Coastal hazard zones and home values: The quest for an unbiased treatment effect

Transcription

1 Coastal hazard zones and home values: The quest for an unbiased treatment effect Christine Blinn Rob Johnston Klaus Moeltner Presented at the W3133 Meeting, Carlsbad, CA, Feb , 217 Funding by the Northeast Sea Grant Consortium is gratefully acknowledged 1 / 26

2 Intro Impact of location in SFHA on home values (CT,RI,MA,NH,ME) 2 / 26

3 Intro Impact of location in SFHA on home values (CT,RI,MA,NH,ME) Existing lit: -1% to -11% loss - not much known about New England 2 / 26

4 Intro Impact of location in SFHA on home values (CT,RI,MA,NH,ME) Existing lit: -1% to -11% loss - not much known about New England Here: 2 Counties in MA - interesting case studies 2 / 26

5 Intro Impact of location in SFHA on home values (CT,RI,MA,NH,ME) Existing lit: -1% to -11% loss - not much known about New England Here: 2 Counties in MA - interesting case studies Residential sales / 26

6 Intro Impact of location in SFHA on home values (CT,RI,MA,NH,ME) Existing lit: -1% to -11% loss - not much known about New England Here: 2 Counties in MA - interesting case studies Residential sales Focus on 1% annual flood risk Special Flood Hazard Areas (SFHAs) 2 / 26

7 Intro Impact of location in SFHA on home values (CT,RI,MA,NH,ME) Existing lit: -1% to -11% loss - not much known about New England Here: 2 Counties in MA - interesting case studies Residential sales Focus on 1% annual flood risk Special Flood Hazard Areas (SFHAs) Matching with Bayesian regression adjustment 2 / 26

8 Intro T i = E (y 1 (x i )) E (y (x i )) 3 / 26

9 Intro T i = E (y 1 (x i )) E (y (x i )) basic estimate: ŷ (x i ) = y (x j ) ˆT i = y 1 (x i ) y (x j ) ( ) E ˆT i = E (y 1 (x i )) E (y (x j )) 3 / 26

10 Intro T i = E (y 1 (x i )) E (y (x i )) 4 / 26

11 Intro T i = E (y 1 (x i )) E (y (x i )) regression-adjusted estimate: ŷ (x i ) = y (x j ) + ˆµ (x i ) ˆµ (x j ) ˆT i = {y 1 (x i ) y (x j )} {ˆµ (x i ) ˆµ (x j )} ( ) E ˆT i = {E (y 1 (x i )) E (y (x j ))} {E (ˆµ (x i )) E (ˆµ (x j ))} 4 / 26

12 Intro Estimation risks and responses Mis-specified hedonic price function / Use non-parametric matching instead 5 / 26

13 Intro Estimation risks and responses Mis-specified hedonic price function / Use non-parametric matching instead Space-invariant temporal shocks / Force matches to be within 1 calendar year 5 / 26

14 Intro Estimation risks and responses Mis-specified hedonic price function / Use non-parametric matching instead Space-invariant temporal shocks / Force matches to be within 1 calendar year Time-invariant spatial confounders / Include SZ effects in aux. regression 5 / 26

15 Intro Estimation risks and responses Mis-specified hedonic price function / Use non-parametric matching instead Space-invariant temporal shocks / Force matches to be within 1 calendar year Time-invariant spatial confounders / Include SZ effects in aux. regression Biased standard errors / Use Bayesian framework for regression adjustment 5 / 26

16 Matching Balanced Matching Pick control homes such that treatment becomes independent of observables 6 / 26

17 Matching Balanced Matching Pick control homes such that treatment becomes independent of observables Same / similar distribution of x for both groups 6 / 26

18 Matching Balanced Matching Pick control homes such that treatment becomes independent of observables Same / similar distribution of x for both groups Less specification burden for the auxiliary regression 6 / 26

19 Matching MatchingVariables structural nature urban age elevation interstate (miles) sqft beach (miles) pricipal artery (miles) lot coast, estuary (miles) high-density development (miles) bedrooms reservoir, lake (miles acres of industrial land 1m bathrooms river (miles) acres of agricultural land 1m acres of open land 1m sale year forced to be within 1 year on either side of treated 7 / 26

20 Intro Matching Regression Adjustment Results Interior Analysis Matching Legend matched controls available controls treated ± cutoff =.25 miles from coast / estuary cutoff =.5 miles from coast / estuary cutoff = 1. miles from coast / estuary 8 / 26

21 Matching 9 / 26

22 available controls available controls available controls available controls available controls matched controls matched controls matched controls matched controls matched controls Matching 4 age 2 age square footage (1s) lot size (1s of sqft) square footage (1s) lot size (1s of sqft) bedrooms bedrooms bathrooms treated bathrooms treated 1 / 26

23 available controls available controls available controls available controls available controls matched controls matched controls matched controls matched controls matched controls Matching 6 elevation (1 meter resolution) 2 elevation (1 meter resolution miles to beach miles to beach miles to coast/estuary miles to coast/estuary miles to lake/reservoir miles to river treated miles to lake/reservoir miles to river treated 11 / 26

24 available controls available controls available controls available controls available controls matched controls matched controls matched controls matched controls matched controls Matching 3 miles to principal artery 3 miles to principal artery miles to high-density development acres of ind. land, 1 meters acres of ag. land, 1 meters acres of open land, 1 meters treated miles to high-density development acres of ind. land, 1 meters acres of ag. land, 1 meters acres of open land, 1 meters treated 12 / 26

25 Regression Adjustment Regression Adjustment, Plymouth mean coefficient estimates variable difference mean p > b*mean age sqft lot bedrooms bathrooms elev ISMi PAMi beami hidmi coaestmi reslkmi rivmi ag ind op total: / 26

26 Regression Adjustment Regression Adjustment, Suffolk & Norfolk mean coefficient estimates variable difference mean p > b*mean age sqft lot bedrooms bathrooms elev ISMi PAMi beami hidmi coaestmi reslkmi rivmi ag ind op total: / 26

27 Regression Adjustment PLYMOUTH: 9% of matched pairs are in the same SZ 15 / 26

28 Regression Adjustment PLYMOUTH: 9% of matched pairs are in the same SZ average spatial distance between pairs = 2.49 miles 15 / 26

29 Regression Adjustment PLYMOUTH: 9% of matched pairs are in the same SZ average spatial distance between pairs = 2.49 miles 15 / 26

30 Regression Adjustment PLYMOUTH: 9% of matched pairs are in the same SZ average spatial distance between pairs = 2.49 miles SUFFOLK / NORFOLK 89% of matched pairs are in the same SZ 15 / 26

31 Regression Adjustment PLYMOUTH: 9% of matched pairs are in the same SZ average spatial distance between pairs = 2.49 miles SUFFOLK / NORFOLK 89% of matched pairs are in the same SZ average spatial distance between pairs = 1.69 miles 15 / 26

32 Results Coastal Results adjusted ATT hedonic ATT mean lo hi Ts origcs mcs PLYMOUTH ** SUFFOLK & NORFOLK ns / 26

33 density Results.18 Coastal estimation results Plymouth Norfolk/Suffolk ATT ($s) 17 / 26

34 Intro Matching Regression Adjustment Results Interior Analysis Results ATTi< ATTi> ± 18 / 26

35 Intro Matching Regression Adjustment Results Interior Analysis Results ATTi< ATTi> ± 19 / 26

36 Intro Matching Regression Adjustment Results Interior Analysis Interior Analysis cutoff: > 1. miles from coast / estuary Legend matched controls available controls treated ± 2 / 26

37 Interior Analysis 21 / 26

38 Interior Analysis ˆT coast = T coast + A coast, A coast ˆT int = T int + A int, A int T coast T int A coast A int, so (in real terms) T int + A int T coast 22 / 26

39 Interior Analysis ˆT coast = T coast + A coast, A coast ˆT int = T int + A int, A int T coast T int A coast A int, so (in real terms) T int + A int T coast So ˆT int = upper bound for T coast, in real terms. 22 / 26

40 Interior Analysis ˆT coast = T coast + A coast, A coast ˆT int = T int + A int, A int T coast T int A coast A int, so (in real terms) T int + A int T coast So ˆT int = upper bound for T coast, in real terms. Also: ˆT coast ˆT int = lower bound for A coast, in real terms. 22 / 26

41 Interior Analysis Combined Results adjusted ATT hedonic ATT mean lo hi Ts origcs mcs PLYMOUTH (row 1 = coastal, row 2 = interior) ** ns SUFFOLK & NORFOLK (row 1 = coastal, row 2 = interior) ns *** / 26

42 density density Interior Analysis.2 Coastal estimation results Plymouth Norfolk/Suffolk ATT ($s).2 Interior estimation results Plymouth Norfolk/Suffolk ATT ($s) 24 / 26

43 density Interior Analysis.18 Amenity effect Plymouth Norfolk/Suffolk ATT ($s) 25 / 26