Take the Q Train: Value Capture of Public Infrastructure Projects · 2019-10-21 · Take the Q Train: Value Capture of Public Infrastructure Projects Arpit Gupta y Constantine Kontokosta

Take the Q Train: Value Capture of PublicInfrastructure Projects ∗

Arpit Gupta † Constantine Kontokosta ‡ Stijn Van Nieuwerburgh §

September 16, 2019

Abstract

We analyze the impact of the Second Avenue Subway (Q-train) construction on local

real estate prices, which capitalize the benefits of transit spillovers. We find evidence

of higher real estate prices in the vicinity of areas served by the new Q-train, relative

to other areas in Manhattan’s Upper East Side. Only 30% of the private value created

by the subway leads is captured through property taxes, and is insufficient to cover

the cost of the subway. Value capture through targeted property tax increases can help

close the funding gap.

Keywords: Public Finance, Land Value Taxation, Urban Development, Value Cap-

ture

∗We gratefully acknowledge support from the Lincoln Institute, as well as conference participants at theCREFR Spring Conference, and the PREA Institute Conference in NYC. We thank Yuan Lai and MichaelLeahy for superb research assistance.†Department of Finance, Stern School of Business, New York University, 44 W. 4th St, New York, NY

10027, [email protected].‡Marron Institute for Urban Management, New York University, 60 5th Ave, 2nd Floor, New York, NY

10011, [email protected].§Department of Finance, Graduate School of Business, Columbia University, Uris Hall 809, 3022 Broad-

way, New York, NY 10027; NBER; CEPR; [email protected].

1

Electronic copy available at: https://ssrn.com/abstract=3466847

1 Introduction

The growth in major urban centers generates demand for infrastructure improvements,an important component of which is public transport infrastructure. However, the costsof building these urban public transportation improvements is high in the United States.Subway investments, which offer the prospect of carrying the most passengers in thedensest locations, carry particularly large costs. Recent extensions of the 7 and the Qsubway lines in New York City cost about $2.5 billion per mile of construction. Thesehigh costs make it essential to measure the benefits of additional subway construction inorder to assess whether this construction is worthwhile.

Prior literature has documented several potential benefits of subway constructionprojects like the 2nd Avenue Subway (or Q-train), on which we focus our analysis. Thisline can be expected to improve access to workplaces and amenities due to shorter com-muting times (Kahn and Baum-Snow, 2000, 2005; Severen, 2018). Improvements in com-muting times itself can boost labor force participation, particularly among women (Black,Kolesnikova, and Taylor, 2014). Additionally, the reduced traffic congestion on roadsand other public bus transportation lines can be expected to reduce pollution (Ander-son, 2014). In our context, congestion on the 4-5-6 subway line which runs parallel to thenewly constructed Q line can be expected to improve. Other associated benefits of transitlinkages include less drunk driving (Jackson and Owens, 2011), improved retail, as wellas noise and crime reductions around stations (Bowes and Ihlanfeldt, 2001). However,these diffuse and varied benefits are difficult to quantify, and so complicate a straight-forward cost-benefit calculation. The public return to infrastructure investment, definednarrowly as the user fees net of operational expenditures and more broadly to include theincremental property, sales, and labor income tax revenues, captures only a part of the to-tal benefit of infrastructure investment because it ignores these positive externalities theinfrastructure generates for the private sector. A failure to appropriately account for thesebenefits will result in important infrastructure investments remaining unfunded.

Our analysis provides an alternate measure to assess the benefits of infrastructure im-provements. We take the approach that real estate values in the vicinity of public trans-portation hubs capitalize the present value of all future benefits that accrue to householdsand business from transportation gains. To perform this calculation, we must measurehow residential and commercial real estate asset values change after the extension of pub-lic transportation.

We focus on the 2nd Avenue Subway construction as it represents the most substan-tial investment in public subway infrastructure in the United States in the past several

2


decades. We define geographical areas that are “treated” by the subway extension inthree ways. Our baseline treatment definition considers the 2nd Avenue corridor. Thesecond treatment is defined as the area within a 0.3 mile walking distance from one ofthe three new subway stops that were added as part of the expansion. The third treat-ment definition considers buildings whose distance to the nearest subway station on anysubway line is reduced after the opening of the new 2nd Avenue subway stations. Wecompare the changes in real estate values in the treated areas with the changes in corre-sponding control areas on the Upper East Side in a difference-in-difference setup. Ouranalysis in the time-series accounts for both anticipatory effects of the construction, aswell as the possible negative disamenities resulting from the construction process itself.

Our key result is that we find evidence that the construction of the 2nd Avenue Sub-way has had any measurable improvements in real estate values. Our benchmark difference-in-difference specification estimates a 10.6% increase in prices when comparing the tenyears before 2013 to the 6 years after along the avenue corridor of the subway itself. Sincethe subway opened in 2017, this allows for substantial anticipation effects. Allowing forsix additional years of anticipation effects during the construction phrase raises the es-timate to 15.4%. We also estimate specifications which control more finely for buildingamenity effects through the use of building fixed effects or unit-specific characteristicsthrough a repeat-sales approach. Though smaller, at 2–6% price increases, these estimatesstill suggest substantial value creation in the area around subway construction.

A key challenge with out identification approach is to identify the relevant set of prop-erties which are potentially affected by the subway construction. Our baseline approachselects all properties between 59th street and 100th street, and between First and Third Av-enues (the “2nd Avenue corridor”) in comparison with all other properties on the UpperEast Side. We conduct several robustness tests on our treatment definition, and find verycomparable results. One alternate approach identifies the treatment area on the basis ofwalking distance to the new subway stops, using a 0.3mi walking distance limit to iden-tify treated properties. This approach yields very comparable results, of price impactsbetween 4–8%. Another treatment definition focuses on the change in walking distanceas a result of the subway construction, and so identifies several properties that are dis-tant from the subway itself but experienced a gain in walking times. This definition, too,yields similar price effects of between 3–9%.

While our estimates are consistent with prior literature, as documented in the follow-ing section, we emphasize that these results suggest that a considerable component of thepublic investment in the subway system have accrued to private landlords and owners.To the extent that the property tax system is able to recoup some of these expenses, this

3


provides a natural mechanism for local governments to finance these investments. How-ever, there are good reasons to think that existing property tax systems will be incomplete.We show that the New York City property tax code only recuperates about 30% of marketvalue increase in present value. As a result; though the subway itself generated morevalue than the already quite high cost of construction—this value is largely accruing toprivate landowners, rather than the city government.

This motivates the possibility for additional value capture taxes which may help re-coup an additional component of the investment cost, and thereby make possible addi-tional public infrastructure investments. Our findings, therefore, are quite policy relevantgiven ongoing debates in New York City on the future extension of the 2nd Avenue Sub-way line, the repair of the L line, and the East Side access. They also have ramificationsfor the broader debate on how to finance an upgrade to U.S. infrastructure assets and howto provide new infrastructure in developing countries whose governments have limitedborrowing and taxation capacities. Given that infrastructure projects entail enormous ex-penditures of public resources, it is essential to have a full accounting of the total benefitsresulting from these infrastructure expansions, which our work helps to provide.

2 Literature Review

Bom and Ligthart (2014) conduct a meta-regression analysis, summarizing decades of re-search measuring the effects of public infrastructure spending on economic output. Theauthors find that on average for the United States, a 1% increase in the public capital stock(about $134 billion in 2015) would increase private-sector economic output by 0.03% in theshort term ($12.7 billion) and by 0.12% in the long term ($18.7 billion). The CongressionalBudget Office estimates the effect of the same-size expansion at 0.06% ($9.2 billion). Theliterature finds a wide range of estimates for the return on infrastructure investment, de-pending on the assumptions made on the efficiency of an expansion of the public capitalstock, the strength of the crowd-out effect on private investment, and the timing vis-a-visthe business cycle.

The literature tends to find modest to strong increases in the value of residential estateafter the completion of transportation linkages. Recent studies have found price premi-ums for real estate surrounding transit hubs of between 2% and as much as 45% for resi-dential and between 8% and 40% for commercial properties. In a study of San Jose, Cali-fornia, Cervero and Duncan (2002) find that commercial properties within 0.25-mile of astation that was part of the regional commuter system achieved $25 per square foot rental

4


premiums. Hess and Almeida (2007) find a 2-5% premium, on average, for homes proxi-mate to a light rail station in Buffalo, New York. Similar work in Chicago finds that pricesof residential properties closer to a new rapid transit line appreciated approximately 7%more than properties further away during the period from 1986 to 1999 (McMillen andMcDonald, 2004). Earlier work on Chicago is by McDonald and Osuji (1995) and morerecent work by Diao, Leonard, and Sing (2017). The direct price effects of new subwaystations have also been studied for Toronto (Dewees, 1976), Taipei (Lin and Hwang, 2004),and sixteen cities among which Atlanta, Boston, Chicago, Portland, and Washington DCby Kahn and Baum-Snow (2005). Other papers have examined the impacts of subwayconstruction on the broader subway network. Fesselmeyer and Liu (2018) finds a 1.5-2%increase in real estate prices around pre-existing subway lines in Singapore, comparedwith 2-3.2% direct effects on the new lines.

A few studies have identified negative relationships between distance to transit sta-tions and prices, based on the assumption that negative externalities of transit stations(e.g. crowding, noise, crime) are then reflected in lower relative real estate values (Bowesand Ihlanfeldt, 2001; Pan and Zhang, 2008).

Our paper is the first to study recent subway expansion in New York City. As argued,this expansion was the most expensive per-mile expansion in U.S. transportation history.The urban density and pre-existing transportation network make for a different and inter-esting context, as does the interaction between the ownership market (condos and coops)and the rental market.

3 Background

Elevated rail lines were formerly running on 2nd and 3rd Avenues in New York, as apart of citywide system of “el” trains operated by privately managed and jointly fundedcompanies. This network was gradually replaced with underground subways. A pro-posed 2nd Avenue Subway, in particular, was a major component of a proposed subwayexpansion as a part of a fully publicly owned and operated managed entity, the Indepen-dent Subway System (IND). Ultimately, the IND, along with two other private companies,were combined and placed under government control. The elevated 2nd and 3rd Avenuelines were torn down in anticipation of a new underground 2nd Avenue Subway. How-ever, construction hit numerous difficulties across several decades, including the GreatDepression, World War II, and the NYC funding crisis in the 1970s.

The most recent, and successful, period of construction on the 2nd Avenue Subway

5


Figure 1: Time Line of Construction

is highlighted in Figure 1. MTA approval for the subway and funding for design andengineering began in 2000, while a bond issue for its funding was approved in 2005. Cru-cially, the Department of Transportation authorized funding for Phase 1 construction in2006. This was followed by the beginning of construction in 2007. Construction of thesubway tunnel was completed in 2011. By 2013, it was clear that the end of construc-tion was on the horizon and a Community Information Center was opened. The grandopening of the subway was on January 1, 2017. The three new Q-line stops have been inoperation since that date.

Figure 2 highlights the subway line itself in the context of the local area. The Q lineruns for 8.5 miles, including the 1.8 mile stretch of the completed Phase 1 2nd AvenueSubway extension between 59th Street and 96th Street. The construction included threenew subway stations on 2nd Avenue at 72nd Street, 86th Street, and 96th Street.

4 Empirical Specification

One key empirical challenge is that the value of real estate depends on a myriad of factorsbeyond the opening of a new subway line. Other changes in the local economic environ-ment may confound the effects from the transit improvements on real estate values. To

6


Figure 2: Subway Map on the Upper East Side of Manhattan

address this challenge, we propose a differences-in-differences analysis, comparing val-uations on the 2nd Avenue corridor before and after the subway extension, relative tooutcomes in a control group.

In our baseline specification, we define the treatment group to be all the land parcelsbetween 59th street and 100th street and between First Avenue and Third Avenue, takingthe midpoint of the avenues as the demarcation line. This is what we call the 2nd Avenuecorridor. Our control group consists of three corridors that make up the rest of the UpperEast Side. The Lexington Avenue corridor is the collection of parcels between 59th streetand 100th street and between Third and Park Avenues. The Madison Avenue corridor isthe collection of parcels between 59th street and 100th street and between Park and FifthAvenues. Finally, the York Avenue corridor is the collection of parcels between 59th streetand 100th street to the East of the midpoint of First Avenue. Because of the geography ofManhattan, this is a smaller area.

This choice of baseline treatment and control group is driven by a trade-off betweenminimizing the treatment effect on the control group and maximizing the similarity interms of common drivers of real estate valuations. By differencing out trends in real es-tate values in the control group, we remove common drivers in real estate value thataffect the entire area (Upper East Side) and are left with the pure effects of the subway

7


extension. The Lexington Avenue corridor is geographically the closest to the 2nd Av-enue and may be affected the strongest by the neighborhood trends that affect real estatevaluation on 2nd Avenue other than the subway extension. However, the Lexington Av-enue control group may also be directly affected by the subway extension. Residents inthe Lexington corridor benefit from the new subway line, either because it directly short-ens their commutes or because it alleviates congestion on the 4-5-6 subway line, whichruns under Lexington avenue and parallel to the Q line. The resulting improvement intransportation from the 2nd Avenue subway extension would support real estate valuesin the Lexington Avenue corridor. Removing those effects would tend to bias downwardour estimate of the value created by the subway extension. A countervailing effect thatwould tend to bias our treatment effects estimation upward is that the subway expansionmay have made 2nd Avenue more competitive in terms of attracting residential, retail,and other commercial tenants away from Lexington Ave.

Residents living in the York Avenue corridor also potentially benefit from the Q lineextension. Indeed, for most of them, the new 2nd Avenue subway stations are the clos-est ones. We consider York Avenue corridor residents to be in the control group in ourbaseline specification because they are fairly far from the new subway stations, but studydifferent treatment definitions below where this group is part of the treatment group.

Figure 3 indicates the buildings where we have at least one apartment transaction inour sample. Apartments in treated buildings are colored in bright yellow while buildingsin the control sample are in light yellow. The large stars indicate subway stations (and alltheir entrances), including the three new stops on the 2nd Avenue subway.

A second important research design question is when to draw the demarcation linebetween the before and after period. The subway went into operation on January 1st,2017. While there was considerable uncertainty about the exact opening date until thelast minute, eventual project completion was long anticipated. Construction started inApril 2007. In 2011, the original 2013 completion date was pushed back to December 2016.Tunnel excavation began in May 2010 and blasting concluded in March 2013. Forward-looking developers and property owners willing to tolerate the inconvenience of the con-struction project could capture some of the potential future benefits by acting prior to thesubway opening. These anticipatory effects may be reflected in real estate prices, whichare forward looking in that they reflect the discounted value of future rents. In our bench-mark analysis, we strike a middle ground and take January 1st 2013 as the demarcationline between the before and after. This allows for four years of anticipation effects priorto the inauguration of the new subway line. A subway community information centerwas opened in the middle of 2013, signaling that project completion itself was no longer

8


Figure 3: Map of Baseline Treatment and Control

9


in doubt. This choice also provides a large enough sample in the before and after period.This specification can be expressed as:

ln(yit) = α + γ1 · Treatmentit + δ1 · Postit + β1 · Treatment× Postit + X ′it · θ + εit

In which yit reflects the sale price of a unit i in period t in real terms. The key parameterof interest is β1, which corresponds to the relative price increase on properties in thetreatment definition (for instance, the 2nd Avenue corridor), in the period 2013-2018.

To investigate the presence of additional anticipation effects, several empirical specifi-cations split the “Pre” period into January 2003-December 2006 and January 2007-December2012. We call the latter period the Construction Period. In those specifications, real estateprices in the Construction and Post periods are estimated relative to the omitted 2003-06period. This specification can be expressed as:

ln(yit) = α + γ1 · Treatmentit + δ1 · Postit + β1 · Treatment× Postit + X ′it · θ+ δ2 ·Construction Periodit + β2 · Treatment×Construction Periodit + εit

The additional parameter of interest is β2, which corresponds to the relative priceincrease in the construction period (2007-12) relative to the earlier period (2003-06); whichis a period why may incorporate anticipatory price effects or disamenity effects resultingfrom the construction itself.

The controls X include: indicators for condo and studio, the year built, number ofbedrooms, floor, walking distance to Grand Central, walking distance to Central Park; aswell as indicators if these variables are missing. In our repeat-sale specification, we subseton transactions which have a prior sale, and include the log sale amount of the prior saleas an additional covariate.

5 Data

For the purposes of our project, we build a new data set of all residential transactions onNew York City’s Upper East Side from January 2003 until March 2019. The two primarydata sources are the New York City deeds records and StreetEasy. The deeds recordshave information on the the sale price, sale date, address, as well as a tax ID (the BBLcode). From StreetEasy we collect information on all past residential real estate sales onthe Upper East Side via web scraping. We add properties between 96th Street and 100thStreet, which StreetEasy considers to be part of East Harlem. We also eliminate properties

10


that are above 100th Street along Fifth Avenue, which StreetEasy considers to be part ofthe Upper East Side. StreetEasy has apartment unit and building characteristics, whichare absent in the deeds records.

We obtain the exact address of the building, latitude and longitude, apartment unitname (e.g. 17A), the number of bedrooms, number of bathrooms, an indicator variable forcondo, an indicator variable of coop, an indicator variable for studio, the square footage ofthe unit, the year of construction of the building, the transaction date, and the transactionprice. We infer the floor of the unit based on the apartment unit name. There is also a textfield describing the transaction in more detail.

We only use transactions of condo and co-op units. We eliminate transactions that arecommercial space, storage units, maid’s rooms, parking spots, garage based on the textfield. We eliminate units that have zero bathrooms and zero bedrooms but are not studios.Importantly, we remove all “sales” which are neither reported as “sold” nor as “closingregistered.” Cross-checking against the deed records database indicates that these “sales”are not sales but merely removed listings.

We express all transaction prices in real terms by scaling by the Consumer Price In-dex based in December 2017. We then eliminate all transactions with a real price below$100,000 and above $10 million. Transactions below $100,000 (in 2017 dollars) are unlikelyto be arms-length transactions for actual apartment units. Transactions above $10 millionare unlikely to be affected by the 2nd Avenue subway and will distort sample averages.The final sample contains 49,673 transactions.

Table 1 illustrates basic summary statistics from our data. The top panel reports prop-erties on the 2nd Avenue Corridor, which are treated according to our baseline treatmentarea definition. The bottom panel reports properties in the baseline control group (Madi-son Ave, Lexington Ave, and York Ave corridors). We have 19,941 sales in the treatmentgroup and 29,732 in the control group, so that % of transactions are treated observations.

The average property on 2nd Ave costs $1.09mi, is about 1039 square feet large, costs$1062 per sqft, has 1.5 bedrooms bathrooms, and is in a building that is 46 years old atthe time of transaction. The treatment group has 40% condos and 60% coops. Buildingsin the control group cost substantially more. The typical sale price is $1.84mi or $1244persqft, are 200 sqft larger, have 1.9 bedrooms and 1.8 bathrooms, and are older (59 years).There is a smaller fraction of studios (5% vs. 9%), while the condo-coop breakdown tiltsmore towards coops at 30%-70%.

Our difference-in-difference analysis accounts for the level differences in prices acrossthe two sets of properties. However, if there are changes over time in property charac-teristics of transacted properties which differ between treatment and control group, then

11


Table 1: Summary Statistics

Panel A: Treatment Group

N Mean St.Dev p1 p25 p50 p75 p99saleprice 19941 1090000 1020000 189000 509000 761000 1280000 5520000sqft 13355 1039.486 670.708 392 670 850 1250 3158ppsf 13330 1062.365 442.955 332.336 779.949 979 1277.826 2444.582bedrooms 19918 1.501 0.968 0 1 1 2 4bathrooms 19384 1.495 0.825 1 1 1 2 5condo 19941 0.375 0.484 0 0 0 1 1coop 19941 0.625 0.484 0 0 1 1 1studio 19941 0.092 0.289 0 0 0 0 1building age 19941 45.791 24.388 1 28 44 57 105vintage2 19941 0.059 0.235 0 0 0 0 1closest pre 19941 0.324 0.114 0.057 0.245 0.313 0.395 0.551closest post 19941 0.183 0.084 0.007 0.111 0.186 0.247 0.364dist change 19941 0.14 0.128 0 0.011 0.112 0.249 0.429treat2 19941 0.803 0.398 0 1 1 1 1treat3 19941 0.79 0.408 0 1 1 1 1treat4 19941 0.728 0.445 0 0 1 1 1

Panel B: Control Group

N Mean St.Dev p1 p25 p50 p75 p99saleprice 29732 1840000 1790000 203000 646000 1180000 2330000 8730000sqft 15527 1271.255 862.084 379 725 1050 1569 4034ppsf 15449 1243.767 610.658 335.328 838.746 1101.92 1472.258 3381.886bedrooms 29678 1.882 1.063 0 1 2 2.192 5bathrooms 28880 1.831 1.03 1 1 1.5 2.5 5condo 29732 0.304 0.46 0 0 0 1 1coop 29732 0.696 0.46 0 0 1 1 1studio 29732 0.053 0.223 0 0 0 0 1building age 29732 59.009 27.97 1 42 56 83 109vintage2 29732 0.041 0.198 0 0 0 0 1closest pre 29732 0.343 0.223 0.022 0.162 0.283 0.503 0.851closest post 29732 0.266 0.143 0.022 0.158 0.247 0.357 0.603dist change 29732 0.078 0.127 0 0 0 0.13 0.429treat2 29732 0.219 0.414 0 0 0 0 1treat3 29732 0.341 0.474 0 0 0 1 1treat4 29732 0 0 0 0 0 0 0

that could affect the estimate of the subway extension. Therefore, our main specificationswill control for property characteristics. We focus on whether we observe convergence inprices. If the value gap for the 2nd Avenue Corridor is driven by scarce access to publictransportation options, we expect price convergence after subway construction.

12


6 Results

6.1 Corridors: Baseline Treatment and Control

Table 2 highlights our main effects. Recall that our estimation follows a difference-in-difference specification. The dependent variable is the natural log of the real sale pricefor apartment units including condos and coops. The key independent variables are thedifference-in-difference interactions. The Post variable captures the price impact afterJanuary 2013, and so accounts for any general time-series increase in price; relative to theentire pre-period of January 2003-December 2012. Our specifications suggest that thesetime trends are generally important. In column 1, for instance, the coefficient on Postis .0903, suggesting that the post-period is associated with log prices that are higher bynearly 9% in real terms on average. This variable accounts for the general increase invaluation of UES apartments. Comparably, the treatment coefficient captures the valuedifferential associated with being “On 2nd Avenue” in general. As discussed in the datasection, this effect is quite negative. Properties in the 2nd Avenue corridor generallytransact for 47% less than properties in the control group, i.e, in the rest of the Upper EastSide, without considering additional controls.

Table 2: Main Difference-in-Difference Results - Baseline Treatment Definition

(1) (2) (3) (4) (5)VARIABLES Log Price Log Price Log Price Log Price Log Price

Post x On 2nd Ave 0.138*** 0.106*** 0.0484*** 0.154*** 0.0640***(0.0154) (0.00985) (0.00877) (0.0115) (0.0104)

Constr. Period x On 2nd Ave 0.0972*** 0.0302***(0.0118) (0.0105)

Post 0.0903*** 0.114*** 0.112*** 0.176*** 0.165***(0.00982) (0.00627) (0.00553) (0.00737) (0.00656)

On 2nd Ave -0.469*** -0.204*** -0.252***(0.00927) (0.00633) (0.00875)

Constr. Period 0.116*** 0.0991***(0.00741) (0.00660)

Observations 49,673 49,673 49,673 49,673 49,673R-squared 0.068 0.622 0.732 0.628 0.735Controls NO YES YES YES YESBuilding FE NO NO YES NO YES

Notes: Post is an indicator variable for the period after January 1st 2013. Constr. Period is an indicator variable for the constructionperiod between January 1st 2007 and December 31, 2012. On 2nd Ave is an indicator variable for a unit located in the Second AvenueCorridor as defined in the main text. Controls include: an indicator variable for a condo transaction; an indicator variable for a studio;number of bedrooms; number of bathrooms; the floor of the building; the year of construction; distance to Central Park; distance toGrand Central Terminal; as well as indicators if the control variables are missing. Standard errors in parentheses.*** p<0.01, ** p<0.05, * p<0.1.

13


Our key variable of interest is the interaction variable of Post × On 2nd Ave. Thiscaptures the differential price impact of being on the 2nd Ave corridor after 2013, thetime period when subway completion was either imminent or achieved. This periodcaptures at least some of the anticipatory effects of subway completion on real estatevalues, namely those between January 1st 2013 and subway opening on January 1st of2017. It also contains the subsequent price effects in 2017, 2018, and the first quarter of2019. The coefficient on the interaction term in column 1 suggests that the 2nd AvenueSubway resulted in a statistically significant price rise of 13.8% for properties transactingon the avenue. This number is in line with the estimates from the literature discussedabove and suggests that the construction of the subway was associated with a substantialincrease in value. In other words, we observe convergence in prices. Subway constructioncloses nearly 1/3 of the gap in valuations between the 2nd Ave corridor and the rest ofthe Upper East Side.

This treatment effect persists in column 2, in which we add a number of important con-trols to account for the differences in property characteristics documented above. Con-trols include an indicator variable for a condo transaction; an indicator variable for astudio; number of bedrooms; number of bathrooms; the floor of the building; the yearof construction; distance to Central Park (an important recreational amenity); distance toGrand Central Terminal (an important central business district); as well as indicators ifthe control variables are missing. These control variables boost the R2 value from 6.8% incolumn 1 to to 62.2% in column 2. The lower coefficient (in absolute value) of “On 2ndAve” indicates that about half of the unconditional difference in valuations between thetreatment and control group disappears once we control for characteristics. More impor-tantly, the estimate of Post × On 2nd Ave remains large and precisely estimated at 10.6%.It indicates even faster convergence of property prices than in column 1: nearly 1/2 ofthe price difference between 2nd Ave properties and properties in the rest of the UES iseliminated around the time of subway completion.

One possibility is that there are additional property characteristics beyond those in-cluded in column 2, and unobserved to us, that matter for real estate values. If the preva-lence or the valuation of such latent characteristics changes differentially for treatmentand control groups, then we could incorrectly attribute to the second avenue subwayconstruction what instead are compositional changes to the pool of units transacted. Oneconservative way of dealing with this concern is to include building fixed effects. We es-timate such a specification in column 3 of Table 2.1 This specification is comparing trans-

1The coefficient on the treatment variable itself is not separately identified from the building fixed effectsso we drop it in the specifications with building fixed effects.

14


actions in the same building. Our sample is dominated by transactions in large buildings;92% of observations are in buildings that contain at least five transactions in the Pre andat least five transactions in the Post period. Thus, we should have enough power to iden-tify the building fixed effects accurately. Adding building fixed effects increases the R2

to 73.2%. After controlling for building fixed effects, property values are 4.8% higher onsecond Avenue in the Post relative to the Pre period and relative to the control group. Theestimate is significant at the 1% level and economically large.

6.2 Additional Anticipation Effects

We consider the possibility of additional anticipation effects as far back as 2007 when thedecade-long subway construction endeavor first got under way. We include an indicatorvariable “Constr. Period” which takes the value of 1 for transactions between January2007 and December 2012, allowing for six more years of potential anticipation effects. Thisbeing also the period of heaviest construction, it is plausible that this period experienceda reduction in property values due to disamenities (noise, pollution, closure of retail)related to the construction activity itself. The interaction effect of Constr Period×On 2ndAve estimates the net effect of additional anticipation and disamenities on prices in the2nd Ave corridor, relative to the omitted category of 2003-06. The coefficient on Constr.Period itself shows the price impact associated with this period in general, relative to theomitted category of 2003-2006. Under this specification, the Post×On 2nd Ave coefficientmeasures the price change between the period 2013-2019 and the earlier period 2003-2006(rather than relative to 2003-2012 in columns 1 and 2).

Column 4 of Table 2 shows that the construction period was associated with a sub-stantial increase in real estate values in general on the Upper East Side. Prices were 11.6%higher in real terms in 2007-12 relative to 2003-06, after controlling for property character-istics. Properties on the 2nd Ave corridor appreciated by 9.7% relative to properties in thecontrol group. The point estimate is statistically significant and demonstrates the pres-ence of additional anticipation effects, strong enough to outweigh the disamenity effectsfrom construction.

In the Post period, properties on 2nd Ave are 15.4% more valuable than in 2003-06, rel-ative to the control group. In sum, subway construction triggered an initial appreciationof 9.7% in 2007-12 and a further appreciation of 5.7% (15.4%-9.7%) in 2013-2019.

Figure 4 illustrates this result graphically. It plots the coefficient estimates from a dy-namic difference-in-differences specification, in which each calendar year is allowed tohave its own treatment effect. Breaking up the baseline treatment effect year-by-year sug-

15


Figure 4: Dynamic Treatment Effects - Baseline Treatment

gests that we do not see an initial price effect in the years 2005–2007, our control period,in the treatment corridor. We begin to see positive price coefficients in our constructionperiod of 2007–2012; and see even higher coefficients after the beginning of our “post”period from 2013 onwards. The sharp rise in 2013, at which point a community centerwas built and final construction seemed imminent, provides additional justification forour decision to use this year as the cutoff for the post period. This dynamic analysisfurther suggests that we see even higher estimates for 2017–2019, after the subway hasbecome fully functional. At this point, prices appear to have stabilized and suggest thatthe market has largely priced in the subway construction impact.

In column 5 of Table 2, we add building fixed effects to the specification of column4. The early anticipation effect is smaller at 3.1% but remains statistically different fromzero. However, property values in the Post period remain 6.4% higher than in the 2003-

16


06 period on 2nd avenue, compared to the control group. This is an economically andstatistically significant difference.

6.3 Repeat Sales

In Table 3, we perform a repeat-sales analysis. This commonly used approach in realestate valuation compares the prices of properties with the previous price paid for thesame property. It has the virtue of holding all else constant about properties, with thelimitation that we are only able to analyze properties that do, in fact, repeatedly transactin this period. The analysis repeats the full-sample analysis on the subset of apartmentsthat transacts at least twice.2 In all, we have 17,859 repeat sales representing 36% of thetotal number of transactions.

The repeat-sales results in a smaller estimate of the treatment effect. The main spec-ification in column 2, with controls included, results in a 4.2% value creation estimatefrom the subway extension, compared to a 10.6% effect for the full sample. Adding build-ing fixed effects in column 3 lowers the point estimate on the interaction effect to 1.5%,compared to 4.8% for all transactions.

In column 4 we add the log real sale price of the previous transaction of the sameunit, i.e., from the first leg of the repeat sale. The previous price helps control for apart-ment unit or building characteristics not already included in the control variables. Thetreatment effect attenuates further to .4% and is no longer significant. One explanationfor the attenuation after including the lagged price is that unobserved building or apart-ment characteristics upwardly biased the treatment effect in column 2. Note however,that these missing characteristics would have had to differentially impact valuations inthe treatment and control groups after 2013 relative to before 2013. For example, if thereare more swimming pools in buildings on 2nd Ave after 2013 than before 2013, but nochange in swimming pools for the control group, then the estimated relative value cre-ation on 2nd Ave could be due to swimming pools rather than to the subway. This isunlikely to be a concern for the repeat sales sample since we are effectively controlling forany fixed characteristics such as swimming pools by looking at sales of the same unit inthe same pre-existing building. An alternative interpretation of the attenuation is that weare over-controlling by including lagged prices. Lagged prices could (partially) incorpo-

2When determining whether a transaction in our 2003-2019 data set is a repeat sale, we look for transac-tions in StreetEasy before January 2003 to avoid additional selection on properties that transact twice withinthe 2003-2019 time frame. Despite the limited data coverage prior to 2003, this results in several hundredadditional repeat sales. Also, if a property is the subject of two (or more) repeat sales, both (all) repeat-salestransactions for which the second leg of the trade pair is in our sample period 2003–2019 enter the repeatsales subsample.

17


rate the value created by the subway extension. We would then be attributing to missingcharacteristics what is really a subway treatment effect. This concern is more severe themore recent is the previous transaction. Given the evidence we found for early anticipa-tion effects, even repeat-sale transactions where the prior transaction took place as earlyas 2007 could suffer from the over-controlling problem.

Columns 5–7 revisit the breakdown of the Pre period into the (omitted) 2003-2006and the construction period from 2007-2012. For this subsample of repeat sales, the earlyanticipation effect may well be the net effect of a strongly negative disamenity effect anda modestly positive anticipation effect. The repeat-sale sample clearly suggests a strongerrole for disamenities related to construction in the 2007-2012 period. The interaction effectPost x On 2nd Ave is 3.4% (insignificant) in column 5 and essentially zero in columns 6and 7. In sum, relative prices for repeat sales on 2nd avenue fell from the baseline 2003-2006 period to the 2007-12 period, and then recovered in the 2013-19 period to 2003-06levels. The same concern of over-controlling applies to the analysis in columns 5–7.

The complement of the repeat-sales subsample, namely units that transact only oncein the sample, shows much stronger price gains from subway construction than the fullsample.

Table 3: Repeat Sales Subsample - Baseline Treatment Definition

(1) (2) (3) (4) (5) (6) (7)VARIABLES Log Price Log Price Log Price Log Price Log Price Log Price Log Price

Post x On 2nd Ave 0.000755 0.0442*** 0.0176 0.0200** 0.0323* -0.000470 -0.00550(0.0227) (0.0123) (0.0110) (0.00908) (0.0177) (0.0163) (0.0130)

Constr. Period x On 2nd Ave -0.0134 -0.0230 -0.0389***(0.0186) (0.0165) (0.0137)

Post 0.0380** 0.101*** 0.104*** 0.0526*** 0.187*** 0.166*** 0.0238***(0.0149) (0.00818) (0.00728) (0.00603) (0.0121) (0.0110) (0.00904)

On 2nd Ave -0.342*** -0.164*** -0.182*** -0.152*** -0.157***(0.0161) (0.00913) (0.00672) (0.0156) (0.0115)

Constr. Period 0.119*** 0.0848*** -0.0382***(0.0126) (0.0111) (0.00938)

Lagged Log Price Residual 0.608*** 0.616***(0.00509) (0.00518)

Observations 16,883 16,883 16,883 16,883 16,883 16,883 16,883R-squared 0.052 0.723 0.819 0.850 0.726 0.820 0.851Controls NO YES YES YES YES YES YESBuilding FE NO NO YES NO NO YES NO

Notes: Post is an indicator variable for the period after January 1st 2013. Constr. Period is an indicator variable for the constructionperiod between January 1st 2007 and December 31, 2012. On 2nd Ave is an indicator variable for a unit located in the Second AvenueCorridor as defined in the main text. Control variables are the same as in Table 2. Standard errors are in parentheses.*** p<0.01, ** p<0.05, * p<0.1.

18


6.4 New Buildings

One possibility is that the subway extension had different effects on newer buildings. Wedefine newer buildings to be those constructed after January 2003. The variable “vin-tage2” in Table 1 shows that 5.9% of units transacted in the treatment group are in newerbuildings compared to 4.1% in the control group.

Table 4 estimates the triple interaction effect Post x On 2nd Ave x Vintage2. We finda 7.6% larger appreciation for units in newer buildings in the treatment area after sub-way construction than for older buildings. The appreciation is 10% for older buildingsand 17.6% for newer buildings. The additional 7.6% is precisely estimated despite therelatively small share of transactions in buildings built after 2003.

Table 4: Heterogeneous Treatment for New vs. Old Buildings

(1)VARIABLES Log Price

Post x On 2nd Ave 0.0999***(0.00990)

Post x On 2nd Ave x Vintage 2 0.0758**(0.0347)

Post x Vintage 2 -0.126***(0.0294)

Post 0.119***(0.00629)

On 2nd Ave -0.205***(0.00625)

Vintage 2 0.515***(0.0155)

Observations 49,673R-squared 0.632Controls YESBuilding FE NO

Notes: “Vintage2” is an indicator variable which is 1 for units in buildings constructed in 2003 or later and zero otherwise. All othervariables are as in Table 2. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

6.5 Unpacking the Control Group

In Table 5, we revisit our main specification but unpack the control group into its con-stituent corridors. Since the omitted corridor is the Madison Ave corridor (spanning fromFifth Ave to Park Ave), all changes are measured relative to the Madison Ave corridor.Since this corridor is the farthest removed from the 2nd Ave subway and since it containsvery wealthy residents unlikely to use public transportation, this is a natural choice for

19


the omitted category. We continue to see our main treatment effect: prices appreciatedby 11.8% more in the 2nd Ave corridor in the Post period relative to the Madison Avecorridor. In contrast, we see no change for the Lexington Ave corridor and a small (andstatistically imprecisely estimated) change for the York Ave corridor. This justifies ourchoice of combining Madison, Lexington, and York Ave corridors in the control group.Column 1 also shows that property prices on 2nd Ave were the lowest in the pre pe-riod, followed by York Ave, Lex Ave, and Madison Ave. The null effect on Lexington Avesuggests that the potential benefits from reducing congestion on the 4-5-6 line did not ma-terialize, or were offset by increased by reductions in prices due to increased competitionin the real estate market on 2nd Ave.

Column 2 shows a strong 17.4% capital gain on 2nd Ave, relative to Madison Ave andrelative to the pre-construction era of 2003-06. The gain of 11.% in the construction periodagain underscores early anticipation effects. Lexington Ave shows no change in eitherperiod, relative to Madison. In contrast, property prices on York Ave appreciate in the2007-12 period relative to Madison Ave (4.9%). The area continued to improvement rela-tive to Madison in the Post period so that prices caught up further (6.7%). This suggeststhat York Ave may have been at least partially affected by the subway extension. Westudy this possibility in detail below.

7 Alternative Treatment Definitions

7.1 Distance to New Stations

One drawback of our baseline definition of treatment is that we assume that all propertiesalong the 2nd Avenue Corridor are equally treated by new subway construction. Thismay not be the case if areas far from the subway stops, along 2nd Ave, do not find muchof a benefit from using the new subway. To analyze this possibility, we consider a secondtreatment definition which includes all properties which are within 0.3 miles of one ofthe three new 2nd Avenue subway stops. Distance is defined by walking distance ascalculated by Google Maps.3 If these are the properties which benefit the most from thesubway construction, they should expect the greatest property price appreciation. But, itis also possible that the disamenities during the construction period were greatest closeto the subway stops.

3For each one of our buildings, we feed in the street address into the Google Maps API and obtain thedistance to each subway station entrance (multiple per station) on the Upper East Side, to Central Park, andto Grand Central Terminal.

20


Table 5: Unpacking the Control Group

(1) (2)VARIABLES Log Price Log Price

Post x On 2nd Ave 0.118*** 0.174***(0.0142) (0.0166)

Post x On Lexington Ave 0.00152 0.000104(0.0158) (0.0185)

Post x On York Ave 0.0410** 0.0675***(0.0159) (0.0186)

Constr. Period x On 2nd Ave 0.111***(0.0167)

Constr. Period x On Lexington Ave -0.00360(0.0185)

Constr. Period x On York Ave 0.0491***(0.0186)

Post 0.100*** 0.154***(0.0121) (0.0141)

On 2nd Ave -0.476*** -0.530***(0.0137) (0.0161)

On Lexington Ave -0.214*** -0.211***(0.0109) (0.0146)

On York Ave -0.408*** -0.431***(0.0194) (0.0215)

Constr. Period 0.100***(0.0140)

Observations 49,673 49,673R-squared 0.626 0.632Controls YES YESBuilding FE NO NO

Notes: “Post” is an indicator variable for the period after January 1st 2013. “Constr. Period” is an indicator variable for the constructionperiod between January 1st 2007 and December 31, 2012. “Within .3 Miles” is an indicator variable which is 1 for a transaction locatedwithin 0.3 miles of one of the three new subway stations on the Second Avenue subway and 0 otherwise. Controls include: anindicator variable for a condo transaction; an indicator variable for a studio; number of bedrooms; number of bathrooms; the floor ofthe building; the year of construction; distance to Central Park; distance to Grand Central Terminal; as well as indicators if the controlvariables are missing. Standard errors in parentheses.*** p<0.01, ** p<0.05, * p<0.1.

Table 1 refers to this alternative treatment definition as “treat2”. It shows that 80.3%of the transactions on the 2nd Avenue corridor and 21.9% of the transactions in the Madi-son, Lexington, and York Ave corridors fall within 0.3 miles of one of the new subwaystations. In other words, this treatment is strongly but not perfectly correlated with ourbaseline treatment. Figure 5 shows the treated and control buildings. The 0.3-mile dis-tance requirement draws diamond-shaped areas around the three new subway stations.

Table 6 revisits our main difference-in-differences estimation for this alternative treat-ment definition. The structure of this table is identical to that of Table 2 for the baselinetreatment definition based on the corridors. In our favorite specifications in columns 2and 3, we find a strongly positive and statistically significant increase in value due to thesubway for those properties that are within 0.3 miles of one of the three new Q-line sta-tions. The headline increase is 8.1% while the increase with building fixed effects is 4.2%.

21


Figure 5: Treatment Based on Distance to New Stations

The comparable numbers for the baseline treatment were 10.6% and 4.8%. This compari-son suggests that properties in the 2nd Avenue corridor that are not within 0.3 miles froma new station benefitted slightly more from the subway than properties in the LexingtonAve or York Ave corridors that are within 0.3 miles of a new station.

Columns 5 and 6 suggest that the value gain in the Post period reflects the continuationfrom a relative appreciation during the construction period. Prices in the treatment area

22


Table 6: Treatment Based on Distance to New Stations


Post x Within .3 Miles 0.0604*** 0.0809*** 0.0422*** 0.0290*** 0.126*** 0.0603*** 0.00443(0.0153) (0.00975) (0.00862) (0.00900) (0.0114) (0.0102) (0.0130)

Constr. Period x Within .3 Miles 0.0863*** 0.0317*** -0.0379***(0.0116) (0.0103) (0.0137)

Post 0.113*** 0.119*** 0.112*** 0.0467*** 0.182*** 0.164*** 0.0187*(0.0104) (0.00660) (0.00582) (0.00635) (0.00772) (0.00686) (0.00953)

Within .3 Miles -0.360*** -0.135*** -0.139*** -0.182*** -0.115***(0.00921) (0.00613) (0.00656) (0.00860) (0.0114)

Constr. Period 0.119*** 0.0970*** -0.0367***(0.00783) (0.00691) (0.00989)



Notes: “Post” is an indicator variable for the period after January 1st 2013. “Constr. Period” is an indicator variable for the constructionperiod between January 1st 2007 and December 31, 2012. “Within .3 Miles” is an indicator variable which is 1 for a transaction locatedwithin 0.3 miles of one of the three new subway stations on the Second Avenue subway and 0 otherwise. Controls include: anindicator variable for a condo transaction; an indicator variable for a studio; number of bedrooms; number of bathrooms; the floor ofthe building; the year of construction; distance to Central Park; distance to Grand Central Terminal; as well as indicators if the controlvariables are missing. Standard errors in parentheses.*** p<0.01, ** p<0.05, * p/<0.1.

appreciated by 12.6% (6%) more than the control group, relative to the 2003-06 period,in the specification without (with) fixed effects. Columns 4 and 7 are estimates from ourrepeat sale sample. These two estimates are mixed with column 4 suggesting a 2.9% priceappreciation and column 7 none.

Further investigation, reported in Appendix Table 10, breaks down the treatmentgroup into units that are between 0 and 0.10 miles, between 0.10 and 0.20 miles, andbetween 0.20 and 0.30 miles from a new Q-line station. The 8.1% price gain in the mainspecification (column 2) results from a large and precisely estimated gain of 8.2% in prop-erties between 0.2 and 0.3 miles away from the station, and a zero gain closer by (the 1.8%gain in the 0-0.1 mile ring is not significant). This analysis also shows a price depressionclosest to the station during the construction period. This is exactly where we expect thedisamenities from construction to show up. In sharp contrast, prices in the 0.2-0.3 milering appreciate 8.2% during the construction period and an additional 3.9% (for a totaleffect of 12.1%) in the Post period.

23


7.2 Closest Subway Station Becomes Closer

We explore an additional alternative treatment definition which places greater weight onperipheral properties which experienced possibly large gains in transit access. For everyapartment in our sample, we compute the distance to the nearest subway station on anyline serving the Upper East Side, both before and after the addition of the three stationson the Second Avenue subway line (8 stations in total). Distance is calculated as walkingdistance based on Google Maps. We calculate the reduction in distance to the nearestsubway station entrance triggered by the opening of the three new Q-line stations.

Table 1 reports that for the average unit in the 2nd Ave corridor, the closest station was.32 miles away before the Q-line extension and .18 miles after, for an average distance re-duction of .14 (225 meters). For the residents of the other three corridors, the averagereduction was smaller at .08 miles (129 meters). The latter is the combination of a zeroreduction for all residents of the Madison corridor and most residents of the Lexingtoncorridor, on the one hand, and a large reduction for the residents on the York Ave corridor,on the other hand. We define an apartment as treated if there is a strictly positive distancereduction to the nearest subway station on the Upper East Side. Table 1 refers to this al-ternative treatment definition as “treat3”. It shows that 79% of the transactions in the 2ndAvenue corridor and 34.1% of the transactions in the Madison, Lexington, and York Avecorridors are in a building which experiences a change in distance to the nearest station.Again, this treatment is strongly but not perfectly correlated with our baseline treatment.Figure 6 shows the treated and control group buildings according to this second alterna-tive treatment definition. The largest change with the baseline and the first alternativetreatment is that all properties east of Second Avenue are now treated.

Table 7 shows the difference-in-difference estimates. For our main specifications incolumns 2 and 3, we find a similar effect from the subway extension: 9.1% without and3% with building fixed effects. In columns 5 and 6 we find a significantly positive effect inthe Construction period, and a continuation in the Post period. Prices in properties withan improvement in distance to the closest subway station end up 14.6% (5.3%) above2003-06 levels in the specification without (with) building fixed effects. Columns 4 and 7are estimates from our repeat sale sample. These two estimates are mixed with column 4suggesting a 1.9% price appreciation and column 7 none.

Further investigation, reported in Appendix Table 11, breaks down the treatmentgroup into units that experienced a reduction in distance (i) between 0 and 0.10 miles,and (ii) greater than 0.10 miles. The latter group consists mostly of units east of 2nd Ave.The 9.1% overall price effect is the average of estimated gains in the former group, and

24


Figure 6: Treatment Based on Change in Distance to Nearest Station

4.2% in the latter group. While one might think that units experiencing a larger gain are“more intensively” treated, we find that the gains are largest for those who experiencea modest reduction in distance. This includes residents on Third Ave and even some on2nd Ave. For several far east residents, it is possible that the 2nd Ave subway remainstoo far away to be useful. Far east residents may continue to use alternate transportationoptions.

25


Table 7: Treatment Based on Change in Distance to Nearest Station


Post x Change in Dist. 0.0652*** 0.0906*** 0.0304*** 0.0191** 0.146*** 0.0534*** -0.00420(0.0149) (0.00972) (0.00859) (0.00907) (0.0114) (0.0102) (0.0131)

Constr. Period x Change in Dist. 0.108*** 0.0428*** -0.0356**(0.0116) (0.0103) (0.0138)

Post 0.101*** 0.108*** 0.115*** 0.0496*** 0.161*** 0.163*** 0.0221**(0.0107) (0.00702) (0.00620) (0.00682) (0.00824) (0.00733) (0.0102)

Change in Dist. -0.532*** -0.155*** -0.127*** -0.212*** -0.104***(0.00892) (0.00792) (0.00792) (0.00987) (0.0123)

Constr. Period 0.100*** 0.0891*** -0.0359***(0.00840) (0.00737) (0.0106)



Notes: “Post” is an indicator variable for the period after January 1st 2013. “Constr. Period” is an indicator variable for the constructionperiod between January 1st 2007 and December 31, 2012. “Change in Dist.” is an indicator variable which is 1 for a transaction of aunit for which the distance to the nearest subway station became smaller after the addition of the three new subway stations on theSecond Avenue subway and 0 otherwise. Controls include: an indicator variable for a condo transaction; an indicator variable fora studio; number of bedrooms; number of bathrooms; the floor of the building; the year of construction; distance to Central Park;distance to Grand Central Terminal; as well as indicators if the control variables are missing. Standard errors in parentheses.*** p<0.01, ** p<0.05, * p<0.1.

7.3 All of the Above

A third alternative treatment definition combines the first three treatments. We consider aunit treated if it is treated under all three previous definitions. Table 1 reports that 72.8%of units on the 2nd Ave corridor satisfy this requirement (“treat4”) and none of the unitson the other corridors, by construction. About 29.2% of the overall sample receives thiscombination treatment. Figure 7 shows the treatment and control groups according tothis combination treatment definition. This treatment isolates properties on the 2nd Avecorridor, close to a new subway station, for which one of the new stations is the closestsubway option (i.e., there is a positive change in distance).

Table 8 shows the difference-in-difference estimates. For our main specifications incolumns 2 and 3, we find a 9.4% and 4% subway effect, both of which are preciselyestimated. In columns 5 and 6 we find a significantly positive effect in the Construc-tion period, and a continuation in the Post period. Prices in the combination treatmentgroup post-subway construction end up 12.8% (5.8%) above 2003-06 levels in column 5(6). Columns 4 and 7 are estimates from our repeat sale sample. These two estimates are

26


Figure 7: Combination Treatment

mixed with column 4 suggesting a 1.7% price appreciation and column 7 none.The analysis in this section confirms large and robust estimated effects from the Q line

subway extension.

27


Table 8: Combination Treatment


Post x All Treats 0.101*** 0.0942*** 0.0402*** 0.0172* 0.128*** 0.0584*** -0.00940(0.0167) (0.0107) (0.00948) (0.00969) (0.0125) (0.0112) (0.0138)

Constr. Period x All Treats True 0.0701*** 0.0364*** -0.0409***(0.0128) (0.0114) (0.0146)

Post 0.113*** 0.128*** 0.119*** 0.0547*** 0.200*** 0.174*** 0.0226***(0.00906) (0.00579) (0.00509) (0.00552) (0.00680) (0.00605) (0.00827)

All Treats -0.446*** -0.165*** -0.150*** -0.199*** -0.124***(0.0101) (0.00679) (0.00721) (0.00940) (0.0122)

Constr. Period 0.135*** 0.100*** -0.0430***(0.00688) (0.00610) (0.00855)



Notes: “Post” is an indicator variable for the period after January 1st 2013. “Constr. Period” is an indicator variable for the constructionperiod between January 1st 2007 and December 31, 2012. “Change in Dist.” is an indicator variable which is 1 for a transaction of aunit for which the distance to the nearest subway station became smaller after the addition of the three new subway stations on theSecond Avenue subway and 0 otherwise. Controls include: an indicator variable for a condo transaction; an indicator variable fora studio; number of bedrooms; number of bathrooms; the floor of the building; the year of construction; distance to Central Park;distance to Grand Central Terminal; as well as indicators if the control variables are missing. Standard errors in parentheses.*** p<0.01, ** p<0.05, * p<0.1.

8 Value Capture

In this section, we take our baseline estimates for the value created by the subway basedon the observed transactions and use them to compute the aggregate value creation forthe stock of residential real estate on the Upper East Side. We then use property tax datato compute how much of this value creation flows back to the city in the form of highertaxes. We find that while there is an overall gain, there is a significant public shortfallcompared to the cost of the subway extension.

8.1 Baseline Valuation of the Stock of Real Estate

We start by valuing the stock of real estate in the treatment area in the period before sub-way construction. We choose 2012 as a base year, the last year of our baseline definitionof the “Pre” period. This stock consists of owner-occupied residential real estate, renter-occupied residential real estate, and commercial real estate.

28


8.1.1 Owner-occupied Residential Buildings

Imputing the value of owner-occupied residential real estate occurs in three steps.

Step 1: Transacted Units For each apartment in the baseline treatment area for whichwe observe at least one sale, we use the dynamic difference-in-differences specificationwith controls to impute an annual valuation for the year 2012. The imputation uses theactual apartment unit and building characteristics alongside the estimated coefficients.Since the regression specification includes a condo indicator variable, valuation differ-ences between coop and condo units are taken into account.

Step 2: Other Units in Coop Buildings with Transactions Even though we observemore than 16 years of transactions in a liquid market, many coop units never transact inour sample. Based on building-level data, we know how many units there are in eachcoop building and therefore what fraction f of units we are missing. We obtain the val-uation of the entire building by multiplying the cumulative value of the units for whichwe have trades by 1+ f . The underlying assumption is that the average characteristics ofthe missing coop units are the same as those of the transacted units.

Step 3: All Other Units Based on a master list of all tax identifiers (Borough-Block-Lot or BBL codes) from the New York City Department of Finance, we obtain a list ofall condo units and coop buildings in the Second Avenue corridor and a 2012 estimatedmarket value (EMV). After comparing this master list against our transactions data, weobtain the BBLs for which we see no transactions. Each condo unit has its own BBLwhereas all units in a coop building share the same BBL. For each condo unit and coopbuilding valued in steps 1 and 2, we calculate an EMV multiple. The EMV multiple isthe ratio of our 2012 valuation to the 2012 EMV in the tax roll data. We then average theEMVs separately for condos and coops and for each tax block. there are 83 tax blocks inour Second Avenue treatment area. The 2012 value of a missing condo unit is its 2012EMV from the city records times the average EMV multiple for condos in that tax block.The value of a missing coop building is the 2012 EMV for that coop building times theEMV multiple for coop buildings in that tax block.

8.1.2 Renter-occupied Buildings

Next, there is a large stock of rental buildings to consider. After all, the home ownershiprate in Manhattan is only 23%. We collect data from StreetEasy on all rental buildings in

29


the Upper East Side. For each apartment unit in the rental data (with a rental listing atsome point between 2001 and 2019) we obtain characteristics such as exact location (intreatment area or not, distance from Central Park, distance from Grand Central), num-ber of bedrooms, number of bathrooms, studio flag, floor, as well as building amenities(doorman, bike room, gym, elevator, laundry room, concierge, live-in super, pool, storageroom, roof deck, children’s playroom, parking), year built and the total number of unitsin the building. We then obtain a 2012 value for each observed unit by combining its owncharacteristics and the dynamic difference-in-difference coefficients from the condo andcoop transactions. We set the condo flag equal to 0.5, assuming that rentals are value atthe average of coops and condos. To obtain the total value of the building, we scale upthe cumulative value of the transacted units by 1 + f , where f is the fraction of missingunits in the rental listings data.

For every rental building valued, we compute the EMV multiple as the ratio of our2012 valuation to the city’s 2012 EMV. We average the EMV ratios for rental buildings bytax block. We value the renter-occupied buildings (BBLs) for which we have no StreetEasyrental data by multiplying their EMV from the tax roll data by the EMV multiple for rentalbuildings in that tax block. As explained below, our valuation approach is consistent withNew York City’s DOF approach which values all owner-occupied buildings as if theywere rental buildings.

8.1.3 Commercial Properties

The final property type is commercial, non-residential real estate: retail, office, and indus-trial properties. Since the 2nd Ave corridor is largely a residential neighborhood, by farthe dominant commercial real estate is street-level urban retail (shops and restaurants),followed by parking garages. Since we observe very few transactions data for these prop-erties and lack sufficient building characteristics for the transactions we do observe, weexclusively use the EMV approach. Specifically, we calculate the average EMV ratio ineach tax block pooling all condo, coop, and rental BBLs. We then value the commercialBBLs as the product of their 2012 EMV by the average EMV multiple in the correspondingtax block.

8.2 Tax Pass-through

To assess the amount of property taxes that typically passes through to the city govern-ment in response to property appreciation, we make use of tax assessment records forNew York City. For owners of condos and co-ops, the city assess property taxes on a por-

30


tion of the property’s market value, the so-called assessed value. This assessed value iscalculated using several steps.

First, the property’s “NYC market value” is calculated as follows. The city imputesthe annual Net Operating Income (NOI) per sqft based on comparable rental buildings,typically the average of three buildings that are geographically close to the building inquestion, of similar size and similar vintage. This annual NOI is then divided by a caprate, a ratio of NOI to price, to produce the NYC estimate of market value (EMV). Thecity’s records indicate that the cap rate was set uniformly at 12.42% in January 2018. Thetrue market cap rate at that time was around 4%, so that NYC’s EMV is about three timessmaller than the actual market value.

Next, the property assessed value is set at 45% of the NYC market value, and ownerspay a 12.9% tax rate on the assessed value, minus exemptions. Absent exemptions, thetax rate is 5.8% of NYC market value. Changes in property taxes over time are graduallyphased in over a five year period.

While we do not observe exemptions, we have tax paid in 2015 for all properties. Thisdata suggests a non-trivial role for exemptions, and indicates that actual tax paid is 4.8%of the city’s assessment of market value or about 1.0% of true market value. We assumethat marginal increases in property values due to the subway creation result in the sameincrease in property taxes as the average increase in value.

Adopting that estimate of tax pass-through, we start with a simple example based ona typical condo building in the 2nd Avenue corridor. Suppose a building has 90 units,and a total of 140,000 sqft. Suppose the true market value is $175 million, or $1,250 persqft. This is the observed average square foot price in the treatment area in Table 1. Givena NOI of $50 per foot, this valuation corresponds to a 4.0% cap rate. The NYC assessmentof market value is based on a 12.42% cap rate and will be $37.65 million, or $269 per sqft.The assessed value will be $16.94 million, or $14 million after the 17.5% condo abatement,a form of exemption. Tax paid is $1.8 million yearly, which is 4.8% of NYC market valueand 1.5% of true market value as mentioned above.

Suppose now that the 2nd Ave subway increases the value of this building by 10.8%,the point estimate in column 2 of Table 2, or $18.9 million. The NYC market value willincrease by $4.1 million, or an increase in assessed value of $1.8m. By our assumptionon average tax paid, taxes paid will increase annually by $194,609 in year 5 and beyond(and gradually be phased in before that). Assuming a government discount rate of 3%,corresponding to NYC’s municipal bond yield, the subway results in $5.78 million inextra tax revenue in present value. The estimate of value capture, or how much of theprice increase accrues to the city government is $5.78 / $18.9 m = 30.6%.

31


We then adopt these estimates to the Second Avenue corridor definition in the Up-per East Side, in Table 9. We estimate the total 2012 market value of real estate in ourtreatment area at $87 billion across several categories of property ownership, as detailedabove. This table shows the estimated increase in market value across several of ourspecifications from Table 2, which are displayed again for convenience in the first row.The maintained assumption is that the value gain from subway construction was uni-form across property types. We estimate that the subway construction led to a total valueincrease of between $4.2–13.2 billion, depending on the specification. However, we esti-mate that the city itself is able to capture only 30.6% of this value in the form of highertaxes, as discussed above. This table displays our estimates of the amount captured bythe city government in present value terms from increased property taxation under therow “Property Tax Receipts.” They range from $1.3–4.1 billion. We contrast these num-bers with the construction cost of $4.5 billion; and show in the last row the shortfall inrevenue. Even though the value generated from subway construction was substantialenough to exceed the (very large) subway construction cost, the gains largely accrued toprivate owners of condo and co-op units and landlords managing rental and commercialreal estate properties. The city suffered a substantial shortfall, especially under the moreconservative value gain estimates.

Table 9: Estimates of Value Creation

(2) (3) (4) (5)Value Add Under: Value in 2012 Standard Controls Building FE Constr. Period Constr. Period

+ Building FE

Treatment Effect: 0.106*** 0.048*** 0.153*** 0.064***(0.00985) (0.00877) (0.01154) (0.01039)

Owner-Occupied Residential ($b) 33 3.46 1.58 5 2.1

Renter-Occupied Buildings ($b) 30 3.21 1.46 4.64 1.94

Commercial Non-residential ($b) 12 1.27 .58 1.84 .77

Total ($b): 75 7.9 3.62 11.48 4.8Property Tax Receipt ($b): 2.42 1.11 3.51 1.47Net Gain to Govt ($b): (2.08) (3.39) (.99) (3.03)

32


8.3 Value Capture Through Micro-targeted Property Taxes

An alternative mechanism of property taxation would levy on each individual propertythe maximal amount of property taxes which would leave owners no worse off, and so ex-tract the entire value generation possible for the city government. Assuming that the cityhas the ability to levy micro-targeted property taxes would allow the city to recoup thefull increase in value creation. Strikingly, nearly all of our estimates of the value gain fromthe Second Ave Subway construction itself exceed the cost of construction. Our estimatessuggest that while the cost of construction of the subway is quite high; so is the valuegeneration, at least in densely populated areas such as the Upper East Side. These gainsare sufficiently valuable that city governments could self-finance infrastructure expensesif they were able to better appropriate the gains from construction, gains that currentlyaccrue instead to private landowners.

8.4 Value Capture in Practice

Alternative value capture policies have succesfully been used as infrastructure financingmechansims in cities such as Hong Kong and Tokyo. Although Manhattan lacks the landtenure system of Hong Kong or the strong eminant domain powers used by Tokyo, itdoes share the key feature that allows both of these cities’ value capture policies to work,high population density. While the specifics of these value capture policies may not beapplicable to the United States, they do provide successful examples of value capture asa financing tool.

Hong Kong’s MTR Corporation, the transit authourity, is regularly profitable withlow fares and a low portion of their revenues from fares (Leong, 2016). As Cervero andMurakami (2009) explain, Hong Kong uses the ‘Rail + Property’ system for rail develop-ment and value capture. When building or expanding a rail line, MRT purchases landfor development from the Hong Kong Government at ”before-development” prices withexclusive development rights. After construction, they sell the land to private developersfor a higher ”after-developemnt” price and capture the value delta. They sometimes alsorecevie a portion of future revenues from the developed properties or a portion of theproperties directly (Cervero and Murakami, 2009, pp. 2024).

Tokyo has a long history of financing transit development with value capture policytools. For most of the 20th century, the main tool was land readjustment in which, ’land-holders give up their property and in return receive parcels that are roughly half the sizeof their original parcels, but that enjoy full infrastructure services (e.g., railway stations,roads, water, and electricity). The remaining land is used for roads and public spaces

33


such as parks and is also sold to cover railway development costs’ (Ingram and Hong,2011, pp. 291). This is a process called codevelopment. More recently, land readjustmenthas been used to clear the path for new rail lines, but without the rail company retainingthe surrounding land that allows them to develop or collect retail revenue (Ingram andHong, 2011, pp. 292). Still, the municipal rail authority is sold the land where the rail islaid at pre-transit prices.

9 Conclusion and Next Steps

Though public transit is essential to manage urban commuting, and is associated with awide array of potential benefits, new subway construction costs have risen to enormousamounts. In order to justify further construction with these costs, transit must demon-strate significant returns either directly or through the capitalization of externalities inreal estate prices.

We find evidence of such capitalization using a difference-in-difference framework.Our baseline estimates compare the increase in real estate prices on the 2nd Avenue cor-ridor where the subway was extended, relative to other areas in the Upper East Side.We contrast a number of treatment definitions, including living near the subway stopsthemselves, experiencing an improvement in commute time, or a combination of thesetreatments. We control finely for other aspects of property price valuation through build-ing fixed effects. Our estimates suggest price appreciation of treated properties benefitingfrom Subway expansion of about 5–10% in our benchmark specification.

These estimates suggest substantial externalities resulting from subway expansion,capitalized into prices. However, these benefits largely accrue to private landlords. Weestimate that the city itself will only recoup a fraction of this increase in the form of futureproperty taxes, suggesting considerable scope for additional value capture taxation whichmay provide the basis for future infrastructure funding.

34


References

Anderson, M., 2014, “Subways, Strikes, and Slowdowns: The Impacts of Public Transiton Traffic Congestion,” American Economic Review, 104(9), 2763–796.

Black, D. A., N. Kolesnikova, and L. J. Taylor, 2014, “Why Do So Few Women Work inNew York (And So Many in Minneapolis)? Labor Supply of Married Women acrossU.S. Cities,” Journal of Urban Economics, 79, 59–71.

Bom, P. R., and J. E. Ligthart, 2014, “What Have We Learned from Three Decades ofResearch on the Productivity of Public Capital?,” Journal of economic surveys, 28(5), 889–916.

Bowes, D. R., and K. R. Ihlanfeldt, 2001, “Identifying the Impacts of Rail Transit Stationson Residential Property Values,” Journal of Urban Economics, 50(1), 1–25.

Cervero, R., and M. Duncan, 2002, “Transit’s Value-Added Effects: Light and CommuterRail Services and Commercial Land Values,” Transportation research record, 1805(1), 8–15.

Cervero, R., and J. Murakami, 2009, “Rail and Property Development in Hong Kong:Experiences and Extensions,” Urban Studies, 46(10), 2019–2043.

Dewees, D. N., 1976, “The Effect of a Subway on Residential Property Values in Toronto,”Journal of Urban Economics, 3(4), 357–369.

Diao, M., D. Leonard, and T. F. Sing, 2017, “Spatial-difference-in-differences Models forImpact of New Mass Rapid Transit Line on Private Housing Values,” Regional Scienceand Urban Economics, 67, 64–77.

Fesselmeyer, E., and H. Liu, 2018, “How much do users value a network expansion?Evidence from the public transit system in Singapore,” Regional Science and Urban Eco-nomics, 71, 46–61.

Hess, D. B., and T. M. Almeida, 2007, “Impact of Proximity to Light Rail Rapid Transit onStation-area Property Values in Buffalo, New York,” Urban Studies, 44, 1041–1068.

Ingram, G. K., and Y.-H. Hong, 2011, Chapter 1: Land Value Capture: Types and Outcomes.Lincoln Institute of Land Policy.

Jackson, C. K., and E. G. Owens, 2011, “One for the Road: Public Transportation, AlcoholConsumption, and Intoxicated Driving,” Journal of Public Economics, 95(1), 106–121.

35


Kahn, M. E., and N. Baum-Snow, 2000, “The Effects of New Public Projects to ExpandUrban Rail Transit,” Journal of Public Economics, 77(2), 241–63.

, 2005, “Effects of Urban Rail Transit Expansions: Evidence from Sixteen Cities,1970-2000,” Brookings-Wharton papers on urban affairs, pp. 147–206.

Leong, L., 2016, “The ‘Rail plus Property’ model: Hong Kong’ssuccessful self-financing formula,” https://www.mckinsey.com/

industries/capital-projects-and-infrastructure/our-insights/

the-rail-plus-property-model, Accessed on 2019-09-10.

Lin, J.-J., and C.-H. Hwang, 2004, “Analysis of Property Prices before and after the Open-ing of the Taipei Subway System,” Annals of Regional Science, 38(4), 687–704.

McDonald, J. F., and C. I. Osuji, 1995, “The Effect of Anticipated Transportation Improve-ment on Residential Land Values,” Regional Science and Urban Economics, 25(3), 261–278.

McMillen, D. P., and J. McDonald, 2004, “Reaction of House Prices to a New Rapid TransitLine: Chicago’s Midway Line, 1983–1999,” Real Estate Economics, 32(3), 463–486.

Pan, H., and M. Zhang, 2008, “Rail Transit Impacts on Land Use: Evidence from Shanghai,China,” Transportation Research Record, 2048(2048), 16–25.

Severen, C., 2018, “Commuting, Labor, and Housing Market Effects of Mass Transporta-tion: Welfare and Identification,” Working Paper (Federal Reserve Bank of Philadelphia).

36


https://www.mckinsey.com/industries/capital-projects-and-infrastructure/our-insights/the-rail-plus-property-model



A Appendix

Table 10: Within Distance Broken Down


Post x Within 0 - .1 mi -0.0185 0.0181 0.00586 0.00681 -0.00626(0.0317) (0.0202) (0.0176) (0.0237) (0.0208)

Post x Within .1 -.2 mi -0.00491 -0.00550 -0.00332 -0.00491 0.00841(0.0217) (0.0139) (0.0123) (0.0165) (0.0146)

Post x Within .2 -.3 mi 0.0809*** 0.0816*** 0.0429*** 0.121*** 0.0574***(0.0164) (0.0104) (0.00919) (0.0122) (0.0110)

Constr. Period x Within 0 - .1 mi -0.0195 -0.0221(0.0243) (0.0214)

Constr. Period x Within .1 -.2 mi -1.13e-05 0.0242(0.0168) (0.0149)

Constr. Period x Within .2 -.3 mi 0.0818*** 0.0280**(0.0124) (0.0111)

Post 0.114*** 0.126*** 0.115*** 0.194*** 0.168***(0.00999) (0.00638) (0.00562) (0.00748) (0.00664)

Constr. Period 0.127*** 0.0975***(0.00758) (0.00669)

Within 0 - .1 mi -0.0902*** -0.0544*** -0.0414**(0.0193) (0.0123) (0.0176)

Within .1 -.2 mi -0.201*** -0.0408*** -0.0431***(0.0132) (0.00864) (0.0125)

Within .2 -.3 mi -0.282*** -0.116*** -0.157***(0.00977) (0.00639) (0.00913)


37


Figure 8: Within Distance Broken Down Coefficient Plot

38


Figure 9: Change in Distance Broken Down Coefficient Plot

39


Figure 10: Treatment Effect Broken Down By Number of Bedrooms Coefficient Plot

Notes: This figure plots the interaction coefficients from a specification where the ”treatment” and ”treatment x post” variables arereplaced with several interaction variables of treatment x number of bedrooms dummy and several treatment x number of bedroomsdummy variable x post variables. This allows us to show a heterogenous treatment effect for apartments of different sizes. The resultssuggest larger apartments in the Second Avenue Corridor experienced a larger relative gain to equivalent units in the control areaafter the Second Avenue Subway.

40


Figure 11: Average Estimated Market Value Multiples by Tax Block and Property Type

Notes: The estimated market value multiple is the valuation we assign a unit or building divided by the city’s estimated market value.For each tax block, we take the mean of the mutltiples for each property type. This figure plots the distribution of those averages forcondos,coops, and rental buildings.

41


Table 11: Change in Distance Broken Down


Post x Chg. dist 0-0.10mi 0.124*** 0.176*** 0.0592*** 0.326*** 0.148***(0.0219) (0.0144) (0.0131) (0.0160) (0.0148)

Post x Chg. dist > 0.10mi 0.0438*** 0.0549*** 0.0198** 0.141*** 0.0823***(0.0163) (0.0106) (0.00932) (0.0117) (0.0104)

Constr. Period x Chg. dist 0-0.10mi 0.309*** 0.170***(0.0148) (0.0135)

Constr. Period x Chg. dist > 0.10mi 0.161*** 0.116***(0.00962) (0.00850)

Post 0.101*** 0.108*** 0.115*** 0.109*** 0.115***(0.0107) (0.00701) (0.00620) (0.00696) (0.00618)

Chg. dist 0-0.10mi -0.482*** -0.231*** -0.385***(0.0129) (0.00969) (0.0122)

Chg. dist > 0.10mi -0.554*** -0.103*** -0.188***(0.00979) (0.00886) (0.0102)


A Data Appendix

Our data begins with 66,934 unique observations from condominum and cooperativebuildings on the Upper East Side downloaded from StreetEasy.com. From the site directly,we start with the date and price of the sale, the building address and year of construction,the unit, number of bedrooms and bathrooms in the unit, unit square footage, and thenumber of units and stories in the building. Additionally, they provide a listing status,which provides a description of the sale status. From the Google Maps API, we have thecoordinates for each building and their walking distnace to each subway entrance in theUES as well to Central Park and Grand Central Station.

We begin by dropping any observations from units likely to be non-residential usingkey terms in the unit variable such as ”commerical”,”storage,” and ”medical.” We alsodrop observations with the term ”multi” which indicates the sale of more than one apart-ment unit. This cuts our sample size to 65,398 sales.

Next, we create two variables for street and street number from the address field.These are standarized so numbered streets contain only digits and avenues only letters.We then perform manual adjustments to the these fields to ensure buildigs with multipleaddresses due to multiple entrances are treated as a single building. The unit variable

42


is cleaned by removing common prefixes such as ”#” and non-alphanumeric characters.Some mannual adjustments are made in cases when a unit is clearly mislabled due to atypo or extra letter.

Next, we merge our Street Easy data with a set of 43,609 condo and coop sales in theUES from 2003-2018 from New York City deeds records. These deeds records are sim-ilarly cleaned and given a street and street number variable. We merge these sampleson the year,month, unit, street, and street number variables. Before merging, we ensurethese 4 variables uniquely identify each observation by dropping duplicates on these vari-ables. We drop 499 of these duplicates from the deeds records and 1069 from the StreetEasy data. The merge results in a common sample of 64,689 observations. 21,586 are un-matched from the Street Easy data, 4907 are unmatched from the deeds data, and 38,196succefully matched.

We next fill in missing unit level information with information found in previous salesof the same unit. We enforce consistency across sales of the same unit for bathrooms andbedrooms. We then drop observations from the Street Easy data with the sale description”No Longer Available on StreetEasy” or a sale price of 0 which are not genuine transac-tions. This cuts our sample down to 64,687 observations.

To ensure there are no duplicate sales, we drop any sale of a unit within 30 days ofa sale of the same unit. This leaves us with 60,091 observations. We use the sale price

43


from the deeds records when available, convert these prices to 2017 dollars, and dropany observation above $10 million and below $100 thousand dollars in real terms whichbrings the sample size down to 57,564.

Finally, we assign each of these sales to a cooridor group based on its location in theUES. Cooresponding to the middle of the respective cooridor, these groups are MadionAve, Lexington Ave, Second Ave, and York Ave. Sales in the Second Ave corridor areassigned our first treatment designation. Observations within 0.3 miles of a second av-enue subway entrance are assigned our second treatment definition and sales of unitsthat had a change in distance to nearest subway entrance after the Second Ave Subwayare assigned our thrid treatment. Treatment 4 is assigned to those observations in all 3treatment groups. Becuase of data coverage we keep only sales taking place in 2003 andbeyond. Our final sample size is 49,673

Valuation After creating our unique sales dateset, we move on to the task of valuingthe stock of real estate in the treatment area/Second Avenue Cooridor. We begin with the2012 NYC tax rolls which contains a list of 1,048,575 Bourough-Block-Lot (BBL) numberswhich unqiely identifies either a condominum unit or cooperative/rental/commericalbuilding. Each BBL has an owner name, estimated market value, and tax class given.

We then cut this list down to only 10,103 BBLs from tax blocks in the Second AvenueCooridor and assign each of them a 2012 valuation. To do so, we do the following:

Use our sales data and run a dynamic differnce-in-differnce regression using yearfixed effects with 03-06 as our excluded control period. We then use the coefficients fromthis regression and predict a sales price for each unit. We then subtract the relevant yearcoefficient for each unit and add in the 2012 coefficient to get a predicted 2012 valua-tion. Becuase all units for a cooperative building share a single BBL, for each cooperativeBBL,we sum the value of each unit and divide this sum by the percent of units in thebuildings we’ve observed trade and get the 2012 valuation for the entire building.

Next, we use the coefficents from the regression ran on the sales data and peform thissame procedure laid out above for coop building on a dataset of rental building apartmentrentals. We then merge this list of BBLs and their estimated 2012 value with our tax rolldata.

At this point, roughly 45% of our BBLs have an estimated 2012 value. We must imputea value for the remaining 55%. To do so we do the following:

For each BBL with a valuation, calculate the EMV multiple ( 2012ValueEstimatedMarketValue )

44


Seperately for condo units, coop buildings, and rental buildings cacluate the averageEMV multiple by tax block. Each unvalued condo unit, coop building, and rental buildingis given an imputed value by multiplying its EMV by its cooresponding block level EMVmultiple average.

For remaning unvalued BBLs we use the average EMV multiple for the block. Finally,any remaining unvaled BBLs are valued using the sample average EMV multiple.

We then assign each BBL one of three types (owner occupied, rental, or commerical)on the basis of its tax class. Total value is summed across each type.

45


Figure 12: Mean Sale Price by Group and Year

46


Figure 13: Mean sale price by Group and Year

47


Table 12: Repeat Sales: Renovation Expenditure as Control

(1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES Log Price Log Price Log Price Log Price Log Price Log Price Log Price Log Price

post treat 0.0200** 0.0231** 0.0290*** 0.0321*** 0.0191** 0.0233** 0.0172* 0.0202**

(0.00908) (0.00910) (0.00900) (0.00902) (0.00907) (0.00910) (0.00969) (0.00970)

Post 0.0526*** 0.0496*** 0.0467*** 0.0433*** 0.0496*** 0.0452*** 0.0547*** 0.0519***

(0.00603) (0.00607) (0.00635) (0.00639) (0.00682) (0.00687) (0.00552) (0.00555)

treat -0.182*** -0.182*** -0.139*** -0.139*** -0.127*** -0.129*** -0.150*** -0.151***

(0.00672) (0.00671) (0.00656) (0.00655) (0.00792) (0.00793) (0.00721) (0.00721)

miss yb = o, - - - - - - - -

lagged log price residual 0.608*** 0.606*** 0.610*** 0.608*** 0.614*** 0.611*** 0.612*** 0.609***

(0.00509) (0.00510) (0.00507) (0.00509) (0.00506) (0.00508) (0.00507) (0.00508)

Condo x Renovation Expenditure 1.59e-07 1.87e-07 1.97e-07 2.16e-07

(6.74e-07) (6.75e-07) (6.76e-07) (6.75e-07)

Coop x Renovation Expenditure Per Unit 7.29e-07*** 7.83e-07*** 8.36e-07*** 7.84e-07***

(1.68e-07) (1.69e-07) (1.69e-07) (1.68e-07)

Observations 16,883 16,883 16,883 16,883 16,883 16,883 16,883 16,883

R-squared 0.850 0.851 0.850 0.850 0.849 0.850 0.850 0.850

Controls YES YES YES YES YES YES YES YES

)

Notes: The table above has standard repeat sale regressions for each of our four treatment defintions to the left of a regression thatadds renovation expenditure as a control for housing quality. ”Coop x Renovation” is the total ammount of renovation expentiturefor a coopertive unit’s building in the period between the last sale of the unit adjusted for the number of units in the building (coop-erative buildings have construction permits aggregated across all units in reporting). ”Condo x Renovation” is the total ammount ofrenovation expentiture for a condominum unit in the period between the last sale of the unit. Standard errors in parentheses.*** p<0.01, ** p<0.05, * p<0.1.

48


Table 13: Construction Activity as Outcome Variable

Log Estimated Cost Log Number Permits Issued Log Number New Constr. Jobs Log Number New Res. Buildings

Post x On Second Ave .0647784 .0157471 .0405556 .112019

(0.06) (0.26) (0.22) (0.41)

On Second Ave -.4466768*** -.9423494*** -.8276411*** -.1141597

(0.04) (0.12) (0.11) (0.21)

Post .287891*** .4742796* .2089062 .2021231

(0.03) (0.18) (0.16) (0.28)

r2 .0186646 .6122481 .5950243 .0397035

N 19175 60 60 50

Notes: ”Log Estimated Cost” is the natural log of the estimated inital cost of a particular permited construction job. ”Log NumberPermits Issued” is the natural log of the ammount of permits issued in a treatment-year. ”Log Number New Constr. Jobs” is thenatural log of the number of construction jobs with permits issued for it in a treatment-year (each job may recieve multiple permits).”Log Number New Res. Buildings” is the natural log of the number of new residential buildings that receive a permit in a treatment-year. Standard errors in parentheses.*** p<0.01, ** p<0.05, * p<0.1.

Table 14: Valuation Source Breakdown

Owner-Occupied Renter-Occupied CommericalPercent of All BBLs 81.24 9.77 8.99

Percent Valued from Data 40 83 1.5Percent Imputed from EMV Averages 60 17 98.5

Notes: This table breaks down the source of the figures used for our valuation in Section 8 and table 9. Row 1 shows the share of our10,113 borough-block-lot numbers from each of our 3 property types. Rows 2 and 3 show the share of the BBLs from each type valuedfrom characteristics we observe in our data vs. our imputation procedure used in cases they are unavailable.

49


Table 15: 10th Percentile of Income by Census Tract

Tract 2005-2009 Avg. 2013-2017 Avg. Change

Control

106.02 $30,746 $39,246 $8500114.01 $10,793 $24,234 $13,441114.02 $20,229 $27,698 $7469116 $27,130 $19,867 $−7263120 $32,800 $43,640 $10,840122 $24,315 $33,996 $9681124 $25,796 $14,783 $−11,013128 $37,463 $20,180 $−17,283130 $37,754 $20,844 $−16,910132 $18,703 $19,335 $632136 $21,351 $18,920 $−2431140 $35,763 $30,553 $−5210142 $49,733 $34,111 $−15,622144.01 $34,372 $27,253 $−7119144.02 $36,539 $31,112 $−5427148.01 $23,848 $25,428 $1580148.02 $28,533 $22,208 $−6325150.01 $39,081 $14,100 $−24,981150.02 $24,459 $36,337 $11,878152 $7432 $7140 $−292158.01 $29,974 $43,861 $13,887158.02 $14,336 $19,155 $4819160.01 $34,958160.02 $10,442 $23,774 $13,332162 $4852 $6727 $1875166 $9601 $7889 $−1712

Treatment

168 $3404 $3777 $Control Area Avg. $24,978 $23,699 $−1279110 $24,044 $17,383 $−6661118 $25,320 $22,401 $−2919126 $23,172 $26,091 $2919134 $22,773 $15,959 $−6814138 $20,450 $24,867 $4417146.01 $28,640 $19,500 $−9140146.02 $35,603 $30,513 $−5090154 $18,001 $23,707 $5706156.01 $15,384 $16,855 $1471156.02 $5638 $10,547 $4909164 $4447 $4762 $315Treatment Area Avg. $20,316 $19,326 $−990

50




Control

106.02 $59,490 $66,854 $7364114.01 $32,640 $94,672 $62,032114.02 $52,356 $64,461 $12,105116 $45,842 $45,716 $−126120 $82,045 $80,397 $−1648122 $49,420 $67,233 $17,813124 $56,456 $62,371 $5915128 $73,173 $71,171 $−2002130 $93,791 $72,702 $−21,089132 $42,684 $51,837 $9153136 $47,900 $56,221 $8321140 $82,895 $71,251 $−11,644142 $80,162 $77,313 $−2849144.01 $67,374 $53,204 $−14,170144.02 $55,248 $66,355 $11,107148.01 $65,057 $78,245 $13,188148.02 $64,276 $55,530 $−8746150.01 $82,731 $80,947 $−1784150.02 $81,920 $91,599 $9679152 $26,918 $23,978 $−2940158.01 $71,157 $87,044 $15,887158.02 $42,317 $44,860 $2543160.01 $103,985160.02 $47,808 $56,854 $9046162 $12,035 $13,749 $1714166 $15,351 $15,247 $−104

Treatment

168 $9604 $9428 $Control Area Avg. $57,209 $59,971 $2762110 $55,785 $48,801 $−6984118 $57,125 $61,454 $4329126 $54,085 $67,200 $13,115134 $54,450 $38,673 $−15,777138 $54,580 $70,725 $16,145146.01 $56,619 $67,208 $10,589146.02 $59,419 $56,493 $−2926154 $41,324 $53,111 $11,787156.01 $40,033 $49,978 $9945156.02 $16,104 $15,686 $−418164 $16,506 $11,344 $−5162Treatment Area Avg. $46,002 $49,152 $3150

51


Table 17: Median Income by Census Tract


Control

106.02 $103,533 $134,943 $31,410114.01 $106,618 $233,646 $127,028114.02 $125,978 $155,353 $29,375116 $74,231 $84,038 $9,807120 $157,601 $187,292 $29,691122 $149,808 $156,827 $7,019124 $90,592 $125,466 $34,874128 $156,339 $165,049 $8,710130 $229,632 $190,208 $-39,424132 $75,587 $89,282 $13,695136 $88,244 $105,867 $17,623140 $162,037 $153,906 $-8,131142 $1626,79 $208,024 $45,345144.01 $123,750 $103,646 $-20,104144.02 $88,968 $126,442 $37,474148.01 $145,972 $154,306 $8334148.02 $110,811 $124,250 $13,439150.01 $144,492 $149,861 $5,369150.02 $183,846 $205,145 $21,299152 $74,741 $85,428 $10,687158.01 $131,360 $152,784 $21,424158.02 $89,494 $96,471 $6,977160.01 $228,750 . .160.02 $98,669 $118,317 $19,648162 $25,214 $32,594 $7,380166 $29,987 $40,593 $10,606

Treatment

168 $24,286 $24,777 $491Control Area Avg. $109,297.52 $120,263.98 $10,966.46110 $99,238 $114597 $15,359118 $120,030 $122,537 $2,507126 $94,591 $133,878 $39,287134 $92,009 $96,591 $4,582138 $92,587 $132,087 $39,500146.01 $87,946 $111,853 $23,907146.02 $114,464 $95,227 $-19,237154 $79,601 $93,078 $13,477156.01 $89,518 $89,057 $-461156.02 $34,770 $56,458 $21,688164 $29,745 $27,116 $-2,629Treatment Area Avg. $92,552.147 $105,606.26 $13,054.116

52




Control

106.02 $200,389 $235,752 $35,363114.01 $403,011 $554,821 $151,810114.02 $385,677 $388,639 $2962116 $150,240 $173,489 $23,249120 $375,938 $445,077 $69,139122 $453,157 $532,812 $79,655124 $150,424 $248,138 $97,714128 $420,372 $611,934 $191,562130 $760,805 $686,721 $−74,084132 $147,805 $150,040 $2235136 $190,496 $191,186 $690140 $473,816 $489,787 $15,971142 $512,067 $793,185 $281,118144.01 $200,458 $308,379 $107,921144.02 $185,636 $218,412 $32,776148.01 $352,044 $440,424 $88,380148.02 $200,184 $337,513 $137,329150.01 $528,564 $369,078 $−159,486150.02 $503,195 $581,481 $78,286152 $154,031 $163,981 $9950158.01 $332,851 $297,559 $−35,292158.02 $130,655 $200,022 $69,367160.01 $518,450160.02 $321,167 $462,351 $141,184162 $47,220 $64,760 $17,540166 $61,950 $91,438 $29,488

Treatment

168 $48,889 $73,670 $Control Area Avg. $304,055 $350,410 $46,355110 $162,244 $229,493 $67,249118 $246,962 $316,079 $69,117126 $168,952 $238,490 $69,538134 $150,427 $246,309 $95,882138 $150,473 $200,153 $49,680146.01 $150,107 $200,351 $50,244146.02 $158,402 $150,465 $−7937154 $146,354 $150,246 $3892156.01 $116,341 $150,418 $34,077156.02 $112,119 $182,685 $70,566164 $98,512 $73,666 $−24,846Treatment Area Avg. $150,990 $194,396 $43,406

53




Control

106.02 $505,536 $508,023 $2487114.01 $826,692 $946,255 $119,563114.02 $742,514 $736,120 $−6394116 $301,545 $447,893 $146,348120 $750,504 $839,698 $89,194122 $798,059 $1,018,246 $220,187124 $424,982 $589,930 $164,948128 $771,388 $1,047,453 $276,065130 $1,286,260 $1,160,878 $−125,382132 $328,039 $365,850 $37,811136 $524,801 $523,756 $−1045140 $846,595 $913,096 $66,501142 $937,199 $1,360,642 $423,443144.01 $486,461 $690,941 $204,480144.02 $542,455 $639,651 $97,196148.01 $650,832 $824,339 $173,507148.02 $519,993 $722,285 $202,292150.01 $1,106,725 $702,760 $−403,965150.02 $936,955 $976,880 $39,925152 $366,394 $412,110 $45,716158.01 $826,994 $624,404 $−202,590158.02 $305,465 $417,296 $111,831160.01 $907,023160.02 $711,320 $897,416 $186,096162 $72,656 $90,160 $17,504166 $102,494 $148,150 $45,656

Treatment

168 $148,700 $148,911 $Control Area Avg. $619,577 $682,813 $63,236110 $450,713 $544,466 $93,753118 $640,341 $729,262 $88,921126 $499,502 $519,635 $20,133134 $399,933 $589,230 $189,297138 $404,790 $434,331 $29,541146.01 $415,378 $501,004 $85,626146.02 $493,799 $419,259 $−74,540154 $307,073 $344,895 $37,822156.01 $176,960 $387,048 $210,088156.02 $244,263 $399,369 $155,106164 $148,780 $148,700 $−80Treatment Area Avg. $380,139 $456,109 $75,970

54


Table 20: 90th Percentile - 10th Percentile of Income by Census Tract


Control

106.02 $474,790 $468,777 $106,122114.01 $815,899 $922,021 $106,122114.02 $722,285 $708,422 $−13,863116 $274,415 $428,026 $153,611120 $717,704 $796,058 $78,354122 $773,744 $984,250 $210,506124 $399,186 $575,147 $175,961128 $733,925 $1,027,273 $293,348130 $1,248,506 $1,140,034 $−108,472132 $309,336 $346,515 $37,179136 $503,450 $504,836 $1386140 $810,832 $882,543 $71,711142 $887,466 $1,326,531 $439,065144.01 $452,089 $663,688 $211,599144.02 $505,916 $608,539 $102,623148.01 $626,984 $798,911 $171,927148.02 $491,460 $700,077 $208,617150.01 $1,067,644 $688,660 $−378,984150.02 $912,496 $940,543 $28,047152 $358,962 $404,970 $46,008158.01 $797,020 $580,543 $−216,477158.02 $291,129 $398,141 $107,012160.01 $872,065160.02 $700,878 $873,642 $172,764162 $67,804 $83,433 $15,629166 $92,893 $140,261 $47,368

Treatment

168 $145,296 $145,134 $−162Control Area Avg. $594,599 $659,114 $64,515110 $426,669 $527,083 $100,414118 $615,021 $706,861 $91,840126 $476,330 $493,544 $17,214134 $377,160 $573,271 $196,111138 $384,340 $409,464 $25,124146.01 $386,738 $481,504 $94,766146.02 $458,196 $388,746 $−69,450154 $289,072 $321,188 $32,116156.01 $161,576 $370,193 $208,617156.02 $238,625 $388,822 $150,197164 $144,333 $143,938 $−395Treatment Area Avg. $359,823 $436,783 $76,960

55




Control

106.02 $402,003 $373,080 $−28,923114.01 $720,074 $712,609 $−7465114.02 $616,536 $580,767 $−35,769116 $227,314 $363,855 $136,541120 $592,903 $652,406 $59,503122 $648,251 $861,419 $213,168124 $334,390 $464,464 $130,074128 $615,049 $882,404 $267,355130 $1,056,628 $970,670 $−85,958132 $252,452 $276,568 $24,116136 $436,557 $417,889 $−18,668140 $684,558 $759,190 $74,632142 $774,520 $1,152,618 $378,098144.01 $362,711 $587,295 $224,584144.02 $453,487 $513,209 $59,722148.01 $504,860 $670,033 $165,173148.02 $409,182 $598,035 $188,853150.01 $962,233 $552,899 $−409,334150.02 $753,109 $771,735 $18,626152 $291,653 $326,682 $35,029158.01 $695,634 $471,620 $−224,014158.02 $215,971 $320,825 $104,854160.01 $678,273160.02 $612,651 $779,099 $166,448162 $47,442 $57,566 $10,124166 $72,507 $107,557 $35,050

Treatment

168 $124,414 $124,134 $−280Control Area Avg. $501,680 $551,870 $50,190110 $351,475 $429,869 $78,394118 $520,311 $606,725 $86,414126 $404,911 $385,757 $−19,154134 $307,924 $492,639 $184,715138 $312,203 $302,244 $−9959146.01 $327,432 $389,151 $61,719146.02 $379,335 $324,032 $−55,303154 $227,472 $251,817 $24,345156.01 $87,442 $297,991 $210,549156.02 $209,493 $342,911 $133,418164 $119,035 $121,584 $2549Treatment Area Avg. $295,185 $358,611 $63,426

56




Control

106.02 $72,787 $95,697 $22,910114.01 $95,825 $209,412 $113,587114.02 $105,749 $127,655 $21,906116 $47,101 $64,171 $17,070120 $124,801 $143,652 $18,851122 $125,493 $122,831 $−2662124 $64,796 $110,683 $45,887128 $118,876 $144,869 $25,993130 $191,878 $169,364 $−22,514132 $56,884 $69,947 $13,063136 $66,893 $86,947 $20,054140 $126,274 $123,353 $−2921142 $112,946 $173,913 $60,967144.01 $89,378 $76,393 $−12,985144.02 $52,429 $95,330 $42,901148.01 $122,124 $128,878 $6754148.02 $82,278 $102,042 $19,764150.01 $105,411 $135,761 $30,350150.02 $159,387 $168,808 $9421152 $67,309 $78,288 $10,979158.01 $101,386 $108,923 $7537158.02 $75,158 $77,316 $2158160.01 $193,792160.02 $88,227 $94,543 $6316162 $20,362 $25,867 $5505166 $20,386 $32,704 $12,318

Treatment

168 $20,882 $21,000 $118Control Area Avg. $92,919 $107,244 $14,325110 $75,194 $97,214 $22,020118 $94,710 $100,136 $5426126 $71,419 $107,787 $36,368134 $69,236 $80,632 $11,396138 $72,137 $107,220 $35,083146.01 $59,306 $92,353 $33,047146.02 $78,861 $64,714 $−14,147154 $61,600 $69,371 $7771156.01 $74,134 $72,202 $−1932156.02 $29,132 $45,911 $16,779164 $25,298 $22,354 $−2944Treatment Area Avg. $64,638 $78,172 $13,534

57


Documents

Take the Q Train: Value Capture of Public Infrastructure Projects · 2019-10-21 · Take the Q Train: Value Capture of Public Infrastructure Projects Arpit Gupta y Constantine Kontokosta