is the property tax a benefit tax? the case of rental · property tax rate for all property....

IS THE PROPERTY TAX A BENEFIT TAX? THE CASE OF RENTAL HOUSING* ROBERT 1. CARROLL* & JOHN YINGER**

Abstract - Is the property tax a benefit tax? This paper tests the benefit-tax hypothesis for rental housing. In the case most favorable to this hypothesis, namely with mobile tenants, relatively high property taxes cannot be directly reflected in rents but can be indirectly reflected to the extent that they lead to relatively high-quality public services. This indirect shifting equals tenants’ willingness to pay for the public service increment. Even in this case, however, the property tax is not a benefit tax unless the tax rate is set so indirect shifting equals a landlord’s tax increase. This second condition is not met in the Boston area in 1980; on average, indirect shifting covers only about $0.7 7 of a $1.00 increase in property taxes.

INTRODUCTION

Scholars disagree about whether a local property tax is a benefit tax, in the sense that the tax burden coincides with benefits from the services it finances.

‘OffIce of Tax Analysis, U S Department of the Treasury, Washington, D.C. 20220 **Center for Policy Research. Syracuse Unwerstty. Syracuse, NY 13244-l 090

This debate is important because a benefit tax causes no efficiency loss and sat- isfies the widely held principle that the people who benefit from a service should pay for it. Mieszkowski and Zod- row (1989, p. 1140) argue, for example, that “the benefit view obtains only if zoning . . . is sufficiently binding” and then “reject the assumption of perfect or binding zoning and conclude that a national system of property taxes is dis- tortionary.” In contrast, Fischel (1992, p. 171), building on the work of Hamilton (1975), claims that “the property tax becomes merely a fee for local public services and thus has no dead weight loss.” This paper provides evidence on this debate for the case of rental housing.’

A necessary condition for the property tax to be a benefit tax is that those who bear a higher property tax burden than others in the same urban area must receive correspondingly higher benefits from public services. Indeed, the core of the benefit-tax concept is that a house- hold must pay a higher property tax “fee” in order to obtain higher-quality public services. The claim that the property tax is a benefit tax, therefore, can be tested by determining whether the

295

burden of property tax differences across communities In an urbarl area falls on the people who benefit -from the accom- panying differences in public service quality. To put it another way, thus clarm can be tested with a model of the balanced-budget incidence Iof rental property tax differentials with,in an urban area.’ This paper develops and tests such a model

As first pointed out by Crrr (1968, 1970), the incidence of the property tax on rental housing can be estimated, at least in part, by examining the relatronship between property taxes and apartment rents3 The only way for landlords to shift the tax onto tenants is to charge relatively high rents in communities where property taxes are relatrvely high. This focus on property tax differentials cannot determine the incidence of the average property tax rate, but is sufficient for a test of the benefit-tax notion.

Existing studres of the relationship between rents and property taxes Implicitly assume that tenants are not fully mobile. Renters do not pay property taxes directly, and mobile renters would not be willing to pay a higher rent simply because property taxes are higher. Renters do consider public service quality when deciding where to live, however, so that differentials in public service quality, which are made possible in part by differences in property tax rates, can be a source of tax shifting even with fully rnobile renters. This general equilibrium effect, which IS the focus of this paper, has not been considered in any previous research.

Section two of this paper reviews prior research on the inciidence of property taxes on rental housing. Section three develops a model of balanced-budget incidence that combines an analysis of renters’ decisions, an equation to deter- rnine the value of an apartment build-

Ing, a community’s budget constraint, and a simple analysis ofrthe determi- nants of Imedian rent. This model can be tested with (3 hedonic equation for rent determinatioln and a measure of public service quality. The fourth section presents our strategy for obtaining the required measure of service quality and estimating lo hedonic equation and describes our data. All these components are assembled in the fifth section, which presents the estimation results and discusses their implications for the incidence of property tax differentials and for the benefit view of the property tax. The final section summarizes our main conclusions.

‘THE LITERATURE

Table 1 depicts the principal features and conclusions of studi~es examining the Incidence of the property tax on rental housing. These studies do not reach a consensus about the portion of the tax that IS shifted onto renters.

In additioil to neglectin the general equilibriurn effect mode ed 4 below, these studies encounter several methodologrcal problems, all of which we address in our own Iregressions. First, several studies employ an Inadequate set of housing at- trrbutes alId are therefore subject to omitted variable bias. Second, most studies use expenditure as a measure of public ser’vrce quality. As explained below, this approach is, at’ best, a rough approximation. Third, most researchers inclucle only education and omit sanita- tion, police and fire protectron, transpor- tation, and other public services, all of whrch may influence renters’ brds for housing.

Another problem common to the studies is that sorne of the expl$natory variables, such as the tax rate, may be simultaneously determined with rent, thereby Introducing stmultaneity bias. As shown

TABL

E 1

EMPI

RICA

L ST

UDIE

S OF

TH

E IN

CIDE

NCE

OF

PRO

PERT

Y TA

X DI

FFER

ENTI

ALS

ON

RENT

AL

HOUS

ING

Orr

(196

8)

Hein

berg

an

d Oa

tes

(197

0)

Orr

(197

0)

Hym

an

and

Paso

ur

Dusa

nsky

, In

gber

, an

d (1

973)

Ka

ratja

s (1

981)

Incid

ence

of

Ta

x Fa

ils

to

find

evid

ence

of

Fa

ils

to

find

evid

ence

of

Sh

iftin

g of

46

pe

rcen

t of

Sh

iftin

g of

60

pe

rcen

t of

sh

iftin

g.

shift

ing.

ta

x. ta

x.

Stud

y Ch

arac

teris

tics

Unit

of

obse

rvat

ion

Geog

raph

ic ar

ea

Tim

e pe

riod

2 Sa

mpl

e siz

e

Mod

el

Spec

ifica

tion

Depe

nden

t va

riabl

e

Prop

erty

tax

varia

ble(s

)

Publ

ic se

rvice

va

riable

(s)

Stru

ctur

al

char

acte

ristic

s Lo

catio

nal

char

acte

ristic

s

Estim

atio

n Pr

oced

ure’

Mun

icipa

lity

Mun

icipa

lity

Mun

icipa

lity

Mun

icipa

lity

Mun

icipa

lity

Bost

on

MSA

Bo

ston

M

SA

Bost

on

MSA

No

rth

Caro

lina

Suffo

lk Co

unty,

NY

19

60

1960

19

60

1970

19

70

31

23

31

115

62

Med

ian

gros

s re

nt

per

Med

ian

gros

s re

nt

per

med

ian

room

s of

al

l m

edia

n ro

oms

of

all

hous

ing

units

. re

ntal

un

its.

Equa

lized

pr

oper

ty ta

x Eq

ualiz

ed

prop

erty

tax

rate

fo

r al

l pr

oper

ty.

rate

fo

r al

l pr

oper

ty.

Educ

atio

n ex

pend

iture

s Ed

ucat

ion

expe

nditu

res

per

pupi

l. pe

r pu

pil.

2 OLS

2 OLS,

2S

LS

Med

ian

gros

s re

nt

and

med

ian

gros

s re

nt

per

med

ian

room

s of

al

l re

ntal

un

its.

Equa

lized

pr

oper

ty ta

x ra

te

for

all

prop

erty.

Ed

ucat

ion

expe

nditu

res

per

pupi

l.

2 3 OLS,

2S

LS

Med

ian

gros

s re

nt.

Med

ian

gros

s re

nt.

Equa

lized

pr

oper

ty ta

x ra

te

for

all

prop

erty.

M

unici

pal

and

educ

atio

n ex

pend

iture

pe

r ca

pita

. 1 6 OL

S

Shift

ing

of

betw

een

62

and

110

perc

ent

depe

ndin

g on

as

sum

ptio

n of

tim

e ho

rizon

an

d di

scou

nt

rate

.

Equa

lized

pr

oper

ty ta

x ra

te

for

all

prop

erty.

Ed

ucat

ion

expe

nditu

res

per

pupi

l.

2 ZSLS

, 3S

LS

‘OLS

=

ordin

ary

leas

t sq

uare

s, 2S

LS =

tw

o-st

age

leas

t sq

uare

s,

and

3SLS

=

thre

e-st

age

leas

t sq

uare

s.

Ibelow, correcting flor this simultaneity can alter the results substantially. Finally, despite the availability of general functional forms for hedonic regressions, these studies use fairly restrictive forms.

PROPERTY l-AXES, SERVICE QUALITY, RENTS, AND APARTMENT VALUES

This section analyzes the extent to which property tax differentials across communities are shifted onto renters. The model has ttio principal components: the renter’s choice of a community and the landlord’s valuaition of an apartment building. These two components <are tied together with two lothers: the community budget constraint and the determi- nants of median rent.

The Renter?; Choice of a Community

The starting pornt of our analysis IS a renter’s maximization problem Following Yinger (1982, 1985), we assume that each renter selects a community based on rent and public service level. Property taxes do no1 enter the renter’s problem directly because renters do not pay a property tax bill. The renter’s problem IS to select a composite good, housing, and a level of public services to

Maximize O(L’, H, 5) Subject to Y z= Z + P(S) H,

vvhere Z = a composite good wit~i a price of unity, H = units of housing services, 5 = quality of’ local public services per capita, )’ = renter’s income, and F’(S) = price per unit of housing services, which IS a function of 5. Note that R(S) = P(S)H equals apartment rent4

In his role as a mover, a renter chooses S indirectly through his choice of a corn munity; that is, renters select their pre- fierred quality of public services by “voting with therr feet.” Although the

nroperty tax does not appear in the renter’s problern, the phblic service level, 5, clearly depends on tt-/e property tax rate (among other things), so that property taxes can have an ihdlrect influence on P and hence on rentlers’ choices.

-The key first-order condlitions of this problem concern the chpices of Z and 5. In particular, the renter Isets

7he left s de of equation 2 is the mar- (ginal rate of substitutioti between 5 and I?, or, because Z IS the numeraire, the marginal oerlefit from S,in dollar terms. -[he right side is the marginal cost of 5, lhat IS, the Increase in the cost of hous- Ing from moving to a ccjmmunity with better publrc services.

I quation 2 15 a maximiz&tion condition for a single renter when P(S) is given. Follovving the logic of ah urban model, thts equation also can bb interpreted as cln equllibriurn condition, for the price of rental housing, assuming perfectly mobile renters. IJnless the price of rental housing varicbs across co/nmunities, that I’;~ across different levels of 5, according to eqiJation 2, some communities will be relatively attractive to rehters and rental prices will be bid up theire.’ The housing market reaches equrlibriom only when renters pay for the privilege of living in a high-service community and are com- pensated for living in a low-service community.’

A Landlord’s Valuation pf an Apartment Building

1 he second tomponent of the model IS the deterrnlnation of malrket value. Be- cause landlords do pay property tax bills, the arnount they are willing to pay for an apartment: building equals the pres-

I IS THE PROPERTYTAXA BENEFITTAX?

ent value of the net rental income it yields minus the present value of the broperty tax payments on the building. Formally, the market value of an apartment building with an expected life of N years is

N P(S)H - T V=c,

n=, (1 + l-1”

where V = market value, T = annual

ment equals the effective property tax rate, t, multiplied by house value, equation 3 becomes7

q P(S)H - tv PWH

V= r Or ’ = (r + t)’

The Community Budget Constraint

The service level and the effective property tax rate are tied together through a community’s budget constraint. By defi-

quality, C. Moreover, spending must equal revenue, which consists of property taxes and miscellaneous other revenue. Thus, a community’s budget constraint is

f = (S)(C) = t\;/ + M (tti+ M)

or 5 = - c

where v = the per capita property value in a community and M = local revenue per capita from sources other than the property tax.

Median Rent

The final component of our model is a simple supply-and-demand analysis of a

community’s median rent, the dependent variable In our empirical work. The median rent in a community is a function of the service quality and property tax rate there, and it equals the price per unit of H In the community multiplied by the community’s median apartment “size,” say I?. Before it can be applied to median rent, however, our theory must be modified, because tenants may not be perfectly mobile and because the median apartment size (unlike the size of a given apartment) may change as the price per unit of H changes8 To account for these possibilities, let us distinguish between housing bids, P, and actual housing prices, say P*. Actual rents, R”, are the product of P* and I?.

Consider the following demand and supply curves:

i& = F/,[P* - P(S)]

h, = &P* (1 - t)].

The direct impact of t on P* and R* can be found by totally differentiating equation 6 with respect to t and P” and setting the change in H, equal to the change in fi,. Let P(U) stand for the price elasticity of demand (s~pply).~ Then these steps lead to

dP* 1 p;r---Z ’ dt P* ,-t-C”

(T

and

E . dR* 1 R:+---= ‘+’

dt R” 1-t-e (T

Equation 7 IS a standard incidence re- Wit. lo The higher (or and the lower the absolute value of p,, the higher the impact of a tax increase on the price of housing and hence the greater the shifting of the tax onto tenants. If tenants are perfectly mobile, that is, if p is infinite, landlords bear the entire burden of the tax increase.

Equation 8 contains a surprise, however. If p is less than minus one, the impact of t on R* is negative. Hence, the estimated value of RF provides a test of the hypothesis that tenants’ demand for housing is elastic. This IF an important test for a theory such as ours that depends on the existence of tenant mobility.

The impact of 5 on P* and R* can be found by totally differentiating equation 6 with respect to 5 and P” and equating the two quantity changes. The results are

q dP* 1 P5 p+--- =--- dS P* , _ !l ’

E’L

where Ps = (dP/d.S)/P, and

These equations provide the key to the indirect shifting of property taxes onto rents, which occurs when property tax iincreases are used to improve service quality. Equation 9 reveals not only that this indirect shifting depends on the extent to which tenants are willing to pay for a service quality increment, as measured by P,, but also that the extent of this indirect shifting actually decreases as

v goes up. In other wotds, the supply elasticity has the opposi\e impact on indirect shifting of the prqperty tax as it has on direct shifting. This result is obtained because indirect Fhifting occurs as the demand curve, not the supply curve, is pushed up. For a givdn increase in the demand curve, the more elastic the supply curve, the smaller the increase in price.

It follows that an increase in CT can either incirease or decre ‘se landlords’ share of a property tax 4 ncrease, de- pending on whether direct shifting or indirect shifting dominates.

Equation 10 reveals thatt, unless housing supply is fixed, one canqot extract the Impact of service qualit on housing bids from a regression of rerjts on service quality without accounti~ng for variation

in mediar apartrnent “s/ze” or /?. A standard nedonic regression for apartment renls controls for fi by including as control variables all obsqrvable housing characteristics, such as the median number of rooms However, our model calls for the reduced-form imlpacts of 5 and t on R* when fi is alloweb to vary across jurisdictions, that is, wit out controlling for housing characteristi s. Results from

a standard hedonic regr ssion will be presented for comparikn purposes.”

The Complete Model

To bring direct and indirect shifting into our analysis, we must rewrite equation 4 based on actual prices, not housing bids, and recognize that chaniges in t affect 5 through tile community ~budget constraint, equation 5. The ~result is

m P” [S(t), tl H \,/ = -----,

rs t

Equation 11 reveals thati a change in the property tax affects the barket value of


an apartment building in three ways: through its impact on the denominator, through its direct impact on P”, and through its impact on 5 and hence its indirect impact on P*. Because the t in the denominator of equation 11 comes directly from the landlord’s tax payment, the first effect can be called the direct tax effect. It always is negative; for a given rental flow, higher property taxes lead to a lower market value. The other two effects, which can be called the direct and indirect rental effects, usually are positive, and can vary from one community to another.‘* Higher taxes lead to better services, for example, but the extent of the increase in services, and hence the extent of the indirect impact on rents, depends on a community’s public service costs.

The net impact of a property tax increase on the value of an apartment building depends on all three effects. At the margin, this net impact is the change in market value relative to a $1 change in the present value of the property tax stream. Totally differentiating equation 11 and the present value of the property tax stream, T/r, we find that13

dV &(I. + t) - tl -= d(T/r) t[c(r + t) + rl

where E = [(aP*/S)(aS/iW + (aP*/at)] x (t/P*), which is the elasticity of P* with respect to t, that is, the sum of the direct and indirect rental effects.

One of the terms in E is (S/at), which can be derived from the budget constraint, equation 5. In differentiating equation 5, we divide property into residential and business; we assume that residential property values are defined by equation 11 with i instead of H; and we allow adjustments in /? in response

to changes in P*. Based on the evidence in Wheaton (1984), we also assume that business rents remain constant.14 With these assumptions, we find that

as rV + t(r + t) VJT: ii = (f + t)(C - G,R,X)

where vR is residential property value per capita.”

Equation 12 describes the incidence of a change in the property tax in one community holding constant the property tax everywhere else. The first requirement for the property tax to be a benefit tax is that any increase in property taxes be fully shifted to tenants. If the expression In equation 12 is zero, that is, if asset values are not affected by a property tax increase, then this full shifting occurs. In this case, the direct and indirect rental effects and the resulting increase in asset value exactly offset the direct tax effect. If this expression is negative, which implies that the rental effects are smaller than the direct tax effect, landlords bear some of the burden of a property tax increase.

Equations 12 and 13 demonstrate that the extent to which property tax increases are shifted to tenants is not the same in every community. A community’s property tax rate, cost index, average property value, and average residential property value, all influence the degree of shifting. At a given tax rate, for example, a tax increase results in a smaller increase in service quality, and hence in a smaller indirect rental effect, in a high-cost community than in a low- cost community. This variation is a significant departure from standard incidence analysis.

The possible outcomes for a community can be summarized by the relationship between E and [t/(r + t)]. A property

tax increase is partially shifted onto tenants if E falls between 0 and [t/(r + t)], exactly shifted to tenants if E equals [t/(r + t)], and overshifted onto tenants if E exceeds [t/(r + t)]. These results indicate that tenants bear the entire burden of the tax increase when the sum of the direct and indirect rental effects, ‘which is measured by E exactly offsets the direct tax effect, which is measured by [t/(r + t)]. A property tax increase is exactly paid by landlords only if E equals <zero, which requires that tenants receive ino benefit from an increment in public *service qualrty (so the indirect rental ef- .fect is zero) and that either p = -m or IV = 0 (so the direct rental effect is zero), and it is overshifted onto landlords If E is below zero. ,4 negative E implies that a property tax increase has such a negative impact on a community’s aver-

age property value that It results in a decline in service quality that is sufficiently large not only to make the indirect rental effect negative but also to make it larger in absolute value than the direct rental effect, which always is positive.16

Although R,* and R-F can be estimated, E cannot be calculated without making an assumption about either p or CT. It will prove useful to consider two extreme cases. The ftrst case is defined by Al. = -03, that is, by a horizontal demand

curve. In this case, which is used in our earlier analysis of renter community

choice, equation 7 reveals that P: = 0. It follows from equation 8 that u = -R: and from equations 9 and 10 that P, = P; = R:/( 1 i- 0). The second case

is defined by u = ~0, that is, by a horizontal supply curve. In this case, equation 7 reveals that )P: = 1 /(l --- t); equation 8 reveals that EA. = [Rf(l -- t) -- 11; equation 9 reveals that P: = 0; and equation 10 reveals that Ps =. -- Rc/p. Substituting these results into equation 13. the definition of E and

equation 12, yields the ~burden on landlords.

With neither p nor (T in~finite and with estimates of R:' and 1 R:, we have two equations, 8 and IO, In three unknowns, P,, p, and (T. Solving equation 8 for p as a function of U, we find that

R:(l - t) -. 1 P = --~- R,* '

tT

If R: IS less than zero (a test of tenant mobility), the demand eilasticity will not be negative as required lunless CT IS greater than the absolute value of R:. With any value of u above -R: we can use equation 14 to findi Al. and equation 10 to find P, as a function of p and (T:

1-z P, = R; --: i 1 We do not simulate with equations 14 and 15, however, beta se cases ‘I and

? 2, defined above, bracket the shifting possibilities.”

Finally, note that equation 12 indncates whether landlords or renters bear the burden of a marginal change in a community’s property tax, but it does not measure the incidence of an existing tax differential. The incidence measure for an existing differential is

landlord share = V(t) - V(i)

T(t) T(i) ’

where the arguments indicate that V and T/r are evaluated at a community’s actual tax rate, t, and at/ the metropolitan average tax rate, t.lq Equation 16 requires an esttmate of the apartment


value that would prevail under the average tax rate, V(t). Our estimating procedure is presented below.

Net Benefits to Tenants

The public services considered in this paper are received by the tenant, not the landlord. For any given increase in t, the net benefits to a tenant equal the present value of the associated increase in service benefits, or (aP/as)(as/at)(H)(dt)/ r, minus the increase in the present value of rents, which is the expression in equation 12 plus one. Using equation 15, the definition of E and the change in t required to alter (tV/r) by $1, we find that this net benefit, NB, is

(rf tJ2[tPT(f- 1) -e(f)]

NB = t[e(r + t) + r]

The second requirement for a property tax to be a benefit tax is that the net benefits to tenants be zero, that is, that the payments made by tenants in the form of higher rents exactly cover the benefits they receive in the form of better services. According to equation 17, this requirement is met if p = --x) (case 1 above), CJ = 0, or E = t(o - p)/[( 1 - t)c - ~1. Under the first two circumstances, the indirect rental effect exactly equals tenant willingness to pay for the service quality increment and the direct effect is zero. In the last circumstance, the value of the service quality increment is not fully reflected in rents, but the direct effect exactly makes up the difference. Finally, note that if (T = ~0 (case 2 above), the net benefits defined by equation 17 must be negative; that is, rents must increase by more than tenants’ valuation of public services.lg

When is the Property Tax a Benefit Tax?

Equations 12 and 13 reveal that a community’s decisions affect the incidence

of the property tax. To maximize their market value, landlords would select the tax rate at which E = [t/(r + t)].‘” At this rate, any additional increase in property taxes would lead to an increase in rents that exactly offset the landlord’s increase in taxes; that is, any additional increase would be borne entirely by tenants, as required for a benefit tax.

However, renters will not in general prefer the same tax rate as landlords. As pointed out by Stiglitz (1983) and Yinger (1985) and confirmed by equation 17, perfectly mobtle tenants are indifferent to property tax increases, regardless of the tax rate.” This indifference at the margin, which also arises with perfectly immobile landlords, is the second requirement for a benefit tax. Without perfect mobility, tenants prefer to increase the tax rate until their net benefit from a further increase equals zero, that is, until the second benefit-tax condition is met. As shown earlier, their preferred tax rate is defined by E = t(a - p)/[(l - t)cr - p].

These results Imply that the property tax is unlikely to be a benefit tax.** If P= --x or if u = 0, tenants’ net benefits equal zero regardless of the local property tax rate. Thus, the two benefit-tax conditions derived here will be satisfied so long as the property tax rate is set at landlords’ preferred level. Landlords do not select the property tax rate, however, and they may have little political power. Indeed, if they are nonresidents, they do not even get to vote. Thus, the tax rate is more likely to be set at the level that renters or homeowners prefer.23 Moreover, without either perfectly mobile tenants or immobile landlords, the property tax rate cannot be set to satisfy the benefit-tax conditions for both landlords and tenants; that is, the two conditions are incompatible.24

303

ESTIMATION STRATEGY

In this section we explain our strategy for estimating both public service quality, 5, and a hedonic equation in which R* is a function of 5 and t. These steps allow us to calculate the net burden of a property tax on both landlords, equation 12, and tenants, equation 1’7. In addition, we describe our simulation procedure for finding V(f) and hence for calculating the incidence of existing tax differences, as given by equation 16.

Public Service Quality

Our procedure for estimating C was developed by Bradbury et a/. (1984) and refined by Ladd and Ylnger (199’1). It recognizes, following Bradford, Malt, and Oates (1969), that the cost of public services depends on environmental factors as well as on the cost of inputs. A city with relatively old housing, for example, must spend more to deliver the same degree of fire protection to Its cite- zens than must other cities. Thus, the procedure begins by estimating, for a Icross section of cities, the impact of environmental cost faictors on city spending, controlling for other factors that influence public service quality, such as income and tax price. A cost index for leach city can then be calculated using the actual environmental characteristics Iof the city and the estimated impacts of these factors on service costs.

To be specific, the analysis begins with a :standard demand function for local pub- ilic services, in which service quality is a function of median income, Y; tax price, which equals the median tax share, V/v, Imultiplied by the (constant) cost per unit Iof service quality, C; and intergovern- Imental aid plus other non-property-tax revenue, M. Assuming a multiplicative form for this demand function, we obtain the following expenditure equation (excluding A/1, for simplicity):2”

Cy+‘.

Following Ladd and Yinlger (1991 L the cost function is assume@ to take the form

where I is an index of iDput costs and C, represents an environmental cost factor, such as the poverty rat+ or the share of housing that is old. Substituting equation 19 into equation 118 and taking log- arithms yields a functio$al form that can be estimated with ordinary least squares. The cost index is comp$ted using equation 19; the estimated brice elasticity, y; and the coefficients of the cost variables.26 The final step id to obtain a measure of service quallity, 5, using equation 5, that is, by dividing each community’s per Its cost irldex.2’

capita expenditure by

The Hedonic Equation

The next step is to use ithe measure of service quality, 5, in a tjedonic equation of the rental housing rr/arket. We obtain R: and R: from the redults of the hedonic. We use the Box-Cox approach, which has been widely employed in the housing literature and @hich allows us to estimate the form instead 0.f

appropliia imposing

te functional it.28

The I’ncidence of Existing Property Tax Differentials

With the cost indexes and the hedonic prices, the incidence of a marginal change in the property tax rate can be calculated directly from ~equations 12 and ‘13. In carrying out these calc:ulations, we set the real initerest rate at ten percent.20

304


Calculating the incidence of existing property tax differentials from equation 16 is more difficult. This incidence depends on V(i), which is the value of V in equation 11 when t is set equal to ?. Finding V(f) is complicated, however, because it requires an estimate of the value of 5 that would exist if t were set to i, S(t), which equals the actual 5 multiplied by the ratio of revenue received with a tax rate of i to actual revenue. Revenue at t depends on the impact of t and 5 on rents, which is specified by the hedonic equation.30 This equation is solved with a simulation procedure that assumes, as does equation 13, that the average size of apartments adjusts to changes in housing price and the rental value of business property remains constant.3’ The final step in the simulation is to calculate the landlords share in equation 16 by substituting S(i) into the estimated hedonic equation.32

The Data

Data for 147 towns and cities in the Boston SMSA in 1980 are used to estimate the hedonic equation. Descriptions of the variables, their means, and their standard deviations are given in Table 2.33 The dependent variable is median gross rent, which includes actual or im- puted utility payments by renters. The explanatory variables are grouped into three categories: public service quality and the tax rate, structural characteristics of housing, and community amenities.

Service quality and tax rate

The two variables most important to this analysis are the public service level, SQUALITY, and the effective property tax rate, TAXRATE. The calculation of SQUALITY is described above. As noted earlier, the coefficient of TAXRATE provides a test of the hypothesis that tenants are mobile.34

Structural characteristics

Rents are a function of the structural characteristics of housing, as well as of public service quality. As explained earlier, these variables are left out of our principal reduced-form equations, but are presented here for completeness. RENTAGE measures the age of the rental housrng stock. As Galster (1987) points out, differences in the age of housing structures typically indicate differences in construction technology, type and efficiency of mechanical systems (plumbing, heating, and wiring), and the time over which the structure has been subject to normal wear and tear and weathering. MEDROOMS indicates the size of rental units. Older and smaller rental units are expected to command lower rents.

The fraction of buildings with two or fewer rental units, UNITS, controls for the size of apartment buildings. UTILITY is the fraction of units that include utilities as part of rent. This variable is not included to correct for the impact of utilities on rents because the dependent variable is median gross rent which, by definition, includes utility payments. In- stead, it controls for the presence of housing characteristics associated with the inclusion of utilities in rents. For example, landlords who live in their buildings may be more likely than the owners of large apartment buildings to include utilities in rent because the two- or three-family houses they occupy are not equipped with separate utility meters. Moreover, several studies have identified a positive relationship between housing upkeep and owner-occupancy. See Gals- ter (1987). Thus, UTILITY may be a good proxy for the presence of well- maintained apartments because of owner-occupancy.

Community amenities

The community amenity variables describe the character and relative location

305

TABLE 2 DESCRIPTION OF VARIABLES USED FOR E:STIMATl0N OF HEDONIC

Ec~u~moN, THEIR MEANS AND STANDARD [SEV~AT~ONS (147 MUNICIPALITIES)

--- ---- Mean

Variable Description (Standard Deviation) --~-~ _____-- -- Dependent

RENT Median gross rent in 1980, annual. $3,627 0-W

SQUALITY Index of public service quality, all municipal 756 services. (If-3

TAXRATE Effective property tax rate for residential housing 2.48 in 1980. (I .09)

Structural Characteristics MEDROOMS Median number of rooms in rental housing 4.19

structures in 1980. (0.46) RENTAGE Fraction of 1980 rental units built before 1950. 0.45

(0.15) UNITS Fraction of structures with two or fewer rental 0.65

units in 1980. (0.18) UTILITY Fraction of rental units that include utility costs as 0.77

part of rent. (0.11) Neighborhood Amenities

DENSITY 1980 population divided by square miles Iof land 2,265 area. (3,255)

RENTER Fraction of year-round housing units that are 0.27 rental units. (0.15)

POPRAT Rate of population change defined as 1980 1.11 population divided by 1970 population. (0.25)

DISTANCE Linear distance to the Boston central business 21.29 district (in miles). (10.34)

RTE128 Linear distance to Route 128 (in miles). 7.94 (6.01) ---~ ------- -...

Source: Sources include the 1980 Census of Population and Housing, Massachusetts Taxpayers’ Foundation, and the Massachusetts Department of Revenue. SQUALITY was estimated from the expenditure deternbnation model.

of communities. These variables are included in our reduced-form equations under the assumption that community characteristics do not change in response to changes in housing price. Densely populated communities, as measured by DENSITY, are likely to have higher congestion, which leads to lower rents, other things equal. Communities with a high percentage of rental housing, as measured by RENTER, tend to be more centralized, lower-income communities with more transient populations, less stable neighborhoods, and less upkeep of properties.35 Consequently, communities with a high proportion of renters are expected to have lower rents. Communities characterized by large in- fluxes of population, as measured by

POPRAT, m&y face temporary upward pressure on rents.36

DISTANC E and RTE128 depict the prox- imity of a community to the central business district (CBD) of Boston and the major beltway for the Boston metropolitan area, respectively. \Iz/e expect rents to reflect the high premium placed on short cornmuting distances. Conse- quently, rents are expected to be higher closer to thte CBD and To Route 128, both of which are employment centers.

RESlJ LTS

This sectjon presents the estimation results for the service quqlity measure and the hedonic equation for rental housing and discusses the implications of these


results for the incidence of property tax differentials.

Expenditure Determination Model

The expenditure determination model differs from that used by Bradbury et al. (1984) only in functional form. The results of the two models are similar, and therefore are only summarized briefly here.37 The town of Lincoln has the lowest cost index at 0.67, while Hull has the highest at 1.63. These figures imply that Hull would have to spend 63 percent more than the average municipality to maintain an average service quality, whereas Lincoln could obtain average- quality services even if it spent 33 percent less than average. Weston has the highest level of public service quality, while Lawrence has the lowest. Boston has the eighth highest cost index in the metropolitan area.

Note that high residential property tax rates do not necessarily lead to high public service quality. Even though Bos- ton has the second-highest property tax rate in the Boston SMSA, for example, its service quality index ranks it only in the top one-third. As explained earlier, a city’s revenue depends on its property tax base and its intergovernmental aid, as well as on its property tax rate, and its ability to translate revenue into services depends on its public service cost index.

Hedonic Equation

The hedonic regression results are presented in Tables 3 and 4. Table 3 contains the regression without structural housing characteristics, which is used for our incidence calculations. Both tables display estimates from several estimation techniques for comparison purposes. The estimates of the Box-Cox (1964) parameters indicate that linear and log-linear functional forms are not appropriate.38

The structural variables tend to have the expected signs and are statistically significant.3g The results for the community amenities are somewhat more mixed. The most interesting results are for DISTANCE and RTE128. The coefficient for DISTANCE is negative, as expected, and statistically significant, but the coefficient of RTE128 is positive and statistically insignificant, perhaps indicating the negative amenities of noise and pollu- tion close to this heavily traveled beltway.

The coefficient of SQUALITY in Table 3 indicates, as expected, that rents are higher in communities with higher- quality public services. Also, as expected, the coefficient of TAXRATE is negative, but it is not always significant. Both coefficients increase in absolute value when housing characteristics are ex- cluded.

These results could be biased because of the endogeneity of the tax rate or service quality variables. We examined this possibility using the specification tests developed by Hausman (1978, 1983), with separate tests for SQUALITY and TAXRATE. The null hypothesis of no specification error is rejected for TAXRATE but not for SQUALITY. Conse- quently, we re-estimate our Box-Cox (1964) equation with nonlinear two- stage least squares (NLTSLS) to control for the endogeneity of TAXRATE. The results of NLTSLS estimation also are re- ported in Tables 3 and 4. The coefficient for SQUALITY is still positive and significant and is similar in magnitude to that obtained from other specifications. The coefficient for TAXRATE is still negative and IS now statistically significant.4’ This result supports the hypothesis that demand is elastic. The NLTSLS results are used in our calculations.

307

TABLE 3 HEDONIC EQUATION WITHOUT STRUCTURAL CHARACTERlSTliCS

Box-dox

Variable OLS 2SLS MLE A-P NL2SiSa A-P NLZSLS”

Al

A2

INTERCEPT

SQUALITY

TAXRATE

DENSITY

POPRAT

DISTANCE

RTEI 28

R2

3,101.4 (421.9)

1.34’ (0.33)

-7,927.g (4,911 .B)

-36.0’ (2 1.4)

308.9 (223.9) -28.1’

(7.9) 6.35

(10.23) 0.22

3,796.8 (525.9)

1.07’ (0.36)

26,142‘ (9,194)

-10.9 (24.8)

193.8 (239.3) -26.7’

(8.3 5.68

(10.75) 0.22

0.63 (0.40) 0.99’

(0.44) 0.046

(CI.021) 0.265’

(0.069) -0.046

(0.030) -0.01 1

(0.014) 0.097

(0.069) -0.136’

(0.047) 01.009

(8.016) 01.22

7 ~~-

0.21 (0.88) l.loc

(0.79) 0.038

(0.026), 0.255’

(0.071)~ -0.045

(0.031) -0.008

(0.018), 0.097

(0.068) -0.127’

(0.060) 0.010

(0.016) 0.21

0.45 (0.99)

1.08 (0.66) 0.054

(0.028) 0.257’

(0.075) -0.148’

(0.047) -0.003

(0.013) 0.097

(0.071) -o.145c

(0.059) 0.011

(0.017) 0.15 -----

aNonlinear two-stage least squares 1s used to control for endogenetty Impkit In the use of th Box-Cox transformation on the dependent variable. Second degree polynomials and cro’ss products of the exogen us variables are used as instruments.

------T----.-

used as instruments. ‘Statistically significant at the 95 percent level (two-tailed); standard errors art: in parentheses,

Incidence of Property Tax Differentials

Table 5 presents frequency dtstributions and summary statistics for the net impact of a $1 property tax increase on landlords, as given by equation ‘12, and on tenants, as given by equation 17. Re- sults are presented for the two cases defined earlier: case 1 sets p == --(m and case 2 sets (7r :== x The landlord burden columns also include case IA, which assumes not only that Al. := ---x but also that u = 0.42

The table indicates that landlords bear a large share of the burden of a property tax increase in all communities in all cases. To be specific, landlords bear, on average, $0.909 for every $1 .OO property tax increase in case 1, $0.892 in case IA, and $0.844 in case 2. The minimum burden in an individual community, which appears in case 2, is $0.747. It follows that property tax rates are set

far from the levels preferred by landlords, and therefore far from the level required for the property tax on rental housrng to be a benefit tax.

The remainrng columns of Table 5 present the net impact of a property tax increase OI tenants, including the value of the resulting service q

u ality improve-

ments. As shown earlier, the assumed infinite elasticity of demand in case 1 implies that the net buirden on tenants is zero. In (case 2, the ne i burden on tenants frorn a $1 .OO increase in property taxes is $0.156, on average, with a mini mum of $0 104 and a ‘maximum of $0.253. So long as tenbnts are highly rnobile relative to land1 0 rds, therefore, the rnet burden of a property tax increase OII tenants is likely to be very small. Tel put it another way, tenant mo bility ensures that, at the margin, the property tax operates li~ke a benefit tax from the point of view of tenants. Ten-


TABLE 4 HEDONIC EQUATION WITH STRUCTURAL CHARACTERISTICS

Variable OLS 2SLS MLE

Box-Cox

A-P NL2SLSa A-P NL2SLSb

Al

A2

INTERCEPT

SQUALITY

TAXRATE

MEDIAN ROOMS

RENTAGE

UNITS

UTILITY

DENSITY

RENTER

POPRAT

DISTANCE

RTE128

R2

-464.5 (468.2)

0.81' (0.24)

-3,656 (3,610)

782.4' (101.3)

-513.2 (315.3)

-1,097.1C (356.5)

2,765.6' (322.4) -48.5'

(22.6) -408.8

(448.5) -193.5

(164.2) -22.1C

(5.9) 2.45

(7.19) 0.63

287.4 (558.7)

0.66' (0.25)

13,813' (6,701)

751.1C (104.7)

-458.2 (319.6)

-1,176.7' (324.4)

2,677.6' (334.7) -25.9

(22.1) -817.2'

(492.9) -286.3

(174.0) -21.1‘

(5.9) 0.34

(7.44) 0.62

0.78' (0.30)

1.06’ (0.40)

-0.260 (0.097) 0.163‘

(0.050) -0.023

(0.022) 0.890'

(0.124) -0.145

(0.090) -0.301C

(0.102) 0.778'

(0.091) -0.013

(0.015) -0.109

(0.127) -0.054

(0.052) -0.105c

(0.035) 0.0039

(0.011) 0.63

0.99’ (0.49)

l.ogc (0.W

-0.252 (0.098) 0.167'

(0.051) -0.021

(0.022) 0.904'

(0.129) -0.146

(0.090) -0.303'

(0.102) 0.775(

(0.091) -0.012

(0.015) -0.116

(0.127) -0.056

(0.052) -o.104c

(0.036) 0.0038

(0.011) 0.62

1.06' (0.52) 1.18'

(0.49) -0.231

(0.101) o.177c

(0.051) -0.072'

(0.030) 0.854'

(0.131) -0.119

(0.094) -0.342‘

(0.107) 0.790'

(0.093) -0.006

(0.012) -0.173

(0.130) -0.054

(0.050) -0.103‘

(0.037) 0.0047

(0.011) 0.61

aNonlinear two-stage least squares is used to control for endogeneity implicit in the use of the Box-Cox transformation on the dependent variable. Second-degree polynomials and cross products of the exogenous variables are used as instruments. bNonlinear two-stage least squares is used to control for the endogeneity implicit in the use of the Box-Cox transformation on the dependent variable, and for the correlation between TAXRATE and RENTER and the error term. Sec- ond-degree polynomials and cross products of the exogenous variables, as well as the variables used to estimate the cost index and percent elderly are used as instruments. ‘Statistically significant at the 95 percent level (two-tailed); standard errors are in parentheses,

ant mobility does not ensure, however, that the tax rate is set so that tenants receive $1 of benefits for every $1 increase in property taxes.

We cannot determine the true values of p and u. If p actually equals -co, the property tax is not a benefit tax, because t is not set at the level landlords prefer, that is, because the first condition is not met. In fact, this condition is not even close to being satisfied as land lords pay $0.909, on average, for a $1 increase in property taxes. Moreover, if 0 actually equals cc, the property tax is not a benefit tax because the two necessary conditions are incompatible. The

average burden on landlords is lower in this case, namely $0.844 for every $1 increase in property taxes, but the average net burden on tenants also is significant, namely $0.156.

Regardless of one’s assumptions about the housing price elasticities, therefore, our analysis of two necessary conditions reveals that the property tax is far from being a benefit tax both on average and in every community. The principal explanation for this outcome is that a $1 .OO property tax increase results in a service quality improvement that is worth far less than $1 .OO to tenants. Either property taxes are not set at an optimal level

309

I I

TABLE 5 BURDEN OF A $1 PROPERTY TAX INCREASE

Burden on Landlords ourden on Tenants

--- ---______

Case 1 Case 1A Case 2 Cas 1 Case 2'" Mean $0.909 $0.892 $0.844 $ .oo T $0.156 Median 0.918 0.905 0.843 0.00 0.15'7 Maximum 0.982 0.974 0.896 0.00 0.253 Minimum 0.703 0.672 0.747 0.00 0.104 -~ ------ -.- Number of Communities with a Burden between $0.00 to 0.20 0 0 0 147 138

0.20 to 0.40 0 0 0 0 9 0.40 to 0.60 0 0 0 0 0 0.60 to 0.70 0 2 0 0 0 0.70 to 0.80 5 6 9 0 0 0.80 to 0.90 41 61 138 0 0 0.90 to 0.95 78 68 0 0 0 0.95 to 1.00 23 10 0 0 0 ----- -____-- ----

Notes: Burden on landlords is defined by equation 12. Burden on tenants IS defined by equatiion 16. Case 1 is defined by p = -m; case 1A adds u = 0, case 2 is defined by CJ T z.

or the marginal benefits to other city residents, presumably homeowners, far exceed their tax costs.

The net impact on landlords of existing property tax differentials, as represented by equation 16, allso is high. (Because the benefits to tenants associated with these tax differentials are difficult to calculate, no attempt was made to calculate the associatecl net Impacts on tenants.) On average, landlords bear 45 percent of existing property tax drfferen- tials. Because of differences across communities in the cost index, in the property tax base, and in the effective property tax rate, however, the burden on landlords varies greatly across communities.43 These results provide further evidence that the property tax on rental housing is not a benefit tax.“4

Conclusion

According to the “‘new view” of the property tax, property tax differentials will be borne by renters only to the extent that renters am-e immobile relative to the other actors in the rental housing market. This paper shows that when associated changes in service quality are recognized, sorne of the burden of these

differentials falls on renters even if they are imobile. This shifting occurs because tenants are willing to pay higher rents to receive the better services purchased by higher property taxes. However, fully mobile tenants (or tenants facing a fixed housing supply) are indifferent to an increase in the property tax because the benefits from the associated service quality increase are exactly offset by an Increase in rent. Thus, the property tax

causes no distortion in the behavior of such1 tenants.

Even with fully mobile tenants, however, the property tax is notIa benefit tax unless the rent increases caused by a property tax Increase (throulgh a service quality increase) exactly equal the increase in the landlord’s tax payment. Otherwise, landlords pay some of the tax increase without receiving the service quality increment, this tax burden on landlords is a source of distortion. In our sample, a $1 .CrO property tax increase results in a rent increase of only about $0.15, on average, even with infinitely elastic housing !supply. The magnitiude of the resulting tax burden on landlords varies from one community to another. As noted earlier, for example, there is more shift-

310

I IS THE PROPERTY TAX A BENEFITTAX?

ing onto tenants in a community with low public service costs than one with high public service costs, all else equal. Nevertheless, rent increases never fully compensate landlords for tax increases, and the maximum rental increase in a single community is only $0.25. This conclusion is confirmed by a simulation of the burdens placed on landlords from existing tax differentials across communities; on average, higher rents offset only 55 percent of the tax differences paid by landlords. In our sample, therefore, the property tax on rental housing falls far short of being a benefit tax, even under assumptions that make the benefit-tax outcome most likely.

ENDNOTES

All views expressed are those of the authors alone and do not represent the views of the Department of the Treasury. We thank James Alm, Gib Metcalf, Douglas Holtz-Eakin, Harvey Rosen, several anonymous referees, and the participants in the public finance seminar at Duke University for valuable comments on earlier drafts of this paper. We are also grate- ful to Katharine Bradbury and Andrew Res- chovsky for providing data. For a recent review of this debate, and an alternative empirical test of the benefit view, see Wassmer (I 993). Balanced-budget incidence has been used to refer to a variety of incidence concepts in the literature. For clarity, we use the definition provided by Musgrave and Musgrave (1989) in which the combined effects of tax changes and benefit changes are taken into account. See also Coen and Powell (1972) and Orr (1972). Throughout this paper, the concept of housing services, H, is used for ease of presenta- tion; equivalent results could be derived, however, with a more general notation in which rents were a function of 5; of other lo- cational variables, such as distance to employment centers; and of structural housing characteristics, such as the number of rooms. In fact, our hedonic regression (described below) uses this type of general form. The application of urban model concepts to public services and rents IS discussed in more detail in Yinger (1985). To the extent that renters in different com-

munlties have drfferent Incomes or tastes, this relatronship may not be exact; housing services will stall command a higher rent in a community with better services, all else equal, but observed differences in rents may not equal any partrcular group’s valuation of public service qualrty. In this paper, we assume that the bids of exrsting residents apply to marginal changes In 5. Instead of assuming that N IS large, r can be replaced with r*, which IS the infinite-horizon discount rate that is equivalent to rate r with lifetime N. See Yinger et al. (1988). Note also that this formulation assumes that the effective property tax rate IS held constant over time, which implies that assessments are kept up to date. We do not consider the possibility that an In- dividual landlord will alter the level of H in her apartments In response to a change in P. This type of response IS possible but IS likely to be small in magnitude. These are not ordinary price elasticities; In- stead, they are elasticities of median H, not total H, with respect to price. This result can be compared to an equation in McDonald (I 993, p. 1 IO), which refers to a unit tax, instead of the ad valorem tax considered here. Technically speaking, both Jo and g can be shown to be equal the sum of an elasticity that applies to the observed components of H and one that applies to the unobserved components of H. A regressron that includes housing characteristics as control variables im- plicrtly sets equal to zero the elasticitres associated with observed components. A regression that excludes housing characteristics can be used to find the total elasticities. As explained below, the indirect effect negatrve under some circumstances.

can be

Equations 7 and 8 are derived by noting that V IS a function of t, totally differentiating V = V(t) and tV = W(t), and rearranging terms. H IS held constant, because we are looking at the taxes on a given apartment building. Wheaton (1984) finds that the rent per square foot for class-A office space in the Boston area in 1980 is not a function of the community’s property tax rate. Thus, his study applies to the same time and place as ours. McDonald (1993) finds that rents do reflect property taxes to some degree, but his study covers intralurisdictional tax differences in Chr- cage. This analysis assumes that an increase in the property tax goes entirely to higher service quality and not to a reduction in other taxes.

311

16

17

18

19

20

21

22

This assumption is reasonable in the case 01 Massachusetts, where municipalities are not allowed to levy any broad-based taxes except the property tax. In anoiher state one might want to amend the model to consider substi- tution between tax sources. Even in Massa- chusetts, there might be some tendency to substitute property taxes for licenses and fees, but we do not consider this posslblllty.

If the decline in service quality is so large that E is less than [-r/(~ + t)], then an increase in the property tax rate actually causes a decline in the property tax payment for a given apartment building. In this extreme case, the sign of the expression in equation 14 must be re- versed, since the only way to raise rnoney IS to lower Vie effective property tax rate. When E is this small, therefore, there always is over- shifting onto tenants.

If the indllect rental effect dominates the direct rental effect, increasing u results in less shifting onto tenants, and setting (T = -I?: results in ihe maximum degree of shifting This value of CT implies that ,LL = --JJ (case 1). If the direct rental effect dominates, setting u = x (case 2) resullr in the maximum degree of shifting onto tenants.

We use the average property tax rate weighted by the share of property value In each community. 1 his formulation is preferred to a simple average because the weighted average will reflect the share of capital In each community.

In this case, E .= t,‘(l - I) = tP:. Plugging this expression Into equation 17 proves the result in the text.

This is easier said than done, E Itself I S a complicated nonlinear function of t and an ex- plicit expression for t cannot be derrved from our equattons.

This staternent does not imply that tenants are indifferent across communities. A tenant selects a community where rents are determined by people in his Income-taste class, that is, where the impact of services and taxes on rents equals his own willingness to pay. See Ylnger (1985). This is, of course, the heart of the benefit-tax idea, namely that taxes serve as a price when people shop for a community.

The two conditions presented here are neccs- sary but not sufficient for the property tax to be a benefit tax. These conditions apply to property tax differentials, not to the average property tax rate in a metropolitan area. Even If the two conditions In the text were met, the prope-ty taxes associated with the aver.

age rate might be paid by landlords, which is another violation of the benefit-tax notion.

Net benefits to tenants from a tax increase equal their willingness tq pay for the associated service quality increment minus the In- crease In rents. The burden on landlords equals their increased tai payment, $1, minus the lnctease in rents. Hehce, the difference between net benefits to ~tenants and net benefits to landlords provideis a benefit-cost test for a tax increase; if this difference IS negative, the costs exceed the benefits. It is ironic that with mobile tenants,, the benefit-cost test will be met only If the tqx rate I S set at the level landlords prefer-even though landlords do not receive any of th benefits. This benefit-cost test applies to b 4 nefits to tenants, not benefits for all residents ~of a community; costs might exceed benefits provided to tenants but be less than behefits provided to all reslden ts.

Ttle two expressions for E cannot both be sat- lsfied with a positive (T aind a negative ,U unless (1 - ~)/r 2 t < (1 + r), which places t far above any observed ialues. The equality applies If and only If (T T =.

For detailed discussions 9f expenditure equations, see lnman (1979), IRubinfeld (1987), or Ladd and Ylnger (1991).

First, the price elasticity I N S used to obtain the CY,‘s (/.e, CY = PJ(r + 1)). Then the cost In- dex, C, I S computed directly from equation 151

This fotmulation assume$ that the error term In the expenditure regre$sion does not reflect management quality. If if did, then ,the service quality medsure would e biased upward in poorly managed cities a

1 d downward in effl-

cient cities To deal with this problem, Iwe also uslzd an alternative cn easure of service quality, namely predicted expenditure (based on our expenditure regr@sion) divided by cost. This alternative me sure yields essentially &ntlcal results. We con

1 lude that our service

quality melsures are no, subject to this bias. Note also that, due to data limitations, this paper follows Bradbury $t al. (1984) by com- puting a general index f$r all services. Ladd and Yinger (1991) are able to compute a cost index fgr general expenditures and separate cost indice’s for police arid fire services.

Recent studies relying or) the Box-Cox (1964) method Include Goodm ‘n (1978, 1988) and Halversen and Pollakow ,” ki (1981). We (apply the Bo:c-Cox transforma ion to both the de-

r pendent and sdme of the explanatory vari- J ables, wtth different val es of h for each side

of the equation. This approach has the advan-

I IS THE PROPERTY TAX A BENEFIT TAX?

tage that widely used log-linear and semi-log models are special cases of our procedure. Box-Cox parameters can be obtained through maximum likelihood estimation (MLE) or nonlinear two-stage least squares (NLTSLS). See Spitzer (1982). Although MLE is most fre- quently used, Amemiya and Powell (1981) recommend NLTSLS because MLE estimates generally are inconsistent because of the endogeneity introduced into the model from the use of the Box-Cox transformation of the dependent variable. Finally, Spitzer (1984) shows that the t-statistics for the Box-Cox MLE are not scale invariant. To solve this problem, we follow Spitzer by dividing each explanatory variable to which the Box-Cox transformation is applied by its geometric mean.

2g As explained in Yinger et al. (1988), this approach states rents in real terms so one must use a real interest rate. They estimate that the real rate was three percent around 1970; real interest rates appear to have increased since then and, in the case of rental housing, should include maintenance, depreciation, and perhaps other factors, although in a model of changes income taxes cancel. A reasonable real rate for 1980 therefore probably would be between five and seven percent. We use a somewhat higher rate to increase the chance that the property tax will be a benefit tax; raising the rate raises the share of the tax borne by tenants. The effect of the interest rate is small, however, and our average burdens on landlords and tenants change by only a couple percentage points when the rate is lowered to five percent.

data set has been supplemented with variables from the 1980 census and other sources.

34 In Massachusetts, 1980 tax payments were

3o The hedonic without structural characteristics is used so that the simulated changes in 5 and t affect both prices and quantities of H.

3’ In addition, we assume that changes in P and H occur in both the rental and sales markets. An appendix with full explanation of our sim- ulations, as well as the simulation program, is available upon request from the authors.

32 For this step we use the hedonic equation with structural characteristics to hold H constant to the extent possible with our data and we ignore the estimated effect of t on rents. Because the coefficients of 5 still reflects changes in the unobserved components of H, this approach should be regarded as an approximation that holds exactly only if p = --m and (T = 0. In this case, the estimated impact of 5 on rents only reflects changes in prices and the direct effect of t on rents is zero. (Hence, one must assume that the estrmated effect reflects some omitted nontax variable).

33 Many of the variables used here were origi- nally collected for Bradbury et al. (1984). This

Factors that alter the median apartment srze, including housing price, may alter the percentage renter in a community. We tested for this possibility with the test developed by Hausman (1978, 1983), using percent elderly as an instrument, and found that percent renter IS indeed endogenous. As a result, we treat percent renter as endogenous In the full hedonrc equation and interpret It as a deter- mrnant of medran apartment size to be ex- cluded from the hedonic equation without structural characteristics. We also discovered, however, that Including percent renter in the hedonic regression without structural charac- terrstics has virtually no impact on our incrdence results.

The vacancy rate also was included in an early versron of the model. Since the hedonic equation IS a reduced form of the supply and demand sides of the rental housing market, the vacancy rate might provide an indication of the availability of rental housing. However, the coefficient for this variable was close to zero and statistically insignificant. Dropping

313

37

38

39

10

11

this variable did not affect the results of the model. Three comrnunrties in the Boston area had some form of rent control. In order to control for the effects of rent control polrcres on rents, we included a dummy vanable for these communities and also deleted them from the zstrmation, but neither approach affected the results of the model Our final re suits Include these comrnunrtres and exclude the dummy variable.

Unlike Brzdbury et al. (1984), we use a multr- plicative functional form, Include 1980 population as an explanatory vanable, arld adjust the cost Index to account for the prrce elastrc- ity. The last step rarses the variance of the cost index Detailed results of our expenditure model are available upon request.

Followrng standarol practice, variables expressed as proportions (MEDROOMS, REN- SAGE, UNITS, UTILITY, and RENTER) are not transformed for the Box-Cox procedure. Note also that ihe MLE estimate of A,, the Box-- Cox (1964) parameter tar the dependent variable, differs somevvhat from the NLTSLS estimates. Th? estimate for h:, the Box--Cox estr- mate for the explanatory variables, is not very sensrtrve to the estrmatron approach, but does change somewhat when, as explaIned below, we control for the endogenerty of RENTER and TAXRATE.

negative because higher~ taxes are correlated with unobserved managbment inefficiency However, our correction’ for this problem, explained In endnote 27, has little impact on this coefficient. This negattve sign also could reflect either the fact th d t renters care about property t,rxes for some reason other than their connection to service levels or a correlation between property t$xes and some variable ornrtted from our epuation. We have no way to determrne whether either of these possrbr It/es applies. We blso asked whether factors that might affecl mobilrty affect the coefficrentrp of SQUALITY and TAXRATE. Inter action terms with Income, poverty, percent black, and percent elderly all were insignrfr- cant.

UTILITY has a posrtrve impact on rents and IS statistrcally significant, but this result IS biased if UTILITY and RENT are simultaneously determined. Ths type of simultaneity could arise If, for example, high-rent apartments tend to have utrlrtres inclucled 111 tent, while low-rent apartments do not. We investigated this SI- multaneity using Hausm,an tests (1978, 1983) and founo no evdence that It exrsts. Variables indicating the general use of particular types of equrpment used for heatrng houstng units or water, or the use of various types of air condrtionrng units were used as additional In- struments for these tests.

I2 Ths subcase 1s contradicted by the rnegative sign of the tax variable unless one assumes that this sign reflects omitted variables, not a demand response to higher housing prices.

” Nine communities have C landlord share above 100 percent and !I6 have negative laidlord shares; that is, jherr service and tax differences from the average community have a very arge positrve imp

a ct on rents.

” These results may have i placations for the r overall bal,rnced-budget incidence of the

property tax. Courant (1 77) shows that the incrdence of tax deviatio 1 s may not cancel out becausle of nonlineatities in the produc- tion functron for real estate This paper shows that the b;rlanced-budgeT incidence of property tar. differentials may Influence the average rncrdence because of the dependence of differentral rncrdence on community characteristics. If landlords bear a~relatively large share of the tax rn communiti ‘s with above-average rates, which we find to

1

e the case, then the burden of the average p operty tax rate un- derstates the average bulrden of the property tax on landlords.

REFERENCES Equation 5 provides gurdance for choosrng the additronal instruments required for these tests. First, the same set of variables deter- mines 5 and t Seclond, only variables relevant for 5 and 1 that dcl not Influence rents should be rnstruments Accordrngly, our instruments were per c-ap/ta income, equalized value pet cap&, the cost Index, state ard per cap&~, federal ge,?eral revenue sharing per cdptd, other federal ard per cap&, aid to regional school districts per cdpitd, and per cdpltd local revenue from nonproperty tax sources The results from these tests are available upon request.

Amemiya, Takeshi and Ja’mes L. Powell. “A Comparison of the Box-Co k Maximum Likelr- hood Estimator and the Non-Linear Two-Stage I east Squares Estimator.” I urndl of fconomet- ncs 17 (December, 1981): ! Sl -81,

Box, G. and 0. Cox. “An Analysis of Transfor- mations.” lourndl of the R+3/ Stdtatml Society, Sener; 5, 26 (196,4) 2 1 I-48.

Bradbury, Katharine, Heldn Ladd, Michael Perrault, Andrew Rescho

t

sky and John Yin- ger. “State ArId to Offset Fi cal Disparitres across Communitres.” National Tad /ourna/ 37 (June, 1984) 151-713.

The coeffrcrent of TAXRATE also might be Bradford, David F., R. A. (vlalt, and Wallace


E. Oates. “The Rising Cost of Local Public Ser- vices: Some Evidence and Reflections.” National Tax Journal 22 (June, 1969): 185-202.

Coen, Robert M. and Brian J. Powell. “Theory and Measurement of the Incidence of Drfferen- tials Property Taxes on Rental Housing.” National Tax /our& 25 (June, 1972): 2 11-6.

Courant, Paul N. “A General Equilibrium Model of Heterogeneous Local Property Taxes.” Journal of Public Economics 8 (1977): 3 13-27.

Dushansky, Richard, Melvin Ingber, and Ni- cholas Karatjas. “The Impact of Property Taxa- tion on Housing Values and Rents.” journal of Urban Economics 10 (September, 1981). 240- 55. Fischel, William A. “Property Taxation and the Tiebout Model: Evidence for the Benefit View from Zoning and Voting.” Journal of Economic Literature 30 (March, 1992): 171-7.

Galster, George C. Homeowners and Neighbor- hood Reinvestment. Durham, NC: Duke Univer- sity Press, 1987.

Goodman, Allen C. “An Econometric Model of Housing Price, Permanent Income, Tenure Choice, and Housing Demand.” Journal of Urban Economics 23 (May, 1988): 327-53.

Goodman, Allen C. “Hedonic Prices, Prrce In- dices, and Housing Markets.” Journal of Urban Economics 5 (October, 1978): 471-84.

Halversen, Robert and Henry 0. Pollakowski. “Choice of Functional Form for Hedonic Price Equations.” Journal of Urban Economics 10 (July, 1981): 37-49.

Hamilton, Bruce. “Zoning and Property Taxes In a System of Local Governments.” Urban Studies 72 (June, 1975): 205-l 1.

Hausman, Jerry. “Specification and Estimation in NonLinear Simultaneous Equations.” In Hand- book of Econometrics, edited by Griliches and In- triligator, vol. 2, 391-445. New York: North- Holland, 1983.

Hausman, Jerry. “Specification Tests in Econo- metrics.” Econometrica 46 (November, 1978): 1251-72.

Heinburg, John D. and Wallace E. Oates. “The Incidence of Differential Property Taxes on Urban Housing: A Comment and Some Further Evidence.” National Tax Journal 23 (March, 1970): 92-8.

Hyman, 0. N and E. C. Pasour, Jr. “Property Tax Differentials and Residential Rents in North Carolina.” National Tax Journal 26 (June, 1973): 303-7.

Inman, Robert. “The Fiscal Performance of Lo- cal Governments: An Interpretative Review.” In Current Issues in Urban Economics, edited by P.

Mieszkowski and M. Straszheim, 270-321. Baltr- more John Hopkrns Unrversrty Press, 1979.

Ladd, Helen F. and John Yinger. America’s Ailing Cities: Fiscal Health and the Design of Ur- ban Policy, updated ed. Baltimore: John Hopkins Press, 199 1

McDonald, John F. “Incidence of the Property Tax on Commercial Real Estate: The Case of Downtown Chicago.” National Tax /ourna/ 66 (June, 1993): 109-20.

Mieszkowski, Peter and George R. Zodrow. “Taxation and the Tlebout Model.” Journal of Economrc Literature 27 (September, 1989): 1098-146

Musgrave, Richard A. and Peggy B. Mus- grave. Public Finance: In Theory and Practice, 5th Edition. New York: McGraw Hill, 1989.

Orr, Larry L. “The Incidence of Differential Prop- erty Taxes on Urban Housing: Reply.” National Tax Journal 25 (June, 1972): 2 17-20.

Orr, Larry L. “The lncldence of Differential Prop- erty Taxes on Urban Housing: A Response.” Na- t/on& Tax Journal 23 (March, 1970): 99-101.

Orr, Larry L. “The lncldence of Differential Prop- erty Taxes on Urban HousIng.” National Tax Journal 2 I (September, 1968): 253-62.

Rubinfeld, Daniel L. “The Economics of the Lo- cal Public Sector.” In Handbook of Public Eco- nomics, Vol. 2, edited by Alan Auerbach and Martin Feldstein, 1-76. New York: North- Holland, 1987.

Spitzer, John J. “Variance Estimates in Models with the Box-Cox Transformation: Implications for Estrmatlon and Hypothesis Testing.” Rev/ew of Economics and Statistrcs 66 (November, 1984): 645-52.

Spitzer, John J. “A Primer on Box-Cox Estlma- tlon.” Revrew of Economics and Sraristlcs (May, 1982) 307-13.

Stiglitz, Joseph E. “The Theory of Local Public Goods Twenty-Five Years after Tlebout: A Per- spective.” In Local Provisron of Public Servrces: The Trebout Model After Twenty-Five Years, edited by George 5 Zodrow, 17-54 New York: Academic Press, 1983.

Wassmer, Robert W. “Property Taxatron, Prop- erty Base, and Property Value: An EmpIrical Test of the New View.” National Tax Journal 66 (June, 1993): 135-60.

Wheaton, William C. “The Incidence of Inter- Junsdlctronal Differences In Commercral Property Taxes ” National Tax Journal 37 (December, 1984): 515-27.

Yinger, John. “Inefficiency and the Median Voter Property Taxes, Capitalization, Heteroge- neity, and the Theory of the Second Best.” In Perspectrves on Local Public Frnance and Public

Policy, edited by John M. Quigley, vol. 2, 3-30. Yinger, John, Howard S. #loom, Axe1 Greenwich, CT: JAI Press, 1985. Bijrsch-Supan, and Helen IF. Ladd. Property Yinger, John. “Capitalization and the Theory of 7bxes and House Values: T/-/e Theory and Estima- Local Public Finance.” louma/ of fol/t/ca/ Econ- tion of Intraprisdictional Prc@erty Tax C’apitaliza- omy 90 (September, 1982) 917-43. Con. New York: Academic #ress, 1988.

is the property tax a benefit tax? the case of rental · property tax rate for all property....

Documents