Do tax cuts
grow the
economy?

Backlog #29: the supply-side claim that cutting the top income tax rate spurs faster economic growth, tested in its plainest form — the top marginal individual income tax rate for every year since the annual real-GDP-growth series begins (Tax Foundation's own historical bracket table, sourced row-by-row to IRS revenue procedures, against BEA's real GDP annual percent change, FRED's keyless CSV endpoint), 1930-2025, n=96 years.

The naive full-sample correlation clears zero — running backward from the claim. Regressing real GDP growth on the top marginal rate: +0.050 points of growth per point of top rate, 95% CI [+0.006, +0.094], p=0.025, R²=0.061 — excludes zero, and the sign is positive: higher top rates, in the raw 96-year record, sit alongside faster growth, not slower. Dropping the six WWII mobilization/demobilization years (1941-46, growth swinging from −11.6% to +18.9%, alongside the highest top rates in US history, 88-94%) barely moves it: +0.040, CI [+0.007, +0.073], p=0.016. Giving a rate change a year to show up in growth, the timing the “tax cuts spur growth” story actually implies, tells the same story: +0.047, CI [+0.001, +0.093], p=0.043.

Splitting the sample at 1981 is where the naive number falls apart. Every top-rate cut anyone actually points to as evidence — 1981-86, 2001-03, 2017-18 — happened after 1981; before that, the top rate spent five decades between 25% and 94% while the country fought a war, rebuilt after it, and only started cutting rates once. Fit the two eras separately: pre-1981 (n=51) alone is even more positive than the full sample, +0.127, CI [+0.010, +0.244], p=0.033 — still excludes zero, still backward. 1981-2025 (n=45), the era that actually contains the cuts, is a clean null: +0.034, CI [-0.068, +0.137], p=0.51 — contains zero, in either direction. Dropping 2008-09 and 2020, the two years a financial crisis and a pandemic swamped everything else in the growth number, pushes it closer to zero still: +0.016, CI [-0.076, +0.109], p=0.73.

Honestly, the formal test for whether the slope itself changed between eras does not clear zero either. A rate×era interaction term: -0.093, 95% CI [-0.249, +0.062], p=0.24 — contains zero, so the desk cannot formally certify that the pre- and post-1981 slopes are statistically different from each other, only that each one, read on its own, tells a different story: one excludes zero (backward), the other doesn't exclude anything.

Read plainly. Ninety-six years of the naive version of this claim run in the opposite direction from the theory, not merely failing to support it — and that shape survives dropping the six most distorted wartime years and giving the rate a year to act. But the wrong-direction result is doing almost all of its work in the 51 years before 1981, an era with essentially no rate-cutting to test in the first place. Restrict the test to the 45 years that actually contain every real-world experiment in cutting the top rate, and the interval says nothing: not that cuts help, not that they hurt, not enough signal in one country's macro history to say which. This is one country's top bracket against one aggregate growth number — it cannot separate the top rate's effect from everything else that changed across 45 years of one economy (monetary policy, globalization, two wars, a financial crisis, a pandemic), and a genuine causal test would need cross-country or cross-state variation this design doesn't have.