Do Tax Increases In New York Cause a Loss of Jobs? Response to the Confusers
By
Moshe Adler
Dept. of Urban Planning
Columbia University
212 873 6803
ma820@columbia.edu
www.columbia.edu/~ma820
The article “Do Tax Increases In New York Cause a Loss of Jobs? A Review of the Evidence,” by Moshe Adler, Oliver Cooke and James Parrott (State Tax Notes, Feb. 4, 2002) examines two studies. The first is by NBER researchers Andrew Haughwout, Robert Inman, Steven Craig and Thomas Luce, “Local Revenue Hills: A General Equilibrium Specification with Evidence from Four U.S. Cities.”[1] The second by David G. Tuerck, Jonathan Haughton, Corina Murg and Sorin Codreanu, “Tax Changes in New York City, The New York City Tax Analysis Modeling Program (NYC-STAMP).” The latter study was commissioned by the Manhattan Institute. Our article shows that both studies suffer from a fatal case of the “omitted variables” syndrome. Almost all the variables that can cause changes in employment except for taxes are missing. And even the relevant tax variables are missing.
Tuerck and Haughton object to the criticism[2] and I respond to their objection below. But it should first be noted that while Haughwout et al. made their data available, Tuerck et al. refused to do so. As shown below, there are reasons to believe that with the correct specification the data they refuse to share does not show that raising taxes causes a loss of jobs.
Tuerck and Haughton claim that:
I. It is unfair to fault Haughwout et al. and Tuerck et al. for specifying only New York City taxes but not taxes in any other jurisdiction. It is unfair because “it would be difficult to assemble a list of jurisdictions that would command wide agreement.”
But who would disagree with including New York City’s obvious competitors, New Jersey and Connecticut? Between 1989 and 1992, private sector employment in NYC fell by 10% at the same time the top personal income tax rate increased by 31% and these observations are in Haughwout et al.’s and Tuerck et al.’s data. But in 1990 New Jersey raised its top rate by a full 90% (from 3.5% to 6.65%). And in 1991 Connecticut instituted an income tax for the first time, with a top rate of 4.5%. These changes must be included in any study of the effect of taxes on employment in New York City.
II. The “Review of the Evidence” falsely depict their model and the Haughwout et al. model as “everything is taxes” models. After all, Tuerck and Haughton argue, the Haughwout et al. model includes in addition to taxes (in New York City only), two other variables: Government transfers and the Dow Jones Index. And their own model also includes, in addition to taxes (in New York City only), two other variables: Government transfers and the U.S. wage rate by sector. The variables that other researchers use to explain changes in employment are superfluous, Tuerck et al. claim.
The Haughwout et al. model contains three explanatory variables: The highest city personal income tax rate in New York City, government transfers to New York City and the Dow Jones Index. In their regression only the highest personal income tax rate is statistically significant. The other two variables are not. Is it wrong to argue that their model, which “proves” that “everything is taxes in New York City,” is an “everything is taxes in New York City” model?
Haughwout et al. draw their data from the years 1970-1997, but as we shall see below their result, that taxes cost jobs, holds only for the 1970s but not for the subsequent period. The 1970s are so prominent in the debate about whether taxes cost jobs, that it is worthwhile to understand in some detail what were the economic conditions at the time. It turns out that in the 1970s New York City underwent such dramatic economic shifts, that even if the Dow Jones and transfer taxes were significant in Haughwout et al. regression, their model would have still qualified as an “everything is taxes in New York City” model. Too many variables that had tremendous effects on employment are omitted.
Most researchers of the 1970s start their analysis with the observation that a fall in employment during the period was not unique to one city but that it afflicted almost all American cities at the same time. In fact, the loss of jobs in New York City during the 1970s is indistinguishable from the loss of jobs in other cities (see previous article for detail), and reflects a geographic restructuring and decentralization of the economy. Between 1970 and 1980 the proportion of the urban population that lives in cities declined from 46% to 40%. During these years the population in the Northeast experienced 0% growth, while the population of the South increased by 20%.
Thus, what needs to be explained is not the loss of jobs in New York City but a monumental change in the location of jobs in America following WWII: the move of jobs and people from cities to suburbs and from the snow belt to the sunbelt. The factors that brought about the shift of people and jobs include the following.
- In the 1970s the growth of manufacturing employment in the U.S. declined significantly. Whereas between 1960 and 1970 the nation gained 2.6 million manufacturing jobs, between 1970 and 1980 the gain was of only .9 million.[3]
- The Northeast relied more heavily on imported oil than the South. After the oil embargo of 1973 the South became a more attractive location for both firms and households.
- Wages and rates of unionization were lower in the South.
- New production techniques in manufacturing required a horizontal layout. But land is cheaper in suburbs than it is in cities.
- Following the development of a national highway system in the 1950s and 60s, trucks came into greater use. This decreased the need of plants to be near train hubs on the one hand, and made cities, with their congested streets, less attractive to manufacturers on the other.
- Starting in the mid-1950s Congress authorized the creation of a national highway system. This made the South more accessible.
- The federal government offered greater subsidies for the construction of new roads and highways but not for the maintenance of existing ones.[4]
- Army Corps of Engineers waterway projects opened the South to international trade.
- The accelerated depreciation that the tax law permitted favored the construction of new plants over the retrofitting of exiting ones.
- Mortgage-interest deductibility encouraged homeownership and suburbanization.
- Following WWII and the G.I. Bill on the one hand, and the launch of the first space satellite Sputnik in 1957, new state universities and colleges were opened everywhere. In the South this meant the development of a new professional workforce.
In their response Tuerck and Haughton are silent about all these factors. They claim, however, that because the wage and unionization gaps between New York City and Southern states preceded the 1970s by at least a century, these gaps are irrelevant for the movement of jobs to the South in the 1970s. They also claim that the preferential taxation of investment in new plants versus existing plants may have favored New York City rather than harm it. And that since mortgages on cooperative and condominium apartments are also deductible, these deductions cannot explain the movement of residents from cities to suburbs Let’s examine these claims.
Jay Helms (1985) has estimated the effect of wages and rates of unionization using data for 1965-1979 and found that while wage differences are indeed not highly significant, rates of unionization are.[5]
Tuerck’s and Haughton’s puzzlement about the role of wages or rates of unionization given that the North-South gap in these variables precedes the 1970s is worthwhile, though. It points to the role of infrastructure. It appears that the reason that jobs did not move to the South earlier is the lack of infrastructure. Good roads, good waterways, a good communication system and a professional workforce for ancillary services such as finance, insurance and law must all exist before industry can move in. It is this infrastructure that was built in the South in the 1950s and 1960s, and made possible the growth in employment.
Tuerck’s and Haughton’s position regarding the preferential tax treatment of investment in new plants versus investment in old plants, is puzzling. They claim that the variable is important in determining employment, but that it was advantageous to New York City. They may be right. But since neither they nor Haughwout et al. included it in their studies, isn’t it an omitted variable, then?
Tuerck and Haughton point out that mortgage deductibility is available not only for single family homes but for apartments as well. Based on this they wish to conclude that mortgage deductibility, even if it did encourage home ownership, did not encourage suburbanization. This is because ownership could just as easily be had in apartments in the city as in single family homes in the suburbs.
Too bad that Tuerck and Haughton were not there to block the millions of Americans who until the late 1980s equated homeownership with single family homes and rushed to the suburbs in order to fulfill that part of the American Dream.[6] In 1970 99% of owner-occupied units were in single family homes.
Let me note that if the task of accounting for all the relevant variables appears daunting, there is a round about way of getting at the effect of taxes without including them in the analysis. This can be done by comparing employment and taxes in New York City with employment and taxes in other cities. (In a model that compares cities to cities, there is no need to include the variables that affect all cities in the same way.) But neither Haughwout et al. (nor Tuerck and Haughton in their study) took this route.
Is it wrong to characterize models which do not include any of the relevant variables but do include the taxes in New York City alone as suffering from a fatal case of omitted variables? Is it wrong to characterize them as “everything is taxes in New York City” models?
III. Haughwout et al.’s study covers the period 1970-1997. The“Review of the Evidence” showed that if the 1970s are removed from the data, the result that taxes cost jobs does not hold. Tuerck and Haughton argue that this not due to the peculiarity of the 1970s, but to the smaller sample size. I therefore ran Haughwout et al.’s regression again with even fewer observations, just 10, for the years 1970-1979. As Table 1 shows, when the data is limited to just these 10 observations, the coefficient for taxes is statistically significant. Nevertheless, when the data has 23 observations without the fiscal crisis years, all the coefficients for taxes are insignificant. Thus, the problem is not a small number of observations. The problem is that the Haughwout et al. study is actually a study of only the years of the fiscal crisis, and that as a model for these years it fails entirely because it omits all the relevant variables. What the Haughwout et al. model captures is the fact that job losses created a fiscal crisis in cities, and that this crisis led to increased taxes.
Table 1: Changes in Non-Farm
Employment Share
New York City 1970-1979,
1978-2000 |
||
|
|
1970-1979 |
1978-2000 |
|
Constant |
-.00129* (-.00028) |
-.00048* (-.00023) |
|
D tax rate |
-.00072* (-000.3) |
-.00019 (-.00034) |
|
D tax rate 1 year lag |
-.00063 (-.00033) |
-.00045 (-.00034) |
|
D tax rate 2 years lag |
-.00042 (-.00030) |
-.00031 (-.00033) |
In the regressions
which included the years 1998-2000--but not in any other regression-- the
variables that are insignificant in the Haughwout et al. were omitted. This is the same procedure that Haughwout
et al. themselves follow: they omit the
property tax variable, for instance, from the regression, simply because it is
insignificant. Omitting insignificant
variables is a standard procedure in regression analysis.
The reason for
omitting these variables when the years 1998-2000 are included is that these
three years are not included in the NBER study and therefore the NBER did not
supply us with data for them. The
government transfers variables is not a government produced variable but a
variable that Haughwout et al. constructed themselves. Since we did not know
how the NBER would construct this variable for the later years, we omitted it,
together with the other insignificant variable, the annual Dow Jones
index.
Whichever the reason
may be, as one would expect and as the “Review of the Evidence” article
actually shows, the omission of insignificant variables had no effect on the
results. Taxes are significant in the regression that include the years
1998-2000 in spite of this “sinister” omission. Once the fiscal crisis years are removed, the tax coefficients
become insignificant not only in the regression that includes 1978-2000, but
also the regression for 1970-1997 that includes all of Haughwout et al.’s
variables.
.
V. The Tuerck et al. study correctly accounts for the restructuring of the securities industry following the crash of 1987.
Since Tuerck et al. do not share their data, it is impossible to know what they do. Nevertheless I take this claim as being supportive of the criticism that Haughwout et al. in their study failed to account for the effect of the crash of 1987.
The fact that Tuerck and Haughton do not supply their data, together with the methodology of their study, raises the possibility that a correct analysis of these data will not bear their results.
Tuerck et al. use a special technique for separating the effect of national economic conditions from local economic conditions on employment. Haughwout et al. include the level of national employment in their model. (They actually use the national employment share of New York City.) This is, of course, the most natural way to account for the effect of national economic conditions on employment. But Tuerck et al. do not include national employment in their model. Instead they include national wages as an indicator of the effect of national economic conditions on employment.
The peculiarity of this method is further evident in the fact that Tuerck et al. also estimate the effect of taxes on wages, and that in these equations they do the opposite: They include the level of national employment as an indicator of the effect of national economic conditions on wages. Thus, they have the data about national employment, but they chose not to use it where it belongs.
It is possible that Tuerck et al. chose this peculiar specification for the following reason. Their data covers the years 1975-1999. Thus it does not include 1970-1974, the years of employment decline that precipitated the fiscal crisis. The Haughwout et al. study shows that once the figures for national employment are included and the 1970s are removed taxes become insignificant. It is therefore possible that when the Tuerck et al. study is run with the correct specification, with national employment as an indicator for national economic conditions, taxes are statistically insignificant.
The Haughwout et al. and the Tuerck et al. studies omit most of the variables that affect employment in a jurisdiction, including taxes in adjacent jurisdictions. Haughwout et al. data shows a correlation between taxes in NYC and employment in NYC during the 1970s but not in 1978-2000. The 1970s were a period of economic restructuring, and job losses in New York City at the time were no different from job losses in other cities. But the Haughwout et al. study does not account for these structural changes. As the current recession in New York City and makes clear, when a city experiences job losses and a decline in tax revenues, it often resorts to raising taxes. This is what happened in the 1970s, and it is clear that Haughwout et al. mistake cause and effect. Tuerck et al. refuse to release their data, and there is reason to believe that their own data belies their claim.
[1] National Bureau of Economic Research, March 2000.
[2] State Tax Notes, August 28, 2002
[3] U.S. bureau of labor statistics.
[4] Franklin J. James, “Economic Distress in Central Cities,” in Cities Under Stress, p. 46.
[5] L. Jay Helms, “The Effect of State and Local Taxes on Economic Growth: A Time Series-Cross Section Approach,” The Review of Economics and Statistics, 1985, pp. 574-582.
[6] The question why Americans equated homeownership with single family homes is explored in Moshe Adler, “The American Dream and Central City Blight Are American,” Journal of Economic Behavior and Organization, 1988, vol. 9, pp. 381-392.