Fishback and Kachanovskaya 1
In Search of the Multiplier for Net Federal Spending in the States
During the New Deal: A Preliminary Report
Price Fishback and Valentina Kachanovskaya
June 2009
Price Fishback is the Frank and Clara Kramer Professor of Economics at the University of Arizona and a Research Associate at the National Bureau of Economic Research ([email protected]). Valentina Kachanovskaya is a Ph.D. student at the University of Arizona ([email protected]). Both addresses are Department of Economics, University of Arizona, Tucson, AZ 85721. The authors would like to thank Shawn Kantor, John Wallis, and Paul Rhode for providing access to data and their comments. We would also like to thank Kei Hirano, William Horrace, and Alfonso Flores-Lagunes for their insights. The research has been funded by National Science Foundation Grants Nos. SES 0617972 and SES 0214483 and funds from the University of Arizona Economics Department. The discussion here does not reflect the views of either of these funding agencies. All errors are the authors’ own.
Fishback and Kachanovskaya 2
Over the past year the U.S. federal government has adopted a sizeable stimulus
package. When added to the budget problems from 2008 and additional spending in
President Obama’s first budget, the federal government’s budget deficit as a share of
GDP has risen near 14 percent of GDP. This is the largest deficit relative to GDP since
World War II, when the U.S. was a command economy engaged in all-out war. The
Obama stimulus package has revived interest in output multipliers associated with fiscal
policy. Throughout the spring of 2009 a range of of multipliers have been suggested.
Mark Zandi (2009) of Moody’s Economy.com suggests a general multipler near 1.5.
Christina Romer, chairperson of the President’s Councel of Economic Advisors, has
provided estimates of multipliers in the 1 to 3 range (Romer 2009, Romer and Romer
2006, and Romer and Bernstein 2009). On the other hand, Robert Barro (2009) suggests
short-run multipliers of less than one, as do Cogan, Cwik, Taylor, and Wieland (2009).
The regional science literature and various economic forecasters, including Zandi (2009),
also offer a wide range of multipliers for states, regions, and communities.
Given the lack of agreement on the size of the multiplier, macroeconomic or
regional, it seems useful to examine the size of the multiplier for state per capita income
from net federal spending in the states during the New Deal from 1933 through 1939. If
there ever was a time when we would expect a large fiscal multiplier it would have been
during the Great Depression when unemployment rates ranged between 14 and 25
percent over the course of the decade. At that time stimulus from net federal spending in
a state was much more likely to add new employment without crowding out private
activity. The multiplier we estimate for individual states will likely be smaller than a
Fishback and Kachanovskaya 3
macroeconomic multiplier for the U.S. because the state economies are more like small
open economies in a world with free trade. Therefore, the state economies are more
likely to rely on importing goods and services from outside the state, which will diminish
the size of the multiplier.
In response to the hard times of the Great Depression, the Roosevelt
administration increased federal government spending as a share of GDP from 4 percent
to 8 percent over an eight-year period. At the macroeconomic level this did not lead to
large budget deficits. E. Cary Brown (1956) and Claude Peppers (1973) have
documented that the federal deficits as a share of GDP were small and fell well short of
being Keynesian policies designed to stimulate the economy.
On the other hand, the distribution of the federal spending varied enormously
across states on a per capita basis. The federal tax burden from the spending also varied
greatly as well. Figure 1 shows the large variation across states in 1935 in per capita
federal grant spending and per capita federal tax receipts in 1967 dollars in each state.1
Just as the Obama administration is implicitly doing today, the Roosevelt
administration ran a large-scale experiment of fiscal stimulus across the states. The
peace-time experiment ran throughout the period from 1933 through 1939 until the
orientation began shifting to preparations to aid allies in World War II and the United
States’ eventual participation. Currently, we have accumulated information on up to
eight years of federal spending in each state and federal tax receipts collected from each
The variation is similar across other years.
1 The variation is even larger when Delaware is added. We left Delaware off the graph to better show the spread across states visually, as Delaware reported federal tax receipts more than $100 per capita higher than in the next highest states.
Fishback and Kachanovskaya 4
state and matched it with information on per capita income in the states that can be used
to estimate the impact of net federal spending on state income per capita.2
Unfortunately, our task of assessing the causal impact of net federal spending in
the states is made more difficult by the Roosevelt administration’s disinterest in
designing the New Deal “experiment” in ways a scientist would prefer. Instead of
randomly distributing the funds across states, the federal government handed out the
money with several purposes in mind, including promoting Relief, Recovery, and
Reform, as well as maintaining the strength of the Democratic Party in holding the
Presidency and a large majority of both houses of Congress. To the extent that the
administration followed through on its pledge to aid troubled areas, there is likely to be a
negative endogeneity bias in Ordinary Least Squares estimates of the net federal spending
multiplier.
3
2 Our longer term goals are more ambitious. We have already collected information on per capita income, per capita federal road spending and loans for Bureau of Reclamation projects back to 1919, and are seeking information by state by the Veterans’ administration, the farm loan programs, and other agencies associated with rivers and harbors.
We use a sparse specification in which we estimate real per capita state
income as a function of real per capita net federal spending in a variety of ways. We then
use a mixture of state fixed effects, year fixed effects, and state-specific time trends to
control for a variety of factors. Since annual information by state on exogenous
correlates is not often available for the period, these measures help control for a variety of
factors for which the best measures we would obtain would be either fixed over time or
would rely on trend interpolations. Finally, we develop an instrumental variable strategy
to work to eliminate any omitted variable bias and endogeneity bias that might be left.
3 There is a large literature on this distribution issue. See Wright, 1974; Wallis, 1987, 1998, and 2001; Couch and Shughart 1998; Anderson and Tollison, 1991; Fishback, Kantor, and Wallis, 2003; Fleck 2001a, 2001b, and 2008; and Stromberg 2004.. The findings vary from study to study, but generally there is evidence that the Roosevelt administration both promoted recovery, reform, and redistribution and played politics with the distribution.
Fishback and Kachanovskaya 5
In the course of the analysis, we have had to make a series of decisions about
adjusting federal spending and tax data from fiscal year to calendar year, the relative
weights to give observations from each state, how to treat federal loans, the differences in
purposes of the grants, and the time period of coverage. As a result, we report a
substantial number of estimates. The estimates that do the most to control for omitted
variable bias and endogeneity bias suggest that an additional dollar of net federal
spending in the states raised per capita income by somewhere between 0 and 62 cents
during the 1930s for estimates based on calendar year interpolations of federal spending
and taxation. Other estimates assume a six-month lag in the impact of net federal
spending by estimating calendar-year state income per capita as a function of fiscal year
net federal spending per capita. They range as high as 0.8 and 1.7. Larger estimates for
analyses based on either the calendar year or the fiscal year can be obtained by removing
controls for state-specific trends or national shocks and assuming that such the changes
across time can be attributed to net federal spending.
Fiscal Multipliers
Macro Multipliers
In the 1970s and early 1980s macroeconomics textbooks reported on estimates of
the Keynesian multiplier. For example, the third edition of Dornbusch and Fischer’s
Macroeconomics in 1984 (p. 148) reports multiplier estimates for an increase in net
government spending of 1.8 from DRI and 0.7 from the Federal Reserve Bank of St.
Louis. Meanwhile, the macroeconomics literature was shifting direction. Hall (1980)
argued that temporary government purchases that are not close substitutes for private
Fishback and Kachanovskaya 6
spending could stimulate the economy by shifting production forward in time but long
run increases would not. Barro (1981) found that the effect of temporary military
spending on consumption exceeds that of permanent military spending; furthermore, non-
military spending, not divided into permanent and temporary, did not have any significant
effect. In late 1980s and 1990s the early works of Barro and Hall were extended by using
mostly one-sector neoclassical growth models with constant return to scale to estimate
fiscal policy multipliers.
Changes in assumptions can lead to quite different effects. In contrast to Barro
and Hall, Aiyagari, Christiano, and Eichenbaum (henceforth ACG) (1992) argued that
fiscal changes could affect long run interest rates and then show both theoretically and
empirically that the impact on output, employment and the interest rate of a persistent
change in government consumption exceeds that of a temporary change. Devereux, Head
and Lapham (1996) modified the commonly used model by assuming imperfect
competition and increasing returns to scale. They found that government spending could
lead to an increase in factor productivity that lowered real wages, raised firm
profitability, and could lead to a stimulus to rather than a crowding out of private
investment.
A series of VAR studies of the macro-economy sought to resolve endogeneity
problems by relying on military build-ups that were said to be unrelated to the state of the
economy or to narratives where it appears that there were no attempts by the government
to respond to economic conditions. Blanchard and Perotti (2002) offer a nice summary
of the literature to that time and find that private consumption is consistently crowded out
by taxation and crowded in by government spending, consistent with the Keynesian
Fishback and Kachanovskaya 7
model. On the other hand, private investment is crowded out by both government
purchases as well as taxation, which is consistent with the neo-classical model.4 Using a
vector auto-regressive framework Afredo M. Pereiral and Rafael Flores de Frutos (1999)
also find that public spending crowds out private spending, leading to multipliers of about
0.65.5
The recent increase focus on the stimulus package has led to a discussion of a
wide range of estimates. Business economists have still been producing estimates of
multipliers at the national and state level. In Congressional testimony Mark Zandi (2009)
has used the Moody’s Economy.com simulation model to report a range of
macroeconomic multiplier estimates for different programs from 0.31 for cuts in
corporate taxes to 1.73 for temporary increases in food stamps. His most cited estimate
was a $1.59 stimulus for GDP based on building bridges and schools. Meanwhile,
Robert Barro (2009, 1981, 1987) has been writing opinion pieces based on his earlier
research that suggest that the short run multiplier is roughly 0.8 for war-time spending
and the long run multiplier is closer to zero (see also Barro 1981, 1987).
Ramey and Shapiro (1998) develop a two-sector dynamic general equilibrium
model in which the reallocation of capital across sectors is costly. The two-sector model
leads to a richer array of possible responses of aggregate variables than the one-sector
model. The empirical part of the paper estimates the effects of military buildups on a
variety of macroeconomic variables and leads to a wide range of findings, some counter
to the literature, some consistent with Keynesian models, and all consistent with their
model.
4 Blanchard, Perotti (2002), p. 1363. 5 Meanwhile, Ohanian (1997) develops a simulation of the military buildup and policies for the Korean war but talks in terms of welfare rather than output measures.
Fishback and Kachanovskaya 8
Christina Romer (2009), the Chairperson of the President’s Council of Economic
Advisors, argues that omitted variable bias is a serious problem in earlier and cites work
with David Romer (2006, 43) that uses narratives to isolates tax changes uncorrelated
with other factors affecting output. They find that tax cuts can raise GNP by multiples of
2 or 3 over a three-year period (Romer and Romer 2006, 43). Her simulations with Jared
Bernstein (2009) suggest a tax cut multiplier of one after a year and a half and a spending
multiplier of about 1.6. In response to this work, Cogan, Cwik, Taylor, and Wieland
(2009) have run simulations based on a new Keynesian model and found much smaller
multipliers than the ones found by Zandi and Romer and her co-authors, smaller even
than the 0.8 estimates discussed by Barro.
Regional Multipliers
On the state and regional level studies of fiscal activity have found mixed results.
Hulten and Schwab (1991) conclude that the link between public infrastructure and
states’ economic growth is weak arguing that the states that saw a large expansion of
public infrastructure in 1970s were not the ones that developed faster during that period.
On the other hand, Munnell (1992 192), Garcia-Mila and McGuire (1992), Costa, Ellison,
and Martin (1987), Blanchard and Katz (1992), Duffy-Deno and Roberts (1991), Fernald
(1991), and Aschauer (1989) find positive effects of public spending on a variety of
dimensions.6
6 Munnell (1992 192) finds a significant effect of public capital on state-level
output, investment and employment growth, although the effects of government spending at the state level are smaller than at the national level. Garcia-Mila and McGuire (1992) constructed a panel of 48 states from 1969 until 1983 to estimate input elasticity coefficients of regional Cobb-Douglas production functions and concluded that
Fishback and Kachanovskaya 9
Meanwhile, regional scientists, business economists, and regional forecasters
have been active in developing models of multipliers that they apply to government
spending and other types of income coming into a location. Mark Zandi’s (2009) makes
predictions about the impact of the new fiscal stimulus package on state employment.
Regional scientists have developed a broad range of theoretical models that lead to
multipliers for net income coming into the region. The models range from the early
Keynesian regional models to input-output models to economic base models to neo-
classical models.7
government provided goods, such as highways and education, have a significant and positive effect on state’s output. Costa, Ellison, and Martin (1987) consider a translog production function and conclude that public capital and labor are complementary inputs. The estimated elasticities of output with respect to public are around one in all states. Meanwhile, Blanchard and Katz (1992) model the effects of negative one-percent employment shocks to a wide range of variables using data from U.S. states from 1947 to 1990 and find sizeable effects on per capita income over an extended number of years.
The empirical work on regional multipliers lead to a broad range of
estimates of multipliers of between 0.5 and two, depending on the technique used. Some
rely on simulations that derive multipliers using input-output models and surveys that
describe the degree to which different industries rely on local labor and external inputs
Duffy-Deno and Eberts (1991) study the effect of the public capital stock on the state’s economic growth, first, without using capital expenditures as a proxy for capital stock, and second, considering public capital both exogenous to the firm and endogenous to the local community positing a simultaneous relationship of public capital and local economic growth. The authors find a positive and statistically significant effect of public capital on state’s economic growth rate.
Assessing a link between public capital and economic growth, Fernald (1999) studies the direction of causation between public capital and productivity and unsurprisingly concludes that road construction (which is one of the biggest components of public spending) causes a surge in productivity in industries with high motor-vehicle use. David Aschauer (1989) also finds that road construction bears the most explanatory power of the change of local productivity, while military spending has almost none. 7 Richardson (1985) surveys all but the neoclassical models. Merrifeld (1987 and 1990) and McGregor, McVittie, Swales, and Yin (2000) for examples of neoclassical multipliers for the economic base.
Fishback and Kachanovskaya 10
and capital. Others rely on Ordinary Least Squares regression estimates (Mulligan (2005,
1987).
There have been some estimates of the impact of New Deal spending on general
economic activity. Fishback, Horrace, and Kantor (2005, 2006) showed a strong positive
influence of public works and relief spending on county-level retail sales and net-
migration. At the same time spending by the Agricultural Adjustment Administration
(AAA) had an adverse effect on retail sales; hence, it may also have had a negative effect
on per retail sales growth and net migration. Garrett and Wheelock (2006) found similar
positive effects of overall New Deal spending in a cross-sectional analysis of the growth
rate in state personal income per capita for the entire period 1933 to 1939 and New Deal
spending during that period. However, neither the Fishback, Horrace, and Kantor
estimates nor Garrett and Wheelock estimates show a short-term fiscal multiplier because
they do not include information on tax receipts from the states and the examine the
impact of spending over a six-year period on retail sales per capita at the end of the
decade.
Studies of labor markets in the 1930s have focused on the impact of relief
spending on labor markets. Neuman, Fishback, and Kantor (forthcoming) examine
monthly data from 1933 through 1940 for over 40 cities and find that relief spending
raised private employment through 1935 but reduced it afterward. Benjamin and
Mathews (1992) find small crowding out effects of private employment from relief jobs
in through 1935 and much larger crowding out effects in the second half of the New
Deal.8
8 We focus on the studies that use panel data here, see Neumann, Fishback, and Kantor (forthcoming for citations to studies relying on cross-sectional estimation.
Fishback and Kachanovskaya 11
Federal Grants and Loans During the New Deal
Understanding the impact of federal spending and taxation during the New Deal
era is complicated by the great diversity of programs during the period. The New Deal
funding programs were divided into the two major classifications: nonrepayable grants
and repayable loans.9
Our main focus in the analysis is nonrepayable grants from the federal
government. About 62 percent of the grants were associated with relief programs. All of
the Works Progress Administration (WPA), Civilian Conservation Corps (CCCG), and
Civil Works Administration (CWA) grants and roughly half of the Federal Emergency
Relief Administration (FERA) were spent on poverty relief projects with work
requirements and could be considered federal expenditures because they produced a good
or service. The Social Security Act Programs (SSA), and the rest of the FERA grants
were New Deal programs that offered transfer payments to alleviate poverty. The
Veterans’ Administration (VA) and Soldiers’ and Sailors’ Homes (SOLD) were grant
programs in place before the New Deal that provided pensions, disability payments, and
living support to military veterans. We treat them as relief in the same vein as grants
from the SSA programs, which were providing matching grants to states that provided aid
to dependent children (ADC), old-age assistance (OAA), and aid-to-the-blind (AB). If
The Office of Government Reports (1940) reported the total
amount spent by each program in each state in each year between July 1, 1932 through
June 30, 1939, which provides the basis for our analysis. The totals and the amounts per
capita for the entire period are reported in Table 1 to get a sense of the size of each
program.
9The Office of Government Reports offered information on the value of housing loans insured by the Federal Housing Administration. Since these loans were private loans, we do not incorporate these into the analysis of net federal spending.
Fishback and Kachanovskaya 12
we performed the analysis for the U.S. as a whole, the transfer payments from the SSA
part of the FERA, the VA, and the SOLD, which account for roughly 20 percent of the
grants, would not necessarily be treated as expenditures because they are net transfers
within the system. However, at the state level these transfer grants become income
within each and we therefore incorporate them into the analysis.
The second major grant category is public works programs, which accounted for
19.4 percent of the grants. The Public Works Administration (PWA) (Federal (F) and
Nonfederal (NF)), Public Roads Administration, Public Buildings Administration (PBA),
Rivers and Harbors Grant (RH) and other smaller programs listed as public works in the
table were not poverty programs. All but the PWA programs were long run federal
programs established before the New Deal. Unlike the work relief poverty programs, the
public works programs could hire from the labor market or the relief rolls, faced no
restrictions on hours worked to limit the amount received by an individual, and paid
hourly wages that were roughly double those on the work relief programs.
Approximately 12 percent of the grants were devoted to agriculture from
programs run by the Agricultural Adjustment Administration (AAA), Soil and
Conservation Service (SCS), Farm Security Administration (FSA), and Agriculturual
Experiment Stations (AES). The AAA was the major New Deal program which was
devoted to payments to farmers to take land out of production. The initial AAA program
was funded with an agricultural processing tax until it was declared unconstitutional in
January 1936.10
10AAA grants per capita were not very strongly correlated with processing tax receipts in cross-sectional correlations. The correlation for 1934 was only 0.034 and for 1935 was 0.1677.
The AAA also administered the replacement program adopted under
the Soil and Domestic Conservation Act of 1936, which continued to make payments to
Fishback and Kachanovskaya 13
farmers to take land out of production without the processing tax. The FSA started
within the FERA relief program and was more a poverty relief program. The SCS began
before the AAA was declared unconstitutional and provided grants for training farmers
about soil conservation techniques.
Dealing with New Deal loans in the analysis is more difficult. It is not clear how
to treat the loans in terms of developing a multiplier. They are not government spending
because at the time the loans were made they all required repayment, and a large share of
the loans were repaid.11
11 There were some cases of loan forgiveness. In the case we know about, the RFC loans offered to cities for poverty relief under the Hoover administration in fiscal year 1933 were eventually forgiven by the Roosevelt administration. The HOLC likely experienced the highest loan default rate because it foreclosed on 20 percent of the mortgages that it supported. Our sense from reading the reports of the various agencies, is that they anticipated repayment and were active in seeking repayment or in the case of default recovery of assets to be sold.
There was a grant feature to the loans to the extent that they
provided subsidies in the form of lower interest and better lending terms. The Home
Owners’ Loan Corporation (HOLC), for example, offered loans at below the interest rates
charged on good loans in the housing market, even though the loans were already
troubled. The HOLC also extended the standard repayment period, and allowed much
smaller down payments relative to the value of the home. The Commodity Credit
Corporation Loans (CCCL) loans were nonrecourse loans that set minimum prices that a
farmer received for his crop, while the Farm Credit Administration (FCA) loans provided
good terms for farm mortgages and short-term loans for crops, seed, and tools. The
subsidies in RFC loans likely varied by type of loan. Given the measurement issues with
loans, we try several ways of dealing with them. As an upward measure of the subsidy
on these loans we add 10 percent of the value of the loans to the federal grants spending
net of taxes. We also run some estimates where we add the full value of the loans to the
Fishback and Kachanovskaya 14
grants. Given the measurement problems with loans, we treat the results incorporating
the loans more as robustness tests of the analysis of New Deal grants minus taxes.
Expectations for the Multiplier.
The multiplier we estimate for the United States is not a national multiplier. The
U.S. states freely trade across state borders and there is a great deal of specialization in
specific goods and services in each state. If thought of as a macroeconomic multiplier, it
would be associated with a series of small open countries receiving grants and paying tax
revenues to a higher authority. At this stage, however, it cannot be seen as a multiplier
for the whole area because the current estimation does not capture the spillover effects of
the net federal spending on other areas of the country.
The coefficient on net federal spending in a Keynesian regional model will be
determined by a series of factors.12
12 The intuitive discussion of the multiplier is based on a Keynesian discussion of consumption and imports. See Cullen and Fishback (2007) and Fishback, Horrace, and Kantor (2005b) for how this works in a simple model. The regional science literature offers a broad array of models that can produce multipliers based on the mix of local versus external consumption and or production. They include Keynesian models, economic base models, input-output models and neoclassical models. Richardson (1985) surveys all but the neoclassical models. See Merrifeld (1987 and 1990) and McGregor, McVittie, Swales, and Yin (2000) for examples of neoclassical multipliers.
It will have positive effects to the extent that the net
spending puts to work resources that would have been unemployed otherwise; to the
extent that the net federal spending is more productive than the private spending that is
replaced by the anticipation of future obligations for taxpayers; to the extent that the net
federal spending produces social overhead capital (like roads, sanitation, public health
programs) that made the inputs in the state economy more productive; and/or leads to
Keynesian multiplier effects as each income recipient purchases goods and services from
others in the state who, in turn, spend their receipts on goods and services produced by
others in the states, and on and on.
Fishback and Kachanovskaya 15
The positive benefits of the multiplier are diminished through a variety of
“leakages” at each stage when the money spent in the process is spent on goods and
services outside the state economy. Much of the federal grant spending on a work relief
program, like the FERA, WPA, or CWA, had small initial leakages because over 80
percent was spent on wages for people in the state. Grants from the Public Works
Administration and Public Roads Administration had larger initial leakages because more
than 50 percent of the monies were spent on materials and equipment imported from
other states. More leakages occurred to the extent that the workers on federal projects
spent their wages on goods and services outside the state. Each leakage reduced the
extent to which the spending could be “multiplied” in a Keynesian sense by additional
spending within the state.
The net federal spending had smaller positive effects on the economy to the
extent that positive net federal spending led people to save in anticipation that they will
have to pay future taxes. The net federal spending would have had an even weaker effect
to the extent that it replaced local production of goods and services. The most obvious
crowding out came from the AAA payments to farmers to take land out of production.
The stated purpose of the act was to reduce output in hopes of raising prices enough to
see an increase in income. In other cases, the federal spending may have replaced state
and local projects that would have been built in the absence of federal spending. The
impact of the reduction in state and local spending was likely to be small because states
were generally required to run balanced budgets. Even when they ran deficits in the early
1930s, the deficits were relatively small as a share of state and local spending.
Fishback and Kachanovskaya 16
Finally, the impact of the net federal spending was reduced to the extent that it
crowded out private production of goods and services in the state. An influx of federal
spending may have bid up local wages in ways that raised the costs of hiring labor to
private producers. It may have also bid up the prices for nonlabor inputs with the same
effect. To the extent that increases in federal spending reduced private activity fully, it
could even have a negative effect on state income if the output from federal spending was
less valuable than the output it replaced.
Empirical Approach
To estimate the multiplier is to use panel data methods. We choose a sparse
specification with real per capita state income (Iit) in state i and year t in 1967 dollars as a
function of real per capita net federal spending in state i and year t ((Git – Tit) as a starting
point.
Iit = β0 + β1 (Git – Tit) + εit.
The net federal spending is calculated as federal nonrepayable grants to the state minus
federal tax revenues collected from the state. To reduce problems with omitted variable
bias, we add three vectors of parameters. First, we incorporate a vector of state fixed
effects (S) to control for factors like geography, state laws, and the basic economic,
cultural, and demographic structure of each state that did not change over time but varied
across states. Second, we incorporate a vector of year fixed effects (Y) to control for
macroeconomic shocks to the entire economy that affected all states in each year. Third,
Fishback and Kachanovskaya 17
we add a vector of state specific time trends (S t) to control for differences in the trend
paths of economic activity that each state was following.
Iit = β0 + β1 (Git – Tit) + S +Y + S t + εit.
Under this specification, the identification of the multiplier β1 for net New Deal spending
comes from the variation from trend across time within states after controlling for
macroeconomic shocks.13
There still remains the possibility of biases from simultaneity and endogeneity.
There is an ample literature about the geographic distribution of New Deal spending that
shows that the Roosevelt Administration tended to distribute more New Deal grants to
areas where income was declining (see Wallis 1998, Fleck 2008, Fishback, Kantor, and
Wallis, 2003). This tendency would impart a negative bias to the multiplier coefficient.
In addition, there are a variety of other factors that may have changed over time in
non-trend ways that also influenced both net fiscal federal spending and per capita
income in the state. There are two problems that arise in trying to control for these other
factors. First, many of the variables that might be included as controls in a productivity
model, such as wages, employment, and interest rates, are themselves components of
personal income. By controlling for them we would be restricting the measure of the
impact of net federal expenditures to the parts of state income for which we have not
13 We have also tried estimating the model while including squared terms. The estimates at the mean of the sample are very similar and there is very little gain from adding the squared terms. In addition, the instruments did not have adequate strength to separate the coefficients for the squared terms. .
Fishback and Kachanovskaya 18
controlled, which is not an appropriate restriction. Second, there are many variables for
which there are not good annual measures at the state level. These would include
measures of consumption and investment that might be used in a Keynesian model. For
consumption, our best estimates would have to be derived from interpolations between
the Cost-of-Living Surveys of 1917-1919 and 1935-36 and the post-war period. In the
early 1900s the U.S. Bureau of the Census routinely described their measures of capital in
manufacturing from earlier periods as inadequate and they stopped collecting and
reporting the information at all after 1919.14
Fourth, the omitted variable that we would most like to include in the analysis is
annual estimates of the size of state and local government net spending in the states
during this period, so that we can measure the effect of federal spending, holding state
and local fiscal activity constant. Without controlling for state and local government, the
federal multiplier estimate we obtain from estimating equation 2 will therefore include
the effects of federal expenditures on state and local spending and taxation, which in turn,
will likely influence per capita personal income. As it stands today, comparable annual
Third, controls for age, race, ethnicity,
population, and the structure of the economy are all available typically only during the
census years and thus we would have to interpolate between census years to provide
values. The state effects essentially control for many of the long-term structural factors
that we could measure, while the state-specific time trends serve to control for the time-
trends that would arise from the interpolations. Essentially, the interpolated measures of
the variables measured in 1930 and 1940 described above would be linear combinations
of the state-specific time trends and/or the state effects.
14The WPA created a series of estimates of capital at the state level in each year that it reported in a mimeo in the National Archives. We have been unable to find any descriptions of the methods used to create these measures, and are therefore hesitant to use them.
Fishback and Kachanovskaya 19
estimates of revenues and governmental cost payments in the states are available only up
through 1931 and after 1936 for all states; therefore, any estimate incorporating controls
for net state spending would miss a very large portion of the New Deal period.
Information is available on cities over 100,000 people throughout the 1930s, but data for
the rest of the governments is available only for 1932. We are in the process of filling the
gaps in computerized information on the large cities and the states.15
As we work to eliminate the biases arising from the issues described above, we
follow an instrumental variable strategy. We need an instrument that varies annually
both across time and across space and is strongly correlated with the net federal grants
but not with the error term in the equation. We developed a hybrid instrument that
interacts two variables. The first variable, which varies primarily across time, is New
Deal grant spending per capita in the eight other census regions outside the region where
the state is located. The variable cross-sectionally as well because it varies across the
nine census regions. Our expectation is that more spending in the entire nation was likely
to lead to more spending within a state.
16
15The federal government stopped collecting the annual information from states for the volume Financial Statistics of the States in 1933 after having collected information from 41 states for 1932. They restarted by collecting the data for 1937 (U.S. Bureau of the Census 1940, p. vi). John Wallis, Richard Sylla, and John Legler have posted information for 16 states for the period 1933 through 1937 with the ICPSR, but it is taking longer than we anticipated to make the data for these states comparable with the federal government’s categories. We are working with John Wallis to collect, computerize, and categorize the information for the remaining states for 1933 through 1937 and for the seven states in 1932 that the Census Bureau had not worked with.
To avoid correlation between the instrument
and the error term of state i in year t, we use the per capita measure for the regions
outside the region where the state was located.
16 If the federal government had established a hard budget constraint nationwide, there might have been a negative relationship between spending in the rest of the regions and spending in the state in question. There did not appear to be a hard spending constraint at the national level because Roosevelt and the Congress often approved additional funds throughout the years and ran budget deficits in most years.
Fishback and Kachanovskaya 20
We focus on grant spending for the instrumental variable rather than grant
spending net of total taxes for several reasons. First, the grant spending version of the
instrument has substantially more strength than the grants less total taxes version.
Second, taxes are tied so closely to personal income that it is difficult to come up with a
tax portion of the instrument that would not also be highly correlated with income, which,
in turn, would increase the likelihood of correlation between the instrument and the error
term in the final-stage equation. The second reason also explains why we have not been
successful at developing an analysis where we look at the grants per capita and federal
taxes per capita separately.
The second variable, which varies primarily across states, is the standard
deviation of the percentage voting Democrat for president in the state between 1896 and
the most recent presidential election prior to year t. The variable varies across time
because each state’s value changes between 1932 and 1933 and again between 1936 and
1937. This is a measure of swing voting in the state that has been found to have strongly
influenced the geographic distribution of New Deal spending in a large number of studies
(Wright 1974; Fishback, Kantor, and Wallis 2003; Fleck 2001, 2008, Wallis 1998, 2001).
Nearly all of these studies find strong positive relationships between the swing voting
measure and the distribution of per capita New Deal spending. By using the measure
calculated up through the most recent presidential election before that year, we eliminate
any contemporaneous correlation with factors that influenced income in the state.17
17 As an example, the instrument for the year 1932 would include the standard deviation of the percent voting for the Democratic presidential candidate from 1896 through 1928.
Given the controls for state time trends and fixed effects, it is unlikely that swing voting
Fishback and Kachanovskaya 21
in the presidential elections prior to the year would have influenced income in the state
except through the New Deal distribution mechanism.
For any state i that is in Census region c the formula for the instrument is:
,,,
,
, ti
cjtj
cjtj
ti stdevdempop
spendspendIV ⋅=
∑∑
≠
≠ (8)
where spendj,t is government spending in Census region j at time t, popj,t is population in
region j at time t, and stdevdemi,t is the standard deviation of the democratic vote in state
i for the presidential elections between 1896 and the most recent presidential election
before year t. The logic behind the hybrid instrument suggests that states with a greater
share of swing voters are more likely to obtain a higher share of national increases in
federal government spending, which leads to higher net federal spending in their state.
Results
We provide a range of estimates under a variety of assumptions related to the
types of federal programs to include in the next spending measure, the weighting of state
observations, and various ways of matching fiscal year federal funds and taxation to the
calendar year state personal income. The estimates vary under different assumptions.
In analyzing the impact of the distribution of New Deal funds net of taxes, it is
important to realize that New Deal dollars and the strings attached to them varied
substantially depending on the type of funds and the programs to which they were
distributed. We start by analyzing the per capita New Deal grants net of federal taxes,
because the New Deal grants were pure expenditures with no repayments involved.
Even for grants there were quid pro quos in terms of matching required by state and local
Fishback and Kachanovskaya 22
government. In the absence of controls for state and local net spending, the coefficient
estimated includes the impact on state per capita personal income that operates through
the effects of net federal spending on state and local spending. State constitutions often
contained constraints that prevented states from running deficits. A number of states ran
deficits in the early 1930s, although these were offset by surpluses in the late 1930s. For
more discussion of the biases that might arise see the appendix that discusses the results
with and without state deficits included.
We have also estimated the analysis by including estimates of the size of New
Deal loan activity. Given the measurement issues with loans, we try several ways of
dealing with them. The first is to treat the loans as incorporating grants through subsidies
on interest rates and favourable terms for lending. We added 10 percent of the value of
the loans to the New Deal grants minus taxes to take into account all of the subsidized
features of the loans. In the second we add the full value of the loans to the New Deal
grants minus taxes per capita as a high measure. Given the measurement problems with
loans, we treat the results incorporating the loans more as robustness tests of the analysis
of New Deal grants minus taxes.
Finally, we try some specifications where we seek to examine the impact of
different forms of New Deal grants on income in the states. The New Deal grants
themselves varied dramatically with respect to their purposes. The Agricultural
Adjustment Administration’s (AAA) grants were explicitly designed to crowd out private
production by reducing the amount of land under cultivation. This role was made even
more obvious in 1933 when the AAA had a late start and therefore was paying farmers to
plow under crops that had already been planted and to kill a large amount of farm
Fishback and Kachanovskaya 23
livestock. As a result, we estimate a specification where we leave the AAA grants out of
the net federal spending measure, and we also estimate specifications where we treat the
programs separately.
Another issue is how the state observations should be weighted. We provide
estimates that are unweighted as well as estimates weighted by the square root of the
population in 1930..18
Finally, we need to deal with the differences in the reporting time frames for
personal income and federal spending and taxes. State income is reported on a calendar
year basis. The New Deal and tax data are reported for fiscal years, such that the values
reported for 1933 are for the year ending on June 30, 1933. We have converted the
grants, loans, and tax data to a calendar year basis using the methods described in the
Data Appendix.
Philosphically, the unweighted measures treat each state as if it
were a separate economy that deserves equal weight, in the way the Senate gives equal
representation to the states. In essence, we are finding an average effect of net federal
spending where each state is treated equally as a political entity. The weighted
estimation gives more weight to states with larger populations in the way the House of
Representative apportions its membership. This weighting scheme provides an estimate
of the average effect that takes into account the fact that larger states make up a
disproportionate share of the overall economy.
In the interpolation we use the second half of fiscal year t and the first half of
fiscal year t+1. This could lead to problems with both measurement error and with
correlation between information in the second half of fiscal year t+1, which is the first
18 When we weight using the interpolated state populations throughout the period, the results are very similar.
Fishback and Kachanovskaya 24
half of the calendar year t+1 and personal income in year t. If the instrumental variable
that we use is uncorrelated with the error in the final-stage per capita income equation,
this may not be a problem. To test the robustness of that assumption, we have estimated
the model as well by estimating a model of personal income per capita in calendar year t
as a function of net federal spending in fiscal year t. This method assumes a six-month
lag in the impact of the net federal spending. A study by Neumann, Fishback and Kantor
(forthcoming) of monthly data on relief spending, private earnings, and private
employment in the 1930s finds that the impact of relief spending on earnings and
employment lasted for six to nine months. We are currently searching published and
unpublished federal government reports by agency in an attempt to find data by quarter or
half year at the state level for the major federal programs in Table 1.
Non-IV Results
The results for a wide range of analyses using calendar-year federal spending are
reported in Tables 2 and 3. Estimates that match the calendar-year state per capita
income estimates with the fiscal year net federal spending are found in Table 4. We offer
calendar year estimates for 1933 through 1939 and 1932 through 1939 and fiscal year
estimates for 1933 through 1939 because the results differ and there are trade-offs
between the various measures. There is likely more measurement error in the 1932
estimates of federal spending than in any of the other years. As seen in Table 1 there
were several grants and two loan programs for which we have data for spending by state
from July 1, 1932 through June 30, 1933. In nearly all cases the value we use for 1932 is
the value for the fiscal year ending June 30, 1933. The RFC loans are an exception
Fishback and Kachanovskaya 25
because the program started in February 1932, the source reported that information and
we could use RFC reports to split the data. We know from looking at the reports of the
Public Roads Administration that the relative spending in fiscal year 1932 and fiscal year
1933 varied across states. In addition, there are likely to be some problems with using
values with information from 1933 for examining 1932 state income. There are two
advantages of using 1932 information if the measurement error is not severe.
Statistically, more observations are better and the hybrid instrument has more strength
because the swing voting portion of the instrument changes value in each state between
1932 and 1933.
The tables show the raw correlation in the panel data set and then uses varying
combinations of state fixed effects, year fixed effects, and state-specific time trends. The
instrumental variables models are estimated using Two Stage Least Square (2SLS) with
robust White-corrected standard errors. The first two column in each Table show the
results for net federal spending per capita in the states, unweighted and weighted and lay
out the range of multipliers obtained using the different sets of controls.19
The raw multiplier estimate shows a strong negative but imprecisely estimated
relationship between real per capita personal income and real net federal spending per
capita. The coefficient of -1.05 in the unweighted estimate in the first column of Table 1
for the 1933-1939 estimation implies that an additional net dollar of per capita federal
grant spending was associated with a reduction of $1.05 cents in per capita personal
income (both in $1967). The coefficient from the weighted analysis is -1.82. The
19To develop the per capita estimates we calculated population in the state as a straight-line interpolation between the 1930 and 1940 populations in the state. We have estimated the model using per capita estimates based on only the 1930 population, and the results are very similar. We have also estimated the model without the state of Delaware, where federal tax receipts per capita of all kinds were very high relative to other states. Again, the results are essentially the same as those we focus on in the paper.
Fishback and Kachanovskaya 26
coefficients from the 1932-1939 estimations in Table 3 are similar at -0.98 in the
unweighted analysis and -1.82 in the weighted analysis, as are the coefficients in Table 4
for the fiscal year net spending specifications.
As we add various combinations of state effects, year effects, and state-specific
time trends, the multiplier coefficient fluctuates up and down, as the analysis controls for
the influences of different omitted factors. A common estimator combines state and year
effects, which leads to imprecisely estimated multiplier coefficients that are less than zero
in all three tables. Adding state-specific time trends raises the multiplier to imprecisely
estimates of 0.26 (unweighted) and 0.09 (weighted) in the 1933-1939 analyses, -0.05 and
-0.06 in the 1932-1939 analyses, and 0.26 to -0.18 in the 1933-1939 fiscal year analysis.
Instrumental Variable Results.
In theory an effective instrumental variable can eliminate the negative omitted
variable bias from omitting state and local spending, measurement error in the right-
hand-side variable, and the negative bias imparted by the way that the Roosevelt
administration distributed more funds to areas where income was falling. Using the
hybrid instrument, we estimated the multiplier for federal net spending using three
different specifications, one with state and year fixed effects, another with state fixed
effects and state-specific time trends, and a final version with state and year fixed effects
and state-specific time trends. The results for the unweighted and weighted analysis are
in the lower section of Tables 2 trhough 4. We only report IV coefficients and standard
errors in situations where we could reject the hypothesis of weak instrument bias if we
are willing to accept up to 15 percent weak instrument bias in comparisons of the
Fishback and Kachanovskaya 27
Keibergen-Paap rank Wald F-statistic with the Stock-Yogo critical values (see notes to
Tables 2, 3, and 4).
The multiplier estimate that controls the most for omitted variable bias as well as
endogeneity and simultaneity bias is the IV specification with state fixed effects, year
fixed effects, and state-specific time trends on the last row. The weighted and
unweighted coefficients for the calendar-year analysis of the years 1933-1939 are similar
in size although neither is precisely estimated. The coefficients imply that a one-dollar
increase in net federal spending raised state per capita income by 52 cents or 44 cents.
The impact is much smaller when 1932 is added to the analysis. Both the weighted and
unweighted estimates suggest that a dollar increase in net federal spending would have
led to a reduction in per capita personal income of 6 cents or less. Both sets of estimates
imply a lack of a multiplier effect and a substantial amount of crowding out of private
activity. The analysis in Table 4 that matches calendar year state per capita income with
fiscal year net federal spending tells a more positive story. The unweighted estimate at
the bottom of the table suggests a statistically significant 84 cent increase in per capita
income for a one-dollar increase in net federal spending. The weighted estimate is the
first to lead to a multiplier effect, implying that a dollar increase in net federal spending
per capita led to a $1.71 increase in per capita personal income.
The differences between these estimates and the other IV estimates is informative
because it shows how sensitive the estimates are to omitted variable bias. If one is
willing to assign all of the impact of state-specific time trends to the changes in net
federal spending in the states, the estimates with state and year fixed effects become
relevant. We ran into problems with instrument strength in the unweighted estimates for
Fishback and Kachanovskaya 28
1933-1939, so we did not report them. The weighted estimates suggest a sizeable
multiplier of 2.51, which is on the high side even for regional estimates. The estimates
are both greater than one for the state and year fixed effects IV estimations for 1932 to
1939 at 1.89 for the unweighted and 1.07 for the weighted. They are even higher in
Table 4 when net federal spending on a fiscal year basis is used in the analysis.
When year fixed effects are left out and state-time trends added to the analysis,
the coefficient estimates fall toward the estimates with state and year fixed effects and
state-specific time trends. In the calendar-year analyses, the values are 1.07 and 1.38 for
the 1933-1939 in Table 2 and 0.47 and 0.78 for the 1932-1939 sample in Table 3. Again,
they are even higher in the fiscal-year analysis in Table 4. These estimates would be
accurate if someone is willing to assign all of the positive benefits of annual nation-wide
shocks to the economy between 1933 and 1939 to the net federal spending. Thus, any
benefits from the expansions of the money supply, the move off of the gold standard, the
rise and demise of the National Recovery Administration, tariff reductions, and other
nation-wide factors would be assigned to the net New Deal spending in the states.
Generally, comparisons of the coefficients across the table suggest that the
endogeneity bias in the analysis is generally negative, as expected. In Tables 2 through 4
comparisons of the Least Squares estimates with no correlates to those with state and year
fixed effects and state-specific time trends show a move from a negative coefficient
nearing -1 or more in the first line to coefficients that are much less negative. When we
remove more of the endogeneity bias using the IV techniques, the coefficients again
become more positive. This can be seen in comparisons of the Least Squares and 2SLS
estimates with the same specifications in the middle and the bottom of the table.
Fishback and Kachanovskaya 29
Taking Loans into Account
Under the New Deal the federal government ran extensive lending programs. In
discussing the distribution of federal funds across states, scholars have differed on
whether loans should be included with grants as a measure of spending. In columns three
and four of Tables 2, 3, and 4 are estimates of the impact of net federal spending when
the coefficient is estimated after adding 10 percent of the value of New Deal loans to
New Deal grants to take into account the potential subsidies from the loans. Columns
five and six contain the results when the full value of the loans is added.
We perform the estimation in the same way as above, including using the same
instrument in the 2SLS estimates. Adding the loans to the analysis appears to increase
the size of the multiplier when comparisons are made between the coefficients in
columns 1 and 3 and between the coefficients of 2 and 4 in the calendar year analyses in
Tables 2 and 3. The addition of the loans has a negative effect on the coefficients in the
fiscal year based analysis in Table 4. The largest of the multipliers in the calendar year
studies that control for the most omitted variable bias is 0.62 for the unweighted
estimates with all loans included for the period 1933-1939 in Table 2. The largest
estimates are 0.77 and 1.44 when loans are included in the fiscal year estimations in
columns 3 and 4 of Table 4.
Grants without the AAA
The multipliers in the first six columns may understate the effects of relief and
public works spending because the grants include the AAA payments, which were
Fishback and Kachanovskaya 30
designed to limit agricultural production. In columns 7 and 8 of Tables 2 through 4 are
the multipliers associated with per capita New Deal grant spending minus the AAA and
minus federal tax collections. In this analysis the instrument was the interaction of the
swing voting measure with real per capita federal grants excluding AAA grants.
Comparisons of the 2SLS estimates in columns 1 and 7 and again in 2 and 8
suggest that the AAA grants had more negative effects than did the other grants, but the
changes in the coefficients are not very large on the bottom row where the most controls
for omitted variable bias are included.20
In this estimate we did not include the AAA
spending per capita, which might lead to omitted variable bias. We don’t believe that the
omitted variable is large. When we regress per capita nonfarm grants minus taxes on per
capita farm grants while including the year and state fixed effects and state-specific time
trends the coefficient very imprecisely estimated and lies between -.12 and 0.09
depending on the time period and the weighting structure.
Multiple Programs
Our ultimate goal is to simultaneously examine the separate impact of several
types of New Deal funds and federal tax collections. Thus far, we have not been able to
develop separate instruments for each of the types of spending that have enough strength
to allow us to identify the separate effects in 2SLS analysis.21
20 When the AAA is removed there are some other farm grants left in the grant measure. When we pull out all farm grants and redo the analysis, the coefficients and standard errors are very similar to those reported to the ones in columns 7 and 8 of Tables ?? and ??.
Absent the instrumental
21 We have made several attempts to estimate separate effects for nonAAA grants, taxes, and AAA grants, but the various tests for instrument strength all suggest that the instruments are very weak, too weak to offer meaningful insights. Among the instruments we have tried for the AAA grants is an interaction between the swing voting measure and the changes in AAA spending outside the region. We have also tried interacting the representation on the House Agriculture Committee at the beginning of the 1933 Congressional Session with the changes in spending outside the region. These cross-sectional interaction
Fishback and Kachanovskaya 31
variable analysis, there are still some lessons that can be learned from a careful
discussion of the results from the OLS analysis with state and year fixed effects and state-
specific time trends. Table 5 shows the results when we include separate variables for
groupings of loans and grants in various categories: relief and education grants, public
works and resources grants, AAA and other farm grants, nonfarm loans, farm loans, and
federal tax collections.22
Remember that this estimation procedure led to a multiplier estimate for net
federal spending that was lower than the multipliers from IV estimation due to
endogeneity. The nature of endogeneity bias likely varies by the type of program, and we
will discuss the direction of the anticipated bias as we discuss the various coefficients.
The estimation procedure is OLS with state and year fixed
effects and state-specific time trends.
Our sense is that the negative endogeneity bias is most severe for relief and
educational spending. The relief programs were designed to distribute money to areas
with high unemployment. Fishback, Kantor and Wallis (2003) analyzed the cross-
sectional distribution of New Deal funds for the whole period from 1933 through 1939
programs with state fixed effects. Essentially, they identified effects based on cross-
sectional variation within counties. To the extent that the cross-sectional distribution
variables have been successful as explanatory variables in cross-sectional analysis. However, none of these variables have been successful here. Similar attempts with taxation have led to no luck. We have also tried a series of political variables describing the party structure in the governor’s office and the state legislature and found little strength. 22The relief and education grants include the following programs using the acronyms in Table 1. WPA, CWA, FERA, CCCR, CCCIS, SSA, VA, BRA, USES, SSS, NG, VE, CAM, SF, OE, SMS, and BFB. We included the National Guard (NG) because it was providing partial employment. The public works grants include the PRA, PWAF, PWANF, RHFC, BR, PBA, PWAH, FSR, FF, ML, LUP, and FWP. The agricultural grants include AAA, FSA, SCS, AEX. The nonfarm loans include the RFC, the HOLC, the PWAL, HOLCT, FRB, DLC, and USHA. The farm loans include the FCA, CCCF, FSAL, REA, and FTP.
Fishback and Kachanovskaya 32
within states is informative about the panel variation in the analysis here, a causal
estimate of a multiplier for relief spending would likely be more positive than the
estimate here. They found that the relief programs generally were distributing funds to
counties within each state with lower incomes, a bigger drop in income between 1929
and 1933, and with higher unemployment. The relief programs did this to a much greater
degree than any of the other types of programs. The endogeneity might account for the
negative coefficients for relief and education programs in all specifications in Table 5.
The public works and resources grants had the only consistently positive effect
of the grant categories. One reason may have been that the public works projects tended
to hire people at full wages, as had been the case before and after the New Deal. The
average hourly wage on PWA projects was double the wage on WPA work-relief
projects, which had the goal of providing the poor and unemployed with enough to get
back on their feet. As a result, the workers on public works projects were each
purchasing a broader range of goods than workers on relief. On the other hand, the
public works projects spent a substantially smaller share on labor and much more on
equipment and raw material that might have been purchased from other states, causing
leakages in the multiplier. We have no strong opinion on the direction of endogeneity
bias for the public works projects. The head of the PWA, Harold Ickes, once expressed
to Congress that he was largely indifferent as to where the money went on the grounds
that a stimulus was a stimulus no matter what part of the country it was spent (Williams
1968, 116). The Public Roads Administration spending was largely determined by a
formula that was based one-third on land area, another third on population, and another
third on the extent of postal roads. The Fishback, Kantor Wallis study suggests that there
Fishback and Kachanovskaya 33
might have been a positive endogeneity bias because public works grants were higher in
counties with higher retail sales in 1929 and less of a drop in retail sales between 1929
and 1933. If the positive bias is correct, the public works projects may have an even
stronger positive effect than what is shown in Table ??.
The coefficient of the AAA and other farm grants suggests a positive but not
statistically significant relationship with per capita income when using calendar year net
federal spending, negative relationships when using fiscal year net federal spending.
The AAA program was paying farmers to take land out of production, but the goal was to
have a nation-wide price increase for farm crops that would cause farm revenue to rise.
There is plenty of narrative evidence, however, that the primary beneficiaries of the
programs were large farmers, and the Fishback, Kantor, Wallis study shows that the
counties with large farms tended to receive more funds. There is also an indirect
evidence that incomes for farm workers fell, as studies of the impact of the AAA at the
county level show that AAA spending was associated with a slight negative effect on
retail sales per capita and led to some outmigration from the counties in the 1930s (see
Fishback, Horrace, and Kantor, 2005, 2006).
Meanwhile, the results in Table 5, show that the farm lending programs had a
positive and statistically significant relationship with per capita income. The Commodity
Credit Corporation nonrecourse loans served to put a floor on farm prices and many Farm
Credit Administration loans aided farmers in obtaining mortgages. We don’t have a
strong sense of the direction of endogenity bias for the farm loans. The Fishback, Kantor,
and Wallis study offers mixed signals. Across counties within states farm loans tended to
go to areas with larger farms in 1929. On the other hand, they also tended to go to areas
Fishback and Kachanovskaya 34
with lower retail sales in 1929 and areas where the drop in retail sales between 1929 and
1933 were greatest.
The coefficient for the nonfarm loans is the smallest in absolute value and also is
statistically insignificant. Our sense is that the endogeneity bias for the coefficient is
positive and thus the effect is more positive that it would be. The vast majority of
nonfarm loans came from the RFC and the HOLC. The RFC loans were often targeted at
banks, railroads, and manufacturers. The Home Owners’ Loan Corporation (HOLC)
loans were targeted at homeowners, roughly 48 percent of the population, who were more
likely to be in the upper tier of the long-term income distribution. Even though the
HOLC refinanced troubled loans, they sought to be selective in choosing the loans to
raise the probability of repayment. The Fishback, Kantor, and Wallis (2003) county
study found that after controlling for state fixed effects, counties with higher incomes and
with a higher share of the population in the upper tier of the national income distribution
tended to receive more RFC and HOLC loans.
The federal tax receipts coefficient is negative, with one exception, although not
statistically significant. We expect that there is a strong positive bias to the tax receipt
coefficient, since roughly 40 percent of internal tax receipts came from highly
progressive income and estate taxes. The income cut-offs at which households began
paying income taxes was $2,000 for a single individual and $5,000 for a family of four,
which were very high levels. As a result, less than 10 percent of American households
paid income taxes during the 1930s. Most of the remaining internal revenues came from
excise and sales taxes, which likely were positively related to per capita income during
Fishback and Kachanovskaya 35
the Depression. During fiscal years 1934 and 1935, the processing taxes collected to
fund the AAA programs accounted for roughly 16 percent.
Conclusions
Among the sea of estimates that we have reported, we trust most the estimates
that incorporate state fixed effects, year fixed effects, and state-specific time trends while
using instrumental variable analysis. Even with all of these controls, there range of
estimates found is quite large. In the estimates using calendar-year net federal spending,
an additional dollar of per capita net federal spending increased income per capita income
in the state by between 2 cents and 52 cents. This is not a multiplier but instead suggests
some degree of crowding out of private activity. The estimates rise to 0.84 and 1.71
when we allow for a six month lag in impact by using fiscal year measures of net federal
spending. Larger IV estimates can be obtained if someone is willing to ascribe the
benefits of state-specific time trends or nation-wide time trends to the distribution of net
federal spending. Given that so many things change year by year in the economy, such
assumptions seem pretty strong.
It may be that different instruments might lead to larger estimates. Without the
instrumental variable analysis, however, the coefficients on net federal spending are
nearly all substantially less than one. The coefficients with the standard package of state
and year effects are all negative, and only the inclusion of state-specific time trends
increases the coefficients to the point where they are close to or above zero.
So what does these multiplier estimates suggest for someone interested in the
modern situation? The New Deal estimates are likely more useful for today than
Fishback and Kachanovskaya 36
estimates based on the situation during World War II because the U.S. was a command
economy mobilized for a full-scale war and devoting more than 40 percent of GDP to
fighting the war. The military determined the allocation of a large number of resources,
wages and prices were controlled, many consumer goods were unavailable or rationed,
and over 10 percent of the labor force had been drafted into the military. Without the
draft and the wage and price controls, the true share of GDP devoted to the war would
likely have been substantially higher.
If there was any time to expect a large peace-time multiplier effect from federal
spending, it would have been during the period from 1932 through 1939. Measured
unemployment rates during this time periods did not fall below 14 percent. Even if we
treating people on work relief as employed lowers the unemployment rates some so that
the lowest rate reached in this period was still a very high 9.1 percent (Darby 1976). As a
result, there were clearly a large number of underemployed resources that could have
been soaked up by federal spending without crowding out private activity. As of the end
of May 2009, the unemployment rate reached over 9 percent for the first time since the
early 1980s, still nowhere near the highs over 20 percent for any measure of
unemployment in 1932 and 1933. We would bet that the impact of net federal spending
in an economy with lower unemployment would be smaller than it was during the New
Deal.
The estimates are imperfect and can be improved in several ways. We are
seeking information on annual military expenditures by state, and are in the process of
filling in gaps on state and local government annual spending and revenues. To the
extent that either of these classes of expenditures are correlated with net federal spending,
Fishback and Kachanovskaya 37
the multipliers we have estimated so far are picking up there effects. See the appendix on
Biases from Omitting State Net Spending. As we were finishing the draft of this paper,
we realized that we can incorporate annual weather information to control for some
additional exogenous variation. Some initial runs suggest not much change in our results,
and we should have a full set of estimates incorporating weather information at the
conference. We also are working to expand the time period back into the 1920s. We
have accumulated annual information on federal tax receipts, road spending, Shepherd-
Towner spending on public health, loans for Bureau of Reclamation projects, state and
city government spending and taxation back to 1919. We are currently searching federal
reports for annual information on Veterans’ Administration spending, military spending ,
and more information on spending for Rivers and Harbors. Any knowledge you can
provide on sources would be greatly appreciated. Finally, we are trying to find ways to
fully separate transfer grants from grants that resembled federal expenditures. Our sense
is that the WPA, CWA, CCC, and half of the FERA relief grants are like federal
expenditures, while the SSA, VA, and the remainder of the FERA grants are transfer
payements.
Fishback and Kachanovskaya 38
References “$75,000,000 for Roads”, The New York Times, November 20, 1921. Aschauer, D. A. (1989). “Is public expenditure productive?” Journal of Monetary
Economics, Vol. 23, pp. 177-200. Aiyagari, S. R., Christiano, L. J., and Eichenbaum, M. (1992). “The output, employment,
and interest rate effects of government consumption,” Journal of Monetary Economics, Vol. 30, pp. 73-86.
Barro, R. J. (1981). “Output effects of government purchases,” The Journal of Political
Economy, Vol. 89, No. 6., pp. 1086-1121. Barro, Robert. “Voodoo Multipliers.” The Economists’ Voice (February 2009). Barro, R. J., and King, R. G. (Nov., 1984). “Time-separable preferences and
intertemporal-substitution models of Business Cycles,” The Quarterly Journal of Economics, Vol. 99, No. 4, pp. 817-839.
Barro, R., Sala-i-Martin, X. (2004). Economic Growth, Second Edition. MIT Press:
Cambridge, Massachusetts. Basavarajappa, K.G. and Bali Ram. “Section A: Population and Migration.” Historical
Statistics of Canada downloaded on June 28, 2009 from http://dsp-psd.pwgsc.gc.ca/Collection/Statcan/11-516-XIE/11-516-XIE.html
Benjamin, Daniel and K.G.P. Mathews. U.S. and U.K. Unemployment Between the Wars: A
Doleful Story, Institute for Economic Affairs, London, 1992 (174 + xvi). Bird, Richard. “Section H: Government Finance. Historical Statistics of Canada
downloaded on June 28, 2009 from http://dsp-psd.pwgsc.gc.ca/Collection/Statcan/11-516-XIE/11-516-XIE.html
Blanchard, Olivier and Lawrence Katz. “Regional Evolutions.” Brookings Papers on
Economic Activity 1 (1992): 1-56. Blanchard, Olivier and Roberto Perotti. “An Empirical Characterization of the Dynamic
Effects of Changes in Government Spending and Taxes on Ouput.” Quarterly Journal of Economics (2002): 1329-1368.
Brown, E. Cary. 1956, “Fiscal Policy in the ’Thirties: A Reappraisal,” American
Economic Review, 46 (December): 857-79. Bureau of Economic Analysis. State Personal Income, 1929-97. Washington, D.C.:
U.S. Government Printing Office, 1999. Cogan, J. F., Cwik, T., Taylor, J. B., and Wieland, V. (2009). “New Keynesian versus old
Keynesian government spending multipliers,” NBER Working Paper N. 14782.
Fishback and Kachanovskaya 39
Costa, J. S., Ellson, R. W., and Martin, R. C. (1987). “Public capital, regional output, and
development: some empirical evidence,” Journal of Regional Science, Vol. 27, No. 3, pp. 419-437.
Couch, J., Shughart, W., 1998. The Political Economy of the New Deal. Edward Elgar, New
York. Cullen, Joseph and Price Fishback. “Did Big Government’s Largesse Help the Locals?
The Implications of WWII Spending for Local Economic Activity, 1939-1958, with Joseph Cullen.” NBER Working Paper Number 12801. Also, University of Arizona Working paper. No. 06-15 2008.
Devereaux, M. B., Head, A. C., and Lapham, B. J. (1996). “Monopolistic competition,
increasing returns, and the effects of government spending,” Journal of Money, Credit and Banking, Vol. 28, No. 2, pp. 233-254.
Devereux, M. B., and Love, D. R. (1995). “The dynamic effects of government spending
policies in a two-sector endogenous growth model,” Journal of Money, Credit and Banking, Vol. 27, No. 1, pp. 232-256.
Dornbusch, Rudiger and Stanley Fischer. Macroeconomics, Third Edition. New York:
McGraw-Hill, 1984. Duffy-Deno, K. T., Eberts, R. W. (1991). “Public infrastructure and regional economic
development: a simultaneous equations approach,” Journal of Urban Economics, Vol. 33, pp. 329-343.
“The Federal Aid Road Act”, Washington D.C., Government Printing Office, 1916. Federal Home Loan Bank Board. Annual Report, 1935. Washington, D.C.: Government
Printing Office, 1935. Federal Works Agency. First Annual Report, Fiscal Year Ended June 1940.
Washington, D.C.: Government Printing Office, 1940. Feldstein, M. (2002). “The role for discretionary fiscal policy in a low interest rate
environment,” NBER Working Paper No. 9203. Fishback, P. V., Horrace, W. C., Kantor, S. (March 2005). “Did New Deal grant
programs stimulate local economies? A study of Federal grants and retail sales during the Great Depression,” The Journal of Economic History, Vol. 65, No. 1, pp. 36-71.
Fishback, P. V., Horrace, W. C., Kantor, S. (March 2005). “Did New Deal grant
programs stimulate local economies? A study of Federal grants and retail sales
Fishback and Kachanovskaya 40
during the Great Depression,” National Bureau of Economic Research Working Paper W8108, 2004.
Fishback, P. V., Horrace, W. C., Kantor, S. “The Impact of New Deal Expenditures on
Mobility During the Great Depression,” Explorations in Economic History,” 43 (April 2006): 179-222
Fleck, R. K., 2001b. Population, land, economic conditions, and the allocation of New
Deal spending. Explorations in Economic History 38, 296-304. Fleck, Robert. “Voter Influence and Big Policy Change: The Positive Political Economy
of the New Deal.” Journal of Political Economy (2008): 1-37. Flores de Frutos, R., Pereira, A. M. (1999). “Public capital accumulation and private
sector performance,” Journal of Urban Economics, Vol. 46, pp. 300-322. Garcia-Mila, T., McGuire, T. J. (1992). “The contribution of publicly provided inputs to
states’ economies,” Regional Science and Urban Economics, Vol. 22, pp. 229-241. Garrett, T. A., Wheelock, D. C. (2005). “Why did income growth vary across states
during the Great Depression?” Federal Reserve Bank of St. Louis working paper no. 2005-013B.
Garrett, T. A., Wheelock, D. C. (June 2006). “Why did income growth vary across states
during the Great Depression?” The Journal of Economic History, Vol. 66, No. 2, pp. 456-466.
Highway Statistics: Summary to 1945. (1947). Public Roads Administration. Federal
Works Agency. United States Government Printing Office: Washington, D. C. Ho, T. (2001). “The government spending and private consumption: a panel cointegration
analysis,” International Review of Economics and Finance, Vol. 10, pp. 95-108. Hulten, C. R., Schwab, R. M. (1991). “Public capital formation and growth of regional
manufacturing industries,” National Tax Journal, Vol. 44, pp. 121-134. Martin, Robert F. “National Income and Its Distribution, 1919-1938. The Conference
Economic Record (September 8, 1939): 81-92. McGregor, Peter, Eric McVittie, J. Kim Swales, and Ya Ping Yin. “The Neoclassical
Economic Base Multiplier.” Journal of Regional Science 40(1) 2000: 1-31. Mulligan, Gordon F. “Employment Multipliers and Functional Types of Communities
Effects of Public Transfer Payments.” Growth and Change 18 (1987): 1-11.
Fishback and Kachanovskaya 41
Merrifield, John. “A Neoclassical Anatomy of the Economic Base Multipier.” Journal of Regional Science 27 (1987): 283-294.
Merrifield, John. “A Practical Note on the Neoclassical Economic-Base Marginal
Multiplier.” Journal of Regional Science 27 (1990): 123-127.. Mulligan, Gordon F. and Lay James Gibson. “A Note on Sectoral Multipliers in Small
Communities.” Growth and Change (October 1984): 1-6. Mulligan, Gordon. “Employment Multipliers and Functional Types of Communities:
Effects of Public Transfer Payments.” Growth and Change (Summer 1987): 1-11 Munnell, A., H. (1992). “Infrastructure investment and economic growth,” Journal of
Economic Perspectives, Vol. 6, No. 4, pp. 189-198. Neumann, Todd, Price Fishback, and Shawn Kantor. “The Dynamics of Relief Spending
and the Private Urban Labor Market During the New Deal..” Journal of Economic History (forthcoming).
Ohanian, L. E. (1997). “The macroeconomic effects of war finance in the United States:
World War II and the Korean War,” The American Economic Review, Vol. 87, No. 1, pp. 23-40.
Peppers, Larry. 1973. “Full Employment Surplus Analysis and Structural Change: The
1930s,” Explorations in Economic History 10 (Winter), 197-210. Ramey, V. A., and Shapiro, M. D. (1998). “Costly capital reallocation and the effects of
government spending,” Carnegie-Rochester Conference Series on Public Policy, Vol. 48, pp. 145-194.
Reading, D. C., 1973. New Deal activity and the states, 1933 to 1939. Journal of
Economic History 33, 792-810. Reconstruction Finance Corporation. Report of Reconstruction Finance Corporation
October 1 to December 31 1932, and February 2 to December 31, 1932. U.S. House Document No. 538, 72d Congress, 2nd Session. Washington: Government Printing Office, 1933a.
Reconstruction Finance Corporation. Report for the Period from the Organization of the
Corporation on February 2, 1932, to June 30, 1932, Inclusive. U.S. Senate Document No. 135, 72d Congress, 1st Session. Washington: Government Printing Office, 1932.
Reconstruction Finance Corporation. Report for the Period Covering Its Operations for
the Third Quarter of 1933, and for the Period from the Organization of the Corporation on February 2, 1932, to September 30, 1933, Inclusive. U.S. House
Fishback and Kachanovskaya 42
Document No. 199, 73d Congress, 2d Session. Washington: Government Printing Office, 1933b.
Richardson, Harry. “Input-Output and Economic Base Multipliers: Looking Backward
and Forward.” Journal of Regional Science 25 (1985): 607-661. Romer, Christina. “The Case for Fiscal Stimulus: The Likely Effects of the American
Recovery and Reinvestment Act. February 2009. Romer, Christina and Jared Bernstein. “The Job Impact of the American Recovery and
Reinvestment Plan. January 2009. Romer Christina and David Romer. “The Macroeconomic Effects of Tax Changes:
Estimates Based on a New Measure of Fiscal Shocks.” NBER Working Paper No. W13264. 2006.
Rosenbloom, J. L, Sundstorm, W. A. (1999). “The sources of regional variation in the
severity of the Great Depression: evidence from U.S. manufacturing,1919-1937,” The Journal of Economic History, Vol. 59, pp. 714-747.
Stromberg, David. 2004. “Radio’s Impact on Public Spending.” Quarterly Journal of
Economics 119 (February): 189-221. U.S. Bureau of the Census. Financial Statistics of States, 1937. Washington, D.C.:
Government Printing Office, 1940. U.S. Bureau of the Census. Financial Statistics of States, 1938. Washington, D.C.:
Government Printing Office, 1941. U.S. Bureau of the Census. Financial Statistics of States, 1939. Washington, D.C.:
Government Printing Office, 1942. U.S. Bureau of Public Roads. Report of the Chief of the Bureau of Public Roads.
Washington, D.C., Government Printing Office, various years. U.S. Office of Government Reports, Report No. 9-1939. Unpublished mimeos. Office of
Government Reports, Record Group 44, National Archives II in College Park, Maryland.
Wallis, J. J., 1998. The political economy of New Deal spending revisited, again: With and
without Nevada. Explorations in Economic History 35, 140-70. Wallis, J. J., 2001. The political economy of New Deal spending, yet again: A reply.
Explorations in Economic History 38, 305-14. Works Progress Administration. Analysis of the Civil Works Program. Washington,
D.C.: Government Printing Office, 1939a.
Fishback and Kachanovskaya 43
Works Progress Administration. Report on the Progress of the WPA Program, June 30, 1938. Washington, D.C.: Government Printing Office, 1938.
Works Progress Administration. Report on the Progress of the WPA Program, June 30,
1939. Washington, D.C.: Government Printing Office, 1939b. Works Progress Administration. Report on the Progress of the WPA Program, June 30,
1940. Washington, D.C.: Government Printing Office, 1940. Works Progress Administration. Final Statistical Report of the Federal Emergency
Relief Administration. Washington, D.C.: Government Printing Office, 1942. Works Progress Administration. Report on the Progress of the WPA Program, June 30,
1940. Washington, D.C.: Government Printing Office, 1940. Wright, G., 1974. The political economy of New Deal spending: An econometric analysis.
Review of Economics and Statistics 56, 30-38. Zandi, Mark. “The Economic Impact of the American Recovery and Reinvestment Act.”
Report. January 21, 2009. Available at www.Moody’s.com.
Fishback and Kachanovskaya 44
Data Appendix. Information on state personal income is from Bureau of Economic Analysis,
1999. We extended the series back into the 1920s using estimates of income by state by Robert Martin (1939). Were merged the income series by running a regression with no intercept for each state from 1929 to 1938 of the form Yt = β Mt, where Y is the BEA estimate of state personal income and M is Martin’s (1939 estimates). The R-squareds from each of the regressions were all above 0.98. When we ran correlations of the growth rates for the overlap periods, they are all over 0.6 and most are over 0.9.
Information on the federal grants and loans was reported by the Office of Government Reports in Mimeographed reports that were originally considered confidential. They were obtained from the National Archives. The information was reported by fiscal year. For each year t the fiscal year ran from July 1 in year t-1 to June 30 in year t. We report results using the fiscal year definition.
In addition, we have run estimates where we have converted the fiscal year measure to calendar years using information on the national totals spent in each half year. At the end we talk about the potential for making these conversations with state level information. Information for the RFC loans was reported for the period February 2 1932 through June 30, 1933. To break the loans into the fiscal years, we assigned 51.3 percent of the loans to the fiscal year ending June 30, 1932 and the remainder to the 1933 fiscal year. The figures were based on national totals reported in Reconstruction Finance Corporation (1932, p. 3; 1933b, pp. 6-7). We have not been able to find state level information that allows us to split the data for each state. To break the loans into calendar years we used additional information on the timing of the loans from Reconstruction Finance Corporation (1933a, p. 6). We assigned 62.0 percent of the loans between February 1932 and June 30, 1933 to calendar year 1932 and 37.987 to calendar year 1933. From later fiscal years we split the amounts 50-50 to convert to calendar year. We think we can improve on those splits nationwide but not at the state level. For later years we split the fiscal year values in half and assigned the halves to the appropriate calendar years.
Information for HOLC loans in each state were reported for July 1, 1933 through December 31, 1934 and for January 1, 1935 through July 1, 1935. In the absence of state specific information, we used national figures on loans from Federal Home Loan Bank Board (1935, p. 67) to allocate 49.2 percent of the loans from July 1, 1933 through December 31, 1934 to the period July 1 1933 through June 30, 1934, and 50.8 percent to the period July 1, 1934 to June 30, 1935. For the calendar year adjustments, all of the lending and investments during the period July 1, 1933 through December 31, 1934 occurred after December 31, 1933; therefore, all of the reported value for July 1, 1933 through December 31, 1934 was assigned to calendar year 1934. The 1935 values for the HOLC programs were the values in the source reported for January 1, 1935 through June 30, 1935 plus half the value reported for the period July 1, 1935 through June 30, 1936. For later years we split the fiscal year values in half and assigned the halves to the appropriate calendar years.
The FERA spending was adjusted from the fiscal year to the calendar year based
on information on the quarterly breakdown of Federal funds distributed after May 1933 for Obligations incurred for general relief, (WPA, 1942, p. 299). We think we can
Fishback and Kachanovskaya 45
improve on this by getting the quarterly breakdowns by states. We also believe we can break the funds into pure transfers and work relief by state.
The CWA federal spending for calendar year 1933 was 245.3/(245.3+590.7) of the amount reported for July 1, 1933 through June 30, 1934 and the amount for calendar year 1934 was 590.7/(245.3+590.7) of the fiscal year amount. All values for calendar years and fiscal years are zero afterword. We based the split on the national split of federal funds on p. 25, Table 11 in WPA (1939a). We will later use the state specific splits.
We split the adjusted the fiscal year amounts to calendar year amounts for grants for the WPA , the Civilian Conservation Corps, the PWA Nonfederal projects, the PWA Federal projects, the SSA and the FSA based on national numbers on the monthly earnings of people or payments to people from WPA 1940, pp. 104-107. We believe that we can do state level splits for the WPA and the CCC and the SSA information. We are less certain about the PWA and FSA payments.
For the PRA, we adjusted the fiscal year spending to calendar year spending on the basis of the number of persons employed. See Federal Works Agency, 1940, pp. 271. For the PBA we used the average number of persons employed on PBA projects from regular funds, from FWA (1940, p. 261). For the USHA loans we used the distribution of people employed by month from FWA (1940, p.345. For the USHA grants after November 1937 we used the distribution of people employed from Federal Works Administration (1940, p. 345.
We are still working on developing splits for the AAA payments. We believe that we can get a sense of the timing of the payments for each year from the annual AAA reports.
Fishback and Kachanovskaya 46
Figure 1 Per Capita New Deal Grants and Per Capita Federal Tax Receipts for the Year 1935 in 1967$ by State, without Delaware
real
per
cap
ita n
ew d
eal g
rant
s,
real per capita total federal ta2.99631 204.434
46.5655
505.682
ct
mema
nhri
vt
njnypa
ilin
miohwi
ia
ks
mnmo
ne
ndsd
vaal
arfl
galams
ncsc tx
ky
md
ok
tnwv
az
co
id
mt
nv
nm
ut
wy
ca
or wa
Delaware had per capita federal tax receipts of $321.16 and per capita federal grants of $67.45 in 1967 dollars for the year 1935.
Fishback and Kachanovskaya 47
Table 1
Total and Per Capita Federal Spending by Program in Contemporay Dollars for the period July 1, 1932 through June 30, 1939
Acronym Amounts
From July 1, 1932 to June 30,
1939 (Millions
$)
Per Capita
Category First Fiscal Year with Significant Spending
Ended Before 1939
TOTAL TAXES COLLECTED FROM STATES
26,061 213.11
NONREPAYABLE GRANTS Works Progress Administration WPA 6,844 55.97 Relief 1936 Veterans' Administration VA 3,955 32.34 Relief Pre 1933 Federal Emergency Relief Administration
FERA 3,059 25.02 Relief 1934 Mar-37
Agricultural Adjustment Administration1
AAA 2,863 23.41 Agriculture 1934
Civilian Conservation Corps CCCG 2,130 17.42 Relief 1934 Public Roads Administration PRA 1,613 13.19 Public Works Pre 1933 Rivers and Harbors and Flood Control RHFC 1,316 10.76 Public Works Pre 1933 Public Works Adminstration--Nonfederal Projects
PWANF 1,032 8.44 Public Works 1934
Civil Works Administration CWA 807 6.60 Relief 1934 Mar-34 Social Security Act SSA 759 6.21 Relief 1936 Public Works Administration--Federal Projects
PWAF 632 5.16 Public Works 1934
Balance from Relief Acts BRA 376 3.08 Relief 1936 Public Buildings Administration PBA 324 2.65 Public Works Pre 1933 Bureau of Reclamation BR 290 2.37 Public Works 1934 Farm Security Administration FSA 273 2.24 Agriculture 1936 National Guard NG 219 1.79 Military Pre 1933 Public Works Administration--Housing Projects
PWAH 129 1.05 Public Works 1935
Soil Conservation Service SCS 100 0.82 Agriculture 1934 Agricultural Extension Work AE 94 0.77 Agriculture Pre 1933 Vocational Education VE 90 0.74 Education Pre 1933 U.S. Employment Service USES 80 0.65 Relief 1934 Indian Service - Civilian Conservation Corps
CCCIS 51 0.42 Relief 1934
Agricultural Experiment Stations AEX 36 0.29 Agriculture Pre 1933
Fishback and Kachanovskaya 48
Forest Service (Roads) FSR 34 0.28 Public Works 1937 Colleges of Agriculture and Mechanical Arts
CAM 24 0.19 Education Pre 1933
Forest Funds FF 17 0.14 Public Works Pre 1933 Mineral Lease Act Payments ML 11 0.09 Public Works Pre 1933 Land Utilization Program LUP 11 0.09 Public Works 1939 State Soldiers' and Sailors' Homes SSS 4 0.03 Relief Pre 1933 Special Funds SF 2 0.02 Miscellaneous Pre 1933 Office of Education--Emergency Relief Act Funds
OE 2 0.02 Education 1936
State Marine Schools SMS 1 0.01 Education Pre 1933 Books for the Blind BFB a) 0.00 Education Pre 1933 Federal Water Project Payments FWP a) 0.00 Public Works Pre 1933 Nonrepayable Grants Total 27,180 222.26 REPAYABLE LOANS CLOSED Reconstruction Finance Corporation RFC 4,782 39.11 All 1932 Farm Credit Administration FCA 3,957 32.35 Agriculture Pre 1933 Home Owners' Loan Corporation HOLC 3,158 25.83 Home
Finance 1934 1936
Commodity Credit Corporation CCCL 1,186 9.70 Agriculture 1934 Public Works Administration PWAL 508 4.15 Public Works 1934 Farm Security Administration FSAL 337 2.76 Agriculture 1934 Home Owners' Loan Corporation and Treasury Investments in Bldg. and Savings and Loans Associations
HOLCT 266 2.17 Home Finance
1934
Federal Reserve Banks. FRB 125 1.02 Finance 1935 Rural Electrification Administration REA 123 1.01 Agriculture 1936 U.S. Housing Authority USHA 56 0.45 Public Works 1939 Farm Tenant Purchases FTP 33 0.27 Agriculture 1938 Disaster Loan Corporation DLC 17 0.14 Relief 1937 Total Repayable 14,549 118.97 Value of Loans Insured by Federal Housing Administration
0 0.00
Title I --Refurbishing and Maintanence Loans
834 6.82 Home Finance
1936
Title II--Home Mortgages. 1,855 15.17 Home Finance
1936
Total Housing Loans Insured 2,689 21.99
aUnder 500,000 dollars.
Fishback and Kachanovskaya 49
Table 2 Estimates of Multiplier for State Personal Income in Response to Net Federal Spending, Calendar Years 1933-1939
Standard Errors Listed Below Coefficients Federal Government Funds Measure Per Capita for Calendar Year in 1967$ Grants minus Taxes Grants plus 10 % of
Loans minus Taxes Grants plus Loans
minus Taxes Grants minus AAA
minus Taxes Unweighted Weighted Unweighted Weighted Unweighted Weighted Unweighted Weighted LEAST SQUARES No Correlates -1.05 -1.82 -1.05 -1.84 -1.01 -1.70 -0.85 -1.31
0.53 0.79 0.53 0.80 0.54 0.80 0.25 0.52
State Fixed Effects -0.24 -0.31 -0.15 -0.08 0.62 1.42 -0.53 -0.95 0.38 0.40 0.43 0.45 0.63 0.58 0.16 0.35
State and Year Fixed Effects
-0.29 -0.27 -0.29 -0.25 -0.26 -0.06 -0.18 -0.22 0.15 0.17 0.15 0.16 0.14 0.15 0.08 0.11
State Fixed Effects and State-Specific Time Trends
0.75 0.79 0.83 0.94 0.98 1.08 0.38 0.18 0.22 0.36 0.25 0.39 0.21 0.26 0.28 0.45
State and Year Fixed Effects and State-Specific Time Trends
0.26 0.09 0.24 0.07 0.08 -0.07 0.19 -0.01 0.20 0.35 0.20 0.36 0.22 0.30 0.17 0.28
TWO-STAGE LEAST SQUARES State and Year Fixed Effects
* 2.51 * 2.60 * * * * 1.01 1.07
State Fixed Effects and State-Specific Time Trends
1.07 1.38 1.10 1.45 1.59 2.57 1.21 1.62 0.46 0.35 0.47 0.37 0.72 0.69 0.52 0.43
State and Year Fixed Effects and State- Specific Time Trends
0.52 0.44 0.53 0.45 0.62 0.59 0.55 0.43 0.42 0.28 0.43 0.28 0.53 0.38 0.45 0.27
Fishback and Kachanovskaya 50
*Weak instrument problem. F-statistic was too low to give credible estimates. Sources: See Data Appendix. The original source reported the federal spending and tax receipt information in fiscal years that ended
on June 30 in each year. We adjusted the information to calendar years using the methods described in the Data Appendix. Notes: This is a balanced panel with information for 7 years for each state. All estimates were made using the STATA 10 reg and
ivreg2 programs. Standard errors for the unweighted least squares estimates are based on White corrections using the robust command with standard errors clustered at the state level. The weight for the weighted least squares estimates is the square root of the 1930 population in the state and standard errors were clustered at the state level. We performed tests on instrument strength by comparing the Kleibergen-Paap rank Wald (KPW) F statistic with the Stock-Yogo critical values for the maximum IV bias that one is will to accept. The critical values for willingness to accept weak instrument bias of 10 percent is 16.38, of 15 percent is 8.96, and of 20 percent is 6.66. We did not report the coefficients and standard errors for the unweighted 2SLS estimates with state and year fixed effects because the KPW F-statistic was less than 2.58 in all cases. The weighted 2SLS estimates with state and year fixed effects are 10.58, 9.92, 3.39, and 4.46 as you move from grants to grants and 10 percent loans, grants plus loans, and grants minus AAA grants. In the estimates with state effects and state-specific time trends the KPW F-statistic exceeded 19 in all weighted and unweighted estimates, suggesting no weak instrument bias if one is willing to accept up to 10 percent bias. In the unweighted estimates with state and year fixed effects and state-specific time trends, the KPW F-statistic exceeds 21.42 for the grants minus taxes and the grants plus 10 percent of loans minus taxes, and is between 13.29 and 14.91 for the grants plus loans minus taxes and grants minus AAA grants minus taxes; the KPW F-statistic exceeds 33 in all of the weighted versions of the specification.
Fishback and Kachanovskaya 51
Table 3 Estimates of Multiplier for State Personal Income in Response to Net Federal Spending, Calendar Years 1932-1939
Standard Errors Listed Below Coefficients Federal Government Funds Measure Per Capita for Calendar Year in 1967$ Grants minus Taxes Grants plus 10 % of
Loans minus Taxes Grants plus Loans
minus Taxes Grants minus AAA
minus Taxes Unweighted Weighted Unweighted Weighted Unweighted Weighted Unweighted Weighted LEAST SQUARES No Correlates -0.98 -1.76 -0.99 -1.78 -1.01 -1.73 -0.87 -1.35
0.55 0.83 0.56 0.84 0.57 0.86 0.27 0.55
State Fixed Effects 0.30 0.34 0.35 0.48 0.70 1.29 -0.39 -0.86 0.53 0.50 0.56 0.55 0.65 0.58 0.13 0.31
State and Year Fixed Effects
-0.14 -0.14 -0.14 -0.13 -0.15 -0.03 -0.12 -0.16 0.11 0.16 0.11 0.16 0.09 0.14 0.05 0.09
State Fixed Effects and State-Specific Time Trends
0.35 0.49 0.39 0.59 0.51 0.73 0.14 0.03 0.23 0.29 0.26 0.32 0.25 0.23 0.23 0.39
State and Year Fixed Effects and State-Specific Time Trends
-0.05 -0.06 -0.06 -0.07 -0.13 -0.12 -0.02 -0.09 0.16 0.23 0.16 0.23 0.14 0.19 0.15 0.19
TWO-STAGE LEAST SQUARES State and Year Fixed Effects
1.85 1.07 1.88 1.10 2.17 1.40 1.85 0.93 0.81 0.40 0.84 0.41 1.15 0.57 0.80 0.35
State Fixed Effects and State-Specific Time Trends
0.47 0.78 0.48 0.82 0.68 1.59 0.55 0.93 0.37 0.31 0.39 0.33 0.58 0.82 0.41 0.35
State and Year Fixed Effects and State-Specific Time Trends
0.02 0.06 0.02 0.07 0.02 0.08 0.05 0.07 0.32 0.25 0.32 0.25 0.35 0.31 0.33 0.25
Fishback and Kachanovskaya 52
Sources: See Data Appendix. The original source reported the federal spending and tax receipt information in fiscal years that ended on June 30 in each year. We adjusted the information to calendar years using the methods described in the Data Appendix.
Notes: This is a balanced panel with information for 8 years for each state. All estimates were made using the STATA 10 reg and ivreg2 programs. Standard errors for the unweighted least squares estimates are based on White corrections using the robust command with standard errors clustered at the state level. The weight for the weighted least squares estimates is the square root of the 1930 population in the state and standard errors were clustered at the state level. We performed tests on instrument strength by comparing the Kleibergen-Paap rank Wald (KPW) F statistic with the Stock-Yogo critical values for the maximum IV bias that one is will to accept. The critical values for willingness to accept weak instrument bias of 10 percent is 16.38, of 15 percent is 8.96, and of 20 percent is 6.66. For the unweighted 2SLS estimates with state and year fixed effects the KPW F-statistic was between 9.13 and 15.46. The weighted 2SLS estimates with state and year fixed effects are 20.9, 21.1, 18.22, and 15.9 as you move from grants to grants and 10 percent loans, grants plus loans, and grants minus AAA grants. In the estimates with state effects and state-specific time trends the KPW F-statistic exceeded 26 in all weighted and unweighted estimates, suggesting no weak instrument bias if one is willing to accept up to 10 percent bias. In the estimates with state and year fixed effects and state-specific time trends, the KPW F-statistic exceeds 20 in all cases.
Fishback and Kachanovskaya 53
Table 4 Estimates of Multiplier for State Personal Income in Response to Net Federal Spending
in the FISCAL YEAR Ending on June 30 for the Years 1932-1939 Standard Errors Listed Below Coefficients
Federal Government Funds Measure Per Capita for Calendar Year in 1967$ Grants minus Taxes Grants plus 10 % of
Loans minus Taxes Grants plus Loans
minus Taxes Grants minus AAA
minus Taxes Unweighted Weighted Unweighted Weighted Unweighted Weighted Unweighted Weighted LEAST SQUARES No Correlates -0.89 -1.65 -1.62 -1.78 -0.77 -1.32 -0.81 -1.59
0.57 0.79 0.77 0.84 0.47 0.57 0.62 0.87
State Fixed Effects 0.45 0.53 0.41 0.48 -0.14 -0.49 0.25 0.17 0.48 0.44 0.40 0.55 0.21 0.23 0.46 0.43
State and Year Fixed Effects
-0.06 -0.16 -0.16 -0.13 -0.04 -0.16 -0.04 -0.15 0.11 0.17 0.17 0.16 0.09 0.14 0.12 0.18
State Fixed Effects and State-Specific Time Trends
0.99 10.22 1.13 0.59 0.46 0.38 1.00 1.25 0.25 0.41 0.37 0.32 0.12 0.15 0.30 0.52
State and Year Fixed Effects and State Specific Time Trends
0.26 -0.18 0.01 -0.07 0.24 0.09 0.28 -0.03 0.22 0.41 0.39 0.23 0.15 0.20 0.22 0.43
TWO-STAGE LEAST SQUARES State and Year Fixed Effects
1.54 2.16 1.50 2.09 1.28 1.68 1.79 2.49 0.47 0.70 0.45 0.67 0.40 0.56 0.55 0.79
State Fixed Effects and State-Specific Time Trends
2.43 3.48 2.25 3.15 1.57 2.04 3.28 5.12 0.34 0.45 0.31 0.39 0.21 0.27 0.55 0.75
State and Year Fixed 0.84 1.71 0.77 1.44 0.45 0.61 * *
Fishback and Kachanovskaya 54
Effects and State Specific Time Trends
0.32 0.78 0.29 0.61 0.16 0.22
Fishback and Kachanovskaya 55
Sources: See Data Appendix. The original source reported the federal spending and tax receipt information in fiscal years that ended on June 30 in each year. We adjusted the information to calendar years using the methods described in the Data Appendix.
Notes: This is a balanced panel with information for 7 years for each state. All estimates were made using the STATA 10 reg and ivreg2 programs. Standard errors for the unweighted least squares estimates are based on White corrections using the robust command with standard errors clustered at the state level. The weight for the weighted least squares estimates is the square root of the 1930 population in the state and standard errors were clustered at the state level. We performed tests on instrument strength by comparing the Kleibergen-Paap rank Wald (KPW) F statistic with the Stock-Yogo critical values for the maximum IV bias that one is will to accept. The critical value for willingness to accept weak instrument bias of 10 percent is 16.38, of 15 percent is 8.96, and of 20 percent is 6.66. For the unweighted and weighted 2SLS estimates with state and year fixed effects the KPW F-statistic was between 12.63 and 15.85 and so meet the criterion if one is willing to accept up to 15 percent weak instrument bias. In the estimates with state effects and state-specific time trends the KPW F-statistic exceeded 33 in all weighted and unweighted cases, suggesting no weak instrument bias if one is willing to accept up to 10 percent bias. In the estimates with state and year fixed effects and state-specific time trends, the KPW F-statistic is roughly 6.84 for the grants minus taxes, meeting the criteria if one is willing to accept 20 percent weak-instrument bias. The values are 8.26 and 9.22 when 10 percent of the loans are added, and rise to over 21.4 when all loans are added. When AAA grants are subtracted in the far right column the values fall below 3.7 and we do not report the coefficients because of potential problems with weak-instrument bias.
Fishback and Kachanovskaya 56
Table 5 Least Squares Estimates for Specific Grant and Loan Categories and Federal Tax Receipts,
Includes State and Year Fixed Effects and State-Specific Time Trends
Unweighted Weighted Unweighted Weighted Unweighted Weighted 1932-1939 1932-1939 1933-1939 1933-1939 1933-1939 1933-1939 Calendar
Year Calendar
Year Calendar
Year Calendar
Year Fiscal Year Fiscal Year
Relief -0.68 -0.70 -0.61 -0.70 -0.35 -0.20 0.32 0.39 0.38 0.39 0.32 0.34
Public Works 0.10 0.27 0.60 0.27 0.39 0.30 0.27 0.31 0.18 0.31 0.11 0.17
Farm Grants 0.49 0.95 0.45 0.95 -0.61 -0.40 0.77 0.67 0.87 0.67 0.75 0.71
Farm Loans 0.60 0.41 0.85 0.41 0.58 0.44 0.27 0.28 0.37 0.28 0.19 0.14
Nonfarm Loans -0.24 -0.26 -0.16 -0.26 0.13 0.23 0.25 0.25 0.31 0.25 0.28 0.30
Federal Tax Revenues -0.22 0.32 -0.38 0.32 -0.37 0.41 0.41 0.63 0.53 0.63 0.91 0.98
Fishback and Kachanovskaya 57
Table W-1 IV Estimates of the Multiplier, Effects of Weather, and Year Fixed Effects from Analyses
with State and Year Fixed Effects and State-Specific Time Trends (Standard Errors below coefficients in italics)
1933-1939 1932-1939 Fiscal Year Calendar Year Calendar Year unweighted weighted unweighted weighted unweighted weighted Kleibergen-Paap rank Wald F-statistic for IV strength
6.62 5.37 20.60 28.52 33.00 48.17 25% 10% 10% 10% 10%
Net Federal Spending 1.14 2.14 0.67 0.61 0.08 0.10
0.41 1.03 0.40 0.32 0.33 0.26
Inches of Precipitation
-0.12 0.50 0.06 0.38 -0.54 -0.43 0.66 0.83 0.73 0.69 0.67 0.60
Inches of Snowfall -0.24 -0.01 -0.32 -0.07 -0.30 -0.18 0.39 0.48 0.38 0.39 0.29 0.30
Days with High Temperatures over 90
-0.04 0.33 -0.21 -0.10 0.08 0.13 0.54 0.65 0.49 0.47 0.44 0.43
Days with High Temperatures between 80 and 90
-0.15 -0.08 0.07 0.09 -0.05 0.25 0.44 0.58 0.39 0.41 0.39 0.37
Days with Low Temperatures Below Zero
0.60 0.93 0.43 0.75 0.14 0.53 0.74 0.95 0.61 0.66 0.55 0.56
Days with Low Temperatures between 0 and 30
0.82 1.03 0.92 0.75 0.57 0.45 0.43 0.66 0.33 0.38 0.31 0.33
Months of Severe or Extreme Drought
-5.31 -2.69 -3.73 -0.96 -2.54 -0.44 2.54 3.16 2.23 2.05 1.86 1.59
Months of Severe or Extreme Wet
-11.66 -11.22 -11.43 -12.24 -6.01 -4.99 4.88 5.97 5.22 4.76 3.87 3.99
Year 1933 26.33 30.48 10.49 10.33
Year 1934 107.98 94.73 108.47 109.52 152.38 149.73 14.06 21.45 14.49 13.54 16.86 16.65
Year 1935 185.15 188.92 199.96 207.79 241.36 242.09 23.36 26.68 22.15 20.37 23.37 23.05
Year 1936 374.83 373.53 383.23 385.45 415.50 413.98 30.96 35.15 28.29 27.97 29.01 29.99
Year 1937 442.85 479.71 442.61 456.21 452.41 464.16 41.45 49.47 40.68 38.76 38.42 38.79
Year 1938 424.74 494.57 377.85 386.01 382.08 383.97
Fishback and Kachanovskaya 58
56.14 79.63 50.32 48.14 46.25 46.74
Year 1939 551.41 600.23 519.09 526.32 520.66 519.52 60.76 77.37 56.70 54.37 51.31 52.09
State Fixed Effects Included Included Included Included Included Included State-Specific Time Trends
Included Included Included Included Included Included
The instrument is the same as in Tables 2, 3, and 4 in the text. The weight in the weighted analysis was the square root of the population of the state in 1930. Of the weather variables only the severe and extreme wetness and drought measures and the number of days with low temperatures between 0 and 30 have precisely estimated impact in some specifications. An additional month of severe or extreme drought reduced per capita income by up to $5.31, while an additional month of severe or extreme wetness reduce per capita income by up to $11.66. The 1930s were known for a large number of droughts, and the average for the period was nearly 2 months of droughts per year. Problems with severe wetness were much less common. Controlling for weather increases the estimates relative to those reported in the main text. We have to be careful about the coefficients in the estimations of state income per capita in the calendar year on net federal spending in the fiscal year because there are more problems with weak instruments than with the other estimates. Table ?? Comparisons of Estimates from Text with Estimates Controlling for Weather. 1933-1939 1932-1939 Fiscal Year Calendar Year Calendar Year unweighted weighted unweighted weighted unweighted weighted Weather Variables Included
1.14 2.14 0.67 0.61 0.08 0.10 0.41 1.03 0.40 0.32 0.33 0.26
Weather Variables Not Included
0.84 1.71 0.52 0.44 0.02 0.06 0.32 0.78 0.42 0.28 0.32 0.25
Kleibergen-Paap rank Wald F-statistic for IV strength Weather Included
6.62 5.37 20.60 28.52 33.00 48.17 25% 10% 10% 10% 10%
Kleibergen-Paap rank Wald F-statistic for IV strength Weather Not Included
6.88 6.84 22.37 98.47 38.63 55.25 20% 20% 10% 10% 10% 10%
Notes. The estimates with Weather Variables not included are from the bottom row of
the first two columns of Tables 2, 3, and 4 in the text.
Fishback and Kachanovskaya 59
Table W-2 Means, Standard Deviations, Elasticities and One-Standard Deviation Effects for the 1933-1939 Unweighted Fiscal Year Estimates in First Column of Table W-1
Elasticity OSD Mean Std.
Dev Net Federal Spending
0.043 0.310 40.11 106.5
Inches of Precipitation
-0.004 -0.004 35.13 14.7
Inches of Snowfall
-0.005 -0.013 23.31 20.89
Days with High Temperatures over 90
-0.002 -0.004 46.65 32.42
Days with High Temperatures between 80 and 90
-0.009 -0.009 66.98 22.65
Days with Low Temperatures Below Zero
0.004 0.019 7.81 12.1
Days with Low Temperatures between 0 and 30
0.063 0.089 81.7 42.1
Months of Severe or Extreme Drought
-0.010 -0.038 1.92 2.79
Months of Severe or Extreme Wet
-0.004 -0.023 0.36 0.774
Real Per Capita Income 1062.67 390.15
Fishback and Kachanovskaya 60
I tried a situation where we interacted the standard deviation measure with federal spending per capita in Canada but the F-statistics were extremely low.