The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks

The paper’s framework shows that because tax changes are taken for many different reasons – some to offset other shocks already affecting output, some for reasons unrelated to current conditions – a regression of output growth on an aggregate tax-change measure is contaminated because “some tax changes are correlated with the error term in this regression,” and the authors argue the bias is “likely to be even larger” for broader measures such as cyclically adjusted revenues, which “include many non-policy movements that may be correlated with other developments affecting output.” The paper also argues that simply adding control variables to such a regression is unlikely to solve the problem, since “it is impossible to proxy for all the information about future output movements that policymakers may have had.”

The paper classifies every significant legislated federal tax change from 1945 to 2007 by its predominant stated motivation, drawn from sources including the Economic Report of the President, presidential speeches, and Congressional committee reports, sorting changes into countercyclical actions, spending-driven actions, deficit-driven actions, and actions taken for long-run growth; only the latter two categories are kept as the paper’s measure of “exogenous” tax changes, since the first two are “clearly actions that are correlated with other developments affecting output.” The paper is explicit that “exogenous” here is a convenience label rather than a strict econometric claim, noting in a footnote that “an equally appropriate terminology would be ‘valid’ and ‘invalid.’”

The narrative analysis identifies 54 quarterly exogenous tax changes, with a mean of -0.03 percent of GDP and a standard deviation of 0.24 percentage points, essentially no serial correlation (Q-statistic p-value of 0.996), and a distribution that is “fairly evenly distributed across the postwar era.” Comparing this series with all legislated tax changes (exogenous plus endogenous), the paper reports that “the two series are nearly identical after 1975” – since countercyclical and spending-driven tax changes were rare after that date – but “wildly different” in the first three postwar decades, with the standard deviation of all legislated changes roughly 50 percent larger than that of the exogenous series alone.

The paper reports that real GDP does not Granger-cause the exogenous tax series, with a p-value of 0.78 for the test that lagged GDP movements predict the tax series, which the authors call “reassuring confirmation that the tax shocks identified from narrative sources are indeed unrelated to past output movements.” Separately, when the two component subcategories are tested, long-run tax changes also show no Granger-causality from output (p = 0.45), while deficit-driven tax changes show only marginal evidence (p = 0.05) that they are “slightly negatively related to lagged GDP changes,” which the paper attributes to low growth periods tending to produce the persistent deficits that give rise to deficit-driven tax increases, rather than to disguised countercyclical motivation.

Using a simple regression of output growth on current and lagged values of the exogenous tax series, the paper estimates that a tax increase equal to 1 percent of GDP produces “a fall in output of 3.08% after ten quarters” (t = -3.53), an effect the paper describes as “very large, sustained, and highly significant.” Adding lagged GDP growth as a control changes this estimate only trivially, to -3.02 percent (t = -2.90), and a two-variable VAR specification yields a very similar maximum effect of -2.93 percent (t = -2.80) after ten quarters, which the paper takes as evidence that “the most basic test for the possibility that what appear as effects of tax changes actually reflect the normal dynamics of the economy provides no support for this view.”

Using the change in cyclically adjusted revenues in the same VAR specification, the paper obtains a maximum output effect of -1.26 percent (t = -2.14), and using all legislated tax changes, -1.86 percent (t = -2.74) – both “noticeably slower and smaller” than the -2.93 percent estimate obtained with the exogenous series. The paper attributes “about two-thirds” of the resulting bias in the cyclically-adjusted-revenue measure to the inclusion of endogenous legislated tax changes, and “about one-third” to the inclusion of non-policy, non-cyclical revenue movements (such as those driven by stock-market fluctuations) that are correlated with future output.

No: the paper reports that a long-run-growth-motivated tax increase of 1 percent of GDP produces an estimated maximum decline in GDP of 2.99 percent (t = -2.92), “virtually identical” to the effect of a generic exogenous tax increase, whereas the point estimates for deficit-driven tax increases are “consistently positive,” with a maximum estimated effect of +2.48 percent (t = 1.03). The paper is careful to caveat this second result, noting that “there are too few tax changes of this type for the effects to be estimated precisely” and that “one should be very cautious in reading anything into such imprecise estimates,” while still describing the results as “suggestive that tax increases to reduce an inherited deficit may be less costly than other tax increases.”

Comparing a measure of tax changes dated at implementation against a present-value measure of news about future tax liabilities dated at the time a bill is passed, the paper finds that the implementation-dated measure’s estimated maximum effect (-3.50 percent, t = -2.22) is “very similar to that for the same specification not including the present-value measure,” while the news-dated measure’s effect is “initially small, negative, and marginally significant” before turning “consistently positive, though… not significantly different from zero,” with a maximum impact of only 1.10 percent (t = 1.17). The paper concludes from this that “output responds to tax changes when they are implemented” and that “the news effects of tax changes may be of secondary importance,” while noting that the positive sign on the later news coefficients is more consistent with an intertemporal-substitution or confidence channel than with the permanent-income hypothesis.

Splitting the sample at the end of 1980, the paper estimates a maximum output decline of 4.29 percent after seven quarters for 1950Q1-1980Q4, versus 3.08 percent after eight quarters for 1981Q1-2007Q4, with the early-period effect also emerging considerably faster (-3.69 percent after one year, versus -0.81 percent in the later period); however, the paper cautions that “the standard errors are large enough that the evidence of a change over time is only modest,” since “the two-standard-error confidence interval for the response for the early period easily includes the point estimates for the later period.” The paper reports that the federal funds rate falls significantly in response to a tax increase in the post-1980 period but not in the earlier period, offering this as “mild support” for the view that more aggressive monetary accommodation of fiscal tightening may be one (unproven) reason the measured output effect has become smaller and slower.

The paper reports that in response to a 1-percent-of-GDP tax increase, “the fall in investment is much larger than the fall in consumption”: the maximum fall in personal consumption expenditure is 2.55 percent (t = -3.06), while the maximum fall in gross private domestic investment is 11.19 percent (t = -3.35), with nonresidential and residential fixed investment separately falling by a maximum of 8.50 percent and 5.84 percent. The paper argues that because conventional models predict a tax increase should lower interest rates and thereby support investment, “the fact that investment falls so strongly… suggests that conventional interest rate effects are not key,” and instead points to the dependence of investment on cash flow and overall economic conditions documented elsewhere in the literature; it also rules out a direct investment-incentive channel, since excluding tax changes that coincided with major changes in investment-tax-incentive rules “in fact strengthens the estimated investment response.”

Within consumption, the paper finds that “expenditures on durables fall substantially more” than nondurables or services – a maximum decline of 8.27 percent (t = -2.86) for durables against 2.27 percent (t = -3.12) for nondurables and about 1.3 to 1.4 percent for services – which the paper attributes to the general cyclical volatility of durables purchases outweighing any offsetting interest-rate effect. For trade, the paper reports that imports fall sharply and significantly (maximum -10.07 percent, t = -3.62) while exports rise, though imprecisely (maximum +4.98 percent, t = 2.08); the paper notes the implied improvement in net exports is “consistent with the tax increase lowering interest rates and hence reducing capital inflows,” but argues that because the import response is so much stronger, “the fall in income may be more important than the interest rate/exchange rate linkage.”

The paper states plainly that its results “are largely silent concerning whether the output effects operate through incentives and supply behavior or through disposable income and demand stimulus,” noting that the persistence of the effects is “suggestive of supply effects” while their speed is “suggestive of demand effects,” and that other studies have found highly persistent output effects from monetary policy, which “necessarily works through demand.” The paper is equally explicit that its results “do not speak to the issue of whether taxes are a more powerful tool of fiscal policy than government purchases,” since conventional estimates of the effects of government purchases “almost surely suffer from omitted variable bias” of their own, and exogenous spending changes of the kind identified by Ramey and Shapiro (1998) “are usually accompanied by substantial changes in taxes,” so they cannot cleanly isolate the effect of purchases alone for comparison.