A Reconsideration of Investment Behavior Using Tax Reforms as Natural Experiments

Cummins, Hassett, and Hubbard argue that standard tests of neoclassical investment models (tax-adjusted q and the user cost of capital) have performed poorly mainly because of severe measurement error in the structural variables, not because investment is genuinely insensitive to the net return to investing. Using firm-level Compustat panel data from 1962-1988, they exploit 13 corporate tax-code changes – and especially four “major” reforms (1962, 1971, 1981, 1986) identified by a large, unanticipated, and short-lived shift in the after-tax cost of investing – as natural experiments: because firms learn major tax parameters with near-certainty once a reform is enacted, isolating the resulting cross-sectional, tax-driven variation in q or the user cost of capital sharply reduces the expectational measurement error that plagues conventional estimates. The resulting coefficient on tax-adjusted q jumps roughly an order of magnitude relative to conventional estimates – from about 0.025 in a standard fixed-effects regression to between 0.47 and 0.87 in the years immediately following each major reform, with no significant effect in non-reform years – implying far more economically plausible capital-adjustment costs (roughly 5 to 12 cents per extra dollar of investment, instead of the one-to-five-dollar estimates common in the literature). A parallel user-cost-of-capital specification yields an average elasticity of investment to the user cost of about -0.66, which the paper uses to predict that the 1962 investment tax credit should have raised the aggregate equipment investment-capital ratio by about 7 percent (versus an actual rise of about 11 percent) – evidence that tax policy is a real, non-trivial, but partial determinant of investment, since the predicted tax effect is only about a quarter of the overall standard deviation of investment in their sample.

Kevin A. Hassett | Macro Paper Warehouse

A Reconsideration of Investment Behavior Using Tax Reforms as Natural Experiments