Mansion Tax - Business Economics and Public Policy Department

Mansion Tax: The Effect of Transfer Taxes on
Residential Real Estate Market
1
Wojciech Kopczuk and David Munroe
Columbia University
Preliminary and incomplete
please contact authors for an updated draft
October 1, 2012
1
Wojciech Kopczuk: Columbia University, email address: wojciech.kopczuk@columbia.edu. David
Munroe: Columbia University, email address: djm2166@columbia.edu.
Abstract
Houses and apartments sold in New York state at prices above $1 million are subject to the socalled 1% ”mansion tax” imposed on the full value of the transaction. This policy generates a
discontinuity (a ”notch”) in the overall tax liability. We rely on this and other discontinuities in
New York and New Jersey (where our data spans the introduction of an analogous ”mansion” tax)
to analyze implications of transfer taxes in the real estate market. Our research design builds on
the recently popularized methodology exploiting ”kinks” and ”notches” to learn about behavioral
responses. Contrary to work on individual responses to taxation (such as to income taxation), the
real estate context brings to the forefront issues of incidence: the response to the tax is generically
a result of the matching process between individual buyers and sellers, with both parties affected
by the outcome.
We find that the incidence of this tax falls on sellers and may exceed 100% of the value of the
tax. Results based on different taxes with different statutory incidence are consistent with this
possibility. Analyzing administrative records of property sales in New York and New Jersey, we
provide evidence indicating that the results are robust across many different types of transactions.
We find that supply-side adjustments to quality of properties may play a role but find no evidence
that results are driven by tax evasion. Using data on real estate listings, we show that sellers
partially internalize the tax when they first post listings, and that this response grows stronger in
subsequent updates to the listings; all happening before bargaining with the ultimate buyers. We
also find evidence that taxation increases discounts from the final asking to sale price, implying
that the presence of the tax shifts bargaining power to the buyers. Finally, there is some evidence
that the tax increases the likelihood that potential sellers leave their real estate agents, suggesting
that incidence is partially borne by intermediaries.
We interpret our results as showing evidence of a market-level distortion. We find indications
that the tax leads to sorting of sellers by their motivation to sell. At the same time, we find that
the presence of bunching induced by the tax reduces the informational content of the price signal
available to buyers by weakening the relationship between listing and sale prices. We argue that
the presence of these market-level distortions can explain counter-intuitive incidence results.
1
Introduction
Purchasing a house or an apartment is a time-consuming and complicated process with large financial stakes, and any inefficiencies or distortions in this market are likely to induce significant
welfare costs. Beyond the price, a typical transaction involves a lot of associated costs including
broker’s fees, inspection costs, legal fees, title insurance, mortgage application and insurance fees,
and moving costs. While some of these costs are payments for services or insurance against various
risks, they all introduce a wedge between the price received by a seller and the acquisition costs paid
by the buyer. This is clearest in the case of transfer taxes that are present in many jurisdictions,
and are imposed on the final sale price. In this paper, we ask how transfer taxes affect functioning
of the real estate market.
Our empirical approach in this paper is to analyze implications of transfer taxes on the real
estate market using variation generated by the discontinuous nature of the taxes imposed in New
York and New Jersey. In a nutshell, these types of taxes are levied as a function of the appropriately
defined purchase price. The so-called “mansion tax” in New York state (since 1989) and New Jersey
(since 2004) applies to residential transactions of $1 million or more. The tax rate is 1% and is
imposed on the full value of the transaction so that a $1 million sale is subject to $10,000 tax
liability, while a $999,999 transaction is not subject to the tax at all. Additionally, in New York
City all real estate transactions are subject to the real property transfer tax (RPTT). The average
rate of this tax changes discontinuously from 1% to 1.425% at the $500,000 mark. Figures 1 and 4
illustrate that the tax affects some aspects of the market: Figure 1 shows that sales in New York
bunch just under $1 million. A similar pattern appears in New Jersey around the time of the
introduction of the tax—Figure 4 demonstrates the onset of bunching stimulated by the tax
Our analysis of the consequences of transfer taxation in New York and New Jersey, allows us to
offer three sets of findings. First, we show that transfer taxes are responsible for distortions to the
distribution of reported sales price. We show that the distortion coincides with the introduction
of the tax and that its presence is very robust across different sources of variation, data sources,
subgroups and methods.
Second, we rely on the heterogeneity in estimates of the size of response to provide evidence that
these effects represent real responses rather than tax evasion. In particular, we find weaker price
responses when there is less room for quality adjustment (such as sales of finished apartments)
and stronger when such adjustments are easier (such as sales occurring before construction is
finalized). On the other hand, we find no substantial difference between cash and mortgagefinanced transactions and no stronger response for transactions between related parties. We also
find moderate effects on time on the market and reduced reliance on real estate agents.
Finally, we show that the tax affects functioning of the market throughout the search process.
We analyze real estate listings data that allow us to trace at which stage of the selling process
responses take place. We conclude that the original asking price (posted before buyers respond) is
already distorted and that the distortion increases with subsequent revisions of the asking price.
We also show that this process corresponds to the ultimate likelihood of sale, revealing sorting
1
of sellers (arguably according to their intrinsic determination to sell). However, while the final
asking price reveals that sellers internalize the tax, it still does not reflect the final incidence: we
find a significant discontinuity at the tax threshold in the extent to which prices are discounted
during bargaining between buyers and sellers . We conclude that the incidence effects unfold in
two different ways. First, the presence of the tax is internalized by sellers who are not directly
responsible for remittance. Second, the presence of the tax appears to increase the bargaining
power of buyers during negotiations with the sellers. The last, and important, piece of the puzzle
that we document concerns informativeness of listings. We show that in the presence of the tax, the
relationship of listing and sale prices is weaker, indicating that the quality of information available
to buyers declines and that the tax is responsible for aggregate (in addition to transaction-specific)
distortion to the efficiency of this market.
The price effects that we find indicate substantial shifting of the tax to the seller. Indeed, our
central estimates indicate that over 200% of the tax is shifted, although a more conservative estimate
that takes into account the possible supply response pushes this number closer to 100%. A number
of other findings are consistent with substantial incidence falling on sellers. As mentioned before,
listings information supports the notion that sellers internalize the tax. Furthermore, our data
covers multiple jurisdictions and different tax instruments with variation in statutory incidence:
we find that in each case responses are consistent with the burden of the tax falling on the supply
side. We note that beyond the usual interplay between tax-sensitivity of the demand and supply
side of this market, the reduced informational context of the listings following the introduction of
the tax is likely to further shift the price response to the supply side.
From the point of view of existing literature, the context we are analyzing is interesting in
several ways. First, the tax is discontinuous, allowing us to identify its effects using a “notch”
design. Second, relative to much of the public finance literature, the bilateral nature of this market
puts at the forefront the equilibrium outcome of the market process—each transaction takes place
between two parties that both may respond to the tax. It is, to our knowledge, the first analysis
of tax distortions in a matching market of this sort and one of the very few emprical papers that
recognizes the incidence effects of taxation in a market with heterogeneous goods (Rothstein, 2010;
Chetty et al., 2011, allow for supply-side responses in labor market context) . Third, the housing
market is interesting in its own right and we uncover new facts about how it operates and the
ability of policy to distort (or influence!) it.
We use discontinuities in the application of the tax to evaluate its economic effects and distinguish between different mechanisms and types of responses that might be present. These taxes
create price discontinuities (these are “tax notches,” see Slemrod, 2010), while the introduction
of the tax in New Jersey creates a time discontinuity.1 Furthermore, the statutory incidence is
different for the mansion tax (which is the statutory responsibility ofthe buyer) than in the case of
the New Jersey realty transfer fee schedule (responsibility of sellers) and RPTT tax (which is the
1
There are also geographic discontinuities that we do not exploit: the RPTT changes discontinuously at the New
York City border and before 2004 the mansion tax changed discontinuously at the New Jersey-New York border.
2
responsibility of a seller, except for new constructions). The volume of transactions is large, allowing us to exploit these sources of variation without strong parametric restrictions. We incorporate
a number of data sources in the analysis, including price and timing of transactions, characteristics
of properties and information from real estate listings.
A small literature focuses on the effect of transfer taxes on the functioning of the real estate
market (Benjamin et al., 1993; Van Ommeren and Van Leuvensteijn, 2005; Dachis et al., 2012). A
more active literature, analyzes the role of real estate agents, attempting to unbundle the effect
of cost from information provision (Levitt and Syverson, 2008; Jia and Pathak, 2010; Bernheim
and Meer, 2012). Finally, a number of (mostly recent) empirical papers incorporate information
available in listings data (Genesove and Mayer, 2001; Haurin et al., 2010; Han and Strange, 2012;
Carrillo and Pope, 2012).
Our paper is also closely related to work on behavioral responses to taxation. As in the work on
responses to income taxation, we are interested in separating real, timing, avoidance and evasion
responses (Slemrod, 1990; Saez et al., 2012). Contrary to that strand of work, however, our context
requires considering the impact on both sides of the market. In recent years, there has been a revival
of work on estimating incidence of specific taxes/transfers (e.g., Doyle and Samphantharak, 2008;
Mishra et al., 2008; Hastings and Washington, 2010; Marion and Muehlegger, 2010). The context of
real estate tax is more complicated because of non-homogeneity of goods, and the closest analogue
to it in the tax literature is work on incidence of income/payroll taxes or credits (Rothstein, 2010;
Saez et al., 2011).2 While a recent but somewhat more established strand of work on behavioral
responses takes advantage of the presence of discontinuities (kinks) in marginal tax rates (Saez,
2010; Chetty et al., 2011), our research is one of the first studies that take advantage of the presence
of tax “notches” — discontinuities in the total tax liability (Slemrod (1987); Sallee and Slemrod
(2012); Kleven and Wassem (2012)).
The structure of the paper is as follows...
2
Policy
Although real estate transfer taxes are common across the U.S., the tax rates in New York and
New Jersey are notable for being discontinuous. These taxes are applied to the sale price of real
property at the time of transaction, and range from as low as no tax in Texas to 0.01% of total
consideration in Colorado to two percent in Deleware. In New York and New Jersey, however, the
tax rates change discontinuously with total consideration, creating corresponding discontinuities in
total tax liability (tax “notches”). Table 1 contains details of the relevant tax schedules.
One notch arising in both states is due to the so-called “mansion tax:” a one percent tax on
the total consideration of homes costing one million dollars or more. Under the mansion taxes of
2
There are also important differences between the two contexts. The occurrence of a real estate transaction is
endogenous to the presence of transfer taxes, whereas the possibility of taxpayers dropping off tax roles is usually assumed away in work on behavioral responses to personal income taxation (Erard and Ho, 2001 is a notable
exception).
3
both New York and New Jersey, buyers’ total tax liability jumps by ten thousand dollars when the
sales price moves from $999,999 (where the tax does not apply) to one million dollars (where the
tax comes into effect). In New Jersey, the mansion tax was introduced on August 1st, 2004, and
covers all residential real-estate transactions in the state. In New York state, the mansion tax was
introduced in 1989 and applies to the sale of individual coop and condo units, and one-, two, and
three-family homes, with few exceptions.3
Real estate sales in New York City and New Jersey are subject to additional taxes with disconinuous average rates. In New York City, the Real Property Transfer Tax (RPTT) applies to
residential sales (as defined for the New York mansion tax) with a rate of one percent if the total
consideration is $500,000 or less, and 1.425% above $500,000. Commercial sales are also subject
to the RPTT at a rate of 1.425% below $500,000 and 2.625% above. Unlike the mansion tax, the
statutory incidence of the RPTT falls on the seller by law, however it is customary for the buyer
to pay the tax when purchasing directly from a sponsor (i.e. purchasing a newly developed condo
or a newly offered coop). Thus, the RPTT is fairly unique tax in that there is variation in the
statutory incidence of the tax.
Residential sales in New Jersey are subject to the Realty Transfer Fee. This transfer fee (or tax)
has a non-linear schedule (see Table 1) that shifts when total consideration is greater than $350,000.
For example, the marginal tax rate for consideration above $200,000 is 0.78% if the total price is
less than or equal $350,000, while this tax rate jumps to 0.96% when the total price is greater than
$350,000. Moving from a price of $350,000 to $350,001 increases the buyer’s tax liability by $630.4
These four real estate transfer taxes offer several sources of variation. Firstly, and most strikingly, these taxes create discontinuities in total tax liability. Moreover, there is variation in the
magnitude of these jumps; increasing the sale price by one dollar can increase the tax paid by
between $628.80 (a change in tax liability at $350,000 for the NJ Realty Transfer Fee) and ten
thousand dollars (one percent change for the mansion taxes). Secondly, the NJ mansion tax varies
over time.5 Thirdly, there is within-state geographic variation in the RPTT—only sales in New
York City (and not the rest of New York state) are subject to this tax. Finally, we have variation in
the statutory incidence of these taxes—the New Jersey RTF and RPTT are remitted by the seller
(for RPTT, this is so except in the case of a sale of newly constructed homes, where it is the buyer
who remits the tax). For the mansion taxes and Realty Transfer Fee, the buyer always remits the
tax. In this paper we focus primarily on buyers’ and sellers’ response to the tax notches associated
with these transfer taxes, which we bolster using the temporal variation in the NJ mansion tax,
and the variation in the statutory incidence of these taxes.
3
Exceptions are as follows. If a residential unit is partially used for commerce, only the residential share of the
total consideration is subject to the tax (although the entire consideration is still used to determine if the tax applies).
Similarly, multiple parcels sold in the same transaction are taxed as one unit unless the parcels are evidently not used
in conjunction with one another. Vacant lots are exempt from the mansion tax, and, finally, any personal effects sold
with the home are deducted from the total consideration for tax purposes (but are subject to state sales taxes).
4
On top of the notch at $350,000 the New Jersey schedule also features small kinks at $550,000 and $800,000. In
this paper, we limit attention to analyzing implications of notches in the transfer tax schedules.
5
Although the other taxes also vary over time, our data does not go back far enough to study this.
4
3
Data
We study both administrative records on real estate transactions in New York state and New Jersey
as well as historical real estate listings in New York City. Sales records, which cover the universe
of recorded real estate transactions in the given geography and time period, come primarily from
three sources: the New York City Department of Finance’s (NYCDOF) Annualized Rolling Sales
files for 2003–2011, real property transfer reports compiled by the New York state Office of Real
Property Services (NYSORPS) for 2002–2006 and 2008–2010, and SR1A property transfer forms
collected in the New Jersey Treasury’s SR1A file. Additionally, we make use of deeds records for
1996–2008 for the five counties of New York City collected by an anonymous private data provider
from the county records.6 We match the Rolling Sales data for Manhattan to a subset of historical
real estate listings in order to get a broader picture of the effect of the tax on sellers’ pre-sale
behavior. Our listing data comes from the Real Estate Board of New York’s (REBNY) electronic
listing service and covers all closed or off-market listings between 2003 and 2010. REBNY is a trade
association of about 300 realty firms operating in New York City and representing a substantial
fraction of listings in Manhattan and Brooklyn. We limit attention to, more complete, Manhattan
listings. REBNY data accounts for approximately 45% of Manhattan sales in the Rolling Sales
files.
The administrative sales records for New York and New Jersey contain details of each recorded
real estate transaction. The NYSORPS records cover all counties in New York state excepting the
five counties of New York City—these sales are covered by the NYCDOF data (and the privately
collected deeds records). Data from the New Jersey Treasury covers sales in all counties of the
state (see the Data Appendix for more details on the data sources). The sales records indicate the
date of each sale and the total consideration paid from the buyer to the seller. Information about
the property being sold includes the the address of the property, the type of property (ex. one-,
two-, or three-family home, residential coop or condo, commercial property, etc.), and the year of
construction of the building (in New York City and New Jersey) or whether the property is newly
constructed (in New York state). Additionally, the NYSORPS data indicate whether or not the
sale is arms-length7 and the privately collected deeds records indicate whether there is a mortgage
associated with the purchase.
For each geography we identify sales that are likely to be subject to each of the transfer taxes.
In New Jersey, we consider all “residential” sales to be taxable (mansion tax and Transfer Fee).
For New York state, we define all single-parcel residential sales of one-, two-, or three-family homes
(including residential condos and seasonal properties) as being subject to the mansion tax. Finally,
6
We are grateful to Chris Mayer and the Paul Milstein Center for Real Estate for granting us access to the New
York City deeds records.
7
NYSORPS defines an arms-length sale as “a sale of real property in the open market, between an informed and
willing buyer and seller where neither is under any compulsion to participate in the transaction, unaffected by any
unusual conditions indicating a reasonable possibility that the full sales price is not equal to the fair market value
of the property assuming fee ownership.” This definition excludes sales between current or former relatives,
between related companies or partners in business, sales where one of the buyers is also a seller, or sales with
“other unusual factors affecting sale price.”
5
in New York City we define all single-unit (non-commercial) sales of one-, two-, or three-family
homes, coops, and condos as subject to the mansion tax.8 In New York City we define commercial
sales as any sale of at least one commercial unit (and no residential units) or a tax class of 3
(utilities) or 4 (commercial or industrial). We identify years since construction as the difference
between the year of the transaction and the year that the property was built Year built is not
indicated in the New York state data, however the data do flag sales of newly constructed homes.
From the REBNY listing service, we observe details of all (REBNY-listed) closed or off market
listings since 2003 (when electronic records are first available). As part of REBNY membership,
realtors are required to post all listings and updates to the REBNY listing service within 24 hours of
putting the home on the market. These data include initial asking price and date of each listing, all
subsequent price updates (and corresponding dates), and whether the property sells or is taken off
the market (and the date of this event). To acquire the final sale price, we match these listings by
precise address (including apartment number) and/or tax lot to the NYCDOF data for Manhattan.
We obtain approximately 90% match rate for listings identified as “closed” in the REBNY data.9
We have 23% match rate for units that were not indicated as closed in the REBNY database and
we interpret these matches as corresponding to either direct sales by owner or with a non-REBNY
agent.
Table 2 presents descriptive statistics from the three sources of housing sales data . Overall,
we have records for 5,705,834 sales (with non-zero sale price, omitting duplicates) spanning 2000–
2011. In all three regions, sales subject to the mansion tax form the majority of our sample and,
on average, sell for a lower price than non-residential sales (except in NY state). Mean price is
generally much higher than the median due to large upper tales in the housing-price distribution.
Unsurprisingly, prices are highest in New York City. Although median (and mean) sale price is
well below the one-million dollar threshold of the mansion tax, the observation counts in the final
column in Table 2 demonstrate that there are still several thousand sales per geography within
$50,000 of the mansion-tax cutoff.
Table 3 presents the number of sales subject to the mansion tax and median prices over time
for the three regions. The growth of housing sales and prices throughout the early 2000s is evident
here, as us the subsequent drop in total sales and median price at the onset of the recession in
2007/2008.
Table 4 presents statistics for the matched REBNY listings data. The sales covered in the
REBNY data have much higher prices, on average, than the general NYC rolling sales data ($1.24
milion versus $658,000) . This is likely a product of the REBNY data only covering sales in
8
While multi-parcel sales in New York state are typically subject to the mansion tax, such a sale may be split
for tax purposes if structures on adjacent parcels are not used in conjunction with or clearly related to one-another.
Since we cannot identify such cases in the New York City and New York state data, we err on the side of caution
and exclude all multi-parcel and partially-commercial sales sales from taxable status.
9
Non-matches fall in a number of categories. Sales in some co-op buildings are missing from the DOF data
(explain). Some transactions contain only street address or non-standard way of specifying apartment number (in
particular, commercial units and unusual properties such as storage units fall in this category). Occasionally, the
same building may have two different street addresses and a unit may be listed differently in the two databases.
6
Manhattan, which is considerably more expensive than the outer boroughs. About 64% of the
homes in our sample end up closing (rather than being taken off the market). At the same time, 8%
of listings do not close, but have a corresponding sale in the rolling sales data, suggesting that these
individuals are selling with non-REBNY realtors or with no realtor at all. Unsurprisingly, homes
that do not close spend longer on the market (200.5 days on average vs. 146 for sold properties).
For properties that close and are matched to the rolling sales data, the average discount between
− sale
the initial asking price and the sale price ( initial
) is about 6%. This discount breaks-down into
initial
a 2% discount from intial asking to final asking price (defined as the price listed immediately prior
to posting a listing status of “in contract”) and a 3% discount from final asking price to sale price.
Note, however, that the median listing in our sample has no price updates between the initial and
final asking prices.
4
Estimates of the effect of the tax
Consider a property that would sell at a price of p absent taxation. When the tax is present, the
property may not sell at all or it may sell at some price p̃. In the population, denote by F (p) the
number of sales at prices lower than p when taxation is not present and by FT (p) the (observed)
number of sales when the tax is present. Because we focus on notches in tax liability, it is possible
that FT (and in practice perhaps also F ) is not differentiable and even not continuous (ie. there
may be mass points, “bunching”).
What effect of taxation should one expect? Consider first the situation when there are no behavioral adjustments (including tax evasion). Assuming that the sale did occur, the effect (presumably
a drop) in the pre-tax price p − p̃ is reflecting the loss in seller’s surplus—the portion of the tax
incidence that is falling on sellers. In the context of the mansion tax, which is our main focus, the
statutory incidence falls on buyers so that the price paid by the buyer is given by p̃ + t(p̃), where
t is the tax liability applying to the final price. In the standard incidence case of a homogeneous
good, p̃ + t(p̃) − p represents the share of the tax born by the buyers.
However, it is natural to think of the housing market as a market where goods are heterogeneous
and changing incentives affect matching between buyers and sellers. Thus, there are two things
that differ here as compared to the standard case. First, the identity of the buyer can change.
Second, the new buyer may have both a different valuation of the house than the original buyer
and a different surplus relative to the house that she would have bought in the absence of the tax.
Hence, p̃ + t(p̃) − p need not represent the change in surplus for any particular buyer. The seller’s
burden, however, p − p̃, is free from this complication and we will focus on the difference in sale
prices in what follows.
4.1
Graphical evidence of the presence of a response
Response to the tax notch is evident in Figure 1, which shows the empirical pattern of taxable sales
in New York. Here, data is grouped into $5,000 bins (top panel) and, to make the figure less noisy,
7
$25,000 bins (bottom panel). There is clear bunching in the sale price just below $1 million and,
arguably, a drop in the volume of sales just above $1 million. Figure 2 shows analogous patterns
at a smaller (0.425% — hence, we are showing only smaller $5000 bins) RPTT notch at $500,000
and also shows some evidence of a response.
Notably, there is also significant bunching at other round price levels (at every $50,000 and, to
a lesser extent, remaining $25,000 multiples).Unlike the bunching at $1 million, this round-number
bunching occurs in the bin above rather than below. It may be that although this observed bunching
below $1 million is consistent with theoretical predictions, it may reflect adjustments to the tax
by very small amounts. Aggregating the data to larger bins in the lower panel of Figure 1 mostly
eliminates such round-number bunching while continuing to indicate that the response covers more
than just the immediate neighborhood of the threshold.
While the amount of bunching at $1 million appears out of line with that at other round
numbers, suggesting that sales are shifting from higher price bins, the one caveat here is that one
might think that $1 million price is more salient than $900,000 or $1.1 million (even in absence of
the tax). However, data for New Jersey spans the reform allowing to verify this concern explicitly.
Figure 3 shows the distribution of sales in New Jersey before the mansion tax was introduced. There
is no indication there that there is $1 million is more salient than other multiples of $50,000. Since
our data spans the introduction of the mansion tax in New Jersey, we can also verify explicitly that
the tax induces bunching by comparing sales in New Jersey before and after the introduction of
the tax. As figure 4 illustrates, the number of sales just below the threshold clearly increases at the
time of the introduction of the tax. Correspondingly, the number of sales above the threshold falls.
Focusing on the region within $10,000 of the threshold, it is evident that the increase in the mass
below the threshold is larger than the shift from just above $1million, providing the first indication
that the apparent effect of the tax may extend beyond 100% of its value ($10,000). Figure 5 shows
monthly distributions of sales in New Jersey, focusing on the $900,000 to $1,100,000 range. There
is no evidence of the distribution being distorted below $975,000, and a clear evidence of a shift of
the mass of sales to just under $1 million from as far above as the $1,025,000 to $1,050,000 range.
Figures 4 and 5 show patterns that may be consistent with anticipation effects — there is an
increase in over-$1 million sales just before the introduction of the tax. This is not surprising: the
tax had been announced and the lengthy process of closing a real estate transaction may allow for
the possibility to speed up the timing of final sale. In Figure 6 we eliminate the timing response by
focusing on sales that occurred at least three months before or three months after the tax. We focus
on the following two years and adjust the period before the tax to obtain the same number of sales
in the $500,000 to $1,500,000 range. The figure plots the difference in log sales, by $10,000 bins,
before and after the tax. By construction, the difference should be centered around zero (modulo
changes in the shape of the distribution and the impact of log transformation, both of which are
expected to be minor). Indeed, it is clear that the difference is about zero away from the threshold
on either side. The evidence of distortion around the $1,000,000 threshold is clearly visible and the
approach allows for more precisely understanding how far the effect on price distribution extends:
8
for the following $80,000 sales after the introduction of the tax appear depressed.
Graphs for both New York and New Jersey indicate clearly the distortion to the shape of the
distribution of prices around the mansion tax threshold and provide evidence that the effect takes
place as the result of introduction of the tax. We note a number of salient aspects of the response.
First, bunching occurs primarily just below the threshold, although it may extend beyond $10,000
or so. Second, the tax is also associated with hollowing out of the distribution above the threshold.
It corresponds to a precipitous drop in sales immediately above the threshold but there is evidence
that the distribution of prices is significantly depressed in an extended range perhaps in excess of
$50,000, far exceeding the tax liability which is of the order of $10,000.
These graphs show features that are inconsistent with the simplest incidence model. Assuming incidence of α and simplifying for the purpose of illustration the tax to be a lump sum of
T = $10, 000, the transfer tax might be expected to reduce sales prices from p to max{p − αT ·
p, $1, 000, 000}. A uniform response would imply that sales from up to αT above the threshold
would shift to the threshold itself, while the distribution at any higher level would shift by αT . If
that was the case, the distribution of prices should not show any significant hollowing out above the
threshold. It is also pretty suggestive from Figures 1 and 4 that the mass at the threshold cannot
be accounted for by sales that would have otherwise occurred within $10,000 of the threshold so
that the implied incidence would have to exceed 100%. Heterogeneity in response might generate
appearance of “hollowing out” of the distribution but would require incidence exceeding 100% for
a substantial number of individuals to match the quantitative effects.
4.2
Quantifying the magnitude of response
We will proceed as follows. First, we will provide a simple framework for understanding and
econometrically quantifying the magnitude of the response. We will then use this framework to
analyze heterogeneity of responses that we will use to discriminate between various possibilities.
Finally, we will incorporate listing information to understand the importance of market and search
distortions.
The basic ideas are illustrated in the top left panel of Figure 7. Absent the tax (and ignoring any
other frictions), prices of seller and buyer are equal to each other so that the possible contract space
is characterized by P B − P S . The presence of the tax distorts the contract space by introducing a
wedge between the two prices above the threshold level so that the contract space is characterized
instead by P B − P S = t · P S · I(P S > H), where t is the tax rate and H is the threshold. Consider
a particular buyer and a particular seller who engage in the transaction absent the tax. We assume
that the bargaining process maximizes a function of their respective surpluses f (P S − C S − G, G −
P B −C B ) where C S and C B are their respective costs,G is the value of the object and f is increasing
and (weakly) convex. Suppose that the same two parties contemplate the transaction when the
tax is present. Under natural assumptions, when the original price is either high or low enough,
the threshold is irrelevant. When the original price is in the neighborhood of the threshold, the
final price of the transaction may either end up at the threshold or become taxable. As long as
9
substitution between the two prices is possible, there is a range of prices that will be dominated
and, if the notch was not present, the marginal transaction at the threshold would otherwise have
occurred at the price strictly higher.
Alternatively, at any transaction price that involves the tax, P S > H, there is a gain in surplus
for the buyer from transacting at the threshold of P S − H + t · P S and the corresponding loss for
the seller of P S − H. This creates a number of possibilities. If transfers are possible, there is room
here for Pareto improving trade. Transfers need not to be one-for-one — an increase in seller’s
surplus by I at the corresponding cost to the buyer of βI, where β > 1 would be Pareto improving
as long as I ≥ P S − H and βI ≤ P S − H + t · P S . Side-payments (tax evasion) are one example.
Reduced investment by the seller (foregoing renovations, removing appliances etc.) is another.
Another possible channel of response involves changes in the outcome of the bargaining process.
As long as the seller is left with positive surplus absent the tax, there is a possibility that the
presence of frictions might leave her with lower surplus. Function f (·) may be interpreted as a
reduced form representing the outcome of the bargaining process between the two parties under
the assumption that the outcomes have to be Pareto efficient (surplus of of each party is maximized
given the choice of the other one).
None of these possibilities explicitly recognizes that the identities of the buyers and sellers may
change or that a transaction may not take place — we will return to these issues below. Still, this
framework provides a useful way of quantifying how large the response is and how it varies across
contexts. Figure 7 illustrates two measures of the effect of the tax. Both of them rely on considering
a marginal transaction that shifts to the threshold. The incidence measure corresponds to the
price (pS ) for the marginal transaction absent the tax: the actual price of this marginal transaction
changes by pS − H and, if this effect represented the full drop in seller’s surplus (absent reduced
investment and/or evasion), it would represent the cost (incidence) to the seller. The gap measure
relies on the price that would have occurred in the presence of the tax but absent the threshold.
For the marginal transaction there are two equilibria — selling at the threshold or at the gap level
— while transactions below/above the marginal end up on either side. The gap quantifies the size
of the missing region and reflects the extent of substitution between buyer’s and seller’s surplus.
We econometrically estimate the extent of the bunching response and corresponding incidence
as follows. We specify a parametric distribution of prices absent the tax as follows:
ln h(p) = g(p) + αD(p)
(1)
and the distribution in the presence of the tax as
ln hT (p) = ln h(p) + βI(p > T )
(2)
where the left-hand side is the log of the probability distribution function at price p, g() is a
parametric function (a polynomial) and D is the set of indicators for round numbers and potentially
an indicator for the price being above the threshold. We estimate this model on the data that
10
excludes some region around the threshold (T − L, T + H), as illustrated in the upper right panel
of Figure 7.We use the predicted values from this regression to obtain a counterfactual estimate of
sales in the omitted region. Given the observed distribution of prices outside of the omitted region,
we estimate equation 1 by maximum likelihood. This procedure yields h̃(p), our estimate of the
empirical distribution function of prices.
Given the estimate of the counterfactual distribution of prices, we can estimate bunching at the
threshold and the implied reduction in prices for the average sale at the threshold. An estimate of
bunching can be found by taking the difference between actual and predicted sales in the omitted
region to the left of the threshold, (T − L, T ):
ˆT
B = FT (T ) − FT (T − L) −
h(p)
T −L
As in other papers that rely on bunching and notches, we use the estimated distribution function
h(p) as the counterfactual for the distribution in the omitted region (T, T + H). Given this counterfactual distribution and the estimate of the excess mass, we estimate where in the distribution
taxpayers at the threshold originated from by solving:
ˆ
I
h(p)dp = B
(3)
T
for I. We will refer to I − T as an estimate of the incidence of the tax. This is illustrated in the
lower left panel of Figure 7: the blue region is the bunching measure and the green area extends
to match the bunching.
In the absence of evasion, this is a lower bound for the estimate of the incidence response. If
the response does not involve re-ranking of prices, individuals at the bunching point will come from
the segment of the distribution just above the threshold and the equation 3 defines the border
of that region. If some transactions do not take place as the result of the tax, this procedure
will an underestimate the true effect, because it mistakenly assumes missing sales end up at the
threshold. The response need not be uniform, in which case we obtain an average for the responders
(elaborate).
The gap estimates would be straightforward to obtain if there were no sales above the threshold.
As we discussed before, empirically sales appear depressed in a significant range above the threshold
but, of course, they are not zero. Hence, instead we proceed by constructing a counterfactual based
on the post tax distribution h̃(p) outside of the missing region and extending it into it. We estimate
the magnitude of the gap by quantifying the number of missing sales relative to the counterfactual
and mapping them into a contiguous region above the threshold, as illustrated in the lower right
panel of Figure 7.
Estimates of gap and incidence are straightforward calculations based on the estimated shape
of the distribution. Our assumed density of the distribution is specified by formulae 1 and 2. We
specify the parametric form by assuming that g(p) is a polynomial (third degree in the baseline,
11
sensitivity checks). For the round-number effects D(p) we proceed as follows. We consider separate
effects for $25,000 and $50,000 multiples. The basic idea is to include the dummies for round
numbers. Because the extent of bunching at round numbers may vary depending on the price
(perhaps $1.2m is not equally salient as $600,000?), we also allow for the interaction of dummies
with the price. Since our objective is to estimate the counterfactual in the omitted region (in
particular, at $1 million), this app(proach amounts to assuming that bunching at other round
prices is a valid counterfactual for the magnitude of bunching in that region — this is not a directly
testable assumption but, as discussed before, data in New Jersey before the introduction of the
mansion tax (Figure 3) provides support for this assumption.
Our actual implementation of the round-number bunching is more involved. Our baseline
approach is to rely on the maximum likelihood estimation and hence specifying the density at any
point. In order to parsimoniously capture various forms of bunching, we introduce “bunching”
regions that for each round number R extend from R − b to R. Within the bunching regions, the
distribution is specified as g(p) + DR + DR · p where DR are the relevant round-number dummies,
while it is g(p) otherwise. In practice, we take b = $1000 — this allows the bunching region to
extend from, for example, $899,000 to $900,000 allowing both for bunching at the $900K level
and just below it (which is of relevance when we turn to listings data). Except for slight gain in
information due to allowing for continuous prices, this approach is very close to binning the data
in $1000 bins.
For completeness, we also estimate analogous specification by OLS — this is standard in the
recent public finance work on notches and kinks —- but we note that any of these approaches
involves specifying the parametric density function and the maximum likelihood estimation is a
natural choice that guarantees that the estimates satisfy law of probability rather than hard-tointerpret mean zero residual restriction. Additionally, by requiring to bin the data, OLS amounts
to throwing out information.
All reported standard errors are obtained by bootstrapping the whole procedure 999 times.
Note that the estimates of incidence and gap may fall into one of the bunching region. When this is
the case, the estimates are not very sensitive — small changes in parameters correspond to staying
in the bunching region. In such cases reported standard errors for estimates of gap and incidence
are small, even though standard errors for the parameters of the parametric density are not.10
4.3
Incidence estimates
Table 5 contains our baseline results for New York City (other jurisdictions follow, we emphasize
NYC because that is where we have listings data to which we will turn next). Our estimate of the
incidence parameter in the baseline specification is $1,020,137: bunching at the threshold is equivalent to all transactions over the following $20,000 shifting to the threshold itself. Taken literally
as an incidence estimate, it corresponds to over 200% incidence of the tax for the marginal transac10
The distribution of incidence and gap may be asymmetric because they may correspond to the bunching regions.
In the next version we will report confidence intervals instead.
12
tion. The corresponding gap estimate is of similar magnitude, indicating that in the presence of the
$10,000 tax, hollowing out of the distribution corresponds to a region equivalent to approximately
$24,000 above the threshold. These estimates are consistent with the graphical evidence presented
below. We will first discuss their robustness and then work on understanding their interpretation.
The baseline estimates were obtained using third-order polynomial and omitting data over the
$990,000-$1,100,000 region. The following five rows present estimates obtained using different order
of polynomials — the results are similar but the inspection of the fit of the data suggests that loworder polynomials cannot capture salient feature of the distribution while high-order polynomials
introduce unrealistic behavior in the omitted region (i.e. out of sample predictions are sensitive).
In the subsequent specifications, we test sensitivity to the choice of the omitted region. The
results (especially gap estimates) are somewhat sensitive to selecting a narrow omitted region —
this is not surprising given indications that the distribution may be distorted over the range of
$50,000 or more above the threshold and the omitted region should extend beyond the range of
distortions. The results are not too sensitive to expanding the omitted region below $990,000. Our
baseline specification relies on the data both below and above the omitted region. The latter is in
fact distorted by the presence of the tax and we rudimentarily control for it by allowing for a level
shift. In further robustness checks, we consider estimating the counterfactual using data below
the threshold only; alternatively we also consider restricting the functional form to be the same
on both side of the threshold. The results are not affected by any of these changes. Finally, we
report the OLS results obtained by binning into $5000 and $10,000 bins and conclude that they
are quantitatively similar.
In Table 6, we report results for other regions subject to the mansion tax, relying on the
alternative datasets. Estimates for New York State and New Jersey after the introduction of the
tax are remarkably similar — within $2,000 — to those for New York City. In contrast, estimates
for New Jersey before the introduction of the tax show no evidence of bunching.
We use information about vintage of property to investigate heterogeneity. We expect that a
property that negotiating a purchase of property before construction is finished allows for significant
response in terms of the level finish, appliances and other amenities; therefore allowing for reducing
the price by reducing the quality. Similarly, older properties may require renovation and hence
allow for some room for response to the tax. In contrast, original sales of apartments or houses
after they have already been constructed and finished may have less flexibility. Our data for New
York City and New Jersey contain information about year of construction. In particular, in New
York City dominated by large apartment buildings, there is a non-trivial number of sales that
occur before construction is finished. We do find that bunching is very large for early sales (before
construction is finished in NYC, the year of construction in NJ( and for sales that occur three or
more years after construction. In contrast, we find that sales that occur soon after construction
— presumably original sales of already fully constructed and equipped properties — show smaller
but still significant (exceeding $10,000 incidence estimate) bunching. We interpret these results
as evidence that supply-side response along the quality/finish dimension is important but cannot
13
account for the full response.
In Tables 7 and 8, we report estimates for other types of taxes. We do not find any evidence of
response to the small ($600) New Jersey threshold. For the $2125 RPTT notch that applies only
in New York City, we find a response for new but not old sales (and we do not find any response
in the rest of New York State where the tax does not apply). These mixed results are consistent
with small magnitude of the tax and its limited salience. However, they also coincide with shifts
in statutory incidence. The RPTT tax on new sales is responsibility of the buyer — just as the
mansion tax, and similarly as in that case we find evidence of a response. The RPTT tax on old
sales and the NJ Realty Fee schedule are responsibility of the seller and in none of these cases do we
find any evidence of response. Putting these results together with our estimates of the magnitude
of the mansion tax, the results so far are consistent with 100% of the tax being shifted to the seller,
with additional response on the quality margin.
Another margin of response that we have not yet considered is tax evasion. Naturally, our
data does not contain direct indicators of tax evasion. However, it does contain some information
that one expects should be related to availability of evasion opportunities. In 9, we split the New
York sample by coop vs condo status. Coop transactions have to be approved by coop boards
that have the power to stop them. In particular, there is anecdotal evidence that coop boards
disapprove transactions that occur below the perceived market price. One might then expect that
if underreporting of the price was the important margin, the extent of bunching in coop apartments
should be smaller than otherwise. This is indeed what we find although not by a huge margin. In
Table 10, we investigate a more direct measure — the nature of transaction. One might expect that
all-cash transactions involving fewer parties (in particular, no financing) and more liquidity would
make it more likely that side payments are made. We find the opposite. In the context of real
estate transactions, tax evasion is certainly possible but one might expect that it is not completely
straightforward: both parties have to agree and money has to change hands at some point during
the long closing process. One would expect that evasion in this context requires an aspect of trust
between the two parties. Our New York State data contains a dummy for whether transaction was
“arms-length” (ie., between related parties). We find no evidence that arms-length transactions
involve more bunching.
In Table 11 we show that our incidence estimates are not spurious by testing if similar response
occurs at other multiples of $100,000 and find no evidence that this is so. In Table 12, we study the
evolution of bunching over time but find no clear association with the state of real estate market.
4.4
Bunching in listing data
Evidence so far shows clearly that transactions taxes distort the distribution of observed prices. We
found some evidence of supply-side response and differences in estimates based on the side of the
market responsible for the tax that are all consistent with incidence falling exclusively on sellers.
Our tests of tax evasion were weak but did not suggest that this is the main force.
In what follows, we turn to information contained in the listings data and necessarily limit
14
attention to New York City where we have it available.
Figure 8 shows the distribution of listing prices around the mansion tax threshold for properties
sold and matched to the tax data, while Figure 9 shows the distribution of listing prices for all
listed properties in Manhattan. Top panels show raw distributions, while the bottom panels show
the distributions adjusted to remove the common round number bunching. There are three prices
shown for sold listings: the initial asking price, the final price in the listing data (that we interpret as
the opening bid when the ultimate buyer is identified) and the sale price. The graph for all listings
naturally does not contain sale prices. There is clear visual evidence of bunching at the threshold
for each of the prices. Bunching appears strongest for the final price than for the initial price,
and perhaps weaker for the sale price. These visual perceptions are confirmed by our estimates in
Table 13 that indicates substantial bunching for both initial and final listing prices, that exceeds
the response at the sale stage.
This evidence casts further doubt on the possibility that the responses that we identified might
be driven by evasion, because these are responses that occur before a buyer who would be willing
to engage in tax evasion is even identified. The response of the listing prices indicates that sellers
internalize the presence of the tax (which is the responsibility of the buyer!) even before meeting
the buyer. In the next section we will try to understand this process in more detail.
Figures 10 and 11 show analogous results for the smaller RPTT threshold. These graphs are
noisier and evidence of response is hard to see; formal estimates do not reveal it. As discussed
before, we found a response to the RPTT only for new sales where it applies to buyers. However,
the number of new sales in the listing data is small and we run into power issues. We will proceed
by analyzing the mansion tax threshold only.
5
Understanding the response
In the previous section, we showed that the distribution of listing prices is distorted already at
the initial stage. Figure 12 show that the effect of the tax is not limited to the initial posting. In
fact, there is clear evidence that discount from the initial price to the final advertised price (i.e.,
before buyer is identified) and to the final sale price increase as the price increase above $1,000,000.
We present the median and 75th percentiles of the distribution of discounts (many listings are not
revised) in the figures. Naturally, the effect is not immediate at $1,000,000 because the tax applies
to the sale price and not the initial price which constitutes the running variable. Figure 13 shows
analogous evidence for discounts from the final to the sale price — i.e., occurring during bargaining
process. There is similar evidence of an increase in discounts above the tax threshold. These
findings suggest that incidence effects uncover throughout the search process — by distorting the
initial prices, subsequent revisions and, finally, during bargaining stage.
In Figure 14, we focus on the mean discounts from the initial price that blur the impact of the
tax but allow for decomposing the response. Roughly half of the response is due to price revisions
and half due to discounts at bargaining stage.
15
In Figure 15, we show the distribution of sale prices conditional on asking price. There is clear
evidence of the distortion to the median and, especially, to the 25th percentile of the prices. In
Figure 16, we superimpose estimated quantile regression lines on both sides of the threshold that
illustrate that discounts from the initial price increase significantly (and apparently permanently)
once the tax threshold is crossed.
5.1
Informational content of prices
The presence of bunching in asking prices suggests that buyers may have less information than
they would have otherwise. More generally, evidence of an increase in discounts suggests that
the relationship between the initial price and the final price may (though does not have to) be
affected. In Figures 17 and 18, we test this possibility by investigating the relationship between
initial and sale prices. Conditional on sale price, the spread of initial prices increases significantly
around the threshold and stays elevated in an extended range. Conditional on the initial asking
price, the evidence of dispersion in the sale prices is very strong and seemingly not limited to the
neighborhood of the threshold. Figure 18 suggests that the tax reduces the information available
to buyers not just around the threshold but permanently.
5.2
Evidence of other responses
Increased frictions may also affect the time that properties spent on the market. In Figure 19 we
investigate this possibility and find that evidence that this is for properties that are initially priced
just over the threshold but no clear evidence otherwise. Another possibility is that properties do not
sell at all. We investigate it on Figure 20 and find counterintuitive result that properties initially
priced over $1 million actually sell at a higher rate than others. Figures 21 and 22 show that this
effect is driven by an unexpected channel: listings that start above $1 million are actually less likely
to sell with the REBNY realtor and much more likely to sell through other channels (directly by
owner or through other realtors).
We interpret these results as showing that sellers who are otherwise affected by the tax, respond
by giving up services of effectively taxable real estate agents and instead substitute their own (nontaxable!) time. Indeed, Figures 23 and 24 show that sales outcome of sellers who stay and those
who do not stay with original realtors are dramatically different: those who sell without a realtor
sell at lower prices with much clearer evidence of pricing below the tax threshold.
6
Conclusions
16
References
Benjamin, John D., N. Edward Coulson, and Shiawee X. Yang, “Real estate transfer
taxes and property values: The Philadelphia story,” The Journal of Real Estate Finance and
Economics, 1993, 7 (2), 151–157.
Bernheim, Douglas B. and Jonathan Meer, “Do Real Estate Brokers Add Value When Listing
Services Are Unbundled?,” Economic Inquiry, 2012. Forthcoming.
Carrillo, Paul E. and Jaren C. Pope, “Are homes hot or cold potatoes? The distribution of
marketing time in the housing market,” Regional Science and Urban Economics, 2012, 42 (1),
189–197.
Chetty, Raj, John Friedman, Tore Olsen, and Luigi Pistaferri, “Adjustment Costs, Firm
Responses, and Micro vs. Macro Labor Supply Elasticities: Evidence from Danish Tax Records,”
Quarterly Journal of Economics, 2011, 126 (2), 749–804.
Dachis, Ben, Gilles Duranton, and Matthew A. Turner, “The effects of land transfer taxes
on real estate markets: Evidence from a natural experiment in Toronto,” Journal of Economic
Geography, 2012, 12 (2), 327–354.
Doyle, Joseph J. Jr. and Krislert Samphantharak, “$2.00 Gas! Studying the Effects of a
Gas Tax Moratorium,” Journal of Public Economics, 2008, 92 (3-4), 869–884.
Erard, Brian and Chih Chin Ho, “Searching for Ghosts: Who Are the Nonfilers and How
Much Tax Do They Owe?,” Journal of Public Economics, July 2001, 81 (1), 25–50.
Genesove, David and Christopher Mayer, “Loss aversion and seller behavior: Evidence from
the housing market,” Quarterly Journal of Economics, 2001, 116 (4), 1233–1260.
Han, Lu and William C. Strange, “Bidding Wars for Houses,” Real Estate Economics, 2012.
Forthcoming.
Hastings, Justine and Ebonya Washington, “The First of the Month Effect: Consumer Behavior and Store Responses.,” American Economic Journal: Economic Policy, 2010, 2 (2), 142–162.
Haurin, Donald R., Jessica L. Haurin, Taylor Nadauld, and Anthony Sanders, “List
Prices, Sale Prices and Marketing Time: An Application to US Housing Markets,” Real Estate
Economics, 2010, 38 (4), 659–685.
Jia, Panle and Parag A. Pathak, “The impact of commissions on home sales in Greater Boston,”
American Economic Review Papers and Proceedings, 2010.
Kleven, Henrik J. and Mazhar Wassem, “Tax Notches in Pakistan: Tax Evasion, Real Responses, and Income Shifting,” March 2012. LSE, mimeo.
17
Levitt, Steven D. and Chad Syverson, “Market distortions when agents are better informed:
The value of information in real estate transactions,” Review of Economics and Statistics, 2008,
90 (4), 599–611.
Marion, Justin and Erich Muehlegger, “Fuel Tax Incidence and Supply Conditions,” March
2010. Harvard KSG and UC Santa Cruz, mimeo.
Mishra, Prachi, Arvind Subramanian, and Petia Topalova, “Policies, Enforcement, and
Customs Evasion: Evidence from India,” Journal of Public Economics, October 2008, 92 (1011), 1907–1925.
Rothstein, Jesse, “Is the EITC as Good as an NIT? Conditional Cash Transfers and Tax Incidence.,” American Economic Journal: Economic Policy, 2010, 2 (1), 177–208.
Saez, Emmanuel, “Do Taxpayers Bunch at Kink Points?,” American Economic Journal: Economic Policy, 2010, 2 (3), 180–212.
, Joel B. Slemrod, and Seth H. Giertz, “The Elasticity of Taxable Income with Respect to
Marginal Tax Rates: A Critical Review,” Journal of Economic Literature, March 2012, 50 (1),
3–50.
, Manos Matsaganis, and Panos Tsakloglou, “Earnings Determination and Taxes: Evidence
from a Cohort Based Payroll Tax Reform in Greece,” Quarterly Journal of Economics, February
2011, 127 (1), 493–533.
Sallee, James M. and Joel B. Slemrod, “Car Notches: Strategic Automaker Responses to Fuel
Economy Policy,” Journal of Public Economics, 2012. Forthcoming.
Slemrod, Joel, “Optimal Taxation and Optimal Tax Systems,” Journal of Economic Perspectives,
Winter 1990, 4 (1), 157–78.
Slemrod, Joel B., “An Empirical Test for Tax Evasion,” Journal of Public Economics, May 1987,
67 (2), 232–38.
, “Buenas Notches: Lines and Notches in Tax System Design,” September 2010. University of
Michigan, mimeo.
Van Ommeren, Jos and Michiel Van Leuvensteijn, “New Evidence of the Effect of Transaction Costs on Residential Mobility,” Journal of Regional Science, 2005, 45 (4), 681–702.
18
A
Data Appendix
New York City Department of Finance Annualized Rolling Sales
The New York City
Department of Finance (NYCDOF) Annualized Rolling Sales files contain details on real-property
transactions for the five boroughs from 2003 to the present (we use the data through 2011). The data
are realeased by the NYCDOF on a quarterly basis and are derived from the universe of transfer-tax
filings (which are mandatory for all residential and commercial sales). Geographic detail for each
sale includes the street address (and zip code), the tax lot (borough-block-lot number), and the
neighborhood (Chelsea, Tribeca, Upper West Side, etc.). The Rolling Sales files contain limited
details about the properties themselves, including square footage, number of units (residential
and commercial), tax class (residential, owned by utility co., or all other property), and building
class category (a more detailed property code—for example, one-family homes, two-family homes,
residential vacant land, walk-up condo, etc.). Transaction details in the data include the sale price
and date. A sale price of $0 indicates a transfer of ownership without cash consideration (ex. from
parents to children).
New York City properties are subject to the mansion tax if they are single-, double-, or triplefamily homes, or individual condo or co-op units. We define taxable sales as those transactions
of a single residential unit (and no commercial units) with a building classification of “one family
homes,” “two family homes,” “three family homes,” “tax class 1 condos,” “coops - walkup apartments,” “coops - elevator apartments,” “special condo billing lots/condo-rental,” “condos - walkup
apartments,” “condos - elevator apartments,” “condos - 2–10 unit residential,” “condos - 2–10 unit
with commercial use,” or “condo coops/condops.” We define co-ops as a building code of “coops
- walkup apartments,” or “coops - elevator apartments.” We define a commercial sale to be a
transaction with at least one commercial unit (and no residential units) or a tax class of 3 or 4.
New York State Office of Real Property Service SalesWeb The New York State Office of
Real Property Service (NYSORPS) publishes sales records for all real-property transactions (excluding New York City) recorded between 2002–2006 and 2008–2010 available through the “SalesWeb”
database. Since deeds are recorded after the sale, this data includes a small number of sales from
2007. The database is compiled by ORPS from filings of the State of New York Property Transfer
Report (form RP-5217).
The NYS deeds records indicate several details about each transaction and property. Transactionspecifc details include the sale price and date, the date the deed was recorded (and recording details
such as book and page number), the buyer’s, seller’s, and attorney’s name and address (often missing), the number of parcels included in the transaction, and details about the relationship between
the buyer and the seller (whether the sale is between relatives, whether the buyer is also a seller,
whether one party is a business or the government, whether the sale is defined by the state as
arms-length11 , etc.). Property details include the square footage, assessed value (for property-tax
11
The data dictionary defines an arms-length sale as “a sale of real property in the open market, between an informed
and willing buyer and seller where neither is under any compulsion to participate in the transaction, unaffected by
19
purposes), address (including street address, county, zip code, school district), and the property
class (one-family home, condo, etc.). We consider as subject to the mansion tax all single-unit sales
with property class equal to one-, two-, or three-family residence or a seasonal residence.
New Jersey Treasury SR1A File
We make use of sales records from the New Jersey Treasury’s
SR1A file for 1996–2008, which contains records of all SR1A forms filed at the time of sale (the form
is mandatory in the state for all residential sales). Each record includes the sale price and date the
deed was drawn, buyer and seller name and address (often missing), deed recording details (date
submitted, date recorded, document number), and whether there are additional lots associated
with the sale. Property details include land value, tax lot, square footage, and property class. We
define taxable sales as those with a residential property class.
New York City County Register Deeds Records
These data are collected from the county
registers for the five counties in New York City: Bronx County, Kings County, New York County,
Queen’s County, and Richmond County. The records were collected by an anonymous private firm
and made available to us by the Paul Millstein Center for Real Estate at the Columbia Graduate
School of Business.
These data include additional detail as compared to the Rolling Sales files, although at the
expense of precision. Prices in this data set are rounded to the nearest $100, which leads to
misallocation of sales to one side of a tax notch. Transaction details include the sale price and
date, an indicator for whether the unit is newly constructed, the number of parcels being sold,
whether the purchase was made in cash (i.e. whether a mortgage is associated with the sale), and
indicators for private lenders and within-family sales. Property details are limited to address, zip
code, and county.
Real Estate Board of New York Listings Service We have collected residential real-estate
listings from the Real Estate Board of New York’s (REBNY) electronic listing service. REBNY
is a trade association of about 300 realty firms operating in Manhattan and Brooklyn. REBNY
accounts for about 70% of all residential real-estate listings in these boroughs. A condition of
REBNY membership is that realtors are required to post all listings and updates to the listing
service within 24 hours.
Using the REBNY listing service, we have collected all “closed” (i.e. sold) or “pernanently off
market” residential listings posted between 2003 (when the electronic listings are first available)
and 2010. REBNY listings include the typical details available on a real-estate listing: asking
price, address, date on the market and a description of the property. Additionally, we observe all
updates to each listing (and the dates of each update), which lets us see how asking prices evolve
and determine the length of time a property is on the market. Finally, we observe the final outcome
of the listing: whether the property is sold or taken off the market.
any unusual conditions indicating a reasonable possibility that the full sales price is not equal to the fair market value
of the property assuming fee ownership”
20
We create several variables for each REBNY listing. We define the initial asking price as the
first posted price on the listing, and the final asking price as the last posted price while the listing is
“active.” We identify the length of time that a listing spends on the market as the number of days
between the initial posting and the date that the listing is updated as “in contract.” We define the
1
discount between two prices as the percent drop in price— p0p−p
, where p0 and p1 are prices and p0
0
is posted before p1 .
One caveat to the REBNY listings is that the price is often not updated at the time of sale.
To overcome this, we match REBNY listings to the NYCDOF data by address and date. Of the
48,220 closed REBNY listings for Manhattan, we achieve a match rate of 92%. At the same time,
of the 23,655 Manhattan listings that are not reported as closed in the REBNY listings database,
we find 7,425 corresponding sales in the NYCDOF data. We treat such matches as an indication
that the property was sold without the REBNY realtor (either sold by the owner or using another
realtor).
21
Table 1: Real Estate Tranfer Tax Schedules
(a) NJ Realty Transfer Fee Schedule
Total Consideration ≤$350,000
0.4%
0.67%
0.78%
$2,105
Total Consideration >$350,000
$0–$150,000
$150,001–$200,000
$200,001–$350,000
Taxes at $350,000
$2,735
$350,001–$550,000
0.96%
$550,001–$850,000
1.06%
$850,001–$1,000,000
1.16%
$1,000,000+
1.21%
Note: Sellers are responsible for remitting the RTF to the government.
(b) Mansion Tax Schedule (NJ & NY)
Total Consideration <$1,000,000
Total Consideration ≥$1,000,000
$0–$999,999
0%
1%
Taxes at $1,000,000
$0
$10,000
≥$1,000,000
1%
Note: In both states, buyers are responsible for remitting the mansion tax to the government.
(c) Real Property Transfer Tax (NYC)
Total Consideration ≤$500,000
Total Consideration >$500,000
Property Type
Residential
Commercial
Residential
Commercial
$0–$500,000
1%
1.425%
1.425%
2.625%
Taxes at $500,000
$5000
$7125
$7125
$13,125
>$500,000
1.425%
2.625%
Note: Sellers are responsible for remitting the NYC RPTT. However, it is customary
for buyers to pay the tax on sponsored sales (ex. new construction, condo conversions).
22
Table 2: Sample Statistics
Number of Sales
Share
Mean Price
Median Price
Total Sales between
$950,000 and
$1,050,000
Total
Taxable
Non-Taxable
Total
Taxable
Non-Taxable
Total
Taxable
Non-Taxable
Total
Taxable
Non-Taxable
Total
Taxable
Non-Taxable
NYC
(2003 – 2011)
633372
392446
240926
100
62
38
944825.4
657774.4
1412405
428068
402000
475000
12561
8133
4428
NYS
(2002 – 2010)
2062229
1483982
578247
100
72
28
204184.7
206706.3
197713.3
85000
112000
28000
8646
6287
2359
NJ
(1996 – 2008)
3010233
2400399
609834
100
79.7
20.3
235141.4
197090.4
384916
149000
145000
169990
12295
7587
4708
Notes: Taxable sales are defined to capture transactions likely to be covered by the mansion tax in each
geography (while non-taxable sales will include cases where tax status is ambiguous due to a combination
of legal exceptions and limited data). NYC data is from the Department of Finance Rolling Sales file for
2003–2011 (taxable sales defined as single-unit non-commercial sales of one-, two-, or three-family homes,
coops, and condos). Data for NYS from Office of Real Property Service deeds records for 2002–2006 and
2008–2010 (taxable defined as all single-parcel residential sales of one-, two-, or three-family homes). Data
for NJ from the State Treasury SR1A file for 1996–2008 (taxable defined as any residential sale).
23
24
N
.
.
.
.
.
.
.
49030
55353
53835
49401
49664
40980
31782
27441
32301
NYC
Med Price Med Real Price
.
.
.
.
.
.
.
.
.
.
.
.
.
.
290000
267017.1
340000
302651
395460
342345.4
444132
371177
480000
391945.3
475000
371548.5
418690.5
328269.3
460000
356305.5
455000
341493.8
N
.
.
.
.
.
.
167417
171622
180127
180519
155726
6059
120061
113362
106880
.
NYS
Median Price
.
.
.
.
.
.
130000
145900
162600
180000
169000
283000
160000
154486
160000
.
Med Real Price
.
.
.
.
.
.
121843.4
133702.1
145194.7
155410
141089.1
228636.5
125202
121030.5
123529.2
.
N
126000
130028
147590
157392
156264
156438
167524
171815
182189
177218
143557
121721
38961
.
.
.
NJ
Median Price
123500
126000
133835
138000
145200
161900
189000
226000
264900
305000
320000
315000
285000
.
.
.
Med Real Price
132758.1
132734.3
138325.1
140158.8
142883.8
154059.7
176913.6
207571
235691.5
263219.3
267671
255736.7
225458.8
.
.
.
Notes: Taxable sales are defined to capture transactions likely to be covered by the mansion tax in each geography (while non-taxable sales will
include cases where tax status is ambiguous due to a combination of legal exceptions and limited data). NYC data is from the Department of Finance
Rolling Sales file for 2003–2011 (taxable sales defined as single-unit non-commercial sales of one-, two-, or three-family homes, coops, and condos).
Data for NYS from the Office of Real Property Services deeds records for 2002–2006 and 2008–2010 (taxable defined as all single-parcel residential
sales of one-, two-, or three-family homes). Data for NJ from the State Treasury SR1A file for 1996–2008 (taxable defined as any residential sale).
Year
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
Table 3: Taxable Sales Over Time
Table 4: REBNY Listings Sample Statistics
Sold
Days on the Market
Initial Asking Price
Final Asking Price
Sale Price*
Discount (First to Sale)*
Discount (First to Final)
Discount (Final to Sale)*
Matched, but not Closed
All Listings
Mean
Median
0.671
1
197.859
110
1604547 899000
1602670 875000
1237652 780000
0.06
0.043
0.019
0
0.03
0.023
0.103
0
n
71875
67550
71875
71875
44320
44320
71875
44320
71875
Closed and Matched Sales
Mean
Median
n
1
1
44320
146.107
80
40680
1384028 825000 44320
1435747 799000 44320
1237652 780000 44320
0.06
0.043
44320
0.012
0
44320
0.03
0.023
44320
0
0
44320
Notes: Data from the Real Estate Board of New York’s listing service; represents all REBNY listings in
Manhattan between 2003 and 2010 that are closed or off market. Sold is an indicator equal to one if the
final status of the listing is “Closed.” Days on the market is calculated as the number of days between the
initial active listing and the final status of “in contract” (if the property sells with REBNY) or “permanently
off market” (otherwise). Initial asking price is the asking price on the listing when first active; final asking
price is the price listed immediately prior to the listing closing or being taken off the market. Sale price is
the price reported in the NYC DOF data and is available only for REBNY listings that have a match in
the DOF data. Discount is defined as (initial price - last price) / initial price. 5,940 listings have invalid
listing and off-market dates (missing or obviously misreported), and are omitted from days on market/sale
calculations. Sold with REBNY is an indicator that a listing has a match in the NYC DOF data, but is
never reported as “Closed” by the REBNY agent. * indicates the statistic is calculated conditional on the
listing closing and matching to the NYC DOF data.
25
26
1020137.099
1022651.161
1021740.212
1019365.681
1018692.079
1016204.914
1016331.357
1016700.363
1018175.902
1020100.292
1021791.952
1021714.491
1023552.716
1024620.193
1023357.325
1016656.579
1018245.351
Baseline: 3rd Order, Omit $990k – $1.1M
1st Order
2nd Order
4th Order
5th Order
6th Order
Omit $990k – $1.01M
Omit $990k – $1.025M
Omit $990k – $1.050M
Omit Price ≥ $990k
Omit $980k – $1.1M
Omit $970k – $1.1M
Omit $960k – $1.1M
Omit $950k – $1.1M
No Discontinuity at Threshold
OLS, $5000 Bins
OLS, $10000 Bins
1002.171
1019.381
1106.111
1459.447
1101.66
979.908
958.485
1021.825
1297.738
1113.779
1126.719
866.317
204.586
845.046
1820.956
1742.113
1043.411
Std. Error
1039172.869
1024316.746
1018037.158
1017633.055
999301.891
999393.565
999960.048
1012703.41
1024350.868
1024343.693
1024518.34
1024991.424
1024977.649
1005032.379
1013332.433
1024169.542
Gap
1194.007
386.877
2211.538
2474.912
1652.231
20.193
194.646
1092.657
384.938
416.717
282.996
966.384
836.057
5120.439
3584.697
686.284
Std. Error
107777
107777
107777
107777
107777
111073
110677
109913
90696
106935
105948
105101
103913
107777
107777
107777
107777
n
Notes: Incidence estimates by MLE using data from NYC Department of Finance Rolling Sales file for 2003–2011. Sample is restricted to all taxable
sales (single-unit non-commercial sales of one-, two-, or three-family homes, coops, and condos) with prices between $510,000 and $1,500,000.
Incidence
Specification
Table 5: Mansion Tax, NYC
Table 6: Mansion Tax, by Region
Sample
NYC
NYC
(Yrs Since Contr.)
NYS (excl. NYC)
NJ Post Tax
NJ Post Tax
(Yrs. Since Constr.)
NJ Pre Tax
NJ Pre Tax
(Yrs. Since Constr.)
All
<0
0
1
2
3
4–6
7+
All
Old
New
All
0
1
2
3
4–6
7+
All
0
1
2
3
4–6
7+
Incidence
1020137.099
1049054.752
1014350.51
1010656.719
1012930.019
1022965.88
1024587.946
1021098.749
1022040.067
1021541.398
1030726.162
1021749.859
1036831.498
1015596.349
1024086.308
1023489.171
1022395.4
1020133.785
999212.833
999510.246
999999.024
999158.962
997134.253
999335.584
999217.855
Std. Error
1112.065
16400.406
8110.852
3161.653
2845.458
5243.869
4903.579
1456.981
1311.569
1380.212
7507.74
1599.151
13676.785
7240.816
7854.407
8373.455
4285.694
1947.966
36.546
2521.52
8195.094
1552.95
2537.63
278.616
45.998
Gap
1024169.542
1026372.814
1013795.203
1034428.244
1020471.916
1023791.783
1040462.957
1024276.017
1028140.388
1026815.57
1049702.23
1024413.21
1024693.428
1031723.732
1014230.067
1024408.158
1024952.859
1024404.738
1006702.44
999767.682
1010920.084
1024102.996
983243.507
1024640.566
1012451.003
Std. Error
703.141
15497.931
15212.777
7078.796
6210.459
8951.187
6185.779
592.98
1988.517
1872.429
3077.446
389.533
13412.089
11380.295
17080.624
11938.152
6122.684
822.455
3126.063
21158.688
12468.02
14120.242
19769.328
6581.474
4308.25
n
107777
604
1095
3642
4635
2345
2554
75674
111517
107535
3982
87823
976
1869
1495
1753
5171
62793
62999
825
985
1222
1424
3787
39017
Notes: Estimates from baseline procedure (3rd-order polynomial, omit $990k–$1M). NYC data is from the
Department of Finance Rolling Sales file for 2003–2011 (taxable sales defined as single-unit non-commercial
sales of one-, two-, or three-family homes, coops, and condos). Data for NYS from the Office of Real Property
Services deeds records for 2002–2006 and 2008–2010 (taxable defined as all single-parcel residential sales of
one-, two-, or three-family homes). Data for NJ from the State Treasury SR1A file for 1996–2008 (taxable
defined as any residential sale). NJ sample restricted to sales recorded between 1996 and 2003 for pre-tax
estimates and post August, 2004 for estimates in presence of the tax. Years since construction defined as
the difference between the year the property was built and the year of sale. New sales in New York state are
defined as any being flagged as new construction.
27
Table 7: NYC Real Property Transfer Tax
NYC
NYS
Sample
New
Old
New
Old
Incidence
501597.724
499614.834
499532.221
499110.756
Std. Error
718.315
28.544
440.783
18.446
Gap
500240.342
499107.35
491151.744
499238.192
Std. Error
1190.69
157.32
3599.767
60.204
n
22626
265543
19687
698862
Notes: Estimates based on 5th-order polynomial, omitting sales between $490,000 and $550,000. NYC data
is from the Department of Finance Rolling Sales file for 2003–2011. Data for NYS from the Office of Real
Property Services deeds records for 2002–2006 and 2008–2010 (excluding NYC) restricted to all single-parcel
residential sales of one-, two-, or three-family homes. Sales in NYC are defined as single-unit non-commercial
sales of one-, two-, or three-family homes, coops, and condos. New sales are defined as any sale occurring
within three years of unit’s construction (in NYC) or any sale flagged as new construction (in NYS, excluding
NYC).
Table 8: NJ Realty Transfer Fee
Omitted Region
$340k–$400k
$345k–$400k
Incidence
349492.126
349469.436
Std. Error
19.804
18.524
Gap
348332.936
348053.374
Std. Error
440.07
435.186
n
421468
426899
Notes: Estimates based on 5th-order polynomial, omitting sales in the indicated region. Data for NJ from
the State Treasury SR1A file for Aug. 2004 through 2008 (taxable defined as any residential sale). Sample
is restricted to taxable sales with prices between $100,000 and $900,000.
28
29
Incidence
1014455.145
1021984.526
1024011.537
Std. Error
2064.824
1177.482
915.067
Gap
1023982.611
1024255.175
1024248.239
Std. Error
2383.789
798.624
815.375
n
28058
79719
67519
Sample
All
Cash
Mortgage
Arms-Length
Non-Arms-Length
Incidence
1017195.518
1014139.99
1018509.338
1021923.036
1023255.82
Std. Error
1607.718
2231.711
2225.505
1367.195
3308.883
Gap
1029198.879
1024450.398
1034432.707
1029302.345
1023398.319
Std. Error
1931.629
828.998
2553.994
2212.865
5594.466
n
79435
29291
50144
100534
10983
Notes: NYC Deeds Records data from deeds records collected by private data provider (taxable defined as any residential sale). Non-arms-length
sales in NYS defined by the Office of Real Property Services as a sale of real property between relatives or former relatives, related companies or
partners in business, where one of the buyers is also a seller, or “other unusual factors affecting sale price” (ex. divorce or bankruptcy).
NY State
NYC Deeds Records
Geography
Table 10: Mansion Tax: Evasion
Notes: Incidence estimates by MLE using data from NYC Department of Finance Rolling Sales file for 2003–2011. Sample is restricted to all taxable
sales (single-unit non-commercial sales of one-, two-, or three-family homes, coops, and condos) with prices between $510,000 and $1,500,000. Coops
are identified as sales with associated building codes equal to “Coops - Walkup Apartments” or “Coops - Elevator Apartments.”
NYC Non-Coops
NYC Coops
Sample
All
All
Old
Table 9: Mansion Tax by Property Type (Coop vs. Non-Coop)
Table 11: NYC Mansion Tax: Placebos
Cutoff
Commercial
600,000
700,000
800,000
900,000
1,100,000
Incidence
999350.137
599257.387
699279.176
799210.939
899169.227
1099059.583
Std. Error
66.102
32.874
30.541
29.639
26.872
146.288
Gap
999738.022
599750.513
698250.242
791895.943
791092.447
1111937.941
Std. Error
1038.92
731.249
1168.353
1983.227
4908.542
1945.735
n
6395
86794
95405
100582
101874
108609
Notes: Data from NYC Department of Finance Rolling Sales file for 2003–2011. Commercial sales are defined
as any transaction of at least one commercial unit and no residential units or a NYC tax class of 3 (utility
properties) or 4 (commercial or industrial properties) and are not subject to the mansion tax.
Table 12: NYC Over Time
Year
2003
2004
2005
2006
2007
2008
2009
2010
2011
Incidence
1013332.409
1022734.837
1029680.354
1019609.237
1016019.041
1016565.392
1021414.821
1015390.475
1019300.361
Std. Error
4028.256
2685.899
4155.561
2741.283
2298.069
2531.723
3080.109
3593.947
3309.653
Gap
1022040.529
1018218.629
1030943.347
1024601.377
1024053.356
1017390.173
1003469.802
1030732.583
1024941.986
Std. Error
5523.747
5680.969
4072.798
2458.913
2890.239
4615.174
5278.356
4766.875
3119.67
n
7646
11174
14381
15636
17331
13464
9048
8847
10250
Notes: Data from NYC Department of Finance Rolling Sales files, restricted to taxable sales (single-unit
non-commercial sales of one-, two-, or three-family homes, coops, and condos) in the given year.
Table 13: Mansion Tax: Listings
Sample
All
Sold
Unsold
Price
First
Final
First
Final
Sale
First
Final
Incidence
1024054.397
1025936.624
1024027.132
1029615.738
1017363.849
1024128.815
1024852.06
Std. Error
1128.428
2124.956
1611.512
3418.527
1919.906
2458.838
3003.655
Gap
1024283.191
1021118.533
1024033.484
1016288.534
1023952.165
1024769.283
1024234.591
Std. Error
633.216
2860.161
2197.155
4097.914
2152.11
2959.917
3425.851
n
37770
37189
26207
25792
25585
7932
7853
Notes: Data from the Real Estate Board of New York’s listing service; represents all REBNY listings between
2003 and 2010 that are closed or off market. Unsold sample defined as all listings with final status not equal
to “closed.” Sold sample defined as all listings that match to a NYC Department of Finance sale record
with final status equal to “closed.” First price is the initial price posted on the listing. Final price is the
last price posted while the listing is active (prior to statust being changed to “in contract” or “off market”).
Sale price is the recorded price from the NYC Department of Finance.
30
0
Number of Sales per $5000 bin
2000
4000
6000
Figure 1: Distribution of Taxable Sales in New York State
750000
1000000
Sales Price
750000
1000000
Sales Price
1250000
1500000
0
Number of Sales per $25000 bin
5000
10000
15000
20000
500000
500000
1250000
1500000
Notes: Plot of the number of mansion-tax eligible sales in each $5,000 (panel a) or $25,000 (panel b) price
bin between $510,000 and $1,500,000. Data from the NYC Rolling Sales file for 2003–2010 (taxable sales
defined as single-unit non-commercial sales of one-, two-, or three-family homes, coops, and condos) and
from NYS from Office of Real Property Service deeds records for 2002–2006 and 2008–2010 (taxable defined
as all single-parcel residential sales of one-, two-, or three-family homes).
31
2000
Number of Sales per $5000 bin
4000
6000
8000
Figure 2: Distribution of Sales in New York City around the $500,000 RPTT tax notch
350000
400000
450000
500000
Sales Price
550000
600000
650000
Notes: Plot of the number of sales in each $5,000 price bin between $350,000 and $650,000. Data from the
NYC Rolling Sales file for 2003–2011. Both commercial and non-commercial sales are subject to the NYC
RPTT.
0
Number of Sales per $5000 bin
500
1000
1500
2000
Figure 3: Distribution of NJ Sales Pre-Mansion Tax
500000
750000
1000000
Sales Price
1250000
1500000
Notes: Plot of the number of mansion-tax eligible sales in each $5,000 price bin between $510,000 and
$1,500,000, prior to the introduction of the tax. Data from NJ deeds records for 1996–2003 (taxable defined
as any residential sale).
32
0
20
Number of Sales
40
60
Figure 4: New Jersey Monthly Sales $990,000 to $1,010,000
2001
2002
2003
2004
2005
2006
2007
2008
Month
$990,000 to $999,999
$1,000,000 to $1,009,999
Notes: Total taxable NJ sales in given price range by month. Data for NJ from deeds records (via NJ
Treasury) for 2000–2008 (taxable defined as any residential sale). Mansion tax introduced in August, 2004.
33
.2
Share of sales $900k -- $1.1M
.4
.6
.8
1
Figure 5: Distribution of Monthly Sales in New Jersey ($900k – $1M)
2001
2002
2003
2004
2005
2006
2007
2008
Date
< 925k
< 950k
< 975k
< 1M
< 1.025M
< 1.05M
< 1.075M
< 1.1M
Notes: Number of taxable sales in given range as a share of total sales between $900,000 and
$1,100,000 by month. Data from NJ deeds records for 2000–2008 (taxable defined as any residential sale).
Mansion tax introduced in August, 2004.
34
-2
log( NJ sales post tax) - log( NJ sales pre tax)
-1
0
1
Figure 6: Distribution of All Sales in New Jersey
500000
750000
1000000
Sale Price
1250000
1500000
Notes: Plot of the difference between the log number of NJ taxable sales per $10,000 price bin
from October 1, 2004 to September 30, 2005 and July 2, 2002 to April 30, 2004. Data for NJ from
deeds records for 2000–2008 (taxable defined as any residential sale). Pre-period (July 2, 2002 to April 30,
2004) selected to balance the total number of taxable sales between $510,000–$1,500,000 between
the pre and post periods.
35
36
Incidence
Seller's price
Threshold
with the tax
Counterfactual
distribution
Buyer's price
Price
Observed distribution
Figure 7: Incidence and gap concepts
Incidence estimate
Source of
bunchers
Bunching
Tax
Without the tax
of buyer's and seller's surplus
leads to maximizing some function
Assume that bargaining process
Gap
Gap estimate
"Missing"
sales
$1,000,000
Omitted region
Counterfactual
distribution
0
.02
Share of Sales
.04
.06
.08
Figure 8: Distribution of Real-Estate Listing Prices in NYC (Sold Properties Only)
750000
Initial Asking Price
1000000
Price
$25,000 Bins
Final Asking Price
1250000
1500000
Sale Price
0
.02
Share of Sales
.04
.06
500000
500000
750000
Initial Asking Price
1000000
Price
$25,000 Bins
Final Asking Price
1250000
1500000
Sale Price
Notes: Data from REBNY listings matched to NYC Department of Finance sales records. Sample
restricted to “sold” listings = last listing status = “closed” and property can be matched to NYC
sales data. Panel a presents a plot of the number of listings per $25,000 bin as a share of all
sales between $500,000 and $1,500,000 (bins centered so that the threshold bin spans $975,001–
$1,000,000). Panel b presents a smoothed plot of the distribution that accounts for round-number
bunching. The log of the per-bin counts from panel a are regressed on a cubic in price and dummy
variables for multiples of $50,000 and $100,000 interacted with the price. Predicted bunching for
round-number bins are then subtracted from the corresponding counts.
37
0
.02
Share of Sales
.04
.06
Figure 9: Distribution of Real-Estate Listing Prices in NYC (All REBNY-Listed Properties)
750000
1000000
Price
$25,000 Bins
Initial Asking Price
1250000
1500000
Final Asking Price
0
.02
Share of Sales
.04
.06
500000
500000
750000
1000000
Price
$25,000 Bins
Initial Asking Price
1250000
1500000
Final Asking Price
Notes: Data from REBNY listings. Sample includes all REBNY-listed sales in the given range.
Panel a presents a plot of the number of listings per $25,000 bin as a share of all sales between
$500,000 and $1,500,000 (bins centered so that the threshold bin spans $975,001–$1,000,000). Panel
b presents a smoothed plot of the distribution that accounts for round-number bunching. The log of
the per-bin counts from panel a are regressed on a cubic in price and dummy variables for multiples
of $50,000 and $100,000 interacted with the price. Predicted bunching for round-number bins are
then subtracted from the corresponding counts.
38
0
.02
Share of Sales
.04
.06
Figure 10: Distribution of Real-Estate Listing Prices in NYC (Sold Properties Only)
350000
400000
450000
Initial Asking Price
500000
Price
$5000 Bins
550000
Final Asking Price
600000
650000
Sale Price
0
.02
Share of Sales
.04
.06
Notes: Plot of the number of listings per $5,000 bin as a share of all sales between $350,000 and
$650,000 (bins centered so that threshold bin spans $495,002–$500,001). Data from REBNY listings
matched to NYC Department of Finance Sales data. Sample restricted to “sold” listings = last
listing status = “closed” and property can be matched to NYC sales data.
Figure 11: Distribution of Real-Estate Listing Prices in NYC (All REBNY-Listed Properties)
350000
400000
450000
500000
Price
$5000 Bins
Initial Asking Price
550000
600000
650000
Final Asking Price
Notes: Plot of the number of listings per $5,000 bin as a share of all sales between $350,000
and $650,000 (bins centered so that threshold bin spans $495,002–$500,001). Data from REBNY
listings. Sample includes all REBNY-listed sales in the given range.
39
0
.05
Median Discount
.1
.15
.2
Figure 12: Median & 75th Percentile Price Discounts by Initial Asking Price
500000
750000
1000000
Initial Asking Price
($25000 bins)
1250000
1500000
First Asking vs. Final Asking Price (P75)
First Asking vs. Sale Price (P75)
First Asking vs. Sale Price (Median)
First Asking vs. Final Asking Price (Median)
0
Median Discount
Final Asking vs. Sale Price
.02
.04
.06
.08
.1
Notes: Plot of the median and 25th percentile discount from initial asking to sale price ( = 1 final/initial) and initial asking to final asking price ( = 1 - sale/initial) per $25,000 initial-askingprice bin. Data from REBNY listings—sample includes all closed REBNY-listed properties in the
range $500,000–1,500,000 that match to NYC DOF data.
Figure 13: Median Price Discount by Final Asking Price
500000
750000
1000000
Final Asking Price
($25000 bins)
75th Percentile
1250000
1500000
Median Discount
Notes: Plot of the median and 25th percentile discount from final asking price to sale price ( =
1 - sale/final) per $25,000 final-asking-price bin. Data from REBNY listings—sample includes all
closed REBNY-listed properties in the range $500,000–1,500,000 that match to NYC DOF data.
40
0
.02
Mean Discount
.04
.06
.08
Figure 14: Mean Price Discounts by Initial Asking Price
500000
750000
First Asking vs. Sale Price
1000000
Initial Asking Price
($25000 bins)
First Asking vs. Final Asking Price
1250000
1500000
Final Asking vs. Sale Price
Notes: Plot of the average discount from initial asking to sale price ( = 1 - final/initial), initial
asking to final asking price ( = 1 - sale/initial), and final asking price to sale price relative to initial
asking price ( = (final - sale)/initial) per $25,000 initial-asking-price bin. Data from REBNY
listings—sample includes all closed REBNY-listed properties in the range $500,000–1,500,000 that
match to NYC DOF data.
41
800000
Sale Price
1000000
1200000
Figure 15: Distribution of Sale Price by Initial Asking Price
800000
900000
10th - 90th Percentile
1000000
Initial Asking Price
1100000
25th - 75th Percentile
1200000
Median
800000
Sale Price
1000000
1200000
Notes: Plot of the median, 10th, 25th, 75th, and 90th percentiles of sale price per $25,000 initialasking-price bin. Data from REBNY listings—sample includes all sold REBNY-listed properties
(matched to NYC DOF) in the range $800,000–1,200,000.
Figure 16: Distribution of Sale Price by Initial Asking Price (with Quantile Regression)
800000
900000
10th - 90th Percentile
1000000
Initial Asking Price
1100000
25th - 75th Percentile
1200000
Median
Notes: Plot of the median, 10th, 25th, 75th, and 90th percentiles of sale price per $25,000 initialasking-price bin. Data from REBNY listings—sample includes all sold REBNY-listed properties
(matched to NYC DOF) in the range $800,000–1,200,000. Lines represent quantile regressions for
42
the given range ($800k–$990k and $1M – $1.2M).
.08
.09
Std Dev of Log(First Asking Price)
.1
.11
.12
.13
Figure 17: Dispersion of First Asking Price, Conditional on Sale Price
500000
750000
1000000
Sale Price
($25000 bins)
1250000
1500000
.08
.09
Std Dev of Log(Sale Price)
.1
.11
.12
Notes: Plot of the standard deviation of the log of initial asking price by $25,000 sale price bin.
Data from REBNY listings—sample includes all sold REBNY-listed properties (matched to NYC
DOF) in the range $500,000–1,500,000.
Figure 18: Dispersion of Sale Price, Conditional on First Asking Price
500000
750000
1000000
First Asking Price
($25000 bins)
1250000
1500000
Notes: Plot of the standard deviation of the log of sale price by $25,000 inital asking price bin.
Data from REBNY listings—sample includes all sold REBNY-listed properties (matched to NYC
DOF) in the range $500,000–1,500,000.
43
0
100
Days on Market
200
300
400
Figure 19: Median Days to Sale by Initial Asking Price
500000
1000000
Initial Asking Price
($25000 bins)
25th Percentile
1500000
Median
.74
Probability of Sale (with or without Realtor)
.76
.78
.8
.82
.84
Notes: Plot of the median and 25th percentile of days to sale per $25,000 initial-asking-price bin.
Data from REBNY listings—sample includes all REBNY-listed properties in the range $500,000–
1,500,000. Days to sale defined as the number of days between initial listing of the property and
buyer and seller entering into contract (defined as final status = “in contract”). Unsold properties
are assigned a value of 999 days on the market.
Figure 20: Probability that Listed Property Sells by Initial Asking Price
500000
750000
1000000
Initial Asking Price
($25000 bins)
1250000
1500000
Notes: Plot of the share of REBNY-listed properties that close or are matched to a NYC DOF
44
sale per $25,000 bin. Data from REBNY listings—sample includes all listed properties in the range
$500,000–1,500,000. “Sold” defined as any property with a final listing status of “closed” or any
listing that matches to NYC DOF sales.
.6
Probability Listing Closes
.65
.7
.75
Figure 21: Probability that Listed Property Sells with the agent by Initial Asking Price
500000
750000
1000000
Initial Asking Price
($25000 bins)
1250000
1500000
.05
Prob. of Leaving Realtor
.1
.15
.2
Notes: Plot of the share of REBNY-listed properties that close in the REBNY database per $25,000
bin. Data from REBNY listings—sample includes all listed properties in the range $500,000–
1,500,000.
Figure 22: Probability of Selling Without REBNY by Initial Asking Price
500000
750000
1000000
Initial Asking Price
($25000 bins)
1250000
1500000
Notes: Plot of the share of REBNY listed properties that are sold in NYC DOF data, but are not
sold in the REBNY listing per $25,000 initial-asking-price bin. Data from REBNY listings—sample
includes all REBNY-listed properties in the range $500,000–1,500,000.
45
800000
Sale Price
1000000
1200000
Figure 23: Distribution of Sale Price by Initial Asking Price if Sell with REBNY
800000
900000
1000000
Initial Asking Price
25th - 75th Percentile
1100000
1200000
Median
800000
Sale Price
1000000
1200000
Notes: Plot of the median, 10th, 25th, 75th, and 90th percentiles of sale price per $25,000 initialasking-price bin. Data from REBNY listings—sample includes all sold REBNY-listed properties
(matched to NYC DOF) in the range $800,000–1,200,000.
Figure 24: Distribution of Sale Price by Initial Asking Price if Sell without REBNY
800000
900000
1000000
Initial Asking Price
25th - 75th Percentile
1100000
1200000
Median
Notes: Plot of the median, 10th, 25th, 75th, and 90th percentiles of sale price per $25,000 initialasking-price bin. Data from REBNY listings—sample includes all sold REBNY-listed properties
(matched to NYC DOF) in the range $800,000–1,200,000.
46