15
As a way to address the data limitation associated with research focused at the national level,
another line of research uses time series models to study the economic effects of changes in fiscal
policies on counties, states, and regions in the United States.
20
That research generally estimates
demand multipliers—often called subnational or local multipliers to reflect that they are not
generated by countrywide data—that are considerably larger than studies for the nation as a whole,
typically in the range from 1.5 to 3.4.
21
A subset of such research uses data available at the state
level to study the economic effects of ARRA (or some of its provisions).
22
Analyzing variations in
the allocation of ARRA funds across states, such research estimates demand multipliers that range
from zero to 3.4, with the majority of estimates close to 2.0.
23
However, demand multipliers estimated using data from counties, states, and regions are of limited
applicability when the ultimate aim is to calculate the economic effects of changes in fiscal policies
for the entire U.S. economy. One reason is that the estimation of such local multipliers cannot
account for spillovers from recipient states to other states (such as shifts in resources from other
20
The literature using state and local data in such time series models is diverse. For example, Clemens and Miran (2012) use
differences in state budget practices, including variations in their balanced-budget requirements; Nakamura and Steinsson
(2011) study regional variations in military spending by the federal government; Reingewertz (2011) uses variations in the party
affiliation of states’ Congressional delegations; Shoag (2011) considers state-level variations in returns of state pension funds;
and Serrato and Wingender (2011) use county-level variations in federal spending allocated on the basis of population estimates.
21
Clemens and Miran (2012) and Conley and Dupor (2011) are exceptions. Clemens and Miran estimate multipliers that are
generally less than 1. As they argue, their methodology captures the tendency for deficits to crowd out private spending, a
feature ignored by most research estimating subnational multipliers. Conley and Dupor find that ARRA decreased private
employment enough to fully offset the increases in state and local government employment it created, implying a subnational
multiplier of zero; however, their estimates are measured with so much imprecision that their results are not very helpful in
uncovering the economic effects of ARRA.
22
See Chodorow-Reich and coauthors (forthcoming), Wilson (forthcoming), Conley and Dupor (2011), and Feyrer and Sacerdote
(2011).
23
State and local data have also been used to investigate the direct effects of ARRA. For example, Taylor (2011) and Cogan and
Taylor (2010) find that most of the ARRA grants to states and localities were used to decrease net borrowing rather than to
increase purchases. As a result, they conclude that the increase in government purchases due to ARRA grants given to states was
close to zero. Chodorow-Reich and coauthors (forthcoming) offer a different perspective, concluding that at least some of the
ARRA funds they examined (Medicaid matching funds) were used to avoid deeper cuts in spending and employment. (From
January 2009 through December 2011, the workforce of state and local governments contracted by about 17,000 employees per
month, on average, or about 621,000 employees for that period. In comparison, over the 10-year period prior to 2009, state and
local governments added roughly 21,000 employees per month to their payrolls.) An unresolved question in the literature on
ARRA’s direct effects on state and local spending is the extent to which states and localities could have borrowed to finance
certain expenditures. For a discussion of the fiscal issues faced by local governments after the recent recession, see
Congressional Budget Office (2010b).