Presentation on theme: "1 Hedonic Property Value Studies of Transportation Noise: Aircraft & Road Traffic Jon P. Nelson Department of Economics Pennsylvania State University Workshop."— Presentation transcript:
1 Hedonic Property Value Studies of Transportation Noise: Aircraft & Road Traffic Jon P. Nelson Department of Economics Pennsylvania State University Workshop on Regulation of Airport Noise ECORE, December 10, 2007 ULB, Brussels
2 Introduction – Objectives of the Survey Discuss methods used in recent hedonic price studies of airport noise. –Five issues for methodology & econometrics Compare hedonic price methods to stated preference methods as a means of valuing noise damages. –Brief summary of stated preference methods & results Summary of recent aircraft noise damage values. – Compare with earlier meta-analyses (Nelson 2004) and other estimates (Navrud 2002, etc.)
3 Outline of Presentation Hedonic price (HP) model – basic concepts & output Five issues: –Extent of the market or market segmentation –Spatial linkages & econometrics –Housing market adjustments and information (“dynamics”) –Noise measurement & annoyance indices –Advantages & limitations of the HP model Stated preference (SP) survey studies –Summary of three studies applied to airport noise Summary of empirical estimates of noise damages –Compare to earlier results & discuss benefit transfer issues
4 Hedonic Price Model – Basic Concepts Products are “bundles” of characteristics or attributes. –Markets impute implicit prices to each characteristic – Hedonic Price –Historical antecedents: Hass 1922, Waugh 1928, Court 1941, Griliches 1971, Rosen 1974. Empirical work on housing – Ridker & Henning 1967. Examples: –Automobile is a combination of engine size & type, weight, styling, etc. –Housing is bundle of structural, location & environment attributes, measured as amenities or disamenities Econometric methods are used to “unbundled” the market price. –First-stage estimation obtains the marginal hedonic price function (typically non-linear for environmental attributes) for each attribute –Second-stage estimation obtains an (inverse) market demand function for an attribute or willingness-to-pay (WTP) schedule
5 HP Model and Property Values Revealed Preference Methods – housing & rental markets are (weakly) complementary to nuisance avoidance & mitigation. –Absent an explicit market, indirect methods are required to value damages & individual willingness-to-pay to avoid damages –If houses with different noise levels were valued the same, relocation of individuals would establish a noise-discount gradient Estimate: PV = F (S, L, Noise Exposure) –PV = property values, S=structural attributes, L=locational attributes –ln(PV) = a + b(S) + c(L) + d(N) + , where Noise Depreciation Index (NDI) as summary (Walters 1975) –NDI = Pct. change in PV for a decibel (dB) change in noise exposure, e.g., a dB change in the Day-Night Sound Level (DNL or Ldn) –NDI = d 100 = Marginal WTP for localized change in noise exposure
6 Noise Depreciation Index Consider two identical houses: –One located close to a busy airport (60-65 DNL zone) & a comparable house located in an ambient noise area (50-55 DNL zone) –10 dB difference is a doubling of perceived loudness (log scale) Suppose that: –Noisy house is valued on the real estate market at US$180,000 and the quiet house is valued at $200,000, so capitalized discount is $2000 per dB –NDI = ($2000/$200,000) 100 = 1% per dB per property Data requirements for HP model: –Sample of real estate values and associated characteristics (living space, number of bathrooms, measures of access to work, noise index, etc., etc.) Nelson (2004) – meta-analysis of 33 NDI estimates for 23 airports: –Wt. mean NDI of 0.59% per dB (std. dev. = 0.04), median = 0.67%, and a wt. meta-regression estimate of 0.67% (std. error = 0.20). Weights are inverse std. errors of individual NDIs. Meta-analysis based one “best estimate” NDI per study –Moderator variables – mean property value (income proxy), sample size, & dummies for accessibility, linear model*, country *, census data, year
7 Housing Market Segmentation What is the appropriate market size for HP analysis? –Do households choose over the entire market? –Basic problems: hedonic price function is non-linear & noise has to vary Today – large metro datasets & GIS methods –Day et al. (2007), 10900 obs. for Birmingham, UK (submarkets by ethnicity, age, wealth, size of property, location) –Homogeneity Tests (Chow, Tiao-Goldberger, etc.) Ex. 1: Baranzini & Ramirez (Geneva) –Private sector rents: NDI = 0.66% per dB (std. error skipped hereafter) –Public sector rents: NDI = 0.79% –Background noise level = 50 dB for Lden (skipped hereafter) Ex. 2: Day et al, Bateman et al. (cluster analysis) –Glasgow: NDI = 0.40% (4 submarkets; only one significant) –Birmingham: NDI = 1.60% and 0.63% (8 submarkets; two significant)
8 Spatial Econometrics How does the NDI change as more spatial linkages are incorporated? –Residuals in HP models are (positively) spatially-correlated due to common attributes and/or omitted spatial variables or endogeneity –Results in biased standard errors and/or biased coefficient estimates Spatial-lag (SLD) and spatial-error dependence (SED) models –GMM estimator. Weighted neighbor matrix for regressors and/or residuals (distance-decay weighting by Tobler’s first law of geography). Ex.1: Salvi (Zurich); SLD + SED –NDI = 0.75% per dB (close to existing estimates) Ex. 2: Cohen & Coughlin (Atlanta); SLD + SED –NDI = 1.4 to 2.1%, but based on only 19 properties out of 508 obs. –Airport accessibility enhances property values
9 Housing Market Adjustments (“Dynamics”) How does the NDI change in the face of new or better information? –Suppose that housing choices are affected by imperfections in the housing market due to limited and/or misleading information about housing attributes, such as noise levels. (Do people error in only one direction?) Do general housing market conditions matter? –It might be that the noise discount is eliminated by “irrational exuberance,” but HP studies are now available for four decades & many areas Ex. 1: Jud & Winkler (Greensboro–Winston Salem, NC) –Extensive newspaper coverage of an expanded air-cargo hub (Fed Ex) –Properties close to the airport sold at 0.2% discount prior to & 9.4% after the news. Market did adjust, but perhaps more than actual noise change Ex. 2: Pope (Raleigh-Durham, NC) –Using state full-disclosure law, R-D imposed a program of informing prospective buyers about noise levels (binding on sellers & agents) –NDI was 0.25% before the program & 0.39% after (+55%)
10 Alternative Noise & Annoyance Indices Past HP studies of airports rely on a cumulative (average) noise indices, such as Ldn, Lden, & Leq, expressed in 5-dB increments. –Which noise measure is most useful for policy decisions? Break the index into component parts (e.g., number of events, time above 75 dB; nighttime noise level, etc.) –Ex. 1: Levesque (Winnipeg); NDI = 1.30% Measure Ldn at each property using a noise simulation model –Ex. 2: On-going California study; NDI = 0.74 to 0.92% Use dummy variables for each noise contour –Ex. 3: Cohen & Coughlin (Atlanta); NDI = 0.74 to 0.91% Use the noise exposure data and existing Schultz-curve studies to estimate a percent highly-annoyed index for each property. –Ex. 4: Baranzini et al. (Geneva) for traffic noise – construct (1) actual Ldn; (2) perceived Ldn; & (3) perceived annoyance. Survey respondents tend to overest. actual noise levels, especially at lower levels.
11 Advantages & Limitations of the HP Model Advantages: –Uses market behavior where individuals voluntarily make actual exchange decisions using money & real resources; –Not subject to numerous survey biases; –Damage values have been obtained for a large number of airports & are reasonably robust over space & time; –WTP values can be calculated using an appropriate discount rate; –Housing markets sort individuals according to noise sensitivity, which is itself a socially efficient means of limiting noise damages. Limitations: –Not entirely sure what is being perceived & valued (annoyance, health effects, visual, safety, air pollution, costs of moving, etc.); –Choice bundle is complex, e.g., access, so specification matters; –Housing market information or conditions may matter.
12 Stated Preference (SP) Methods Survey approach to valuing public goods –Using a constructed market, respondents are asked to accept (or reject) hypothetical changes at given price –Obtained result is a WTP (or WTA) value or function for a given scenario Many variations depending on: –No. of choice dimensions in the scenario –Type of payment vehicle (tax, energy price, etc.) Hundreds of survey studies exist (Carson’s bibliography has 5000 entries), but relatively few for noise exposure, especially aircraft –Harder to elicit values for intangible nuisance compared to values associated with “tangible” goods, such as green space or cleaner water
13 Examples of Survey Studies of Aircraft Noise Ex. 1 – Feitelson et al. (Dallas- Ft. Worth) –How much would you be willing to pay for a house or apartment if located in a quiet area, rather than close to the airport or under the flightpath? –NDI = 1.5% for houses; NDI = 0.9% for apartments; sharp rise past 70 dB Ex. 2 – Frankel (Chicago) – not in references (see Nelson 2004) –Survey of real estate agents and appraisers – asked to estimate the pct. discount that an average property is diminished by aircraft noise – For 60-70 dB, NDI = 0.64% to 0.71%; 70-77.5 dB, NDI = 1.36% to 1.56% Ex. 3 – Wardman & Bristow (Manchester, Lyon, Bucharest) –Noise evaluated along with nine other quality of life variables & local tax. Noise as number of movements per hour (20, 30), categorical noise levels (type of plane), & time (weekday, weekend, daytime, evening, night). –Time of Day results: Weekday (6pm-10pm), 4.25 cents per movement (Manchester), 7.65 cents (Lyon), and 0.95 cents (Bucharest). Sunday, 6.94 cents (Manchester), 2.94 cents (Lyon?), and 1.31 cents (Bucharest).
14 SP Advantages & Limitations Advantages: –Very flexible, context can be controlled; –Ex ante and ex post policy changes can be valued; –Strong link with preferences, in theory. Limitations: –Results are not very robust; –Choice surveys are subject to several well-known biases, such as Hypothetical bias (protest responses, zeros, DK), Strategic bias (free-rider problem), Embedding/Scope bias (WTP should be size dep.), Sample selection bias (SP estimate can be more or less than HP). Are SP estimates greater than HP estimates for WTP? –Carson et al. (Land Economics 1996) – meta-analysis for 83 studies and 555 estimates; the SP/HP ratio is about 0.62 (so WTP for SP < HP) –Gen (GA Tech diss., 2004) – meta-analysis for 337 SP and 252 HP estimates; SP/HP ratio is about 0.44 (so WTP for SP < HP) –Three SP & HP studies for noise – all possible results (sample selection?)
15 Conclusions Noise discount has probably risen some over time (positive income elasticity): –Airport noise – mean NDI = 0.92%, median = 0.74; Nelson (2004) = 0.67% –Traffic noise – mean NDI = 0.57%, median = 0.54%; Bertrand (1997) = 0.64; Nelson (1982) = 0.40%. Three major applications: –Cost-benefit analyses of specific noise mitigation and abatement projects –Total social-cost evaluations of different transportation modes (“full-cost”) –Models of alternative policy instruments (noise and congestion taxes) Benefit transfer issues: –General problem in environmental economics is the use of a WTP value for a given study area (or mode) for policy evaluation for another location –Both unit value transfers and function transfer are possible –This paper and my earlier meta-analysis provide data for such transfers for all three types of applications