This article appeared in the September 9, 2002 issue of State Tax Notes
David G. Tuerck,
Suffolk University
Jonathan Haughton, Suffolk University
The following relates to the article in the December 2002 issue of State Tax Notes entitled, “Do Tax Increases in New York City Cause a Loss of Jobs? A Review of the Evidence” (revised version of December 11, 2001), by Moshe Adler, Oliver Cook and James Parrott of the Fiscal Policy Institute (FPI).
That article offers a highly critical review of two otherwise unrelated tax studies. The first, “Local Revenue Hills: A General Equilibrium Specification with Evidence from Four U.S. Cities” (NBER, March 2000), was published by a group of academic researchers, including Andrew Haughwout, an economist now working for the Federal Reserve Bank of New York. [1] The second, Tax Changes in New York City: The New York Tax Analysis Modeling Program (NYC-STAMP), was issued in September 2001 by the Manhattan Institute for Policy Research. [2]
The study by Haughwout, et al. was designed primarily to chart the tax revenue burdens of four large cities against their theoretical point of diminishing returns, as tracked on the so-called “Laffer curve.” The study also used regression analysis to estimate the impact of tax changes on employment in two of the cities, New York and Philadelphia.
The Beacon Hill Institute's State Tax Analysis Modeling Program (“STAMP”), the model used as the basis for the Manhattan Institute study, was first created by the Beacon Hill Institute (BHI) in 1994 to measure the effects of income tax cuts in Massachusetts. Since then, BHI has applied STAMP to more than 14 states. In constructing STAMP, BHI uses standard statistical methods to estimate the significance of relationships between economic variables (such as employment) and tax variables (such as the average marginal income tax rate). The variables for NYC-STAMP are drawn from city economic and tax data from 1975 through 1999.
The FPI report claims that both the Manhattan Institute report and the Haughwout et al. study “explain the changes in employment...[in New York City] by changes in taxes” (p. 2). FPI derides this as a simplistic, “everything is taxes” (p. 2) approach.
Contrary to what FPI implies, neither the Manhattan Institute report nor the Haughwout study asserted that changes in tax policy had been the sole determinant of changes in city employment. The Haughwout study includes two non-tax independent variables – government transfers to New York City, and the Dow Jones Index. As shown in Table 1 below, the NYC-STAMP model used in the Manhattan Institute report includes both a measure of government transfer payments and a national economic indicator, either the U.S. wage rate or U.S. employment, as shown.
Both studies indicate that tax policy played an important, but not exclusive, role in determining changes in city employment. The Manhattan Institute's NYC-STAMP model indicated that tax cuts had generated 80,000 jobs, or about one-fourth of the city`s private sector employment growth since 1997. The Haughwout study suggested that 16 percent of the decline in New York City's share of national employment from 1970 through 1997 could be attributed to the rise in the city's income tax rate.
Despite the evidence that both studies do acknowledge the role of non-tax factors in determining employment trends in New York City (a fact that the FPI authors acknowledge only in a footnote), the FPI authors build on their erroneous claim to assert that the studies suffer from “omitted variable syndrome,” in which “correlation is being mistaken for causation.” The real issue is whether and which of these factors are relevant or useful in a model designed to assess short- and medium-term changes in the city's employment levels.
The FPI authors identify a number of explanatory variables that, they argue, the Haughwout and Manhattan Institute studies should have included. We contend that the variables they identify are not patently better than those already those included in the studies and may not even be relevant to the time period of interest.
For example, one of the supposedly omitted explanatory variables identified by FPI – lower wages in the South –– has been a settled economic fact of life for at least a century, including the period of New York's greatest growth and economic pre-eminence. The question then is what made the wage gap suddenly relevant in the 1970s? Might the city's rising income taxes during this period have played a role? The same point applies to unionization – another of the explanatory variables proposed for inclusion by FPI, which has historically been lower in the South than in the Northeast, not just in the period of concern for these studies. [3]
FPI also cites alleged federal policy biases against New York as a variable that contributed to decreases in employment. However, as with North/South wage rate and unionization disparities, these policies have been in place for decades, and their effects are at most ambiguous.
Another supposedly omitted variable identified by FPI is the federal tax treatment of depreciation, which has varied over time. However, it is not clear how variations in federal law, as it applies to depreciation, should be expected to affect investing in New York City. A lower cost of capital might tilt investment toward capital-intensive sectors for which the City has an advantage; indeed, the city enjoyed a surge of job growth after federal tax depreciation was markedly accelerated for a time in the 1980s. Or it might favor other, less congested and lower-wage parts of the country. One simply doesn't know. Nor do mortgage and tax subsidies for home ownership create an inherent bias away from New York City apartments toward single-family homes. Interestingly, FPI's laundry list of omitted variables ignores one factor that surely did have a significant effect on the City's housing market: rent control.
In identifying these variables, FPI simply goes on a fishing expedition. It simply suggests that there is some variable or group of variables that, if included, would have dominated taxes in explaining New York City economic activity. Never mind that FPI cannot show convincingly how, for better or worse, these variables should have been expected to affect that activity or whether their inclusion would have improved (and they almost certainly would not have improved) the robustness of the estimate.
FPI also criticizes the Manhattan Institute study specifically for not including tax rates from other jurisdictions. While taking this step might have improved NYC-STAMP, it might just as well have made no difference. [4]
If this criticism is considered valid, the difficulty arises of determining what the relevant other jurisdictions should be. Clearly New Jersey and Connecticut. But New York City competes with jurisdictions across the country – indeed the world – for workers and capital. So what about the Massachusetts rate? The Florida rate? The Iowa rate? It would be difficult to assemble a list of jurisdictions that would command wide agreement. Indeed, any such construct, no matter how successfully applied, would be sure to garner criticism as ad hoc and inadequate.
Having raised the problem of omitted variables and made their suggestions as to what was lacking in the studies, the authors of the FPI report sought to replicate the results of the Haughwout study using the same data set. They report their exercise confirmed the finding of the Haughwout study that the tax variables are statistically significant using data for 1970-1997 – but not if the sample period is confined to 1978-1997, and not if the variables that measure government transfers and the Dow Jones Index are excluded.
In short they find that if they alter the model by truncating the data set or by omitting variables of their choosing, the model is not robust. This is not a surprise. It is an accepted fact that the statistical significance of any econometric model will be substantially altered if that model is re-estimated in such an altered fashion. The loss of statistical observations that would result from such alteration is to be avoided.
Shortening the data set and removing selected variables is curiously inconsistent with FPI's concern about “omitted variables” in the Haughwout and Manhattan Institute studies. After arguing that the studies under consideration should include more variables and consider a longer time period, it is unpersuasive to say that there are no tax effects on employment because the models do not hold up when re-estimated with a shorter time period or with certain variables omitted. The FPI analysts seem to be arguing simultaneously that the Haughwout study is based on insufficient data – and that it can be disproven if based on less data.
It would have been more interesting to see a re-estimated version of the Haughwout et al. model that included some of the other variables, such as wage or unionization rates, that the authors of the FPI report consider to be important. Unfortunately, despite the relative ease of construction of the requisite data sets, such an exercise was not offered.
As it happens, the NYC-STAMP model is based on data for 1975-1999 – closer to the time period to which FPI wishes to confine the Haughwout study. Yet, as reported by the Manhattan Institute, NYC-STAMP does find a statistically significant tax effect on employment during this period. Truncating the NYC-STAMP data set on terms more agreeable to FPI might alter this finding as with the manipulation of the Haughwout study; but as before, such manipulation would be inappropriate by econometric standards.
The final section of the FPI report argues that the drop in employment in New York City that occurred in 1989-92 was due largely to the aftereffects of the stock market crash of 1987. Banks and security firms tightened their belts, which hit the city particularly hard. As cited in the FPI report, “between 1988 and 1991, Wall Street employment fell 16 percent and real earnings dropped 12 percent (p.11).”
In fact, the NYC-STAMP estimates are based on data for six sectors, one of which is Finance, Insurance and Real Estate (FIRE). The model's wage and employment equations include a measure of national wage rates, which pick up the effects of sectoral shocks such as the shakeout in the banking and securities industry after 1987. The STAMP model found significant tax effects even with the inclusion of this variable. In other words, the FIRE sector did contribute significantly to the city's job losses. But so did the city's own income tax hikes – which were imposed in advance of and not merely “in response to” the city's biggest employment losses, as inaccurately reported by FPI in the report.
In short, the FPI report does not provide a serious criticism of either the NYC-STAMP model or the research by Haughwout et al. State and local taxes do matter.
-30-
[1] [1] Andrew Haughwout, Robert Inman, Steven Craig and Thomas Luce, “Local Revenue Hills: A General Equilibrium Specification with Evidence from Four U.S. Cities,” NBER Working Paper 7603.
[2]The study and the NYC-STAMP on which it is based were prepared by the Beacon Hill Institute at Suffolk University in Boston under contract and in consultation with the Manhattan Institute.
[3] In any event, it would probably be inappropriate to include both relative wage rates in the South and the relative unionization rates, because the two are almost certainly correlated. That said, the NYC-STAMP model does control for wage rates (by sector) elsewhere in the U.S., although not specifically those prevailing in the South.
[4] In constructing its STAMP model for the state of Virginia, BHI experimented with a specification that included a weighted average of sales tax rates in border states, in addition to the sales tax rate in Virginia itself. However, this variable did not have a statistically significant effect and did not materially alter the other coefficients in the equation. Changes in the sales tax within Virginia dominated the effects of changes in the sales tax among border states.