Using Accounting Information to Predict Aggressive Tax Placement Decisions by European Groups

Abstract

Aggressive tax planning (ATP) consists in taxpayers’ reducing their tax liability through arrangements that may be legal but are in contradiction with the intent of the law. In particular, ATP by multinational groups (MNE) is a source of major concern. In this paper we consider the MNE’s decision to locate or to maintain a company in a tax haven as a relevant symptom of ATP. The research question we want to address is whether this decision can be predicted using publicly available accounting information. We use the ORBIS database and we focus on European MNEs. We observe that, in 2021, slightly less than 40% of European MNEs have a company located in a tax haven. Thus, for a tax authority, it would be difficult, without a specific analysis, to identify riskier MNEs. We find that a random forest model that uses accounting information for years between 2015 and 2019 predicts reasonably well the decision to locate (or maintain) a company in a tax haven in 2021. Using this model in 2019, a tax authority could have identified almost 80% of European MNEs that were going to locate or maintain a company in a tax haven in 2021. We observe that the most important variables for prediction are those associated with the size of the group, its positive profitability, and its financial structure, while individual time-invariant features are less relevant. We also find that the predictive performance of the model is maximized when the information is taken from the time subset 2017-2019 and that most important predictors for the risk of using tax havens are also good predictors for the level of intensity of such a use, as measured by the share of subsidiaries located in tax havens. The main policy implication of these results is that (European) non-tax havens could effectively anticipate (and prevent) the decision to locate (or maintain) companies in tax havens, and shape their policies accordingly, with particular reference to cooperative compliance schemes. These policies are more credible in the context of renewed international cooperation in the design of corporate tax rules, and in particular, of the implementation of Pillar Two within the European Union.

Keywords: Aggressive Tax Planning, European Multinationals, Machine Learning