-in estimator16 log(P |p)∧

to construct empiricalrelative deviation curves

foptimist(po) := exp(log(P |po)∧

− log(P |po = 0)∧)

and

fpessimist(pp) := exp(log(P |pp)∧

− log(P |pp = 0)∧)

.

A value of po close to zero describes a person that is not or almost not optimistic. The degreeof optimism then increases with po. The case po = 1 refers to fully convinced optimists.Analogously, pp ranges between pP = 0, which describes individuals that are not pessimistic atall, and pp = 1, which indicates fully convinced pessimists.These relative deviation curves hence measure the change in the magnitude of the effect withincreasing probability mass allocated to either positive (foptimist) or negative (fpessimist) scenar-ios relative to fully convinced optimists or pessimists. It is important to keep in mind that aperson allocating zero probability to, for instance, positive scenarios may allocate probabilitymass to both negative and stable scenarios.Figure 7 shows the deviation curve for 2012 and 2016. The magnitude of the effect associatedwith optimism (measured as the vertical distance between foptimist and the y = 1 line) is largerthan the effect associated with pessimism (correspondingly measured as the vertical distancebetween fpessimist and the y = 1 line). Also, we find that “much more pessimism” is neededto substantially impact the WTA as the curve only really starts sloping downwards once theallocated probability is larger than roughly 20%. This all points towards loss aversion.For 2016, we find strict loss aversion as for any value of p ∈ [0; 1], the effect associated withoptimism is larger than the effect associated with pessimism. For 2012, for p close to zeroresults are ambiguous. Only for p > 0.4, we can really sepak of loss aversion.To assess the overall effect, we compare all distances foptimist(p) − 1 to all distances 1 −fpessimist(p). We hence compute the surface between the curves and the y = 1 line via nu-merical integration:

Aoptimist =∫ 1

0(foptimist(p)− 1) dp and Apessimist =

∫ 1

0(1− fpessimist(p)) dp.

15Again, we estimate the model via penalized least squares. For the smooth term we choose thin plateregression splines of dimension 7 as basis functions (see Wood, 2003).

16Note that the properties of the log-normal distribution imply that the exponent of the plug-in estimator isan unbiased estimator for the median.

19

Figure 7: Deviation Curves.

Notes: Relative deviation curves foptimist(po) and fpessimist(pp) for po, pp ∈ [0; 1]. The left panel refersto 2016 and the right panel to 2012. The degree of optimism and pessimism increases with in po andpp, respectively. The vertical distance between the curve and the y = 1 line indicates the instantaneouseffect of optimistic or pessimistic expectations on WTA.

In general, we call a population loss averse, if

S = Aoptimist

Apessimist

> 1.

The larger S, the larger is the degree of loss aversion. Note that strict loss aversion impliesAoptimist > Apessimist

17 and hence S > 1. However, S > 1 also covers a less strict definition ofloss aversion, where for some p ∈ [0; 1], foptimist(p) − 1 may be smaller than 1 − fpessimist(p).The population is still called loss avers, if the overall size of the effect associated with optimismis larger than the overall effect associated with pessimism.For 2016, we find Aoptimist = 3.45% and Apessimist = 7.40%, and hence S = 2.15. Hence,our aggregate measure yields a large degree of loss aversion. Optimism is roughly twice asimportant as pessimism.In 2012, loss aversion is not strict in the above sense. But the overall measure finds againoverall evidence for loss aversion. S = 3.01 when summing over the entire domain. A moreconservative approach would suggest to sum only over the domain, where fpessimist < 1 asotherwise the interpretation is inexact. This is true for p ∈ [0.386; 1].18 Then, S = 1.89, a valuevery close to 2016: again, optimism is roughly twice as important as pessimism. Note, thatin 2012 the scenarios are “framed” in a more extreme way, as cut-off points are at 10% ratherthan 5%.Overall, we hence conclude that loss aversion dampens the effect of negative expectations onadjustments of the WPA. This finding also implies that a large share of optimists may lead tolarger price effects than a large share of pessimists.

17This follows from the monotonicity of integrals.18Inferred from numerical intersection of the fpessimist and y = 1 lines with a step-size of ε = 0.0001.

20

6 Summary and Conclusions

We argue that subjective data is a valuable source to study housing markets. We make use ofowner-reported values of her home collected in the Italian SHIW and the European HFCS andconstruct subjective residential property price indices.We apply a suitably adjusted hedonic and two variations of the repeat-sales methodology. Theresulting indices are compared to objective counterparts. Particularly for Italy, where a rich setof survey data is available, we find strikingly similar patterns in the subjective and objectiveindices. In particular, annualized changes match almost perfectly.Still, even with the very limited information available in the HFCS we are able to compile quitereliable indices. We emphasize that with every additional survey wave that becomes available,the accuracy of the index improves. Also, the index is constructed in a way such that itpotentially grows into the past with every future survey wave. The resulting long-term indicesincrease the information on price developments in some countries where long-term informationis currently missing.From these analyses we conclude that subjective data is well suited to study housing marketdynamics. The additional information available in surveys enables a more detailed analysisthan any study relying on market data only.For Italy, we further study subjective, owner-stated house prices and the same owner’s beliefson future price changes. We estimate a subjective hedonic equation, which includes statedexpectations on future house price changes as an explanatory variable.We argue that an owner-reported house price elicits the owner’s willingness-to-accept price(WTA). We find strong evidence that optimistic expectations increases owners’ stated WTA,whereas pessimistic expectations coincide with lower WTAs. Results are robust across variousdefinitions of optimism and pessimism. We find stronger effects when the survey directlyasks for expectations about the future price development of the owner’s home or a home inthe neighbourhood as compared to general beliefs about the housing market. These findingsstrongly support the existence of an individual relationship between prices and expectations onthe supply side of the housing market.Additionally, we document that the positive adjustment coinciding with optimism is largerthan the negative adjustment associated with pessimistic views. We relate this finding to lossaversion.

21

References

Agarwal, S. (2007). The impact of homeowners’ housing wealth misestimation on consumptionand saving decisions. Real Estate Economics, 35(2):135–154.

Bailey, M. J., Muth, R. F., and Nourse, H. O. (1963). A regression method for real estate priceindex construction. Journal of the American Statistical Association, 58(304):933–942.

Cannari, L., D’Alessio, G., and Vecchi, G. (2016). House prices in Italy, 1927 – 2012. BancaD’Italia Questioni di Economia e Finanza (Occational Papers), No. 333.

Cannari, L. and Faiella, I. (2008). House prices and housing wealth in Italy. In Householdwealth in Italy. Banca D’Italia.

Case, K. E. and Shiller, R. J. (1987). Prices of single family homes since 1970: New indexesfor four cities. New England Economic Review, Sept/Oct:45–56.

Case, K. E. and Shiller, R. J. (1989). The efficiency of the market for single-family homes.American Economic Review, 79(1):125–137.

Chan, S., Dastrup, S., and Ellen, I. G. (2016). Do homeowners mark to market? a comparisonof self-reported and estimated market home values during the housing boom and bust. RealEstate Economics, 44(3):627–657.

Choi, J. H. and Painter, G. (2018). Self-reported vs. market estimated house values: Are home-owners misinformed or are they purposely misreporting? Real Estate Economics, 46(2):487–520.

Clark, A. E. (2001). What really matters in a job? hedonic measurement using quit data.Labour economics, 8(2):223–242.

D’Ambrosio, C., Clark, A. E., and Barazzetta, M. (2018). Unfairness at work: Well-being andquits. Labour Economics, 51:307–316.

de Haan, J. and Diewert, W. E., editors (2013). Handbook on Residential Property PricesIndices (RPPIs). Methodologies and Working Papers. Eurostat, Luxembourg.

Doiron, D., Fiebig, D. G., Johar, M., and Suziedelyte, A. (2015). Does self-assessed healthmeasure health? Applied Economics, 47(2):180–194.

Gallin, J., Molloy, R., Nielsen, E. R., Smith, P. A., and Sommer, K. (2018). Measuring aggregatehousing wealth: New insights from an automated valuation model.

Goodman Jr, J. L. and Ittner, J. B. (1992). The accuracy of home owners’ estimates of housevalue. Journal of Housing Economics, 2(4):339–357.

Heston, A. and Nakamura, A. O. (2009). Questions about the equivalence of market rents anduser costs for owner occupied housing. Journal of Housing Economics, 18(3):273–279.

Hill, R. J. (2013). Hedonic price indexes for residential housing: A survey, evaluation andtaxonomy. Journal of Economic Surveys, 27(3):879–914.

Hill, R. J. and Scholz, M. (2017). Can geospatial data improve house price indexes? a hedonicimputation approach with splines. Review of Income and Wealth, doi:10.1111/roiw.12303.

22

Hill, R. J. and Syed, I. A. (2016). Hedonic price-rent ratios, user cost, and departures fromequilibrium in the housing market. Regional Science and Urban Economics, 56:60–72.

Himmelberg, C., Mayer, C., and Sinai, T. (2005). Assessing high house prices: Bubbles,fundamentals and misperceptions. Journal of Economic Perspectives, 19(4):67–92.

Jorda, O., Schularick, M., and Taylor, A. M. (2017). Macrofinancial history and the newbusiness cycle facts. NBER Macroeconomics Annual, 31(1):213–263.

Kahneman, D., Knetsch, J. L., and Thaler, R. H. (1990). Experimental tests of the endowmenteffect and the coase theorem. Journal of Political Economy, 98(6):1325–1348.

Kain, J. F. and Quigley, J. M. (1972). Note on owner’s estimate of housing value. Journal ofthe American Statistical Association, 67(340):803–806.

Knoll, K., Schularick, M., and Steger, T. (2017). No price like home: Global house prices,1870–2012. American Economic Review, 107(2):331–353.

Liberini, F., Redoano, M., and Proto, E. (2017). Happy voters. Journal of Public Economics,146:41–57.

Lyons, R. C. (2018). An index of Dublin housing prices since 1900. Mimeo.

Pace, K. (1998). Appraisal using generalized additive models. Journal of Real Estate Research,15(1):77–99.

Piazzesi, M. and Schneider, M. (2009). Momentum traders in the housing market: surveyevidence and a search model. American Economic Review, 99(2):406–11.

Poterba, J. M. (1984). Tax subsidies to owner-occupied housing: An asset-market approach.The Quarterly Journal of Economics, 99(4):729–752.

Senik, C. (2014). Wealth and happiness. Oxford review of economic policy, 30(1):92–108.

Shiller, R. J. (2015). Irrational Exuberance. Princeton University Press, 3rd edition.

Stiglitz, J. E. (1990). Symposium on bubbles. Journal of Economic Perspectives, 4(2):13–18.

Thaler, R. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior& Organization, 1(1):39–60.

Urry, H. L., Nitschke, J. B., Dolski, I., Jackson, D. C., Dalton, K. M., Mueller, C. J.,Rosenkranz, M. A., Ryff, C. D., Singer, B. H., and Davidson, R. J. (2004). Making alife worth living: Neural correlates of well-being. Psychological science, 15(6):367–372.

Vermeulen, P. (2016). Estimating the top tail of the wealth distribution. American EconomicReview: Papers & Proceedings, 106(5):646–650.

Wallace, N. E. and Meese, R. A. (1997). The construction of residential housing price indices:A comparison of repeat-sales, hedonic-regression, and hybrid approaches. The Journal ofReal Estate Finance and Economics, 14(1/2):51–73.

Waltl, S. R. (2016). Variation across price segments and locations: A comprehensive quantileregression analysis of the Sydney housing market. Real Estate Economics, doi:10.1111/1540-6229.12177.

23

Waltl, S. R. (2019). Multidimensional wealth inequality: A hybrid approach toward distribu-tional national accounts in europe. Mimeo.

Ward, G. (2015). Is happiness a predictor of election results? CEP Discussion Paper, No.1343.

Willig, R. D. (1976). Consumer’s surplus without apology. The American Economic Review,66(4):589–597.

Wood, S. N. (2003). Thin-plate regression splines. Journal of the Royal Statistical Society (B),65(1):95–114.

Wood, S. N. (2006). Generalized Additive Models: An Introduction with R. Chapman and Hall/ CRC, Boca Raton, Florida.

24

Appendix

A Measuring expectations from surveys

From 2010 onward, the SIHW experimented with different ways to elicit expectations aboutfuture price changes. In each wave, a question was included, but the presentation differed acrosswaves.

The 2016 wave. In 2016, the question referred to expected price changes of the household’smain residence. Respondents are asked to specify their full distribution of beliefs. A surveyexperiment was run to test for framing effects.Half of the respondents were asked the following question:

In your opinion, how will the value of your principal residence change? Assign atotal of 100 points among the options below according to which you think is morelikely: give high points to the most likely and low points to the least likely. Overthe next 12 months the value of your residence will:

• Decrease by over 5%• Decrease between 5 and 2%• Fluctuate between -2% and +2%• Increase between 2 and 5%• Increase by over 5%

For the other half, the exact amounts of a, say, 5% increase were computed and presented tothe respondent. The amounts use the self-reported value of the property as basis.

The 2014 wave. In 2014, the question also refers to price changes of the household’s mainresidence. Respondents have three options to choose from:

In your opinion, at the end of 2015, the value of your main dwelling will be

• higher,• the same,• lower.

The 2012 wave. In 2012, the question relates to expected changes in house prices in arespondents neighbourhood. A survey experiment was run, where two equally sized groupsreceived quite different choice sets.The first groups was asked to fully specify their distribution of beliefs:

Twelve months from now, the price of a house in your neighbourhood will be (dis-tribute 100 points)

• much higher than today (by 10 per cent or more)• somewhat higher (2 to 10 per cent)• basically the same (no more than a 2 per cent increase or decrease)

25

• somewhat lower (2 to 10 per cent)• much lower than today (by 10 per cent or more)

The other half was first asked to reveal their general beliefs on the likelihood of a future priceincrease. Depending on the provided answer, a more specific follow-up question is asked:

Now we would like to ask your opinion on some future developments. For each ofthe following statements, please assign a score on a scale of 0 to 100 points accordingto your own view of their probability: a low number if there is little chance of thishappening and a high one if there is a good chance.Twelve months from now, the price of a house in your neighbourhood will be higherthan today, even by just one euro (on a scale of 0 to 100).

(a) [If the stated probability is greater or equal to 50%:]Supposing it is higher, it will be at least 5 per cent higher (on a scale of 0 to100).

(b) [If the stated probability less than 50%:]Supposing it is lower, it will be at least 5 per cent lower (on a scale of 0 to100)

The 2010 wave. In 2010, the question refers to house prices in general. Respondents areconfronted with two questions, and they had the option to answer with “Don’t know,” whichled to low responses. In contrast to other survey waves, the question was exclusively framed ina negative way. Respondents were confronted with the following question:

1. On a scale from 0 to 100, what is the likelihood that in a year’s time houseprices will be lower than today?

2. What is the likelihood that they will fall more than 10%?

B Subjective and objective RPPIs

The objective RPPIs are predominantly taken from the BIS data warehouse, that includesRPPIs for a long list of countries. We give preference to indices representing the entire countryand including all types of dwellings (apartments and houses). For some countries more thanone index fulfills these criteria whereas for other countries no index exists that has such a broadcoverage. Table 3 summarizes details about all objective RPPIs used in this article.Figure 3 and Figure 8 graphically compare objective and subjective indices. Table 4 reportsthe full set of subjective index series.

26

Tabl

e3:

Obj

ectiv

ere

siden

tialp

rope

rty

pric

ein

dice

s.

IDC

ount

ryC

over

age

Freq

uenc

yT

ime

fram

eQ

ualit

yC

ompi

ler

Sour

ceSo

urce

IDA

djus

tmen

t(if

sour

cedi

ffere

ntto

com

pile

r)AT

1A

ustr

iaA

lldw

ellin

gs;

Qua

rter

ly20

00–

2017

Hed

onic

NC

BB

ISQ

:AT

:0:1

:0:0

:6:0

Who

leco

untr

yB

E1

Bel

gium

All

dwel

lings

;Q

uart

erly

2005

–20

17H

edon

icN

SIB

ISQ

:BE:

0:1:

0:1:

6:0

Who

leco

untr

yB

E2

Bel

gium

Exist

ing

dwel

lings

;Q

uart

erly

1973

–20

17G

eogr

aphi

cal

NC

BB

ISQ

:BE:

0:1:

1:0:

0:0

Who

leco

untr

yst

ratifi

catio

nB

E3

Bel

gium

Div

erse

Ann

ual

1878

–20

12D

iver

seK

noll

etal

.(20

17)

CY

1C

ypru

sA

lldw

ellin

gs;

Qua

rter

ly20

02–

2017

Hed

onic

NC

BB

ISQ

:CY

:0:1

:0:0

:6:0

Who

leco

untr

yD

E1

Ger

man

yA

llow

ner-

occu

pied

Ann

ual

1975

–20

15D

iver

seN

CB

BIS

A:D

E:0:

1:0:

0:6:

0dw

ellin

gs;W

hole

coun

try

DE

2G

erm

any

All

dwel

lings

;Q

uart

erly

2000

–20

17H

edon

icN

SIB

ISQ

:DE:

0:1:

0:1:

6:0

Who

leco

untr

yD

E3

Ger

man

yD

iver

seA

nnua

l18

70–1

938,

Div

erse

Kno

llet

al.(

2017

)19

62–

2012

EE1

Esto

nia

All

Dw

ellin

gs;

Qua

rter

ly20

05–

2017

Stra

tifica

tion

NSI

BIS

Q:E

E:0:

1:0:

1:1:

0W

hole

coun

try

ES1

Spai

nA

lldw

ellin

gs;

Qua

rter

ly19

95–

2018

Stra

tifica

tion

Gen

eral

BIS

Q:E

S:0:

1:0:

3:1:

0W

hole

coun

try

Gov

ernm

ent

ES2

Spai

nA

nnua

l19

00–

2016

Fran

cisc

oA

mar

alM

acro

hist

ory

Dat

abas

e(J

orda

etal

.,20

17)

FR1

Fran

ceA

lldw

ellin

gs;

Qua

rter

ly20

06–

2017

Hed

onic

NSI

BIS

Q:F

R:0

:1:0

:1:6

:0W

hole

coun

try

FR2

Fran

ceD

iver

seA

nnua

l18

70–

2012

Div

erse

;Mai

nly

Kno

llet

al.(

2017

)R

epea

t-sa

les

GR

1G

reec

eA

lldw

ellin

gs;W

hole

Qua

rter

ly19

94–

2017

No

NC

BB

ISQ

:GR

:1:1

:0:0

:1:0

coun

try

excl

.ca

pita

lG

R2

Gre

ece

Flat

s;C

apita

lQ

uart

erly

1997

–20

17N

oN

CB

BIS

Q:G

R:3

:8:0

:0:1

:0ci

tyan

dsu

burb

sH

U1

Hun

gary

All

dwel

lings

;Q

uart

erly

2007

–20

17H

edon

icN

SIB

ISQ

:HU

:0:1

:0:1

:1:0

Who

leco

untr

yIE

1Ir

elan

dA

lldw

ellin

gsM

onth

ly20

05–

Feb

2018

Hed

onic

NSI

BIS

M:IE

:0:1

:0:1

:0:0

Who

leco

untr

yIE

2Ir

elan

dD

ublin

only

;A

nnua

l19

50–

2014

Hed

onic

Lyon

s(2

018)

IT1

Ital

yA

lldw

ellin

gs;

Qua

rter

ly19

90–

2017

NC

BB

ISQ

:IT:0

:1:0

:0:6

:0

27

Who

leco

untr

yIT

2It

aly

Who

leco

untr

yA

nnua

l19

27–

2012

Div

erse

Can

nari

etal

.(20

16)

LU1

Luxe

mbo

urg

All

dwel

lings

;A

nnua

l19

74–

2015

NSI

BIS

A:L

U:0

:1:0

:1:0

:0W

hole

coun

try

LV1

Latv

iaA

lldw

ellin

gs;

Qua

rter

ly20

06–

2017

Hed

onic

NSI

BIS

Q:L

V:0

:1:0

:1:6

:0W

hole

coun

try

MT

1M

alta

All

dwel

lings

;Q

uart

erly

2000

–20

17N

oN

CB

BIS

Q:M

T:0

:1:0

:0:0

:0W

hole

coun

try

NL

1N

ethe

rland

sA

lldw

ellin

gs;

Qua

rter

ly20

05–

2017

NSI

BIS

Q:N

L:0:

1:0:

1:6:

0W

hole

coun

try

NL

2N

ethe

rland

sEx

istin

gsin

gle-

fam

ily;

Qua

rter

ly19

95–

2017

NSI

BIS

Q:N

L:0:

2:1:

1:0:

0ho

uses

;Who

leco

untr

yPL

1Po

land

Flat

s;Q

uart

erly

2010

–20

17St

ratifi

catio

nN

SIB

ISQ

:PL:

0:8:

0:1:

6:0

Who

leco

untr

yPL

2Po

land

Sing

le-fa

mily

hous

esA

nnua

l20

03–

2016

Aver

age

pric

eN

SIB

ISA

:PL:

0:2:

0:1:

1:0

Who

leco

untr

yPT

1Po

rtug

alA

lldw

ellin

gs;

Qua

rter

ly20

08–

2017

Hed

onic

NSI

BIS

Q:P

T:0

:1:0

:1:6

:0W

hole

coun

try

PT2

Port

ugal

Ann

ual

1971

–20

16Fr

anci

sco

Am

aral

Mac

rohi

stor

yD

atab

ase

(Jor

daet

al.,

2017

)SI

1Sl

oven

iaA

lldw

ellin

gs;

Qua

rter

ly20

07–

2017

Hed

onic

NSI

BIS

Q:S

I:0:1

:0:1

:6:0

Who

leco

untr

ySK

1Sl

ovak

iaA

lldw

ellin

gs;

Qua

rter

ly20

05–

2017

Per

Squa

reN

CB

BIS

Q:S

K:0

:1:0

:2:1

:0W

hole

coun

try

met

erpr

ices

EA1

Euro

Are

aEA

-17;

Aver

age

Qua

rter

ly19

80–

2018

Div

erse

ECB

RPP

.Q.I6

.N.T

D.0

0.3.

00ov

erna

tiona

lind

ices

Not

es:

The

tabl

epr

ovid

esov

ervi

ewin

form

atio

nab

out

the

obje

ctiv

ein

dice

sus

edin

this

artic

le.

Long

serie

sus

ually

incl

ude

stru

ctur

albr

eaks

due

toch

ange

sin

met

hodo

logi

esan

d/or

data

sour

ces.

In-d

epth

met

hodo

logi

cald

escr

iptio

nsar

epr

ovid

edby

the

resp

ectiv

ein

dex

com

pile

rs.

NSI

refe

rsto

Nat

iona

lSta

tistic

alO

ffice

and

NC

Bto

Nat

iona

lCen

tral

Ban

k.In

the

case

ofse

vera

lind

ices

,pre

fere

nce

has

been

give

nto

long

serie

sw

ithth

ebr

oade

stco

vera

gein

term

sof

type

sof

dwel

lings

and

geog

raph

ical

cove

rage

.

28

Figure 8: Subjective versus objective indices.

29

30

Notes: Subjective (S-RPPI) versus objective indices. All indices are nominal and normalized withrespect to the year 2012 or, in case of quarterly data, the average in 2012. Details about the objectiveindices are reported in Table 3. More country results are shown in Figure 3.

31

Tabl

e4:

Subj

ectiv

ere

siden

tialp

rope

rty

pric

ein

dice

s(S

-RPP

I).

Year

ATB

EC

YD

EES

FRG

RIT

LUM

TN

LPT

SISK

EEH

UIE

LVPL

1940

0.7

1941

0.2

1942

0.4

1943

0.4

1944

0.1

1945

0.2

6.7

1946

2.5

2.2

1947

1.1

5.2

1948

0.0

13.5

1949

2.9

11.7

1950

0.4

2.0

8.0

1.0

1951

0.2

16.3

1.8

0.8

1952

0.5

18.3

9.4

0.6

1953

10.1

0.5

1.3

1.3

1.4

1954

5.4

0.6

4.0

3.7

1.0

1955

6.3

0.3

1.1

6.6

2.3

3.1

1956

8.7

0.6

1.0

5.0

1.6

1.7

1957

8.2

0.8

2.3

4.7

1.5

1.0

1958

6.5

0.8

1.1

8.8

2.3

1.2

1.2

1959

5.7

0.5

4.9

8.1

0.8

23.2

7.8

1.4

1960

17.8

10.8

1.3

3.5

8.7

14.2

2.2

10.6

3.5

0.9

1961

24.4

11.2

1.0

4.4

5.3

4.7

1.1

10.8

2.7

1.0

1962

24.0

10.0

25.9

1.2

4.1

4.4

3.0

2.0

18.2

9.7

1.1

1963

3.8

7.2

22.2

1.0

1.5

5.6

3.9

1.5

30.0

1.5

1.4

1964

18.8

10.4

36.3

1.4

3.2

7.2

5.5

1.0

14.4

2.9

1.6

1965

19.9

12.9

33.6

1.1

2.7

6.5

8.2

23.7

1.7

14.9

2.8

1.5

1966

5.3

7.6

36.2

1.8

1.3

5.2

11.2

16.6

1.5

11.2

2.0

2.9

1967

31.2

13.4

34.0

1.6

3.3

8.4

12.9

16.7

2.8

5.0

1.1

1.7

1968

21.2

11.2

5.3

38.8

1.4

3.0

6.7

10.3

14.7

2.7

21.4

2.3

2.5

1969

31.8

11.0

4.2

29.1

1.8

5.0

8.4

8.4

8.1

2.1

13.1

2.6

2.6

1970

24.5

12.0

8.9

37.0

2.3

7.0

8.6

12.5

10.5

10.9

14.3

2.4

6.4

3.0

3.4

1971

18.1

13.7

8.4

41.3

1.9

11.6

3.1

10.7

14.7

2.7

12.4

1.8

4.8

3.8

3.9

32

1972

25.3

14.2

7.3

48.5

2.4

17.2

3.6

8.3

13.4

8.2

17.1

2.2

16.4

5.6

3.5

1973

31.5

17.3

10.9

48.0

2.7

10.3

3.1

10.3

14.9

5.3

14.8

2.5

12.4

5.9

3.9

1974

33.2

18.8

11.8

40.7

3.1

9.2

7.3

10.4

10.6

11.1

20.6

3.4

21.7

6.1

4.4

1975

30.1

18.4

7.8

48.2

3.9

15.2

8.6

15.6

10.2

10.5

18.1

4.3

16.0

3.1

6.1

1976

40.4

20.2

18.0

45.3

3.7

14.1

5.5

13.1

12.8

11.9

25.9

5.0

14.5

19.5

7.6

1977

48.3

22.3

10.6

46.9

6.0

11.5

4.7

13.3

13.2

9.4

26.1

6.3

10.6

6.3

7.0

1978

27.8

25.8

15.0

48.2

5.6

16.3

5.7

18.3

14.3

7.4

35.1

5.4

15.1

8.6

9.2

1979

35.1

25.4

22.3

62.5

6.8

17.9

8.2

17.2

18.9

13.7

33.3

8.0

21.4

6.8

13.8

1980

38.3

26.8

18.2

71.5

8.6

27.9

12.9

25.6

15.9

15.7

31.5

11.3

10.8

10.4

14.1

1981

47.1

26.1

30.3

64.8

9.7

29.0

11.1

27.0

13.1

8.6

29.5

11.6

12.4

12.2

18.4

1982

51.6

22.5

33.5

57.1

10.0

13.5

12.1

24.2

14.6

12.8

33.3

12.3

16.8

7.9

20.2

1983

38.9

26.3

22.9

69.6

11.7

35.3

14.1

25.0

19.8

16.6

26.6

19.1

16.4

6.2

15.8

1984

46.3

27.6

24.3

67.1

12.6

29.6

17.4

31.0

22.0

10.6

31.8

22.7

14.1

12.7

21.6

1985

55.2

24.6

26.5

62.7

13.9

20.1

18.3

32.0

17.2

13.0

30.7

25.5

13.5

6.6

19.5

1986

37.6

33.8

28.7

68.3

13.9

31.1

11.7

32.4

20.9

19.0

31.1

24.3

23.4

28.2

19.3

1987

46.4

28.7

28.1

67.7

16.8

28.8

32.5

34.6

21.3

14.7

32.9

26.2

13.7

13.0

20.9

1988

50.1

31.0

24.4

72.5

18.2

48.2

22.5

37.8

26.0

15.4

31.5

32.6

14.9

6.6

19.7

1989

46.9

34.8

32.8

79.3

18.8

38.5

25.5

39.9

20.1

12.1

39.8

36.3

11.9

18.6

22.3

1990

46.5

35.6

35.1

66.7

24.1

28.0

33.5

40.6

33.0

14.9

36.6

43.2

21.4

15.6

23.5

28.0

1991

42.0

41.2

53.3

77.7

28.3

34.9

35.3

45.4

33.3

19.5

35.5

46.2

63.9

16.6

45.4

29.4

1992

51.5

40.3

48.5

70.3

31.7

23.8

40.7

48.3

38.6

22.9

38.5

43.4

56.7

20.6

14.8

36.1

33.8

1993

63.0

39.3

56.2

85.7

30.6

40.6

34.4

51.2

40.8

24.1

41.6

49.0

31.4

15.6

10.0

33.0

33.4

59.6

1994

58.9

43.1

46.2

86.0

32.7

30.4

40.5

53.0

36.5

38.3

50.0

61.0

61.7

17.6

16.5

36.7

33.5

42.8

1995

57.2

45.0

56.0

90.1

36.8

54.6

39.3

59.0

39.3

37.0

50.7

54.9

63.5

23.8

9.6

38.4

35.9

43.4

1996

72.0

42.8

54.3

94.5

37.7

35.6

39.9

51.6

43.7

36.3

55.0

60.1

59.4

17.4

10.7

28.7

43.2

56.1

27.5

1997

59.8

47.2

58.5

85.6

37.8

43.3

51.8

61.8

39.6

31.1

56.4

62.6

91.6

25.5

11.5

25.3

44.7

22.8

49.7

1998

69.0

51.5

55.8

96.7

38.5

38.1

55.5

63.1

38.1

52.6

58.0

67.7

78.9

20.2

19.8

30.8

61.4

34.8

47.6

1999

77.3

52.9

48.2

94.6

38.9

41.8

52.3

66.7

47.3

55.9

67.4

73.8

87.5

26.3

14.5

36.4

74.7

39.8

35.2

2000

81.3

57.8

68.6

103.

050

.054

.774

.873

.554

.945

.079

.779

.182

.939

.325

.764

.384

.538

.254

.020

0171

.657

.964

.092

.051

.956

.278

.380

.752

.761

.985

.282

.667

.635

.927

.058

.110

1.4

37.2

67.9

2002

76.1

62.7

61.3

90.6

58.5

55.1

92.1

79.6

63.1

61.6

95.2

84.6

88.8

37.3

39.0

84.6

97.4

45.7

32.9

2003

75.9

63.6

65.3

100.

865

.854

.985

.585

.771

.185

.090

.584

.672

.742

.440

.484

.611

9.1

70.7

54.6

2004

90.8

73.4

72.9

95.7

78.0

63.3

89.9

104.

267

.195

.799

.290

.690

.954

.145

.795

.813

3.4

57.2

44.9

2005

84.1

80.5

85.9

99.0

78.8

77.1

88.6

99.3

71.3

97.2

105.

192

.169

.761

.273

.210

0.3

153.

376

.652

.120

0692

.881

.386

.098

.589

.587

.894

.911

1.8

80.7

96.1

101.

097

.271

.664

.865

.610

1.8

171.

910

8.2

68.4

33

2007

91.0

91.3

99.5

101.

288

.198

.388

.411

3.9

82.0

89.5

102.

997

.983

.662

.510

6.9

123.

817

6.6

133.

355

.520

0895

.497

.910

4.5

92.7

90.9

88.7

95.4

123.

185

.512

1.9

109.

097

.799

.368

.480

.258

.817

2.0

152.

261

.020

0993

.710

0.2

112.

810

0.4

114.

594

.311

1.2

114.

188

.394

.010

4.2

95.2

91.0

93.8

86.5

88.9

133.

313

3.7

61.7

2010

88.6

117.

213

5.5

101.

092

.992

.093

.817

0.9

102.

813

2.5

112.

911

6.0

131.

589

.510

4.8

75.6

111.

387

.710

8.9

2011

96.3

97.7

146.

010

9.5

95.0

87.3

124.

312

1.7

95.3

85.6

122.

187

.783

.795

.997

.299

.110

7.7

120.

410

5.7

2012

100.

010

0.0

100.

010

0.0

100.

010

0.0

100.

010

0.0

100.

010

0.0

100.

010

0.0

100.

010

0.0

100.

010

0.0

100.

010

0.0

100.

020

1311

7.9

103.

010

9.4

107.

598

.510

0.1

105.

210

2.7

103.

699

.710

3.1

98.3

114.

312

9.9

70.6

99.0

106.

092

.720

1412

3.7

120.

110

1.2

121.

410

6.8

100.

112

6.3

116.

212

2.8

102.

310

5.8

156.

310

0.8

134.

214

1.3

34

Notes: The figures compare S-RPPIs estimatedfrom pooled HFCS data to objective, transactions-based RPPIs (O-RPPIs). All indices are nominaland normalized with respect to the year 2012 or,in case of quarterly data, the average in 2012. Theeuro area figure also shows the coverage of the indexmeasured as share of EA GDP (ranging between10.4% and 97.9%).

C More than economic conditions

Expectation son future house price growth could just reflect overall economic conditions. Peopleliving in economically low performing regions may just in general be more pessimistic thanpeople living in economically strong regions.For 2016, we analyze the regional variation in expectations, which is depicted in Figure 9. Inall regions, people jointly believe that stable prices are more likely than rising prices and thatfalling prices are more likely than rising prices.19

Figure 9: Expectations Across Regions.

Notes: The figures show the probability mass jointly allocated to a decreasing, stable or increasingprice scenario by all respondents in a region in 2016.

Pessimistic expectations are indeed more common in economically less performing regions: theshare of pessimists is significantly larger in the south than in the center or north. We do notfind an association with the share of optimists.As an example, Figure 10 plots the share of pessimists in 2016 against (youth) unemploymentrates. However, macro-economic conditions explain only a small fraction of the variation in theshare of pessimists (regardless of the specification, we achieve an R2 or roughly 20%), whichclearly shows that pessimistic outlooks are not only driven by general economic conditions.

19Across regions, Italians are, however, undecided whether falling or stable prices are more likely: in particularin Liguria and to a lesser extend also in Abruzzo and Calabria people are jointly more pessimistic. There, 62%(Abruzzo: 55%, Calabria: 55%) of the probability mass is allocated to falling price scenarios. On the contrary,in Campania and Valle d’Aosta the scenarios of rising prices receive a probability of 27% each. In Trentino-AltoAdige (76%) and Marche (55%) beliefs are strong that prices will remain unchanged.

36

Figure 10: Pessimists and Unemployment.

Notes: The figure relates the share of pessimists in 2016 to regional (youth) unemployment rates(Source: EUROSTAT).

37