Petro Lisowsky (Boston University, Department of Accounting) has posted "Seeking Shelter": Empirically Modeling Tax Shelters and Examining Their Link to the Contingent Tax Liability Reserve on SSRN.  Here is the abstract:

Using confidential tax shelter data obtained from the IRS, this paper develops an empirically valid model for inferring the likelihood that a firm engages in a tax shelter. The Treasury (1999) white paper on tax shelters is used as a conceptual guide in developing publicly available financial statement proxies for the characteristics of tax shelters. Results show that tax shelter likelihood is positively related to the presence of subsidiaries located in tax havens, material foreign operations, prior-year reported effective tax rates, financial complexity, litigation losses, and use of promoters. Validation tests on out-of-sample tax shelter observations indicate the likelihood model can be used generally by researchers, investors, and tax administrators to infer tax shelter likelihood. The use of publicly available information as inputs is central to the model’s usefulness.

This paper also reports the first clear empirical link between the contingent tax liability reserve, or tax cushion, and tax shelters. Prior research finds general evidence that the tax cushion is increasing in IRS audit adjustments, suggesting the tax cushion is increasing in tax risk. Tax shelters have been suspected as a reason for this association, but until now no clear empirical link had been established. Results also show that the tax cushion is positively related to financial earnings management. Taken together, the findings suggest that the tax cushion may be subject to both tax and financial reporting pressures.