You are on page 1of 35

Public Finance Reviewhttp://pfr.sagepub.

com/
Are Workers ''Ricardian''? Estimating the Labor Supply Effects of State Fiscal Policy
Karen Smith Conway
Public Finance Review 1999 27: 160
DOI: 10.1177/109114219902700203
The online version of this article can be found at:
http://pfr.sagepub.com/content/27/2/160

Published by:
http://www.sagepublications.com

Additional services and information for Public Finance Review can be found at:
Email Alerts: http://pfr.sagepub.com/cgi/alerts
Subscriptions: http://pfr.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://pfr.sagepub.com/content/27/2/160.refs.html

>> Version of Record - Mar 1, 1999


What is This?

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

PUBLIC
Conway
FINANCE
/ ARE
REVIEW
WORKERS RICARDIAN?
This research investigates whether workers perceive current deficits as implied
Abstract future taxes ( la Ricardian equivalence) and therefore consider current deficits
when formulating their expectations about future net wages. A life cycle model of
labor supply that permits workers to be aware of the governments budget constraint
over time is constructed and then estimated using microlevel data and state fiscal
policy variables. The results reveal that state government budget deficits tend to
increase male labor supply more (or decrease it less) the more the state relies on
individual income taxes. Such an effect is what would be predicted with Ricardian
workers.

ARE WORKERS RICARDIAN?


ESTIMATING THE LABOR SUPPLY
EFFECTS OF STATE FISCAL POLICY
KAREN SMITH CONWAY
University of New Hampshire

Under Ricardian equivalence, the private sector is aware of


the governments intertemporal budget constraint and correctly perceives debt-financed spending as implied future taxes; thus,
under certain restrictive conditions, a reduction in taxes, holding
spending constant, will have no real effect on the economy (Barro
1974). The Ricardian equivalence result typically requires that all
taxes be lump-sum taxes. However, the basic assumption underlying
Ricardian equivalence has broad implications for intertemporal labor
supply behavior when taxes are instead distortionary. Specifically, if
workers perceive current deficits as implied future taxes, they should
consider them when formulating their expectations about future net
wages.1 And if they expect their future taxes to be higher (and therefore their future net wages to be lower), they may work more now and
AUTHORS NOTE: I am grateful to Michael DeSimone for his invaluable research assistance
and to Jim Ziliak, David Bradford, and participants in the UNH Economics Seminar for their
helpful comments. I also thank the numerous state tax officials who assisted me in my calculation of state income tax rates.
PUBLIC FINANCE REVIEW, Vol. 27 No. 2, March 1999 160-193
1999 Sage Publications, Inc.

160

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

161

work less in the future (or intertemporally substitute their labor supply). Postponing taxes on labor income via deficit finance could be a
desirable policy because it increases current labor supply, which
should also boost current savings (Hansson and Stuart 1987). If workers are Ricardian, their current labor supply depends not only on
current wages, income taxes, and perhaps government spending but
also on the level of the deficit and their expectations about how the
government will pay it off (Conway 1994).
A life cycle model of labor supply that permits workers to be aware
of the governments intertemporal budget constraint is constructed
and then estimated using microeconomic labor supply data from the
Panel Study of Income Dynamics (PSID) for prime-aged men. To estimate such a model, there must be information on workers who pay different amounts of taxes, receive different amounts of public-sector
spending, and face different levels of government budget deficits or
surpluses. State government policy provides this necessary dimension
and is an interesting laboratory for testing the underlying assumptions
of Ricardian equivalence because most state constitutions mandate a
balanced budget in the intermediate run. Thus, if a state has a budget
deficit one year, taxpayers know that the deficit must be paid off in the
2
near future. In addition, states cannot resort to printing money to
finance a deficit and may not have the same access to credit markets as
the federal government (Poterba 1997). Individuals may also be more
aware of and affected by the fiscal policies that occur closest to home.
On the other hand, taxpayers can avoid repaying a current state deficit
by moving to another state. Also, unlike the federal government, many
state governments have budget surpluses that may arguably have a
weaker effect than a budget deficit. These latter two differences suggest that the labor supply effects of the budget balance is smaller at the
state level than the federal level. However, state laws that require balanced budgets may make workers more Ricardian if and when a state
3
budget deficit/surplus appears.
This research contributes to our understanding of the full impact of
fiscal policy on labor supply behavior by first deriving the theoretical
labor supply effects of a budget deficit and showing how they depend
on the tax structure. I then investigate empirically whether workers are
aware of the governments budget constraint and use it in making
work decisions. In so doing, this article adds a new kind of empirical

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

162

PUBLIC FINANCE REVIEW

evidence to the current debate over whether Ricardian equivalence


best represents reality.4 Given the sensitivity of the existing evidence,
which uses macroeconomic time-series data (e.g., Barth et al. 1991), a
new approach using different data should be useful. The results presented here suggest that state budget deficits/surpluses, when combined with information about the states current tax structure, have an
effect on the labor supply of prime-aged men consistent with
Ricardian behavior.

A THEORETICAL MODEL OF INTERTEMPORAL


LABOR SUPPLY AND RICARDIAN BEHAVIOR

Are workers aware of the governments budget constraint over


time? Quintieri and Rosati (1988) explore the theoretical labor supply
effects of several policy changes, including deficit-financed ones,
when workers are Ricardian. Hansson and Stuart (1987) analyze the
welfare costs of deficit finance in a general equilibrium model with
Ricardian agents and show that postponing taxes on labor via deficit
finance may be a desirable policy. Fremling and Lott (1989) show how
a deficit alters the current labor supply and savings of Ricardian workers when the deadweight losses from distortionary taxes are considered, and they demonstrate that only a fraction of taxpayers must be
Ricardian in order for Ricardian equivalence to result. All of these
studies emphasize the importance of knowing whether workers are
Ricardian or Keynesian in determining the outcome of government
policies.
Although many theoretical studies make the assumption of
Ricardian agents, empirical labor supply studies have failed to explore
it. To undertake such an empirical investigation, one must first construct a theoretical framework that allows for more complicated tax
structures than those typically considered. In this section, I consider
the effects of deficit finance on current labor supply under a variety of
possible tax structures and methods of repayment.
For simplicity, I consider a two-period perfect foresight model,
similar to Hansson and Stuart (1987) and Blomquist (1985), in which
the individual maximizes utility over two periods,

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

U (T h1 ,C1 , G1 ) + V (T h2 , C 2 , G2 ),

163

(1)

subject to his or her lifetime budget constraint,


A0 + W1h1 + Y1 (1 + 1 )C1 f1 t1[]
+

W 2 h2
Y
(1 + 2 )C 2 f2 t2 []

,
+ 2
1+ r
1+ r 1+ r
1+ r 1+ r

(2)

where the subscripts 1 and 2 denote the present and future, C is consumption, T is the total amount of time available for work and leisure,
h is hours of work, G is government spending, r is the real interest rate,
A0 is beginning of period assets, W is the gross (before-tax) hourly
wage, and Y is exogenous income.5 The tax system is represented by ,
f, and t[], and anticipated changes in the future system are shown by
and . reflects all broad-based consumption taxes (or a weighted
average of all specific good taxes), f reflects all lump-sum taxes and
fees, and t[] is the income tax function. The utility functions U and V
are strictly concave, twice differentiable, increasing in C and G, and
decreasing in h. I assume that utility is separable over time to simplify
the comparative statics. Future utility is discounted by . Because I
focus only on the labor supply behavior of prime-aged men, I ignore
the possibility of corner solutions.
Ricardian behavior occurs as workers make predictions about the
future tax system (and expenditures). For instance, if the government
currently runs a deficit, Ricardian workers should increase their predictions of future taxes. This is shown through an increase in and/or
, which results in a proportionate increase in t[] and/or and f,
respectively. Thus, the effects of deficit finance on current labor supply depend not only on whether workers are Ricardian but on how they
think the deficit will be paid off. I turn to several special cases of the
model.
A TAX ON LABOR (AND EXOGENOUS) INCOME

Most theoretical models of Ricardian workers assume a proportional wage tax. Hansson and Stuart (1987) consider the welfare costs
of deficit finance used to delay proportional taxation of labor supply
and compare them to deficit finance used to delay capital taxation.

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

164

PUBLIC FINANCE REVIEW

Quintieri and Rosati (1988) make the assumption of a proportional


wage tax in their derivation of balanced-budget, labor supply effects
of different government policies. It appears implicitly in the argument
put forth by Fremling and Lott (1989), although they also discuss possible distortions in the capital market. I begin with this simple case
because it lays the groundwork for the more complicated ones.
If the tax system consists only of a proportional wage tax, the workers problem described in equations (1) and (2) becomes much simpler. Specifically, and f equal zero, and t[] = tWh. The relevant wage
for optimization is the net wage, or w = (1 t)W. Deficit finance
increases current labor supply, as seen by the comparative static result,

tW h
h1 1
=
t2W 2 H 31 + 2 2 2 H 51 > 0 ,

H 1 + r
1+ r

(3)

where |H| > 0 is the determinant of the bordered Hessian, |Hij| is the
determinant of the ijth minor of H, and is the Lagrange multiplier.
Both |H31| and |H51| are positive, so that h1 / is positive. The first
term is the intertemporal substitution effectcurrent leisure is relatively more expensive than future leisure, so the individual works
more in the present period. The second term is the income effect of the
tax increase on current labor supply, which is also positive. Intuitively,
a budget deficit increases the Ricardian individuals predicted future
tax, thereby decreasing the future net wage. This reduction in future
wages makes the individual feel poorer and makes current labor supply more financially rewarding than future labor supply. Thus, a deficit increases a Ricardian workers current labor supply when the tax
system consists only of a proportional wage tax.
A progressive wage tax is one in which t[] = t[Wh], and the first and
second derivatives of t, t and t, respectively, are both positive. Allowing the tax rate on labor income to increase with income does not substantively alter the predictions of the model. The comparative static
result for h1 / is the same as in equation (3), except that in the first
term, t2 is now t 2 , and in the second term, t2W2h2 is equal to t2[W2h2].
Increasing a future progressive wage tax has the same effect as
increasing a future proportional onefuture wages decrease, and the
individual feels poorer. Including exogenous income, Y, in the progressive tax function (i.e., t[] = t[Wh + Y]) does not substantively

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

165

change how a deficit affects a Ricardian workers current labor supply.6 Thus, allowing for a progressive wage tax and/or including
exogenous income in the tax function does not change the result that a
deficit increases a Ricardian workers current labor supply.
A TAX ON ALL INCOME

Blomquist (1985) shows how a nonlinear tax function that includes


all sources of income, including asset income, greatly complicates the
analysis and overturns much of the conventional wisdom regarding
intertemporal labor supply behavior. Blomquist discusses the labor
supply response to changes in present or future gross wages (i.e.,
h1 / W1 and h1 / W 2 ) in the presence of a nonlinear income tax.
Although some of the intuition put forth by Blomquist applies here,
my focus is how a proportionate shift in the tax system affects current
labor supply.
Specifying a tax on all income makes the budget constraint nonseparable over time, as seen by examining the present and future tax
functions,
t1[]= t1[W1h1 + Y1 + rA 0] ,

(4)

and

t2 [] = t2 W2 h2 + Y2 + r(W1h 1 + Y1 + A0 t1[W1h 1 + Y1 + rA0 ] ( 1 + 1 ) C1 f1 ) . (5)

For instance, how much a worker earns and saves in the present period
influences his future tax bill and, in a progressive tax system, his
future marginal tax rate. The decision to work in this period versus the
future depends not only on the two wages but on the cost of saving current income to be consumed later. This can best be seen through the
first-order conditions for current labor supply and consumption,
respectively,
t r
U l + W1(1 t1 ) 1 2 = 0
1+ r

and

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

(6)

166

PUBLIC FINANCE REVIEW

t r
U c (1 + 1 ) 1 2 = 0,
1+ r

(7)

where Ui denotes the first partial derivative of the utility function with
respect to good/factor i, and l denotes leisure. The relevant marginal
wage or true price of current leisure considers both present and future
taxes. For instance, every hour of current labor supply yields
W1 (1 t1 ) in current earnings, but if those earnings are saved, future
taxes increase by t2 rW1 (1 t 1 ) . Hence, the gross wage must be
deflated by this amount to reflect the true price of leisure. Likewise,
such a tax reduces the true price of current consumption by the amount
that each dollar consumed today will reduce future tax payments.
Because this is only a two-period model, the first-order conditions for
future labor supply and consumption are as typically derived.
Equations (6) and (7) reveal that increasing future income taxes
reduces the true prices of current labor supply and consumption, as
well as that of future labor supply, making the effect on current labor
supply much more difficult to determine. The comparative static
result is
h1

t2 r(1 + 1 )
t []
1 t2 rW1(1 t1 )
t2 W 2
H
+
H
+
|
|
|
|
|H 31| + 2 |H 51| , (8)
11
21

1+ r
1+ r
1+ r
|H | 1 + r

where all terms are as previously defined. Comparing this expression


to equation (3), we see that the last two terms, the intertemporal substitution effect and the income effect, are essentially the same and therefore imply an increase in current labor supply as before. However, the
determinants of the minors, |Hij|, are now sufficiently complicated so
that neither |H31| nor |H51| can be signed.
The first two terms are new and result from the reduced true prices
of current labor supply and consumption. The first term is negative, as
|H11| < 0 is a principal minor. This is the own substitution effect; the
true price of current labor supply has decreased, so the individual
reduces current labor supply. The second effect is the intratemporal
substitution effect; the true price of current consumption has also
decreased, and its effect on current labor supply is indeterminate (|H21|
is indeterminate).

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

167

The model simplifies somewhat if the tax function is proportional,


or t = t. As Blomquist (1985) notes, the budget constraint can be
rewritten in terms of the after-tax wages, w; exogenous income, Y =
(1 t)Y; and interest rate, r = (1 t2)r and is no longer nonseparable
over time. A deficit paid off with future income taxes now has the
effect of reducing these three variables, the first two of which have an
unambiguously positive effect on current labor supply. The comparative static result for this model appears likely positive but unfortunately remains ambiguous unless an additional assumption is imposed
(see the appendix). A deficit paid off by increasing a future proportional income tax will not necessarily increase current labor supply,
although such an effect appears likely. This is an important distinction
because many states have income tax systems in which the marginal
7
tax rate is constant (t = 0) for most taxpayers. In addition, the
assumption that a future change in the tax system will come as a proportionate increase ( > 1) is much more realistic if the tax system is
proportional.
Given that I cannot derive unambiguous results for a proportional
tax, it is not surprising that a progressive tax structure yields comparative static results that are impossible to simplify in a meaningful way.
Now, current labor supply and consumption decisions affect not only
the future income tax bill but the future marginal tax rate as well. However, recall that it is not the progressivity of the tax structure, per se,
that makes the current labor supply effect of an increase in the future
tax system (h1@) difficult to predict. Rather, it is the taxation of
asset income. When asset income is taxed, increasing future income
tax rates make the worker feel poorer and reduce the price of future
labor supply, as before, but they now also reduce the price of current
labor supply and consumption. With a proportional tax system, these
effects can be manipulated such that h1@ appears likely positive.
With a progressive tax system, such manipulation is much more
involved and shows little promise.
WHAT EFFECT DO THE OTHER ELEMENTS OF
THE GOVERNMENTS BUDGET CONSTRAINT HAVE?

State governments are not known for their reliance on income


taxes. Unlike the federal government, state governments obtain a

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

168

PUBLIC FINANCE REVIEW

much larger share of their tax revenues from other sources such as
sales and excise taxes, lotteries, and user fees. The labor supply effect
of an increase in these taxes is therefore important to my empirical
analysis. In the above model, these taxes appear as a tax on consumption, , and a lump-sum tax or fee, f, and reflects a proportionate
change in the future levels of these taxes. The comparative static result
for is

C + f
h1 1 2
=

|H 41| + 2 2 2 |H 51| ,
1+ r
|H | 1 + r

(9)

where |H41| is positive in all models except the one with a progressive
tax on all income. The first term reflects that future consumption is
now more expensive and is a negative substitution effect on current
labor supply, as one goal of current labor supply is to finance future
consumption. The second term is the income effect of the tax increase,
which has the usual positive effect on labor supply. Thus, the labor
supply effect of an increase in these kinds of taxes is theoretically
ambiguous.
Finally, a deficit could be paid off with a reduction in future expenditures, G2. The comparative static result for future expenditures is
h1
1
=
(VlG |H 31| + VCG |H 41|) ,
G2 |H |

(10)

Where l denotes leisure and Vij denotes the second partial derivative of
the second-period utility function. The effect of reducing future government expenditures not surprisingly depends on the preference relationship between G, consumption and leisure. For instance, if the utility function is separable in G (as is implicitly assumed in labor supply
studies that omit government spending), then there is no effect on current labor supply. If decreasing G decreases the marginal value of
either consumption or leisure (i.e., V lG or VCG > 0), as appears likely,
then current labor supply will decrease.
In summary, this theoretical framework shows that a deficit affects
the current labor supply of Ricardian workers differently, depending
on the structure of the income tax system and the states reliance on
other sources of revenue. Deficits in states that rely more heavily on

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

169

personal income taxes probably increase current labor supply more


because the effects of other kinds of taxes have two conflicting effects
on labor supply. This is particularly true if asset income is not taxed. If
asset income is taxed, the labor supply effects of a deficit are more difficult to determine but are likely positive in a proportional tax system.
If workers instead believe that deficits will be paid off with a reduction
in future government spending, the effect on labor supply is again
ambiguous but likely either zero or negative.

EMPIRICAL SPECIFICATION ISSUES


AND DESCRIPTION OF THE DATA

The difficulty in estimating a life cycle labor supply equation


derived from a model such as the one written in equations (1) and (2) is
obtaining values for all of the future variables. As outlined in
Blomquist (1985), there are two approaches to this problemeither
model how the expectations of future variables are formed or derive a
supply equation that is conditioned on sufficient statistics, such as
the marginal utility of wealth (518-19). Although the central focus of
this article is whether workers use the fiscal policy variables in forming their expectations, many other factors may help form workers
expectations about future wagestheir age, education, and current
job performance, for example. Rather than trying to specify fully this
expectations process, I instead include both the sufficient statistics
and the fiscal policy variables to see if the latter offer any improvement
over the former. By also estimating the model without the sufficient
statistics, I can examine how important they are to current labor supply and whether including only state fiscal policy variables seriously
misspecifies the expectations process.
Blomquist (1985) discusses two methods that use sufficient statistics: (1) specifying -constant functions (e.g., Heckman and MaCurdy
1980) and (2) two-stage budgeting, developed by Blundell and Walker
(1986). The first method is typically applied to panel data with , or
the marginal utility of wealth, estimated as an individual-specific
fixed effect. Blomquist notes that a nonlinear tax on all income
invalidates the -constant approach because these functions now

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

170

PUBLIC FINANCE REVIEW

contain future prices. The second method, however, remains empirically useful and has the additional advantage that it can use crosssectional data. By viewing the individuals problem as a two-stage
budgeting process, Blundell and Walker derive household labor supplies that condition on the current period allocation out of life cycle
wealth. This methods sufficient statistics are the beginning-of-period
8
assets (henceforth denoted At 1 ) and end-of-period assets (At).
The ability to use cross-sectional as opposed to panel data has
advantages beyond reduced data requirements. My two-period model
is more applicable to cross-sectional data; I observe the present period
with my cross-sectional data, and the future period is simply the
future. Likewise, I do not have to specify when the deficit will be paid
off, only that it will be paid off in the future. If one were to use panel
data, numerous other complicationssuch as dealing with individu9
als who move across states, controlling for changing federal government policy, and modeling individual or time effectswould have to
be addressed. Estimating this model with panel data, properly
accounting for the above-mentioned complications, and exploring the
migration effects of fiscal policy are worthwhile extensions of this
research but beyond the scope of this study. I therefore choose to use a
cross section from the PSID to study whether workers appear to be
Ricardian in their labor supply response to state government policy.
EMPIRICAL SPECIFICATION

I specify the simplest empirical labor supply equation as


h = w + Y + 0 At 1 + 1 At + G + B + Z ,

(11)

where B is the state governments budget balance (revenues minus


expenditures) per capita, Z is a vector of variables affecting tastes, and
all other terms are as previously defined.10 Notice that a deficit (or
debt) means that B < 0. To explore whether the deficit adequately captures the expectations process so that the sufficient statistics are
unnecessary, I also estimate all specifications without the asset
variables.
A nonlinear income tax complicates the empirical model by making w endogenous and because the marginal wage no longer equals the

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

171

average wage. I control for the latter problem by making a virtual


income adjustment to Y (which results from linearizing the budget
constraint), as discussed in Killingsworth (1983). However, this virtual income adjustment is also endogenous. I therefore estimate the
model by two-stage least squares, treating the observed net wage (w),
11
virtual income (Y), and end-of-period assets (At) as endogenous. (At
is a choice variable because it is the assets accumulated by the end of
the period.) I explore whether beginning-of-period assets are endogenous as well. Note that by treating w as endogenous, I am controlling
for the well-known division bias associated with imputed wage
12
measures that are calculated as earnings divided by hours worked.
Likewise, treating Y as endogenous controls for the possibility that the
spouses labor income is simultaneously determined and that any
transfer income is work conditioned. In other words, even without the
nonlinear tax, the wage and nonlabor income are likely endogenous.
The effect of the states budget balance, B, on current labor supply
depends on how workers think future fiscal policy will be affected,
and current state policy seems a reasonable guide. The information the
worker uses is specified here as the states current reliance on income
taxes, which is measured by the proportion of the states revenues
(excluding intergovernmental grants) that come from individual
income taxes, or . The states budget balance is multiplied by this
proportion (and 1 ) to approximate the amount of the deficit/surplus
13
that workers believe will be paid/rebated in the form of income taxes.
Of course, other factors might influence how an individual predicts his
future taxes will be affected, such as the workers age, asset income,
the health of the state economy, and the stringency of the states
balanced-budget rules. These factors are explored in the empirical
section.
A final implication of Ricardian equivalence is that workers internalize government debt/assets. To test for this aspect of Ricardian
behavior, state government net assets are included in one of the specifications. The most general labor supply equation estimated is
therefore
h = w + Y + 0 At 1 + 1 At + G + 1B + 2 (1 ) B + G N + Z ,

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

(12)

172

PUBLIC FINANCE REVIEW

where N is per capita state net assets. Several hypotheses emerge from
this specification. An obvious test is whether the s and G are zero.
Rejecting this hypothesis suggests that workers use the governments
budget constraint and are behaving in a Ricardian manner. If is relevant as workers form their expectations about future wages, then 1 2.
In addition, 1 should be negative. This occurs because a state budget
deficit (B < 0) that workers believe will be paid off with future income
taxes should increase current labor supply. Finally, if workers view the
state governments assets/debt as their own, G should equal 0.
DESCRIPTION OF THE DATA

The survey data used in the empirical analysis are from the University of Michigans PSID for 1980. Poterba (1994) notes that in the
early 1980s, states and localities were in near fiscal balance and
reports that total year-end balances as a percentage of expenditures
during 1978-1993 were highest in 1980. He also finds that unexpected
state budget surpluses lead to much smaller fiscal policy changes than
unexpected deficits. These two results combine to suggest that by
using 1980, I am perhaps less likely to find evidence of Ricardian
behavior than if a year exhibiting more fiscal stress had been chosen.
Any bias associated by choosing 1980 as the year to study should
therefore work against finding Ricardian behavior.
I restrict my sample to males between the ages of 25 and 55 to create a more homogeneous sample and limit my study to (potential) fulltime labor force participants. Due to both data limitations and the difficulties in correctly modeling the labor supply behavior of moonlighters and self-employed workers, such individuals were deleted
from the sample. I also limited the sample to those who were either
14
paid hourly or were salaried. Less than 5% of the men in the sample
had annual hours worked of zero; therefore, a self-selection model
appeared infeasible, and such observations were dropped. Individuals
who lived in more than one state during the survey year were also
dropped. I also eliminate the low-income Survey of Economic Oppor15
tunity subsample and use only the random sample. Table 1 reports
the variable definitions, means, and standard deviations for the
sample.

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

TABLE 1:

173

Definitions, Means, and Standard Deviations of the Variables Used in


the Analysis

Variable Definition

Mean

Annual hours worked


2,138.54
Net hourly wage in dollars
6.40
Net virtual income in dollars
10,292.50
Beginning-of-period assets in dollars
40,424.06
End-of-period assets, deflated to 1980 dollars
39,806.62
1,130.30
State government spending per capita, G a
State capital spending per capita, K
101.18
State budget balance per capita, B b
67.21
NASBO measure of budget balance
33.08
Proportion of state revenue coming from individual
income taxes, c
.178
State net assets per capita, N d
643.68
Education (in years)
13.06
Age, limited to 25 to 55
36.95
Health dummy variable equaling 1.0 if a health
condition severely limits work
.07
Number of children age 17 or younger
1.26
Number of children age 6 or younger
.78
Dummy variable equaling 1.0 if the person is married
.90
Dummy variable equaling 1.0 if the person is white
.92
Dummy variable equaling 1.0 if lives in Northeast region
.21
Dummy variable equaling 1.0 if lives in North Central region
.33
Dummy variable equaling 1.0 if lives in the Southern region
.29
Additional state characteristics
Unemployment rate
7.54%
Change in state personal income 1979-1980
.46%
Urbanized
71.63%
With a high school education
68.05%
Nonwhite
13.38%

Standard
Deviation
502.97
2.64
7,920.04
49,878.77
45,606.50
197.72
42.61
65.58
34.23
.087
370.49
2.61
9.18
.26
1.17
.98
.30
.28
.41
.47
.46
2.16
2.3
12.54
5.68
6.82

SOURCE: National Association of State Budget Officers (NASBO 1981); U.S. Bureau of
the Census (1981).
NOTE: Number of observations = 881.
a. Includes all general state expenditures = direct state expenditures + intergovernmental grants + insurance trust expenditures.
b. Defined as state total revenues minus total general state expenditures (defined in
Note a).
c. Defined as individual income taxes divided by total state revenue, excluding intergovernmental revenue.
d. Defined as total state government cash and security holdings minus total state debt.

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

174

PUBLIC FINANCE REVIEW

The labor supply variable used is annual hours worked. Two hourly
wage measures are available: an imputed wage (earnings divided by
hours worked) and an hourly wage that is reported by the respondent.
Because most research uses the imputed wage, for the sake of brevity,
I report only the imputed wage estimates and discuss any differences
between the two measures. Before-tax nonlabor income includes all
noninterest income of the household minus the husbands labor
income. The PSID does not have much information on consumption
or saving, so the asset variables had to be constructed. I define assets in
a manner similar to Ziliak and Kniesner (1996), who also estimate a
life cycle labor supply model with nonlinear taxes using the two-stage
16
budgeting method and data from the PSID. In particular, assets have
a liquid and an illiquid component. Liquid assets are derived by dividing the husbands and wifes nominal rent, interest, and dividend
income by a nominal interest rate; the passbook savings rate of 5.5% is
used for the first $400 of income, and the average 3-month T-bill rate is
17
used for all income exceeding $400. The illiquid component is the
value of home equity and is defined as the difference between house
18
value and the outstanding principal remaining. End-of-period assets,
At (which refer to the beginning of 1981), are deflated into 1980 dollars using the implicit price deflator for total personal consumption
expenditures.
Estimated federal income taxes and marginal tax rates are available
in the PSID, but it was necessary to calculate federal payroll and state
19
income taxes and marginal tax rates. These (state + federal) tax rates
and tax bills are then used to construct the net (after-tax) wage, w, and
20
the linearized or virtual nonhusband labor, noninterest income, Y. I
follow Conway (1997), who uses the same data to estimate the effect
of government spending on labor supply, in my choice of government
spending measures. My primary measure of G (and the only one for
which results are reported) is total general state government expenditures per capita (thereby including both direct expenditures and intergovernmental grants by the state) because it appears to be the least
21
ambiguous or endogenous measure. Local government spending is
more likely within the individuals control and is much harder to identify correctly for each individual. Federal spending also differs by

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

175

state, but it too is a more ambiguous measure because many federal


expenditures may benefit more than one state. However, because Conway (1997) finds that different categories of spending have different
labor supply effects (e.g., transfer spending is always strongly negative) and that spending undertaken at different levels of government
has a different impact, I explore the sensitivity of my results to the
measure of G used.
The states budget balance, B, is calculated as the states total revenues minus total expenditures, and the states net assets, N, is state
assets minus debt. Both variables are calculated per capita. The states
reliance on income taxation, , is the proportion of the states revenues
(excluding intergovernmental grants, which are arguably out of the
state governments control) that come from individual income taxes.
All of the information needed to construct these variables is found in
State Government Finances in 1980 (U.S. Bureau of the Census
1981), and further details are provided in the notes to Table 1.
This measure of the budget balance may be deficient, however, in
two respects. First, Poterba (1994) argues that social insurance funds
should be removed from the calculation of the budget balance, and he
uses the budget balance as reported by the National Association of
State Budget Officers (NASBO) in their Fiscal Survey of the States
1980-81 (NASBO 1981). Although this measure is highly correlated
with mine, it has a lower mean and exhibits less variance across the
states. I therefore also estimate many of the specifications using the
NASBO measure.
The other possible deficiency is the treatment of capital expenditures. Capital budget balances may have a different effect than operating budget balances on workers expectations because capital expenditures by definition yield benefits and possibly even tax revenues in
the future (as in a toll road or, more generally, any public investment
that increases future economic growth). Thus, a capital budget deficit
may require a smaller increase in future taxes than an operating budget
deficit. Unfortunately, only 29 states had capital budgets in 1962, and
that number increased to only 34 states by 1986; furthermore, states
have differing definitions of capital goods (Poterba 1995, 168-69).

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

176

PUBLIC FINANCE REVIEW

Decomposing the budget balance variable into operating and capital


components is therefore not feasible. Rather, I explore this issue by
treating capital expenditures, for which there are data available, dif22
ferently from other kinds of expenditures. In particular, the most
general treatment of capital expenditures allows the direct effects of
such spending (the coefficient on G) and the budget effects (the coefficient(s) on B) to differ:
h = w + Y + 0 At 1 + 1 At + 1 (G K )

(13)

+ 2 K + 1 ( B + K ) + 2 (1 )(B + K ) + Z ,

where K is capital expenditures, and 1 and 2 measure the direct


effects of noncapital and capital expenditures, respectively, on labor
supply. The differential effects of capital versus operating deficits
comes in through . If there is no difference, then equals zero. If the
direct effects are also the same (1 = 2), then equation (13) reduces to
the usual specification. At the other extreme, if all capital expenditures are viewed as generating discounted future tax revenues equal to
their current cost, then they add nothing to the deficit and should be
omitted from the expenditure side of the balance calculation; hence,
equals 1.0. The model written in equation (13), however, is highly
nonlinear in coefficients on variables that vary only over states, and
estimating it may be asking too much of the data. I therefore also estimate a linear model in which the direct effects are assumed equal (1 =
2), and = 1; this specification essentially redefines the balance to be
total tax revenues minus total noncapital expenditures.23
After all of the exclusionary restrictions mentioned above are
24
imposed, the sample contains 881 observations spanning 43 states.
The sample average state balance per capita is $67, indicating a surplus, but ranges from a deficit of ()$109.50 to a surplus of $338. Likewise, most states assets outweighed their debts, yielding a sample
average of $644 but ranging from $173 to $2,301.50. The proportion
of state revenues coming from individual income taxes ranges from
zero to 34.7%, with a sample average of approximately 18%. The
states contained in the sample therefore vary widely in their budgetary
situations.

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

177

EMPIRICAL RESULTS

Table 2 reports the two-stage least squares estimates of the key


parameters in equations (11) and (12) under a variety of assumptions,
including a base model that omits the deficit variables (column 1). All
of the main models are estimated three ways: (1) including both asset
variables and instrumenting end-of-period assets, (2) including and
instrumenting both asset variables, and (3) omitting the asset variables
under the assumption that the deficit adequately captures how future
predictions are formed. Because the treatment of the asset variables
does not substantively affect the results, only a subset of these estimated models is presented. In general, instrumenting beginning-ofperiod assets produces similar coefficients but larger estimated standard errors, a result consistent with using a less efficient estimator
25
(Method 2 above versus Method 1). I therefore report those models
that instrument only end-of-period assets.
Several results emerge from this table. The wage coefficient is consistently negative and sometimes statistically significant, suggesting a
wage elasticity ranging from .39 to .18, which is at the low end of
the range reported by Pencavel (1986, 69) but is very close to the
unconstrained estimates produced by MaCurdy, Green, and Paarsch
(1990), who also use an IV procedure and a cross section of the PSID.
The virtual income coefficient is always positive yet statistically insignificant. Although it is disconcerting to find a zero or even positive
income effect (h@Y 0), such a result is consistent with other
empirical male labor supply research that deals with progressive
income taxation and does not implicitly impose h@Y to be negative
(see Pencavel 1986, 69; Moffitt 1990; MaCurdy, Green, and Paarsch
1990; Triest 1990). The two asset variables coefficients are of similar
magnitude and usually opposite sign, such that the sum is typically
zero or slightly negative. Neither coefficient is statistically significant.
Thus, the sufficient statistics do not appear to have much empirical
importance, whether or not the deficit is included. This may be due to
the high collinearity between the two asset variables and the virtual
income variable and the fact that at least two of the three must be
instrumented. Nonetheless, economic theory mandates that all three
variables be included (see Note 8).

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

178

PUBLIC FINANCE REVIEW

TABLE 2:

Coefficient

Summary of Two-Stage Least Squares Results (t statistics in


parentheses)

Base

Model 1

Model 2

Model 3 Model 2a Model 3a

The basic models


w
90.19*
87.97* 82.78
77.36
76.38 69.22
(1.89)
(1.65)
(1.55)
(1.43)
(1.59) (1.39)
Y
.024
.026
.022
.019
.020
.018
(1.06)
(1.20)
(1.07)
(.94)
(1.15)
(1.02)
.0027
.0026
.0019
.0016
A t 1
(.88)
(.84)
(.66)
(.63)
At
.0033
.0032
.0022
.0016
(.65)
(.62)
(.44)
(.38)
G
.108
.113
.025
.025
.057
.051
(.96)
(1.00)
(.19)
(.20)
(.48)
(.43)
.037
Balance, B
(.11)
4.167
3.969
3.828 3.577
B
(1.42)
(1.36)
(1.39) (1.28)
.900
.987*
.856
.957*
(1 )B
(1.63)
(1.71)
(1.60)
(1.68)
Net assets
.026
.030
(.34)
(.51)

Coefficient

NASBO
and
NASBO Exclude K Exclude K

Add K

NASBO
and
Add K

Using
Debt

Alternative Measures
of B in Model 2b
w
92.86**
67.23 90.51**
61.42 108.50**
98.75**
(2.00)
(1.33)
(2.02)
(1.20)
(2.30) (2.14)
Y
.023
.009
.006
.013
.010
.023
(1.07)
(.46)
(.30)
(.69)
(.50)
(1.15)
11.065**
1.593 2.901*
.479
5.728
.789**
B
(1.00) (2.04) a
(2.22) a (1.06) (1.72) a (.12)
(1 )B
2.302**
.800**
1.175**
.427
1.077
.143*
(2.14)
(2.66)
(2.84)
(.48)
(.82)
(1.69)
1.648
1.927
K
(.50)
(.72)
1.385*
1.362**
(1 )K
(1.84)
(1.965)

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

179

TABLE 2 Continued

Coefficient
Other Variations
on Model 2b
w

Y
B
(1 )B

X B
X (1 )B

X=
Asset
Dummy

85.03
(1.56)
.026
(1.22)
5.312
(1.62)
.916
(1.62)
2.00
(.95)

X = Age

X=
1/Age

88.26*
84.30
(1.66)
(1.62)
.022
.023
(1.06)
(1.08)
9.740
a
(1.45)
1.919
(1.03)
.148 148.536*
(.74)
(1.68)
.028
31.318*
(.55)
(1.79)

X=
X=
% Change Including
Strict Rule State
State
Dummy Income Variables

70.25
79.97 129.48**
(1.23)
(1.61) (2.26)
.026
.027
.032
(1.28)
(1.14)
(1.15)
5.235
.236 6.97**
(1.01)
(.08)
(2.40) a
2.359
.008
1.65**
(1.34)
(.01)
(2.21)
1.357 213.546
a
(.25)
(1.64) a
1.533
33.785*
(.81)
(1.77)

NOTE: NASBO = National Association of State Budget Officers.


a. The null hypothesis that the B and (1 )B coefficients are equal is rejected at the
10% level or better.
b. The coefficients on the asset and government spending variables are not reported for
the sake of brevity.
* Statistically significant at the 10% level. ** Statistically significant at the 5% level.

The effect of state government policy is also fairly stable over the
specifications. As in Conway (1997), the effect of state government
spending on labor supply is usually negative; however, its magnitude
and statistical significance decrease as Ricardian behavior is permitted. This raises the question of whether state government spending,
when included in isolation, is actually capturing the budgetary situation of the state, or if trying to identify both a state spending effect and
26
a state budget balance effect is simply asking too much of the data.
Turning to the state budget variables, it is evident that the states
current revenue structure is important to the effect of the budget balance on labor supply. In the model that does not include this information (equation (11) and Model 1 in Table 2), the effect of the budget
balance is essentially zero. When included (Models 2 and 3), the coefficients have signs that are consistent with Ricardian behavior. In particular, the portion of the budget balance predicted to be paid/rebated

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

180

PUBLIC FINANCE REVIEW

via income taxes (B) has a negative, but not quite statistically significant, effect. This suggests that a budget deficit (B < 0) in a state that
relies heavily on income taxes may increase current labor supply
just what is expected of Ricardian agents. The other portion of the
budget balance is positive and of about one fourth the magnitude of the
income tax part. Although this pattern is quite consistent across specifications, the individual coefficients are only marginally significant at
best, and one cannot quite reject the hypothesis that these two coefficients are equal.
Another implication of Ricardian equivalence is that workers treat
government debt or assets as if it was their own. Model 3 incorporates
this possibility. The results here are inconclusive; although the
hypothesis that government net assets have the same effect as private
assets cannot be rejected, it is likely because both coefficients are
approximately zero. Imposing the restriction that they be equal had no
real impact on the results (results available upon request). Finally,
omitting the asset variables (Models 2a and 3a) has no impact on the
empirical results. Taken together, the results of Table 2 are weakly
consistent with Ricardian behavior.
The lower two panels of Table 2 report several variations on Model 2.
I do not report the asset and government spending variables coefficients because they are fairly consistent across the specifications and
never statistically significant. The first panel reports results from
models in which alternative measures of B are used. The NASBO
measure, which Poterba (1994) and others prefer, yields similar but
much more statistically significant results. In addition, the magnitudes of the coefficients are greater, and the hypothesis that they are
equal is easily rejected at the 5% level. The next four columns address
the issue of capital versus operating budgets. The nonlinear model
written in equation (13) appears to be asking too much of the data;
none of the key economic variables is even close to being statistically
significant, and the results are therefore not reported. I do report two
more restrictive linear specifications. The first simply excludes capital
expenditures from the calculation of the budget balance and yields
coefficients of smaller magnitudes but otherwise similar results,
regardless of the deficit measure used (columns 2 and 3). The second
adds capital expenditures, multiplied by and (1 ), respectively, to

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

181

the model. If these coefficients are equal to their budget balance counterparts, then excluding capital expenditures from B is supported. If
they are equal to zero, then no special treatment of capital expenditures is required (see Note 23). The results support the idea that capital
deficits have a different effect than operating deficits (i.e., the first
hypothesis is not rejected and the second is) on labor supply. The last
column reports estimates from a model in which state net assets are
used instead of the budget balance. Net assets are a more long-run
measure of a states financial situation and are less prone to temporary,
cyclical fluctuations. The results from this model are essentially the
same, if not stronger, as when the budget balance variable is included.
In sum, the second panel reveals that using alternative measures of the
states fiscal situation indeed tends to strengthen the evidence that
workers are behaving in a Ricardian manner.
The last panel allows the labor supply effects of deficits to depend
on other factors and uses the original measure of the budget balance,
B. (The results using the NASBO measure are very similar, unless otherwise noted.) The first column allows for a differential effect of deficits financed with future income taxes (B) for households that have
asset income. (Recall that it is the taxation of asset income that makes
this variable theoretically ambiguous.) Although the results are consistent with theory in that having asset income dampens the effect of
B, the difference is not statistically significant. As discussed earlier,
an individuals age may also affect his future predicted tax burden.
The next two columns find some support for allowing the effect of the
deficit variables to depend on the workers age. Column 2s results are
theoretically consistent in that the effects of the deficit grow weaker as
27
an individual ages, but the difference is not statistically significant.
The third column allows the size of the coefficients to decrease as a
person ages; this specification improves the statistical significance of
the balance coefficients but otherwise yields similar results.
The last three columns consider the effect of other state characteristics. Poterba (1994) discusses how states vary in their stringency of
rules mandating balanced budgets and finds that such rules matter in
predicting states responses to fiscal crises. One might then expect
workers to be more Ricardian in states with stricter budget rules. In
column 4, I employ the same variable as Poterba (1994) and see if it
28
affects the balance coefficients. I find no evidence that this variable is

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

182

PUBLIC FINANCE REVIEW

importantthe coefficients are of the wrong sign and are statistically


insignificant. Column 5 repeats the same exercise with the percentage
change in state personal income. In this case, the interaction terms
become more important than the original variables, although they
have the same sign and general relative magnitude (although the coefficient on B is now about six times as big as that on (1 )B). Thus,
the stronger a states growth rate, the bigger the labor supply effect of a
deficit will be, a rather counterintuitive result because a high rate of
economic growth should decrease the workers prediction of his
future tax burden. These confounding results could be due to other
state-specific influences, a possibility I explore shortly. They could
also be due to an unfortunate by-product of this specification that
requires, by construction, negative state growth to cause a flip in the
signs of the coefficients (e.g., the true coefficient on B is now positive). In addition, using the NASBO measure did not reveal any such
pattern.
Overall, the results provide some support for the notion that workers use the deficit (and the states reliance on income taxes) to forecast
future variables. Although the statistical significance of the budget
balance variables is marginal, the overall pattern is consistent across
specifications: Deficits in states that heavily rely on income taxes are
more likely to increase labor supply. The statistical insignificance of
the sufficient statistics, the asset variables, is troubling, however. One
might expect them to be unimportant if modeling the expectations
process via the deficit is empirically superior. However, they are also
insignificant when the deficit variables are omitted. One explanation
is that my asset measures, especially when instrumented, are so noisy
and collinear as to render the coefficients statistically insignificant.
Another explanation is that workers are myopic, and the life cycle
model is inappropriate; however, that requires the deficit also to have
no effect. Perhaps, as also suggested by the state personal income
interaction terms, the deficit variables are instead capturing some
other effect. I explore this and other possible explanations next.
A FURTHER EXPLORATION

Could there be an alternative explanation for these results? One


might suspect that the arithmetic relationship between deficits and

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

183

observed labor supply is responsible. Reduced labor supply decreases


state tax revenues in states that tax labor income and thereby induce a
deficit. Changes in labor supply might be causing changes in the deficit, rather than vice versa. Recall, however, that my results suggest that
deficits in states that rely heavily on income taxation increase labor
supply more (or decrease labor supply less) than states that do not, a
result that is directly counter to the arithmetic relationship between
labor supply and deficits.
Another argument is that these results are a peculiarity of my model
specification choices, especially because the statistical significance of
the key coefficients fluctuate and are marginal. To explore this possibility, I reestimate the model by alternately (1) using reported wages;
(2) expanding my sample to include the low-income subsample, both
with and without weighting; and (3) breaking government spending
into categories and including local spending also, as in Conway
(1997). Both (1) and (2) tend to further reduce the statistical significance of the balance coefficients but typically yield estimated coefficients of a similar magnitude. The third exercise tends to strengthen
the results, if anything.
29
I also reestimate the model omitting the six states with = 0. This
omission greatly reduces both the size and significance of the deficit
coefficients, suggesting that the distinction between states with personal income taxes versus those without is important to the labor supply effects of deficit finance. To further explore this, I redefine to be a
30
dummy variable equal to 1.0 if the state has an income tax. This also
tends to reduce the size and significance of the deficit coefficients,
especially that of B. These two exercises suggest that the discrete
(income tax/no income tax) and continuous aspects of the state tax
structure combine to produce important labor supply effects of deficit
finance.
State deficits also might be picking up any number of state characteristics (such as other state taxes or the local economy) that affect
labor supply behavior. The last column in the bottom panel of Table 2
reports estimates from a model that includes other state-specific characteristicsthe state unemployment rate, the percentage change in
state personal income between 1979 and 1980, the percentage urbanized, and the percentage of the state population that is high school

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

184

PUBLIC FINANCE REVIEW

educated and nonwhite. The coefficients actually increase in statistical significance when these variables are included.31
A less restrictive model specification than the one(s) estimated
above would permit state-specific random effects. I reestimate the
model, controlling for random state effects using the two-step method
outlined by Amemiya (1978) and Borjas and Sueyoshi (1994).
Amemiya (1978) proves that this method yields coefficients on the
state-level variables that are algebraically identical to the usual GLS
random effects model, but it is much simpler computationally, especially when the groups are of different sizes as is the case here. (That
is, I do not have the same number of observations for each state.)
Briefly, the method involves first estimating the model with fixed
state effects (accomplished by including J state dummies and excluding the common intercept and all state-specific variables). In the second stage, the estimated coefficients from these state dummies, , are
then regressed on a vector of state-specific characteristics, X, including the state government variables, or
$ = X + = X + u + e,

(14)

where u is the error arising from the fact that only an estimate of is
available, and e is the inherent randomness in the state effects that cannot be explained by X. Notice that u is both heteroskedastic and serially correlated, as the estimated dummy coefficients have different
variances and are correlated with one another. One way to deal with
this is to adjust the standard errors for the known heteroskedasticity
and serial correlation (i.e., Cov( $ ) = ( X X ) 1 X Cov( ) X ( X X ) 1 ).
The other more efficient method, at least asymptotically, is to perform
GLS. Both require an estimate of the covariance matrix of . This
covariance matrix can be written as
Cov ( ) = Cov (u) + 2e I = Cov ($ ) + 2e I ,

(15)

where I is a J J identity matrix. Cov()


$ is available from the first2
stage estimates, but e must be estimated. Amemiya (1978) suggests
the simple variance calculated from estimating equation (14) via ordinary least squares (OLS) as a consistent estimator because asymptotically, Cov()
$ approaches zero (thereby also negating the need for

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

TABLE 3:

185

Summary of Results From the Two-Stage Procedure Controlling for


Group Effects (t statistics in parentheses)a

First-stage coefficients:
w = 73.56 (.84), Y = .0043 (.13), A t 1 = .0017 (.40), At = .0017 (.23)

Model 1
Second-Stage
Coefficients
G
Balance, B
B
(1 )B
Net assets

OLS
.162
(.59)
.915
(.91)

GLS
.031
(.14)
.426
(.55)

Model 2
OLS
.132
(.51)

5.656
(1.37)
1.573
(1.22)

Model 3

GLS
.058
(.26)

OLS
.099
(.39)

5.816* 5.487
(1.73)
(1.33)
1.628*
1.016
(1.67)
(.63)
.094
(.58)

GLS
.064
(.29)

5.811*
(1.72)
1.759
(1.44)
.024
(.18)

NOTE: Tendency of other variables in OLS models: unemployment rate (), % change
state personal income (), urban (+), education (), nonwhite (); no |t statistic| > 1.19.
Tendency of other variables in GLS models: unemployment rate (), % change state
personal income (), urban (+), education (), nonwhite (); no |t statistic| > .86. OLS =
ordinary least squares; GLS = generalized least squares.
a. The OLS standard errors are calculated using the correct standard errors; that is,
Cov ($ OLS ) = (X X ) 1X Cov () X (X X ) 1.
* Statistically significant at the 10% level. ** Statistically significant at the 5% level.

GLS). Borjas and Sueyoshi (1994) suggest an estimate that adjusts for
the fact that the simple variance also includes Cov()
$ by subtracting
the mean variance attributable to .
$ Unfortunately, in my model, this
adjustment yields a negative estimate of 2e , so I must use Amemiyas
estimator, which is likely biased upwards.32
Because of this problem and the somewhat dubious practice of
appealing to asymptotic superiority in a model with a (second-stage)
sample size of 43 observations, I report both the GLS estimates and
the OLS estimates in Table 3. The other state-specific characteristics
included in X are those listed above. The sign and general magnitude
of the budget balance coefficients are unchanged by modeling random
effects, although the statistical significance of these coefficients is
diminished by the asymptotically less efficient OLS method. I also
estimate these models without the asset variables and again obtain

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

186

PUBLIC FINANCE REVIEW

almost identical results. Thus, the empirical support for Ricardian


workers found in Table 2 also survives the inclusion of random state
effects.
EVALUATING THE RESULTS

The results of the empirical analysis therefore suggest that although


the statistical significance of the budget balance variables is sensitive,
the estimated labor supply effects are fairly consistent across different
model specifications and measures of the deficit. Deficits in states that
rely heavily on income taxation increase current labor supply,
whereas deficits in states that do not rely on income taxes decrease
current labor supply. My results (for the models without interactions)
suggest that the proportion () in which deficits will have no effect
(the two parts just offset each other) ranges from approximately .153
to .232. (If capital expenditures are excluded from the balance, then
the proportion rises to around .3.) For the 22 states in my sample with
< .153, a deficit reduces labor supply, whereas for the 8 states with
> .232, it increases labor supply. The labor supply effect of a deficit
can therefore be positive or negative for the remaining 13 states,
depending on which set of estimates are used. Although the coefficients on B and (1 )B suggest rather substantial labor supply
effects (e.g., a one dollar per capita increase in the income tax
financed deficit results in a 5-hour increase in annual labor supply),
33
the combined effect is likely small for most states. However, the fact
that the budget balance, when combined with information about how a
state might be expected to finance/rebate that balance, matters at all to
current labor supply supports the hypothesis that workers are indeed
Ricardian.

CONCLUDING REMARKS

This research investigates a model of labor supply that permits


workers to be Ricardian or aware of the state governments intertemporal budget constraint. The theoretical analysis derives the labor supply effects of deficit finance under a variety of possible tax structures.
The empirical results suggest that state government budget deficits

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

187

increase male labor supply more (or decrease it less) the more the state
relies on individual income taxes, suggesting that workers are
Ricardian and adding a new kind of evidence to the Ricardian equivalence debate. For a number of reasons, however, these results are not
as definitive as one would hope and thus point to the need for additional empirical studies. First, the key coefficients statistical significance varies a great deal. Part of this may be due to the conservative
choice of 1980 as the year of study. Choosing 1980, a very good year
for the states, may very well have resulted in smaller estimated
Ricardian effects than if a year was used when more states were in fiscal distress. The lack of significant income and asset coefficients is
also troubling but entirely consistent with other empirical labor supply
research.
Finally, the results found here for state government policy should
not be directly applied to an analysis of the labor supply effects of federal deficits because of the multitude of differences between federal
and state government policy. These differences affect the likelihood
that the worker will indeed face higher future taxes as a result of a current deficit, a necessary assumption for Ricardian behavior to be evident. Foremost, unlike most states, the federal government does not
mandate a balanced budget and has demonstrated a willingness to run
deficits continually. As noted by Poterba (1997, 79), the federal government also likely has better options for financing a deficit, such as
the ability to print money and greater access to credit markets. On the
other hand, workers can avoid repaying a current state deficit by moving to another state, an option typically not available at the federal
level. Anticipated future tax increases at the state level are also more
likely to be capitalized (Poterba 1997). And states are much more
likely to run surpluses than the federal government, which have been
found to cause weaker fiscal reactions by state governments than deficits (Poterba 1994). Thus, the extent to which a current deficit is
rationally forecasted to cause a future tax increase for the individual
34
likely differs between the states and the federal government. The evidence presented here suggests that workers are using state government deficits/surpluses in a manner consistent with Ricardian behavior. In situations in which workers are Ricardian, Hansson and Stuart
(1987) show that postponing taxes on labor income via deficit finance

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

188

PUBLIC FINANCE REVIEW

may actually be a desirable (and not neutral) policy by increasing both


current labor supply and savings. At the very least, this analysis suggests that mens work decisions are influenced not only by state
income taxes but by other aspects of state fiscal policy as well.
APPENDIX

With a proportional tax system, the consumers budget constraint


can be written more simply as
A0 + (1 t1 )W1h1 + (1 t1 )Y1 C1 (1 + 1 ) f1
+

(1 t2 )W 2 h2
C (1 + 2 )
f2
(1 t2 )Y2

2
+
,
1 + r(1 t2 ) 1 + r(1 t2 ) 1 + r(1 t2 ) 1 + r(1 t2 )

where all terms are as previously defined. The comparative static


h 1 h 1
can be simplified as
result for
=
t 2

Vcl (1 + 2 )( W 2 rW 2 (1 t2 ))

1 + r(1 t )

2
2
2 W (1 t )V + (1 + ) rV

ll
2
2
cc
2
h 1 ~
,
=
t 2
+ ( W h + Y + rC (1 + ) + rf ) V V V 2
(
)
ll cc
cl
2 2
2
2
2
2

where

(U (1 + 1) U cc (1 t1)W1) 1 > 0 ,
~ = lc
|H |
( 1 + r(1 t2 )) 2
and Uij and Vij are the second partial derivatives of the utility functions, and Uii, Vii < 0 and Uij, Vij > 0 for all i j. Recall that , t, and r are
all tax or interest rates and therefore are between 0 and 1.0. This information allows us to determine that all of the terms are positive except
for the one involving V ll , which is negative. Thus, we must assume

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

189

that (1 + 2 ) 2 r V ll is outweighed by the rest of the terms in the equah h


tion to conclude that 1 or 1 is positive.
t 2

NOTES
1. Hansson and Stuart (1987), Judd (1987), and Quintieri and Rosati (1988) are among
those who discuss the theoretical effects on labor supply of distortionary taxation and deficit
finance. To my knowledge, there is no accompanying empirical evidence using microlevel data.
2. Conway (1994, 215) also suggests using state fiscal policy and microeconomic data as
a new avenue for empirically testing the fundamental assumptions underlying Ricardian
equivalence.
3. Using time-series, cross-sectional data, Rogers and Rogers (1993) find that although
few states have a balanced budget in any one year, each state obeys its budget constraint over
time. In addition, a growing literature that examines the effects that state budget rules have on
state fiscal policy (e.g., Poterba 1994, 1995, 1997) offers support that these limitations do indeed
matter.
4. See Barro (1989) and Bernheim (1987, 1989) for further discussion of this debate.
5. For instance, Y could be the labor income of the spouse, assuming that the two individuals make their labor supply decisions independently. This assumption is explored further in the
third section. Y does not include asset income; such income is not exogenous in a life cycle
model.
6. The only additional alteration to equation (3) is the inclusion of Y in the tax function that
is multiplied against |H51|.
7. For instance, in the year used in the empirical analysis1980the highest tax bracket
for married couples began at a household income of $13,000 or less in 19 of the 41 states that had
a broad-based income tax.
8. Blomquist (1985) notes that these measures may be replaced by net savings over the period (S = A (1 + r (1 t ))A 1 ) if the income tax system is proportional. Because all workers
face the nonlinear federal income tax, I must include both measures.
9. As mentioned shortly, individuals who moved across states were deleted from my crosssectional sample. With a longitudinal sample, this omission reduces sample size much more
because the individual is deleted if he moved at any time during the panel.
10. Specifically, Z includes age, age squared, education, education squared, number of children younger than age 6 in the household, number of children younger than age 17, health status,
and marital status. Z also includes a constant. Notice that because all individuals face the same
federal budget balance, its effect cannot be estimated and is instead contained in the common
intercept. This is precisely why state fiscal policy presents an interesting experiment of
Ricardian behaviordifferent individuals face different state budget balances at a given point in
time.
11. The reduced form wage, virtual income, and ending assets equations also include the
cubic polynomial of age and education (as in Mroz 1987), the state unemployment rate, and
dummy variables for the individuals race and region of the country.
12. See Killingsworth (1983) for further discussion of this problem.
t

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

190

PUBLIC FINANCE REVIEW

13. An alternative approach is to calculate a person-specific as the proportion of current


state taxes paid by each individual in income taxes. However, this is the result of the workers
utility maximization problem and is therefore endogenous. Using this measure would add two
more endogenous variables (B and (1 )B) to a system that already has three or four endogenous variables (w, Y, and the asset variables). Furthermore, the goal here is to include the
parameters of the workers problem, not the utility-maximizing solution, in the labor supply
equation.
14. These groups of workers were omitted from the sample because it is very difficult to
measure their true marginal wage. For moonlighters, identifying the marginal job is problematic,
whereas for the self-employed, distinguishing the returns to labor from the returns to capital is
difficult. Likewise, it is difficult to theorize the true marginal hourly wage facing a commission
or piecework employee. In addition, these observations also made up a disproportionate number
of the extreme outliers and those with apparently miscoded data.
15. In the empirical section, I estimate the key models using a bigger data set that includes
the low-income subsample, both with and without using the weights provided by the Panel Study
of Income Dynamics (PSID) to explore the impact that an increased sample size and the use of
weights has on the results.
16. Ziliak and Kniesner (1996), in turn, draw on Zeldes (1989) and Runkle (1991) in their
construction of the liquid component of assets.
17. The average 3-month T-bill rate was 11.5% in 1980 and 14% in 1981.
18. The latter variable is unfortunately truncated at $99,999; dropping observations with
this truncated value reduces the sample by one observation.
19. The employees share of Social Security and Medicare taxes in 1980 was 6.13% with a
ceiling on taxable earnings of $25,900 (Browning and Browning 1994, 408). Statutory tax rates
and brackets, standard deductions, exemptions, and other characteristics of the state personal
income tax systems were found in Significant Features of Fiscal Federalism 1979-1980 Edition
(Advisory Commission on Intergovernmental Relations 1980) and the State Tax Handbook
(1980). Deductibility of federal income taxes for all taxpayers (including nonitemizers) was
accounted for, as well as any universally applied tax credits. If the two documents contradicted
each other or if necessary information was missing, state tax officials were contacted.
20. Specifically, income taxes on before-tax nonhusband labor, noninterest income are estimated using the households average tax rate (estimated total taxes paid divided by total taxable
income) and then subtracted. The virtual income adjustment is calculated by multiplying the difference in the marginal tax rate and the average tax rate by the husbands labor income, or (tm
ta)Wh, where tm and ta are the marginal and average tax rates, respectively. See Mroz (1987) or
Killingsworth (1983) for further discussion.
21. State expenditures were found in Table 9 of State Government Finances in 1980 (U.S.
Bureau of the Census 1981).
22. Total state government expenditures on capital outlays are found in Table 12 in State
Government Finances in 1980 (U.S. Bureau of the Census 1981).
23. An intermediate linear model would be one that restricts the direct effects to be equal but
does not restrict to equal 1.0. This involves adding K and (1 )K as regressors to the model;
their corresponding coefficients equal 1 and 2 , respectively.
24. Not surprisingly, the states not included tend to be states with smaller populations (and
hence fewer initial observations)Alaska, Delaware, Hawaii, Idaho, Montana, North Dakota,
and Vermont. Alaska and Hawaii are frequently omitted from studies of the states anyway
because they are such outliers in so many ways. The remaining five states are somewhat unique
in that they are low-population states, but fortunately in each case, the sample includes similar
states from the same region.

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

191

25. If anything, the coefficients are of a larger magnitude when both asset variables are
included so that the overall significance levels are mostly unchanged.
26. For instance, state government spending and the budget balance are highly correlated;
the Pearson correlation coefficient calculated for these two variables using state-level data (43
observations) is .279, which is statistically significant at the 7.0% level. Likewise, transfer
spending and are very highly correlated.
27. Specifically, the coefficients on B and (1 )B are reduced for each year. For instance,
the coefficient on B is approximately 5.3 for a 30-year-old and 2.34 for a 50-year-old.
28. The Advisory Commission on Intergovernmental Relations (1987) cataloged the rules
and assigned an overall score between 1 and 10 to the stringency of a states rules. The dummy
variable, STRICT, is equal to 1.0 if the score is 6 or higher.
29. This omission is not the same as deleting states without a broad-based income tax. Connecticut, New Hampshire, and Tennessee all have narrowly defined taxes on asset income, so
that , although quite small, does not equal zero for them even though they have no broad-based
income tax.
30. I define this two ways; one is if the original > 0, and the second is if the state has a
broad-based income tax. See Note 25.
31. This result is not confined to this one specification; including the state-level characteristics tended to increase the statistical significance of the deficit variables in most specifications.
32. This likely arises because several states have only one or two observations and therefore
have very large dummy coefficient variances.
33. For example, the labor supply effect of a one dollar per capita increase in the deficit in
the state with the heaviest reliance on income taxes ( = .347) would increase annual labor supply by approximately 1 hour.
34. Or, put differently, the assumptions necessary for the individual to internalize the current deficit differ between state and federal deficits. For example, the former requires assumptions about capitalization and future plans to relocate, and the latter requires assumptions about
infinitely lived consumers or intergenerational altruism.

REFERENCES
Advisory Commission on Intergovernmental Relations. 1980. Significant features of fiscal federalism 1979-1980 edition. Washington, DC: Advisory Commission on Intergovernmental
Relations.
. 1987. Fiscal discipline in the federal system: National reform and the experience of the
states. Washington, DC: Advisory Commission on Intergovernmental Relations.
Amemiya, Takeshi. 1978. A note on a random coefficients model. International Economic
Review 19 (3): 793-96.
Barro, Robert J., 1974. Are government bonds net wealth? Journal of Political Economy,
November/December, 84 (6): 1095-12.
. 1989. The Ricardian approach to budget deficits. Journal of Economic Perspectives,
Spring, 3(2):37-54.
Barth, James R., George Iden, Frank S. Russek, and Mark Wohar. 1991. The effects of federal
budget deficits on interest rates and the composition of domestic output. In The great fiscal
experiment, edited by Rudolph G. Penner. Washington, DC: Urban Institute Press.

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

192

PUBLIC FINANCE REVIEW

Bernheim, B. Douglas. 1987. Ricardian equivalence: An evaluation of theory and evidence. In


NBER macroeconomics annual, edited by Stanley Fischer. Cambridge, MA: MIT Press.
. 1989. A neoclassical perspective on budget deficits. Journal of Economic Perspectives,
Spring, 3 (2): 55-72.
Blomquist, N. Soren. 1985. Labour supply in a two-period model: The effect of a nonlinear progressive income tax. Review of Economic Studies 52: 515-24.
Blundell, Richard, and Ian Walker. 1986. A life-cycle consistent empirical model of family
labour supply using cross-section data. Review of Economic Studies 53: 539-58.
Borjas, George J., and Glenn T. Sueyoshi. 1994. A two-stage estimator for probit models with
structural group effects. Journal of Econometrics 64 (1/2): 165-82.
Browning, Edward, and Jacqueline Browning. 1994. Public finance and the price system. New
York: Macmillan.
Conway, Karen Smith. 1994. Reconsidering the effects of fiscal policy on private sector behavior: A unifying view of neutrality. Public Finance Quarterly 22 (2): 194-220.
. 1997. Labor supply, taxes and government spending: A microeconometric analysis.
Review of Economics and Statistics 79 (1): 50-67.
Fremling, Gertrud M., and John R. Lott, 1989. Deadweight losses and the saving response to a
deficit. Economic Inquiry 27:117-29.
Hansson, Ingemar, and Charles Stuart. 1987. The welfare costs of deficit finance. Economic
Inquiry 25:479-96.
Heckman, James J., and Thomas E. MaCurdy. 1980. A life-cycle model of female labour supply.
Review of Economic Studies 47:47-74.
Judd, Kenneth L. 1987. A dynamic theory of taxation. American Economic Review 77 (2): 4248.
Killingsworth, Mark R. 1983. Labor supply. Cambridge, UK: Cambridge University Press.
MaCurdy, Thomas, David Green, and Harry Paarsch. 1990. Assessing empirical approaches for
analyzing taxes and labor supply. Journal of Human Resources 25 (3): 415-90.
Moffitt, Robert. 1990. Introduction. Journal of Human Resources 25 (3): 313-16.
Mroz, Thomas A. 1987. The sensitivity of an empirical model of married womens hours of work
to economic and statistical assumptions. Econometrica 55 (4): 765-800.
National Association of State Budget Officers. 1981. Fiscal survey of the states 1980-81. Washington, DC: National Association of State Budget Officers.
Pencavel, John. 1986. Labor supply of men: A survey. In Handbook of labor economics, edited
by Orley Ashenfelter and Richard Layard, vol. 1. New York: Elsevier.
Poterba, James. 1994. State responses to fiscal crises: The effects of budgetary institutions and
politics. Journal of Political Economy 102 (4): 799-821.
. 1995. Capital budgets, borrowing rules, and state capital spending. Journal of Public
Economics 56:165-87.
. 1997. Do budget rules work? NBER Working Paper No. 5550, National Bureau of Economic Research (NBER): Cambridge, MA.
Quintieri, Beniamino, and Furio Camillo Rosati. 1988. Fiscal policy and labor supply. Public
Finance 43 (3): 425-39.
Rogers, Diane Lim, and John H. Rogers.1993. An empirical examination of state government
budgets. Unpublished manuscript, Department of Economics, Pennsylvania State
University.
Runkle, David E. 1991. Liquidity constraints and the permanent-income hypothesis. Journal of
Monetary Economics 27:973-88.
State tax handbook. 1980. Chicago: Commerce Clearing House.

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

Conway / ARE WORKERS RICARDIAN?

193

Triest, Robert K. 1990. The effect of income taxation on labor supply in the United States. Journal of Human Resources 25 (3): 491-516.
U.S. Bureau of the Census. 1981. State government finances in 1980. Series GF80, no. 3. Washington, DC: Government Printing Office.
Zeldes, Stephen. 1989. Consumption and liquidity constraints: An empirical investigation. Journal of Political Economy 97:305-46.
Ziliak, James P. and Thomas J. Kniesner, Estimating Life-Cycle Labor Supply Tax Effects,
University of Oregon Working Paper, November 1996.

Karen Smith Conway is an associate professor of economics at the University of New


Hampshire. Her research interests are in the areas of labor supply, public economics,
and applied econometrics.

Downloaded from pfr.sagepub.com at Jazan University on October 29, 2014

You might also like