How reliably do empirical tests identify tax avoidance?

How reliably do empirical tests identify tax avoidance?

By Lisa De Simone, Jordan Nickerson, Jeri K. Seidman, Bridget Stomberg
June 27,2017Working Paper No. 3446

Can common empirical tests reliably identify tax avoidance? This is an important question because our understanding of the determinants of tax avoidance largely depends on results generated using such tests. We address this question by using a controlled environment to examine the effectiveness of empirical tests that use effective tax rates (ETR) and book-tax differences (BTD) as tax avoidance proxies. We seed Compustat data with three tax avoidance strategies and examine how reliably empirical tests identify this incremental simulated tax avoidance, all else equal. We find that power varies with the proxy and the type of tax avoidance. Thus, we offer guidance to researchers in matching specific types of tax avoidance with the most powerful proxy to detect it. We further offer evidence on how research design choices affect power. Results suggest researchers can increase power by eliminating observations with both negative pre-tax book income and negative tax expense, and by using robust regression to address data outliers. In contrast, power is impaired when truncating ETR proxies and when using Execucomp data. We also provide evidence that tests have less power to detect tax avoidance when multi-year ETR proxies are used.