The elasticity of taxable income during the 1990s: new estimates and sensitivity analyses.

• Author: Giertz, Seth H.

1. Introduction

   Economists have long recognized that taxation creates economic inefficiency by distorting relative prices, often between leisure and all other goods in the economy. Even a broad-based income tax can have substantial efficiency costs, so long as leisure remains untaxed. Harberger (1964) uses this as motivation for comparing the efficiency implications of direct versus indirect taxation. In so doing, he shows how labor supply elasticities can be used to measure the efficiency implications of income taxation. (1) Harberger's analysis had a profound influence on public and labor economists, spurring increased research into labor supply elasticities, which were seen as proxies for the efficiency costs from taxation.

   More than two decades after Harberger, economists began to emphasize the variety of other margins over which taxes can distort behavior, in addition to hours worked. These economists noted that these other behavioral responses may too have important implications for efficiency. An important building block in the evolution of this literature was set forth by Slemrod (1990), who presented a hierarchy of behavioral responses to taxation. Slemrod divided behavior into three categories based on their responsiveness to taxes. His least responsive category, real responses, includes individuals changing consumption and work patterns and businesses altering investment and production decisions. Second, in order of responsiveness is circumvention, which includes both illegally (evasion) and legally (avoidance) bypassing the tax system. (2) In the case of evasion, income is concealed from tax authorities--or factors that offset tax liabilities, such as expenses and deductions, are inflated (beyond what is legally permitted). In the case of avoidance, income is shifted in order to receive more favorable tax treatment. Diverting income into a tax-deferred retirement account is an example of avoidance. (3) Included in circumvention is the expenditure (of time and money) necessary to avoid or evade taxation. Third on the hierarchy--and the most responsive--is the timing of income receipt, which includes "pulling" the receipt of income into the previous year or "pushing" it into the next in order to take advantage of more favorable tax treatment (while not altering real behavior). (4) Slemrod further recognizes that the type of behavioral response could have implications for both tax revenue and tax incidence and in some cases efficiency.

   Predating Slemrod's insights, Lindsey (1987) produced elasticity estimates for taxable income--although what Lindsey termed taxable income more closely approximates adjusted gross income (AGI) instead of labor supply. (5) Lindsey's estimated elasticities were much larger than the labor supply literature would have suggested. Lindsey emphasized the revenue implications of the elasticity of taxable income (ETI) but not its efficiency implications.

   Building on Slemrod and Lindsey, Feldstein (1995) produced ETI estimates but went much further than Lindsey in describing the behaviors that could affect taxable income. He argued that many of these behaviors were not captured by labor supply elasticities. Furthermore, Feldstein posited that (income) taxation creates economic inefficiency not only by distorting the relative price between labor and leisure but more broadly by distorting the relative price between goods or activities that are taxed and those that are not taxed since leisure is not the only untaxed activity. For example, in response to taxes, not only work hours but also work effort might change. Compensation can shift between taxed forms and untaxed forms. When tax rates are higher, more compensation is paid in tax-exempt fringe benefits instead of wages. And economic activity may shift from jurisdictions with more burdensome taxes to others where taxes are more favorable; evasion is another response to taxation that confers deadweight loss but does not imply increased leisure. Over the long run, taxes may also distort investment decisions, including how much education to pursue and in what occupations to specialize.

   Next, Feldstein (1999) showed that the ETI along with information on marginal tax rates and income is all that is necessary to calculate changes in both tax revenue and efficiency. (6) In Harberger's model, labor is the only source of income, all income is taxed when earned, and thus taxable income equals labor income; therefore, the ETI with respect to the tax rate is the same as the labor supply elasticity--or at least the elasticity of labor earnings since labor hours and labor earnings may be imperfectly correlated because of factors such as work effort.

   The beauty of Feldstein's (1995, 1999) papers is that it is not necessary to separately measure all the underlying behaviors that influence taxable income. All these behaviors have efficiency implications, and, in equilibrium, the costs associated with a $1.00 change to taxable income, at the margin, are always equal (to the marginal tax rate). However, more recent research highlights instances where the manner in which taxpayers respond could have important (efficiency and revenue) implications. And understanding the anatomy of behavioral responses is important for tax policy design. For example, on efficiency grounds, one may want to tax more heavily income sources that are least responsive to taxation--just as Ramsey (1927) suggests taxing goods proportionally to the inverse of their demand elasticities.

   While Feldstein (1995) spawned dozens of subsequent papers and the ETI is now a central parameter in tax policy discussions, there is no consensus as to the size of the ETI. (7) The earliest studies (Lindsey 1987; Feldstein 1995) suggested an ETI ranging between 1 and 3, although, as discussed later, most now believe that these early estimates are substantially biased upward. While the most widely cited ETI estimates are now much lower, often around 0.4, credible estimates range from 0 to 1--and some studies report estimates in excess of 1 for very high income groups. Where the ETI lies within this wide range of estimates has tremendous consequences for the efficiency and revenue implications from changes to tax rates. In fact, in examining the major components of the "Bush" tax cuts, Giertz (2009) uses a range of ETI estimates from the literature to show how allowing the individual income tax rate cuts to expire--scheduled to occur after 2010--might affect economic efficiency and tax revenues. Giertz finds the deadweight loss per additional dollar of tax revenue from the increase in federal income tax rates to range from $0.19 at an ETI of 0.2 to $1.72 at an ETI of 0.8--that is, an ETI of 0.8 implies nine times the deadweight loss (per dollar of additional federal revenue) than does an ETI of 0.2. Note that at an ETI of 0, this measure equals 0 (i.e., changes to tax rates have no efficiency implications), and once the revenue-maximizing (or Laffer) tax rate is reached, this measure equals infinity. Giertz estimates that an ETI of 0.8 implies a total revenue offset of 63.3% versus 15.8% at an ETI of 0.2 (both compared to the potential gain in federal revenues with no behavioral responses). These estimates include offsets to payroll and state income taxes (whose bases overlap with the federal income tax base).

   The majority of research in the U.S. taxable income elasticity literature has focused on the 1980s tax cuts. However, research that does exist on the 1990s suggests that responses for the two decades may be dissimilar. Giertz (2007) reports estimated elasticities for the 1990s that are over one-third smaller than analogous estimates for the 1980s. Additionally, work by Slemrod (1996), Goolsbee (1999), and Saez (2004) all suggests great variation in the ETI across time periods. However, Chetty (2009a) shows that variation in estimates across tax changes does not necessarily imply that the true ETI is changing over time but could instead reflect costs of reoptimizing. If taxpayers imperfectly optimize, Chetty shows that a range of ETI estimates could all be consistent with a single true ETI.

   This article revisits the 1990s tax increases using panels of Internal Revenue Service (IRS) tax return data for years 1984, 1985, and 1988-1995. These data heavily oversample high-income filers, which research suggests play a critical role in determining overall responses to changes in tax rates (Navratil 1995; Moffitt and Wilhelm 2000; Gruber and Saez 2002; Saez 2004). I find that when employing the richest set of income controls, estimates are larger for the longer term than for the short term. (8) As is common in this literature, sample income cutoffs (or floors) are imposed in order to lessen influences from mean reversion on the estimates. When imposing a $50,000 income cutoff, the income-weighted estimated ETI is 0.33 when measuring responses over three-year intervals compared to 0.19 when measuring behavior over single-year intervals. Including separate controls for both mean reversion and divergence in the income distribution substantially alters estimated ETIs over the longer term but not for the short term. Employing the specification favored by Gruber and Saez (2002) yields a negative estimated ETI when employing single-year differences; when differencing over three-year intervals, Gruber and Saez's specification yields an ETI estimate of 0.54--the largest ETI estimate (from the specifications that include just one tax rate variable among the explanatory variables). An alternative approach that includes lead and lag tax rate changes produces a short-term elasticity estimate of 0.43 and a longer-term estimate ranging from 0.78 to 1.46. Income weighting is important. Unweighted estimates are much smaller than income-weighted estimates and suggest at most a very modest ETI. This also suggests that the ETI increases with income-consistent with several other articles in the ETI...