A working academic paper by an accounting professor and a PhD student at the University of North Carolina at Chapel Hill’s Kenan-Flagler Business School is generating a lot of political interest in Australia.
Kevin S. Markle, who will soon graduate from UNC’s PhD program, and Douglas A. Shackelford, the Meade H. Willis Distinguished Professor of Taxation at UNC Kenan-Flagler, are authors of the working paper, “Do Multinationals or Domestic Firms Face Higher Effective Tax Rates?” A June 2009 draft of the paper is posted on the National Bureau of Economic Research web site at http://www.nber.org/papers/w15091.
Members of the media contacted Shackelford and Markle after an Australian Treasury review into the nation’s tax system cited figures from their academic paper during a policy debate.
The paper's usefulness in formulating policy for one sector in one country should not be overstated, said Shackelford in response to one reporter’s query. “The purpose of the paper is to collect all data about the corporate income taxes paid by all publicly traded firms worldwide and thus provide at a starting point for comparing one country with another.”
“Our research has been pulled into a policy debate currently unfolding in Australia,” said Markle. “We have read accounts of how by both sides of the debate have referenced our paper and members of the media have contacted us asking for our reactions and comments.”
“We stand behind our research,” Shackelford said. “However, it appears to us that a tiny element of a very large study has been taken out of context and that proper consideration has not been given to the methods used to derive the number, the intent of the study, and the limitations of the findings for informing specific policy debates.”
To help resolve misunderstandings, the authors provided the following comments.
- The purpose of our paper is not to study specific industries in specific countries. Nor is it to precisely calculate rates of tax that are paid. Our paper is intended as a broad comparison of effective tax rates across countries. All numbers in the tables in the paper are appropriately interpreted on a relative – rather than absolute – basis.
- The version of the paper cited is a June 2009 draft posted on the National Bureau of Economic Research (NBER) website. This draft is a working paper and has been through no peer review. Posting on the NBER website is not equivalent to publication; it is a venue for in-process research to be circulated for the purpose of sharing ideas and garnering feedback. As with other academic papers on the NBER site, we are in the midst of additional revision before submitting it for publication in a peer-reviewed journal.
- The entire controversy appears to center around the numbers reported in the “Mining” column of Table 4 of our paper. That table reports the results of our estimations of our regression equation on observations grouped by 2-digit NAICS codes and is included in their paper as supplemental analysis to determine whether the cross-country differences we observe when all industries are grouped together exist at the industry level. The data are for 2003-2007. We make an arbitrary cut-off for reporting of 20 observations. Thus, it is possible that the numbers for Australia represent average numbers for as few as four companies over the five years. As such, we reach no conclusions nor make any comments about individual industries in individual countries. Our purpose in producing the table was to make relative comparisons only.
- The data used in our study are from the publicly available financial statements of firms. We do not have data from tax returns or any other proprietary data. In the June 2009 draft, the “tax rate” used was calculated as total tax expense divided by pre-tax income, both taken from the financial statements.
- Our study is of corporate income tax rates only. We do not consider royalties, fees, value-added tax or anything else that does not get included in the total tax expense line on the financial statements. This is a limitation of our study which is due to data availability and we acknowledge it in the paper.
- In the most recent draft of the paper (March 2010), we changed our data source because it enables us to address the questions of our study more effectively. In this version, we do not have sufficient observations to include a number for the mining industry in Australia. This is simply a function of the coverage of the two data providers. The underlying data used in the June 2009 are all publicly available. Anyone interested in replicating the results in that draft would have access to all of the data that we used.
- We have read the analysis of Professor Sinclair Davison posted at http://catallaxyfiles.com/ and do not disagree with his conclusions.
- It appears that people have assumed that the paper is authored by Mr. Markle because his name is listed first. That is incorrect. In our field, the default convention for co-authored work is for names to be listed alphabetically.
- This final point is technical in nature, but since it has been raised multiple times, we include it here.
To derive the numbers that are causing the stir, here is a description of what we did. We will simplify it to assume there are just three countries in the study: Australia, Canada and the United States.
- Go to the database and collect all firms that have the needed financial statement information (total tax expense, pretax income, ETR between 0 and 70%) and have 21 as the first two digits of their NAICS number.
- We then run an ordinary least squares regression on a model that has ETR = f(AUSTRALIA, CANADA, US, year controls, size controls). (We are ignoring the domestic/multinational split, which is just an additional split but uses identical intuition.) The estimate of the coefficient on AUSTRALIA (which is simply an indicator variable =1 for all Australian observations) is reported as the “Australian ETR” (the 17% that folks are focused on).
- Note that a regression framework is used only because it allows us to test for statistical significance of coefficients and to control for factors such as size. The estimates that come out are really just conditional averages.
If the goal were to analyze the real taxes paid by the mining sector in Australia, our approach would serve, at best, as a very preliminary and rough analysis. That is not a goal of our paper.
Even when one fully grasps what is being captured by our methodology, there are several possible sources of error in our measurement that we are forced to accept. The simplest example is an error in the data – the firm reports 1,000 but the database entry is 100. The greatest concern in our context is that the database contains a zero tax expense when the correct number is something positive. We attempt to mitigate concerns by dropping outlier ETRs, but this is not a perfect control.
These types of issues become much more important as the sample becomes smaller. When we are using our full sample, we have some assurance that the effect of such errors will not affect inferences. Down at the industry level, however, where the sample is much smaller, the threat is much larger.
If we had intended our paper to provide definitive answers at the country-industry level, we would have tested the robustness of those results much more thoroughly.