A previous article in this journal, “Optimal Tax Rates: A Review and Critique” (Reynolds 2019), analyzed such U.S. postwar ETI estimates that were being misconstrued as recommendations for a 73–83 percent optimal top tax rate for the federal income tax alone. I surveyed evidence and arguments suggesting that even if top tax rates designed to maximize short‐term revenue might be “socially optimal” in some sense, such high marginal rates would not prove to be economically optimal in terms of the incentive effects on sources of longer‐term expansion of the economy and the tax base. As Goolsbee (1999: 38) rightly emphasized, “The fact that efficiency costs rise with the square of the tax rate is likely to make the optimal rate well below the revenue maximizing rate.”

The early paper by Goolsbee included estimates of the ETI in the 1920s and 1930s. Together with another paper about the ETI during those years, by Romer and Romer (2014), the prewar studies came to nearly the same conclusion as their postwar counterparts did—namely, that hypothetical top tax rates of 74–83 percent could have maximized federal tax revenues during the Great Depression. Unlike Diamond and Saez (2011), however, the prewar studies excluded state and local taxes (which were much larger than federal taxes) and major new federal taxes on payrolls and sales added in 1932–37.

Romer and Romer (2014: 269) use an average of ETI estimates for federal income tax changes from 1918 to 1941 to conclude that “our estimated elasticity of 0.21 implies an optimal top marginal rate of 74 percent” (2014: 269). Yet their estimated elasticity is twice that high for 1932 (0.42) when the top marginal rate was raised from 25 percent to 63 percent. And their elasticity coefficients for major tax changes in 1934–38, they acknowledge, “cannot be estimated with any useful degree of precision” (2014: 266).

Goolsbee (1999: 36) compares 1931 and 1935 to judge how high‐income taxpayers responded to much higher tax rates in 1932 (and higher still in 1934). He concludes the ETI at high incomes was so low that “if there were only one rate in the tax code, the revenue maximizing tax rate given the [low] elasticity estimated … [would be] 83 percent using the using 1931 to 1935 data.”

When discussing a smaller 1936 rate increase, confined to incomes above $50,000, Goolsbee concludes: “Technically, the revenue‐maximizing [single tax] rate would be at the maximum of 100 percent using 1934–38 data, since the elasticity was negative” (ibid). A study of postwar data by Piketty, Saez, and Stantcheva (2014: 252) likewise theorized that “the optimal top tax rate … actually goes to 100 percent if the real supply‐side elasticity is very small.” My review of that paper (Reynolds 2019: 250–54) found their estimated ETI and Pareto parameters to be far below consensus estimates for high incomes and inapplicable to untested tax rates of 83–100 percent. This review of similar prewar studies also finds their ETI estimates implausibly low and the alleged revenue‐maximizing tax rates of 74–83 percent too high.

Romer and Romer (2014) are incorrect in claiming that tax responsiveness was low in the 1920s and 1930s, and Goolsbee is incorrect in making that same claim about just the 1930s. Their erroneous low response estimates lead them to conclude that high tax rates are a good thing. This study finds, instead, that high income taxpayers were very responsive to lower marginal tax rates in the 1920s and higher marginal tax rates in the 1930s.

I find that large reductions in marginal tax rates on incomes above $50,000 in the 1920s were always matched by large increases in the amount of high income reported and taxed. Large increases in marginal tax rates on incomes above $50,000 in the 1930s were almost always matched by large reductions in the amount of high income reported and taxed, with a brief exception connected with the 1937–38 recession, which is investigated in detail.