Finance and Economics Discussion Series
Divisions of Research & Statistics and Monetary Affairs
Federal Reserve Board, Washington, D.C.
The Connection Between House Price Appreciation and Property
Tax Revenues
Byron F. Lutz
2008-48
NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary
materials circulated to stimulate discussion and critical c omme nt. The analysis and conclusions set forth
are those of the authors and do not indicate concurrence by other members of the research staff or the
Board of Governors. References in publications to the Finance and Economics Discussion Series (other than
acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.
The Connection Between House Price Appreciation and Property Tax Revenues
September 12, 2008
Byron F. Lutz
Federal Reserve Board of Governors*
Research Division
20
th
and C Streets, NW, Stop #66
Washington, DC 20551-0001
*The views expressed are those of the author and do not necessarily represent those of the Board
of Governors or other members of its staff.
2
ABSTRACT
This paper explores two aspects of the connection between property tax revenues and house
prices. First, I estimate the elasticity of property tax revenues with respect to house prices. This
elasticity does not necessarily equal 1 as governments may adjust effective tax rates to offset
changes in property values. Second, I examine the timing of the relationship. Institutional
features of the property tax make it unlikely that changes in house prices will immediately
influence tax revenues. The results suggest that the elasticity eventually equals 0.4 and that it
takes three years for house price changes to impact tax revenues.
3
I. Introduction
In the wake of the real estate boom in the first half of this decade, the property tax
revenues of local governments soared, growing 50 percent faster from 2001 through 2005 than
they did from 1996 through 2000.
1
More recently, the downturn in the housing market has
generated concern that property tax revenues will slow or decline.
2
Although many analysts and
commentators have noted the connection between house prices and property tax revenues (e.g.
National League of Cities 2007), there is only limited research on the precise nature of the
relationship. This paper provides evidence on two aspects of the relationship. First, the paper
addresses the question: “when house prices rise, how much do property tax revenues rise?” This
can be viewed as assessing the magnitude of the relationship. Second, the paper addresses the
question: “when house prices increase, how long does it take for property tax revenues to
increase?” This can be viewed as assessing the timing of the relationship.
Understanding how the evolution of house prices influences property tax revenues is
important for at least three reasons. First, the tax plays a central role in financing local public
goods in the U.S. Property taxes account for around three-fourths of local government tax
revenue and a quarter of total local government revenue. They are particularly important for
education as they provide approximately 95 percent of tax revenue for independent school
districts (Evans, Murray and Schwab 2001). Given the magnitude of the tax and the fact that
most local governments must balance their budgets, fluctuations in property tax revenue – driven
by changes in property values – would be expected to influence local government spending
decisions.
Second, the connection between property tax revenues and real estate values likely
influences the ability of the state and local government sector as a whole to weather fiscal crises.
During the state fiscal crisis of 2002 – 2004, localities responded to cuts in state education aid by
increasing property tax revenues in order to prevent cuts in education budgets (Dye and
Reschovsky 2008). Their ability to do so was likely a function of the strong state of the housing
1
Author’s calculation based on the Census Bureau’s State and Local Government Finance Data. The average
annual percent change in property tax revenue from 1996 to 2000 was 4.1 percent; from 2001 to 2005 it was 6.2
percent.
2
For example, in late 2007, the Wall Street Journal noted that “California could lose $2.96 billion [in property tax
revenue] over the next several years because of the housing bust” (Merrick 2007).
4
market at that time. Localities may not be well positioned to offset reductions in state funding
during the current period of slow economic activity given the softening of house values.
Finally, the relationship between the housing market and property taxes may impact the
political viability of the property tax. The share of income devoted to the property tax has risen
sharply in recent years (see figure 1), likely due, in part, to the housing boom,
3
and this appears
to have generated a political backlash. Several states have either enacted, or seriously
considered, significant reforms of their property tax in recent years.
4
Although past research has examined the effect of home price appreciation on property
tax revenues in specific states (e.g. Bloom and Ladd 1982; Cornia and Walter 2006; Dye,
McMillen and Merriman 2006; Ladd 1991), I am unaware of any systematic studies conducted
on the national level. The lack of previous research may reflect the widely held view that the
property tax is a stable revenue source. Indeed, the relative stability of the property tax over the
course of the business cycle is often cited as one of the primary virtues of property taxation (e.g.
Brunori 2003; Giertz 2006). The recent nationwide housing market run-up and subsequent
weakness, however, raises the possibility that the property tax will not be as stable going forward
as it has been in the past.
The topic of this paper has parallels with the literature on the marginal propensity to
consume (MPC) out of housing wealth (e.g. Carroll, Otuska and Slacalek 2006; Case, Quigley
and Shiller 2005; Lehnert 2004; Skinner 1996). This literature seeks to understand how changes
in housing market wealth influence personal consumption decisions. Although the literature is
far from conclusive, MPC estimates are generally around $0.03 – i.e., every additional dollar of
housing wealth leads to an additional 3 cents of consumption. If local public good decisions are
viewed through the lens of the median voter model (Black 1948), this paper can be viewed as the
public goods analogue to the housing market MPC literature (which examines private goods
consumption). When the median voter experiences a wealth shock due to an increase in his
home value, he will vote to increase his consumption of public goods by his marginal propensity
to consume public goods out of wealth. Although there is a large literature on the marginal
propensity to consume public goods out of income – it is generally thought to be equal to around
3
The upturn in the ratio of property tax revenue to personal income starting in 2001, visible in figure 1, is
attributable to faster property tax revenue growth, as well as somewhat slower income growth.
4
For example, in 2008 Florida and Indiana enacted significant property tax limitations and the Georgia state
legislature seriously considered eliminating the tax.
5
5 to 10 cents per dollar of income (Hines and Thaler 1995) – I am unaware of any work on the
public goods MPC out of wealth. The results of this paper can be interpreted as providing such
estimates, subject to a significant caveat: Increases in the value of residential real estate may
increase the share of residential property in the tax base relative to commercial and industrial
property and thereby increase the median voter’s tax price – i.e. the median voter may be
required to fund a higher percentage of public expenditures at the margin. The positive price
shock may partially offset the positive wealth shock, suggesting that the estimates in this paper
should be viewed as lower-bound estimates of the public goods MPC out of housing wealth.
The remainder of the paper is organized as follows. Section II provides background
information; Section III presents the empirical estimates of the connection between house prices
and property taxes, and Section IV concludes.
II. Background
The property tax is assessed on the value of residential real property (i.e. personal real
estate), commercial, business and farm real property, and personal property (e.g. automobiles).
Residential real property, the focus of this paper, accounts for approximately 60 percent of
taxable assessments and is the largest component of the tax base by a significant margin;
commercial, industrial and farm property account for around 30 percent and personal property
accounts for less than 10 percent
5
.
There is significant heterogeneity in the administration of the tax across jurisdictions – a
“bewildering array” of different institutional features (Giertz 2006). Abstracting from this
heterogeneity, property tax revenue can be defined as being equal to the effective tax rate times
the market value of property
*
R
V
τ
=
(1)
where R is property tax revenue; τ is the effective tax rate (which should be distinguished from
the statutory rate), and V is the market value of taxable property. When the market value of
property increases, tax revenue will mechanically increase. Policy makers, however, do not have
5
Author’s calculations based on the 1987 and 1991 Census of Governments’ accounting of the assessed value of
taxable property. Residential real property includes vacant platted land.
6
to accept this mechanical increase; they may choose to offset some or all of the mechanical
change by adjusting the effective tax rate.
6
The change in tax revenue is therefore equal to the
sum of the mechanical and policy offset components:
* *
R
VV
τ
τ
Δ= Δ+Δ (2)
mechanical policy offset
The first question addressed by this paper, “when house prices increase, how much do
property tax revenues rise?”, can be viewed as estimating the average magnitude of the policy
offset. If there is no policy offset, then the elasticity of property tax revenue with respect to
house prices will equal 1 – a one percent increase in house prices will generate a one percent
increase in property tax revenue – and if there is complete policy offset, the elasticity will equal
0. If there is partial policy offset, the elasticity will range between 0 and 1. Table 1 displays the
average annual percent change in property tax revenues, house prices and the effective property
tax rate from 2000 to 2005 – a period of rapid house price appreciation. Although property tax
revenues grew at a brisk pace over this period, they did not rise as quickly as home values.
Policy makers offset some of the mechanical increase by reducing the effective tax rate.
The second question addressed by this paper, “what is the timing of the relationship
between house price appreciation and property tax revenue?”, is motivated by four institutional
features of the property tax likely to generate a delay between changes in the market value of real
estate and corresponding changes in property tax revenues. First, the property tax is assessed in
an inherently backward looking manner, as the current year’s taxes are based on the assessed
value of property in the previous year. This feature of the tax suggests that, at a minimum,
property tax revenue will respond to house price changes with a lag of at least a year. Second,
assessed values often lag market values. In some jurisdictions this occurs by legal mandate. For
example, Maryland reassesses once every three years, and increases in the taxable value of
property are phased in, in equal increments, over a three year period (Bowman 2006). Third,
most states have some form of caps and/or limits on property tax rates, tax revenues or taxable
6
The effective tax rate can be adjusted by altering the statutory tax rate or by altering the way in which taxable
property is assessed for tax purposes.
7
assessments. During periods of rapid house price growth, these limits will prevent assessments
or revenues from growing at the same pace as market values. Michigan provides an example; it
has an assessment growth limit of the lower of five percent or the rate of inflation (Anderson
2006). When the rate of house appreciation exceeds this limit, assessments will rise at a slower
rate than market values and a ‘stock’ of untaxed appreciation will develop. Assessments will
catch-up to market values only when house price growth slows below the limit and the ‘stock’ of
untaxed appreciation is incorporated into taxable assessments. Finally, the tax is poorly
administered in some locales and assessments do not occur in a timely fashion – for example,
Utah once went 20 years without conducting meaningful reappraisals. This “poor”
administration may be intentional in some jurisdictions, particularly those that elect tax
assessors. Such officials may have incentives to delay incorporating changes in market value
into assessed values (Cornia and Walters 2006).
Figure 2 displays the annual growth rate of property tax revenue and house values, as
measured by the OFHEO index, from 1988 to 2008.
7
The growth rate of both series declined
through the early portion of the period. Although house price appreciation reached a trough in
mid-1991, property tax revenue growth did not bottom out until the start of 1995, implying that
property taxes track real estate prices with a considerable lag. In the more recent period, house
price appreciation began falling around the start of 2006, but property taxes have continued to
rise at a reasonably strong pace through the end of 2007.
III. Empirics
Two empirical approaches are used in this paper to estimate the relationship between
house values and property taxes. The first, referred to as the time-series approach, uses quarterly
data aggregated to the national level, and the second, referred to as the micro-data approach, uses
annual data on individual governments.
Time-Series
The estimating equation for the time-series approach is:
7
The property tax data is obtained from the Census Bureau’s Quarterly Summary of State and Local Tax Revenue.
The volatility of the property tax series is partially explained by the fact that it is derived from a relatively small
sample.
8
20
1
ln ln
titit
i
R
VXe
αβ δ
−
=
Δ=+Δ +Δ+
∑
(3)
where R
t
is the National Income and Product Account (NIPA) measure of total, national property
tax revenue at time t; V
t
, the measure of house prices, is the OFHEO purchase-only house price
index; and X
is a vector of twenty lags of the log of personal income as measured in the NIPA.
8
See the Data Appendix for additional information on the data sources.
The β
i
coefficients are the parameters of interest – they measure the elasticity of property
tax revenue with respect to house prices. The magnitude of the elasticity, at a point in time, is
determined by the cumulative sum of the β
i
coefficients up to that period. For example, the
elasticity four quarters after a change in house prices is equal to the sum of β
1,
β
2,
β
3,
and β
4.
The
timing of the relationship between house prices and property taxes is determined by the evolution
of the β
i
coefficients over time.
Figure 3 conveys both the magnitude and timing of the connection between house prices
and property taxes implied by estimating equation (3). It plots the sum of the β
i
coefficients from
the first quarter following a house price change through the 20
th
quarter following the change
(i.e. the end of the fifth year following the change) and the associated 90% confidence interval.
With regards to timing, a change in house values is estimated to essentially have no effect on
property tax revenues in the first two years following the change. The effect of the change in
house prices on property taxes then phases in over the course of the third year and, smoothing
through some oscillation in the estimated effect, holds steady in years four and five. The average
elasticity in years three through five is 0.4 (the dashed line in the figure) – a 10 percent increase
in home values produces a 4 percent rise in property tax revenues. One can infer from this result
that policymakers, on average, choose to offset approximately 60 percent of home value
increases by reducing effective tax rates.
These estimates must be interpreted with caution. Housing prices incorporate both
current and expected future economic conditions. An area which receives a positive economic
shock may simultaneously increase demand for local public goods and bid up house prices, both
8
The full set of data is available from 1975Q1 to 2008Q1. Given the lag structure of equation (3), this allows for
estimation over the 1980Q1 to 2008Q1 period.
9
in response to expected income gains.
9
Such a situation would generate a spurious correlation
between house prices and property taxes. Although this and other endogeneity concerns may be
mitigated by controlling for changes in income (the X vector in equation (3)) and the more
extensive set of control variables employed below, the fact remains that virtually any unobserved
factor that influences housing prices may also alter demand for public goods consumption.
Accordingly, the results of this paper should be viewed as establishing the magnitude and timing
of the correlation between house price appreciation and property tax revenue, not as providing
strictly causal estimates of the relationship.
10
The timing of the relationship between house
prices and property taxes does, however, provide some assurance that the estimated correlation is
not purely spurious. An unobserved shock which altered both house values and public goods
demand would likely generate a change in property tax revenue relatively quickly. It seems
unlikely that a change in public goods demand would take three years to manifest itself.
Table 2 presents additional time-series estimates from a model using annual data, as
opposed to quarterly data. Column (1) presents the basic specification. The sum of the
coefficients on the change in house prices from years t-1 and t-2 is small and imprecise,
suggesting that house prices have little impact on property tax revenues in the first two years
following the change in prices. The sum of the coefficients from years t-1 through t-3, however,
is equal to 0.44 and can be distinguished from zero. The annual model therefore indicates that it
takes 3 years for a change in house prices to influence growth in property tax revenue and that
the cumulative elasticity is equal to around 0.4 – conclusions very similar to those produced by
the quarterly model.
The estimate becomes more precise when the first two lags of the change in house prices
are omitted (column (2)), likely because changes in house prices display a high degree of serial
correlation and the resulting multi-collinearity reduces the precision of the individual estimates.
The increased precision makes it feasible to control for additional variables. Column (3) adds
the lagged change in property tax revenues to the specification in order to control for persistence
in the growth rate of property tax revenue. Column (4) controls for changes in the stock of
9
For instance, suppose a locality receives a new manufacturing plant. This event would increase incomes and
population. Housing prices would be bid up and demand for public goods would increase (in response to the higher
incomes and the need to build infrastructure for the new residents).
10
The ideal strategy would be to find an instrument for housing prices. Finding an appropriate instrument at the
national level – i.e. one that is correlated with house prices, but uncorrelated with public goods demand – is a
daunting, perhaps impossible, task.
10
residential property. The rate of residential construction tends to be elevated during periods of
strong growth in house prices and growth in the housing stock will increase property tax
revenues independently of house price increases. It is therefore possible that the estimated
elasticity between house prices and property tax revenues spuriously reflects (or partially
reflects) changes in the stock of housing. Changes in both the price and stock of commercial
property are also included as controls because they too can influence property tax revenue and
may be correlated with changes in the price of residential real estate. Column (5) controls for
changes in the tax bases of the other major state and local government taxes – personal income
(personal income tax base), personal consumption (sales tax base), and corporate profits
(corporate income tax base). This is to control for the possibility that state and local
governments substitute property tax revenue for other forms of tax revenue during times of fiscal
stress (Dye and Reschovsky 2008). None of these control variables substantively alter the
results. No attempt is made to control for property tax caps or limitations. The effectiveness of
these limitations is already the subject of a large literature (e.g. Dye and McGuire 1997;
Merriman 1987; Fisher and Gade 1991; Preston and Ichniowski 1991). The β
i
estimates capture
the average response of policy makers, both those making unconstrained decisions and those
making decisions constrained by caps and/or limits.
11
Micro-Data
The micro-data regressions utilize a panel of data on individual local governments that
directly raise property tax revenues. The micro-data provides a much greater range of variation
in house prices than the time-series and this allows for testing the hypothesis that local
governments respond differently to unusually large house price increases, unusually small house
price increases and negative house price changes, relative to typical sized house price increases.
For instance, it is possible that governments will offset a greater proportion of very large house
price increases because tax increases beyond a certain size are politically unacceptable. A
drawback of the micro-data approach is that it must be executed over a shorter time horizon than
the time-series estimates (the data permits running the model from 1985 – 2005, compared to
1976 – 2007 for the annual time-series estimates).
11
The enactment of tax limits and caps can be viewed as an endogenous choice to lower the elasticity of property
tax revenue with respect to property values. The β
i
coefficients capture the effect of these decisions, including the
decision of some voters not to enact such limits.
11
The property tax data is drawn from the Census Bureau’s State and Local Government
Finance Data, and the OFHEO all-transactions index is used to measure housing prices. Both
the state-level and the MSA-level version of the OFHEO index are used. Use of the state-level
index provides a larger sample (local governments located outside of MSAs can be included in
the sample) while use of the MSA-level sample provides a greater range of house price changes.
See the Data Appendix for additional information on the data sources.
The micro-data estimation equation is
20 20
,,,,
11
20 20
arg , , , , ,
11
ln ln ln *( )
ln *( arg ) ln *( )
is t i s t g small i s t g
gg
l e i s t g decrease i s t g is t
gg
RV Vsmall
Vle VdecreaseXe
αβ β
ββ δ
−−
==
−−
==
Δ=+Δ + Δ +
Δ+Δ +Δ+
∑∑
∑∑
(4)
where i indexes local governments located in state or MSA s, small is an indicator variable
equaling one if the change in house prices (
,
ln
s
tg
V
−
Δ
) falls in the bottom quartile of house price
increases, large is an indicator variable equaling one if the change in house prices falls in the top
quartile of house price increases, and decrease is an indicator variable equaling one if the change
in house prices is negative. For the state sample, the mean house price change for the bottom
quartile of increases is 1.7%, the mean change for the top quartile is 13% and the mean change
for decreases is -2.4% (the means for the MSA sample are similar). Each observation is
weighted by its share of total property tax revenue collected nationally that year and the standard
errors are clustered by state or MSA. As in the time-series analysis, the elasticities at a point in
time are equal to the sum of the corresponding coefficients.
12
Figure 4 graphically presents the results of estimating equation (4) using the state sample
(Appendix Table A1 presents the full coefficient estimates). The vector of control variables, X,
includes five lags each of the log of population and the log of personal income, both measured at
the state level. In addition to the elasticity estimates, the figure displays the results of two
hypothesis tests for each estimate. The first is the standard test for being able to distinguish the
12
For the large, small and decrease categories, the elasticity at a point in time is the sum of the corresponding main
effect coefficients and the corresponding category coefficients. E.g., the elasticity for a small house price increase
two years after the increase is equal to
12 ,1 ,2
s
mall small
β
ββ β
++ +
.
12
estimate from zero. The null hypothesis is that the elasticity estimate is equal to zero – i.e. that
policy makers completely offset changes in house prices. Estimates for which the null
hypothesis can be rejected are denoted by large shapes, and estimates for which the null
hypothesis cannot be rejected are denoted by small shapes. The null hypothesis for the second
test is that the elasticity estimate is equal to one – i.e. that policy makers engage in no offset and
allow the full value of house price changes to pass into property tax revenues. Estimates for
which this second null hypothesis can be rejected are denoted by an asterisk.
The response of property tax revenue to a normal housing price increase, which
corresponds to an increase that falls within the middle two quartiles of positive house price
changes (encompassing price changes from 3 to 8 percent), is remarkably similar both in timing
and magnitude to the time-series estimates. Property tax revenue does not respond to a change in
housing prices until three years after the change and the average elasticity in years three, four
and five following the house price change is equal to around 0.4. Both null hypotheses can be
rejected for years four and five following the house price change and it is therefore possible to
rule out both full offset and no offset.
The elasticity estimates for large house price increases have a similar time profile but are
smaller in magnitude, equal to around 0.2. When house prices rise by an unusually large
amount, policy makers offset more of the increase than they would for a typical size increase.
Policy makers and/or voters may prefer to avoid very large increases in property tax burdens.
Alternatively, locations prone to substantial house price increases may be more likely to have
property tax limitations in place which prevent large increases in tax bills.
The elasticity estimates for small changes are much larger, equal to around 1 in the fifth
year following a change in house prices. These estimates suggest that when house price
appreciation is anemic, policy makers offset little to none of the increase. It should be noted that
it is generally not possible to statistically distinguish between the “normal increase” elasticity
estimates and the “small increase” and the “large increase“ elasticity estimates.
The elasticity estimates for house price decreases are negative, indicating that policy
makers more than offset the impact of house price declines on property tax revenues. Although
the confidence intervals around these estimates are quite large (see Appendix Table A1), it is
possible to distinguish them from one – policy makers offset house price depreciation by raising
13
effective tax rates such that tax revenues do not fall by the full amount implied by the decline in
house prices.
Figure 5 displays state sample estimates which exclude California.
13
California has very
stringent property tax limitations (Sheffrin 2005) and the response of property tax revenue to
house price changes may differ from the rest of the nation as a result. Figure 6 utilizes the full
state sample and adds to the control vector, X, five lags each of the log of total residential
building permits issued (to control for increases in the stock of residential property) and the
unemployment rate (to control for local economic conditions). Figure 7 displays the results of a
different control strategy. The control vector, X, is replaced by a set of census region-year fixed
effects and local government-specific linear trends. The region-year effects control for any time-
varying factors, such as economic shocks, at the region level; the trends control for any factor,
such as population growth, which evolves in a gradual, linear manner at the level of the
individual locality.
14
Finally, figure 8 displays the results of estimating the model at the MSA
level (with the X vector including the log of population and the log of personal income).
The “normal increase” elasicities are quite robust to the alternative estimation strategies.
The “large increase” and “small increase” estimates are somewhat more variable across the
different specification but produce conclusions generally similar to those discussed above. The
“decrease” elasticities are also similar to those discussed above, with the exception of figure 7,
which employs the region-year and government trend approach. In this instance, the estimates
are positive, large in magnitude, and can be distinguished from zero, suggesting that house price
depreciation produces large decreases in property tax collections. It is possible that the region-
year/trend approach is not as successful as the control variable strategy at controlling for
negative economic shocks, which reduce home prices and also reduce demand for public
services.
IV. Conclusion
The evidence suggests that property tax revenues are quite responsive to changes in
house prices. Although it takes several years for house price appreciation to feed through to
13
The full set of coefficient estimates for Figures 5 – 8 are available from the author upon request.
14
If the region-year, state trend and X variables are all included simultaneously, the estimates become imprecise.
This likely reflects the fact that there is little residual variation in house prices when the specification is so heavily
saturated.
14
property tax revenues, the long-run elasticity is on the order of 0.4. On average, policy makers
are estimated to respond to increasing home prices by reducing effective tax rates so as to offset
60 percent of the increase in tax revenue that would have occurred in the absence of a change in
the effective tax rate. Institutional features of the property tax, such as delays in bringing
assessed values into line with market values and caps and limitations on the tax, likely explain
the lag between house prices and tax revenues (and may also influence the magnitude of the
relationship.)
There is some evidence that the elasticity is smaller for unusually large house price
changes, suggesting that policy makers and voters may prefer to avoid unusually large increases
in property tax bills. Similarly, there is evidence that during periods of unusually sluggish house
price growth the elasticity is larger, suggesting that policy makers and voters may not want tax
revenue growth to slow too much. The evidence on the impact of house price depreciation is
somewhat mixed, but on the whole there is little evidence that house price declines influence
property tax revenues. It appears that policy makers raise effective tax rates to offset declines in
tax revenue arising from downward swings in the housing market. These estimates should be
interpreted cautiously because they come from a sample in which most house price declines are
relatively small. Thus, the results may not accurately predict the response of local governments
to some of the large price declines that have occurred in different parts of the country, given the
political difficulty of increasing tax rates under such circumstances.
This paper’s estimates reflect the typical behavior of the local government sector as a
whole and may not be valid for individual states or communities. For instance, states with
stringent property tax limitations, such as California, may have a smaller elasticity of property
tax revenue with respect to property values than estimated here. Similarly, states such as
Virginia, which have few property tax limitations and bring assessed values into line with market
values relatively quickly, may have a shorter lag between real estate appreciation and tax
revenues than the estimates in this paper would suggest.
If public good allocation decisions correspond to the median voter model, this paper can
be viewed as the public goods equivalent to the literature on the marginal propensity to consume
out of housing wealth (which focuses on private goods consumption). Using sample averages,
the 0.4 elasticity estimate implies that the marginal propensity to consume local public goods out
of housing wealth is equal to 4/10 cent – i.e. each dollar of additional housing wealth increases
15
public goods consumption by $0.004 (see the Data Appendix for details on this calculation). As
a reference point, the marginal propensity to consume out of housing wealth for private goods is
generally thought to be around $0.03 – a little over seven times as large. As discussed more fully
in the introduction, the MPC interpretation is subject to the caveat that this paper’s results may
reflect both a wealth shock and a partially offsetting price shock. The MPC estimate should
therefore be seen as a lower-bound estimate.
Alternatively, if public good allocation decisions correspond more closely to a Leviathan
model in which public officials seek to maximize tax revenues (Brennan and Buchanan 1980),
the results suggest that public officials find it easier to increase revenues when it does not require
a tax rate increase. In fact, the evidence presented here suggests that periods of house price
appreciation allow them to increase revenues while simultaneously lowering tax rates.
16
Data Appendix
Time-Series Regressions (equation (3); figure 3 and table 2)
Data for property tax revenues comes from the Bureau of Economic Analysis (BEA) National
Income and Product Accounts (NIPA) Table 3.3, line 9. Data for house prices is an extended
version of the purchase-only index from the Office of Federal Housing Enterprise Oversight
(OFHEO). The purchase-only index starts in the first quarter of 1991; all values from 1975
through 1990 are from OFHEO’s all-transactions index, normalized to equal 100 in 1991 Q1.
Personal income comes from BEA NIPA Table 2.1, line 1. For the annual models displayed on
Table 2, annual levels of the variables are set equal to the fourth quarter value for the year. The
log change of the variables can therefore be interpreted as a year-to-year percent change. The
net stock of private residential structures and net stock of private non-residential structures come
from BEA Fixed Assets Table 2.1, line 59 and Table 1.1, line 6, respectively. Data for
commercial real estate prices is the weighted average of the office, warehouse, and retail
transaction price indices from the National Real Estate Investor (NREI). Corporate profits (BEA
NIPA Table 1.14, line 32) are before taxes and exclude inventory valuation adjustment (IVA)
and capital consumption adjustments (CCAdj). Personal consumption expenditures are from
BEA NIPA Table 1.1.5, line 2.
Panel Regressions(equation (4);figures 4 – 8)
Government units that had missing or zero values for property tax revenues were excluded from
the sample. Government units with annual changes in property tax revenues exceeding 100%
were dropped from the sample – such changes likely arise from unusual events, such as the
consolidation of localities, unrelated to the connection between house prices and property tax
revenues. For variables available quarterly, annual values were set according to the fiscal year
ending variable provided in the Annual Survey of Governments (ASG) data – i.e. if the fiscal year
ended in the second quarter, then we used the second quarter data point (or the last month in the
quarter) as the value for the year. If the fiscal year ending date was missing, we assumed the
fiscal year ended on June 30, the most common fiscal year end. The log change of the variables
can therefore be interpreted as a fiscal year-to-fiscal year percent change. The MSA-level
regressions, depicted in figure 8, use the MSA boundaries established by OMB Bulletin No. 08-
01 (November 20, 2007) and group individual units in MSAs using the Federal Information
Processing Standard (FIPS) codes provided in the ASG individual data files. This eliminated
government entities outside of MSAs. We utilized the MSA-level house prices (OFHEO—all-
transaction index), personal income (BEA—Local Area Personal Income, CA1-3, 1.0), and
population (BEA—Local Area Personal Income, CA1-3, 2.0). Sample sizes for the state panel
regressions (figures 4 – 7) are on the order of 250,000 observations. The sample size for the
MSA panel regression (figure 8) is approximately 140,000 observations.
Public goods MPC out of housing wealth calculation
The marginal propensity to spend on public goods out of housing wealth is calculated as follows.
R
R
elasticity
V
V
Δ
=
Δ
and
R
MPC
V
Δ
=
Δ
therefore *
R
MPC elasticity
V
=
17
For this calculation, R is aggregate property tax revenue from single family residential real estate
and V is the aggregate value of single family residential real estate. The elasticity is assumed to
be 0.4 based on the evidence presented in Section III. R = ($376 billion * 0.5); the $376 billion
is total aggregate property tax revenue based on the Census Bureau’s 2006 Quarterly Summary
of State and Local Government Tax Revenues and the 0.5 represents the percent of total property
tax payments accounted for by single family homes (author’s calculation from the 1987 and 1991
Census of Governments). V = $18,559 billion; the value of owner-occupied, single family
property in the 2006 American Community Survey.
(376*.5)
* 0.4* 0.004
18,599
R
MPC elasticity
V
== =
18
ACKNOWLEDGEMENTS
I thank the following individuals for helpful comments and suggestions: Nathan Anderson,
Samuel Brown, Jane Dokko, Eric Engen, Glenn Follette, Amanda Kowalski, Andrea Kusko,
David Reifschneider, Dan Sichel, Larry Slifman and participants at the Spring 2008 NTA State
and Local Symposium. Thanks to Brian McGuire, Daniel Stenberg and particularly Samuel
Brown for excellent research assistance. The analysis and conclusions set forth are those of the
author and do not indicate concurrence by other members of the Federal Reserve research staff
or the Board of Governors.
19
REFERENCES
Anderson, Nathan. “Property Tax Limitations: An Interpretative Review.” National Tax
Journal 59 No. 3 (September, 2006): 685-94.
Bowman, John H. “Property Tax Policy Responses to Rapidly Rising Home Values: District of
Columbia, Maryland, and Virginia.” National Tax Journal 59 No. 3 (September, 2006): 717-33.
Black, Duncan, "On the Rationale of Group Decision-Making," Journal of Political Economy
56 No. 1 (1948): 23-34.
Bloom, Howard and Helen Ladd. “Property Tax Revaluation and Tax Levy Growth.” Journal
of Urban Economics 11 No. 1 (January, 1982): 73-84.
Brennan, Geoffrey and James Buchanan. The Power to Tax: Analytical Foundations of a Fiscal
Constitution. New York: Cambridge University Press, 1980.
Brunori, David. Local Tax Policy: A Federalist Perspective. Washington D.C.: Urban Institute
Press, 2003.
Carroll, Christopher D., Misuzu Otsuka and Jirka Slacalek. “How Large Is the Housing Wealth
Effect? A New Approach.” Working Paper, 2006.
Case, Karl E., John M. Quigley, and Robert J. Shiller. “Comparing Wealth Effects: The Stock
Market versus the Housing Market.” Advances in Microeconomics 5 No. 1 (2005): Article 1.
Cornia, Gary and Lawrence Walters. “Full Disclosure: Controlling Property Tax Increases
During Periods of Increasing Housing Values.” National Tax Journal 59 No. 3 (September,
2006): 735-49.
Dye, Richard and Andrew Reschovsky. “Property Tax Responses to State Aid Cuts in the
Recent Fiscal Crisis.” Public Budgeting & Finance, Vol. 28, Issue 2, 2008.
Dye, Richard and Therese McGuire. “The Effect of Property Tax Limitation Measures on Local
Government Fiscal Behavior.” Journal of Public Economics 66 No. 3 (December, 1997): 469-
487.
Dye, Richard, Daniel McMillen and David Merriman. “Illinois’ Response to Rising Residential
Property Values: An Assessment Growth Cap in Cook County.” National Tax Journal 59 No. 3
(September, 2006): 707-16.
Evans, William, Sheila Murray and Robert Schwab. “The Property Tax and Education Finance.”
In Property Taxation and Local Government Finance, edited by Wallace Oates, 209-235.
Cambridge, MA: Lincoln Institute of Land Policy, 2001.
20
Fisher, Ronald and Mary Gade. “Local Property Tax and Expenditure Limits.” In State and
Local Finance for the 1990s: A Case Study of Arizona, edited by Therese McGuire and Dana
Naimark, 449-464. Tempe, AZ: Arizona State University, 1991.
Giertz, J. Fred. “The Property Tax Bound.” National Tax Journal 59 No. 3 (September, 2006):
695-705.
Hines, James R., Richard H. Thaler. “Anomalies: The Flypaper Effect.” Journal of Economic
Perspectives 9 No. 4 (Fall, 1995): 217-26.
Ladd, Helen. “Property Tax Revaluation and Tax Levy Growth Revisited.” Journal of Urban
Economics 30 No. 1 (July, 1991): 83-99.
Lehnert, Andreas. “Housing, Consumption, and Credit Constraints.” FEDS Working Paper No.
2004-63. Washington D.C.: Federal Reserve Board, 2004.
Merrick, Amy. “Property Tax Frustration Builds: States, Cities Revise Strategy as Homeowners
Protest Rising Levies.” The Wall Street Journal (December 18, 2007).
Merriman, David. The Control of Municipal Budgets: Toward the Effective Design of Tax and
Expenditure Limitations. Westport, CT: Greenwood Press, 1987.
National League of Cities. City Fiscal Conditions 2007. Washington, D.C.: October, 2007.
Preston, Anne and Casey Ichniowski. “A National Perspective on the Nature and Effects of the
Local Property Tax Revolt, 1976-1986.” National Tax Journal 44 No. 2 (June, 1991): 123-145.
Sheffrin, Steven. “Proposition 13/Property Tax Caps.” In The Encyclopedia of Taxation and
Tax Policy, edited by Joseph Cordes, Robert Ebel and Jane Gravelle, 321-323. Washington,
D.C.: Urban Institute Press, 2005.
Skinner, J. S. “Is Housing Wealth a Sideshow?” In Advances in the Economics of Aging,
National Bureau of Economic Research Report, 241-268. Chicago: 1996.
21
TABLE 1
ANNUAL PERCENT CHANGES IN PROPERTY TAX REVENUES,
HOUSE PRICES, AND PROPERTY TAX RATE
Variable
2000 – 2005
Average
1) House Prices
7.9%
2) Effective Property Tax Rate
-1.7%
3) Property Tax Revenues
6.1%
Source. OFHEO purchase-only index. BEA NIPA Table 3.3, line 9. Author’s calculations.
The effective property tax rate, row 2, is defined as (0.6*property tax revenue)/(house
prices), where 0.6 reflects the fact that approximately 60% of total property tax revenues
come from residential real estate.
22
TABLE 2
ANNUAL TIME SERIES REGRESSIONS
Dependent Variable: ∆ log(Property Tax Revenue)
#1
#2
#3
#4
#5
1) Constant
0.038**
0.037**
0.027**
-0.046**
0.015
(0.008) (0.007) (0.008) (0.012) (0.017)
2) ∆ log(House Price Index [t-1]) -0.066
(0.206)
3) ∆ log(House Price Index [t-2]) 0.085
(0.269)
4) ∆ log(House Price Index [t-3]) 0.422** 0.462** 0.382** 0.480** 0.418**
(0.193) (0.107) (0.107) (0.199) (0.142)
5) ∆ log(Property Tax Revenue [t-1]) 0.242**
(0.114)
Controls for: Stock of Residential and Commercial
Property, Price of Commercial Property
X
Controls for: Personal Income Tax Base, Corporate
Income Tax Base, Sales Tax Base
X
Observations
29
29
29
28
29
Degrees of Freedom
25 27 26 19 15
R-Square 0.413 0.410 0.498 0.679 0.711
Adj. R-Square
0.342 0.388 0.459 0.545 0.460
Note. Property Tax Revenue is the BEA state and local property tax revenue series. House Price Index is the OFHEO price index for
owner-occupied real estate, purchases only extended. Column #4 controls for the contemporaneous and lag t-1 of the log change in the
following variables: BEA net stock of private residential structures and BEA net stock of private non-residential structures. It also
controls for lags t-1, t-2, and t-3 of the log change in the NREI price index for commercial real estate. Column #5 controls for the
contemporaneous and lags t-1, t-2 and t-3 of the log change in the following variables: BEA personal income, BEA corporate profits
before tax, and BEA personal consumption expenditures.
* Value significant at 10%
** Value significant at 5%
23
APPENDIX TABLE A1
MICRO-DATA ESTIMATES: STATE SAMPLE: X = ∆ ln(Pop), ∆ ln(YP)
Change in House Price
Cumulative
Point Estimate
Upper 95% CI
Lower 95% CI
Normal Increase, Year 1
-0.023
0.073
-0.119
Normal Increase, Year 2
-0.005
0.233
-0.244
Normal Increase, Year 3
0.289
0.652
-0.075
Normal Increase, Year 4
0.399
0.665
0.132
Normal Increase, Year 5
0.536
0.801
0.271
Large Increase, Year 1
-0.035
0.017
-0.088
Large Increase, Year 2
0.040
0.172
-0.093
Large Increase, Year 3
0.169
0.327
0.011
Large Increase, Year 4
0.177
0.280
0.075
Large Increase, Year 5
0.128
0.278
-0.022
Small Increase, Year 1
0.135
0.527
-0.257
Small Increase, Year 2
0.346
1.127
-0.436
Small Increase, Year 3
0.536
1.596
-0.524
Small Increase, Year 4
0.849
1.942
-0.245
Small Increase, Year 5
1.046
1.990
0.103
Decrease, Year 1
-0.034
0.234
-0.301
Decrease, Year 2
-0.131
0.309
-0.571
Decrease, Year 3
-0.114
0.245
-0.474
Decrease, Year 4
-0.218
0.186
-0.623
Decrease, Year 5
-0.352
0.171
-0.875
Note. Cumulative point estimates for normal house price increases are the sum of the β
i
coefficients from the first year following the house price change through year i. For example, the
cumulative point estimate for the second year following a normal house price increase is β
1 +
β
2
.
Cumulative point estimates for other changes in house prices are the sum of the β
i
coefficients for
the normal increase and the sum of the β
i
coefficients for the other changes in house prices. For
example, the point estimate for the second year following a small house price increase is β
1
+
β
2
+
β
small
,
1 +
β
small
,
2
.
Percent of personal income
Source. Census State and Local Government Finances. BEA NIPA Table 2.1.
Figure 1. Property Taxes as a Percent of Personal Income
Year-to-year percent change, 4-quarter moving average
Property taxes
House price index
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
Source. Census Quarterly Summary of State and Local Government Finances. OFHEO purchase-only index, extended.
Figure 2. House Price Appreciation and Property Taxes
Elasticity
Years after property value change
90% confidence interval
Point Estimate
Average magnitude
Year 3 to Year 5 = 0.4
Figure 3. Time Series Regression Estimates
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
Normal Increase
Small Increase
Large Increase
Decrease
*
*
*
*
*
*
*
*
*
*
*
* *
*
Figure 4. Micro-data Estimates: State Sample: X = ln(Pop), ln(YP)
Elasticity
Years after property value change
Note. Large figures signify that the point estimate is statistically different from 0 at the 95% confidence level. Asterisks denote that the point is
statistically different from 1 at the 95% confidence level.
Small Increase
Normal Increase
Large Increase
Decrease
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Elasticity
Years after property value change
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
Note. Large figures signify that the point estimate is statistically different from 0 at the 95% confidence level. Asterisks denote that the point is
statistically different from 1 at the 95% confidence level.
Figure 5. Micro-data Estimates: State Sample, Excluding California: X = ln(Pop), ln(YP)
Normal Increase
Small Increase
Large Increase
*
Decrease
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Elasticity
0.8
1
1.2
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Years after property value change
Note. Large figures signify that the point estimate is statistically different from 0 at the 95% confidence level. Asterisks denote that the point is
statistically different from 1 at the 95% confidence level.
Figure 6. Micro-data Estimates: State Sample: X = ln(Pop), ln(YP), ln(Permits), ln(Unemploy)
Normal Increase
Small Increase
Large Increase
Decrease
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Years after property value change
Elasticity
0.8
1
1.2
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Note. Large figures signify that the point estimate is statistically different from 0 at the 95% confidence level. Asterisks denote that the point is
statistically different from 1 at the 95% confidence level.
Figure 7. Micro-data Estimates: State Sample: Alternate Control Strategy
Normal Increase
Small Increase
Large Increase
Decrease
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Years after property value change
Elasticity
0.8
1
1.2
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Note. Large figures signify that the point estimate is statistically different from 0 at the 95% confidence level. Asterisks denote that the point is
statistically different from 1 at the 95% confidence level.
Figure 8. Micro-data Estimates: MSA Sample: X = ln(Pop), ln(YP)
