Regulatory avoidance responses to private Country-by-Country Reporting

Abstract

This paper investigates regulatory avoidance in the context of private country-by-country reporting (CbCR) introduced as part of the OECD/G20 BEPS initiative. The reporting framework requires multinational companies above a revenue threshold to provide tax authorities with new and detailed information on their global activities, but the data are not made publicly available. I find robust evidence for an increase in mass below the revenue threshold after the introduction of CbCR in line with an avoidance response. Company types for which CbCR would imply relatively high costs including private companies or more tax-aggressive firms show a stronger avoidance response. The heterogeneities found can at least partially be explained by an analysis of increases in tax costs. The finding of regulatory avoidance of multinational enterprises in response to a fixed revenue threshold is of additional relevance in light of the international tax reform agenda which relies on similar thresholds.

Appendices

Appendix 1

See Figs. 8, 9, 10, 11, 12, 13 and Tables 2, 3, 4.

Fig. 8

Template for Table 1 of a CbC report. Notes: This template is taken from OECD (2015), p. 29

Fig. 9

Template for Table 2 of a CbC report. Notes: This template is taken from OECD (2015), p. 30

Fig. 10

Template for Table 3 of a CbC report. Notes: This template is taken from OECD (2015) p. 30

Fig. 11

Visualization of the CbCR process. Notes: Example for the CbCR mechanism based on a parent entity filing obligation for an MNE with ultimate parent entity (UPE) in country B and subsidiaries in countries A and C. In this example, countries A to C are all part of the exchange mechanism. The CbCR process comprises four main steps: (1) The UPE collects all data required for the report, including information from all its subsidiaries. The financial data are aggregated at the country level if an MNE has more than one entity in a country. (2) The UPE files the CbC report with the tax authority in its home country (country B). (3) The tax authority of country B shares the report with all countries in which the MNE is active and that are part of the CbC exchange mechanism (A and C). (4) Tax authorities in countries A, B, and C can use the CbC information to assess the tax risk of the reporting company. While the reports are shared between tax authorities, they are not made publicly available

Fig. 12

Excess mass for different bunching regions. Notes: Excess mass under assumption of different bunching regions. Calculations are based on Eq. (1). The bars indicate the percentage difference between the average density in the pre-reform period of 2010–2014 and the average density in the post-reform period of 2016–2018

Fig. 13

Definition of the treatment and control group. Notes: Definition of the treatment and control group for the DiD estimations. The assignment is based on consolidated group revenue relative to the threshold applicable and the number of cross-border subsidiaries. Axes are not to scale

Table 2 Summary statistics for balanced sample

Table 3 Number of MNEs by threshold applicable

Table 4 Revenue growth of firms in the bunching region

Appendix 2

2.1 Additional analysis of the effects of CbCR on reporting companies

2.1.1 Additional data description

In addition to the exclusion described above, additional observations are excluded from the sample to improve the comparability of companies in treatment and control. I exclude companies with revenues above 25 times the revenue threshold representing the largest 2% of companies. The control group consists of companies with revenues below the threshold but exceeding 25% of the threshold revenue. This cutoff ensures that companies in the control group are not too small to be subject to comparable developments as the treatment group. Robustness and placebo tests reported below also use stricter revenue exclusions and a placebo revenue threshold to ensure that the effects are not driven by differences in firm sizes between the treatment and control group. Companies with revenues exceeding the threshold applicable but without cross-border subsidiaries listed in Orbis, as well as companies with changes in their CbCR obligation after 2016, are excluded from the sample in the baseline estimations. This excludes about 5 and 6% of the base sample and reduces potential attenuation bias. The MNE definition in the main estimations is based on the company structure noted in Orbis as of February 2020. Alternative assumptions are used as robustness checks.

Of the resulting sub-sample, 3,176 company groups are assigned to the treatment group (21,230 company-year observations), 7,385 companies are assigned to the control group (31,901 company-year observations). Summary statistics comparing treatment and control in the pre-reform period are shown in Table 5. By construction, treated companies have higher revenues, pre-tax profits, and tax payments, but consolidated ETRs used to assess changes in tax costs are very similar across both groups. The firm-level data are complemented with information on statutory tax rates from KPMG’s Corporate Tax Tables and EY’s Worldwide Corporate Tax Guides. Country-level data on GDP per capita growth and inflation come from the World Bank’s World Development Indicators database.

Table 5 Summary statistics for treatment and control group (pre-CbCR)

The independent variable used in the regressions is the consolidated effective tax rate. ETRs measure taxes paid over pre-tax profits and are a frequently used ex-post measure of tax avoidance (e.g., Dyreng et al., 2010; Hanlon & Slemrod, 2009; Overesch & Wolff, 2021). The sample only includes companies with positive profits. Negative tax payments are replaced with zeros for the calculation of ETRs. One advantage of using ETRs at the consolidated level is that they reflect all types of tax avoidance, including the strategic avoidance of permanent establishments in high-tax countries, which would not show in unconsolidated data (Beer et al., 2020).

2.1.2 Comparison to earlier results

Section 5 shows that CbCR led to an increase in the ETRs of MNEs subject to the reporting requirement. Similar responses to CbCR are described by Joshi (2020) for a sample of European companies. In a set of DiD estimates, Joshi (2020) reports an increase in ETRs of around 1.5 percentage points following the introduction of CbCR.²¹ The results on regulatory avoidance reported in Sect. 4 suggest that these estimates may suffer from some downward bias due to selection into—or rather out of—treatment since companies close to the threshold are not excluded. The stronger treatment effect for companies with lower ETRs reported in Sect. 5 which at the same time show a stronger avoidance response may aggravate the downward bias if companies close to the threshold are not excluded.

To investigate the effect, I rerun my estimate using a sample only consisting of European companies. As reported in Column (1) of Table 6, the treatment effect for this sample is 2.1 percentage points—more than double the effect for my full sample, and around a third larger than the effect found by Joshi (2020). Column (2) of Table 6 shows that the difference to Joshi (2020) can at least in part be explained by the treatment of companies close to the revenue threshold. If only companies within 10% of the threshold are considered, the treatment effect is insignificant and even yields a negative point estimate. The generally larger effect for EU companies may be driven by the discussion about the publication of CbCR data and the early adoption of parent entity filing in EU countries.

Table 6 Effects of CbCR on consolidated ETRs

Column (3) investigates the timing of the response in more detail by including an individual interaction term for each of the treatment years (2016–2018). This specification can thus be interpreted as an event-study design. The coefficients on the interaction terms do not change sign over the three treatment years but are only statistically significant in 2018. In this last year of the sample period, the treatment effect on ETRs is almost twice as large as the average effect over the full treatment period. This is in line with the results of Joshi (2020) who describes a similar pattern and the finding of an increasing bunching response over the sample period as reported in Sect. 4.2.²²

2.1.3 Testing the identifying assumption

The main identifying assumption for the difference-in-difference estimator used in Sect. 5 is that the treatment and control group would have trended similarly without the introduction of CbCR. The identifying assumption would be violated if other regulatory changes besides CbCR affected the treatment results. While there were no other size-dependent reforms implemented as part of the BEPS project in the sample period, there might have been other shocks affecting treatment and control differently. To validate the assumption of parallel pre-trends, I estimate the following model (under inclusion of the year 2015) and graphically show the treatment effect on the consolidated effective tax rate over time:

$${\text{ETR}}_{ist} = \mathop \sum \limits_{T = 2010}^{2018} \beta_{T} \left( {t = T} \right){\text{CbCR}}_{i} + \gamma_{i} + \mu_{t} \cdot \varphi_{s} + \delta^{\prime } X_{it} + \in_{ist}$$

(3)

All variables are defined as in Eq. (2) of Sect. 5. Figure 14 shows the coefficients ${\beta }_{T}$ for ETRs according to Eq. (3) over the years 2010–2018 with and without additional controls. I consider the parallel trend assumption between the treatment and control group to be satisfied, as none of the coefficients are statistically different from zero before 2016. The fact that the coefficient becomes larger (but remains insignificant) in 2015 points toward an announcement effect, justifying the exclusion of this year. The trend plots also confirm that the effects of CbCR on ETRs took some time to materialize and are strongest in the last year of my sample period.

Fig. 14

Treatment effect on consolidated effective tax rates. Notes: Coefficients on the term $\sum_{T=2010}^{2018}{\beta }_{T}\left(t=T\right){{\text{CbCR}}}_{i}$ from Eq. (3). Outcomes are normalized to zero in 2012 in the middle of the pre-reform period. The dependent variable is the consolidated effective tax rate. The dashed vertical lines indicate the 95%-confidence intervals for the coefficient estimates with standard errors clustered at the company level

To further test the validity of the research design, I conduct a series of placebo tests. The results are visualized in Fig. 15. First, I test the effect of CbCR on financial companies. Generally, financial companies are subject to stricter reporting regimes; in the European Union, multinational banks are even subject to mandatory public CbCR after 2014 (European Commission, 2013). The introduction of private CbCR via the BEPS project in 2016 therefore only leads to minor changes in the reporting requirements of (EU) financial companies. In line with this assumption, the treatment effect of CbCR on consolidated ETRs of all financial companies as well as for EU financial companies only is not statistically different from zero. Second, I use a placebo threshold of 200% of the actual threshold applicable to define the treatment and control group. Again, the coefficient on the treatment effect is not statistically different from zero. Lastly, I define 2013 as a placebo treatment year. This test also yields no statistically significant treatment effect.

Fig. 15

Placebo tests. Notes: DiD coefficients of the treatment effect of CbCR on consolidated effective tax rates for different samples following the estimation structure of Eq. (2). The vertical lines indicate the 95% confidence intervals for the coefficient estimates with standard errors clustered at the company level

2.1.4 Robustness checks

The result of an increase in ETRs of treated companies after the introduction of CbCR reported in Sect. 5 is robust to a large number of checks. Figure 16 summarizes the treatment effects across the different robustness checks. First, the treatment effect of CbCR on ETRs in a balanced sample is 0.974 and thus almost identical to the result for the main sample. Second, I reweigh all observations to account for differences in terms of industry or headquarter country distribution using entropy balancing as suggested by Hainmueller (2012) and Hainmueller and Xu (2013). In a third test, weights are also based on return on assets and leverage in the pre-reform year of 2013. In further tests, I include the previously excluded year 2015, companies with revenues above the threshold but no cross-border subsidiaries listed in Orbis, or companies with revenues above 25-times the threshold revenue. To make sure that selection into treatment is not an issue, I rerun all estimations under exclusion of company groups with revenues within 25% of the threshold applicable. In addition, I use data on the company structure in 2016 taken from an older vintage of Orbis to determine the treatment status. Lastly, I exclude companies with revenues below one-third of the threshold as well as companies with more than three times the threshold revenue to test if effects are driven by differences in firm sizes between the treatment and control group. For all these tests, the treatment effect is close to the baseline estimate indicated by the dashed line.

Fig. 16

Robustness checks. Notes: DiD coefficients of the treatment effect of CbCR on consolidated effect tax rates for different samples following the estimation structure of Eq. (2). The vertical lines indicate the 95% confidence intervals for the coefficient estimates with standard errors clustered at the company level. The dashed horizontal line represents the coefficient from the baseline estimation at 0.975

2.1.5 Effects of CbCR on subsidiaries

The increase in ETRs following the introduction of CbCR observed in the treatment group points toward a reduction in profit shifting. Since extreme ETRs may also result from other factors besides profit shifting (Blouin, 2014; Schwab et al., 2022), I conduct additional tests for changes in tax avoidance based on unconsolidated subsidiary data.

The sample of subsidiaries covers 61,073 entities from OECD countries. Subsidiaries are only included if a majority shareholder exists, if unconsolidated subsidiary revenues exceed EUR 1 million, and if their parent would not be excluded from the regression sample used in Sect. 5. The pre-reform year of 2015 is again excluded. The sample of subsidiaries is split into treatment and control based on the treatment status of their majority shareholder. This assigns roughly 70% of the sample to the treatment group (41,915 subsidiaries, 256,031 entity-year observations). 19,158 subsidiaries make up the control group (94,466 entity-year observations). The share of treated subsidiaries is higher since MNEs in the treatment group are larger than those in the control group and tend to have more subsidiaries listed in Orbis. The regression model used for the subsidiary-level estimations follows the same structure as Eq. (2). Controls, fixed effects, and the clustering of standard errors in the regressions based on unconsolidated data are at the subsidiary level.²³

To assess the impact of CbCR on corporate tax avoidance at the subsidiary level, I first adapt the approach taken by Riedel, Zinn and Hofmann (2015) and investigate whether the tax sensitivity of subsidiary profitability changed due to CbCR. To this end, I add a triple interaction term between ${\text{CbCR}}$, ${\text{post}}2016$, and the statutory CIT rate of the subsidiary country to Eq. (2).²⁴ To account for differences in average subsidiary size between treatment and control, I rely on the return on assets (RoA) as dependent variable instead of profit levels. As shown in Column (1) of Table 7, a one percentage point higher statutory CIT rate in the subsidiary country was associated with a 0.104 percentage point lower return on assets before the introduction of CbCR. The coefficient on the triple interaction term is 0.043 and statistically significant, implying a reduction in the tax sensitivity of subsidiary profitability after the introduction of CbCR by more than 40%. Figure 17 shows that the parallel trend assumption also holds for the regression model at the subsidiary level. The dependent variable in Column (2) of Table 7 is the logarithm of the return on assets allowing to calculate a semi-elasticity. According to this specification, CbCR reduces the semi-elasticity of subsidiary profitability by around a quarter. Second, reduced profit shifting should lead to an increase in the share of profits remaining in high-tax subsidiaries. To test this conjecture, I split the subsidiary sample into high- and low-tax subsidiaries at the median statutory CIT rate and assess whether the share of subsidiary profits in consolidated group profits changed due to CbCR. The importance of an independent analysis of the two groups is shown by Dowd et al. (2017). Results are reported in Columns (3) and (4) of Table 7. The treatment effect for high-tax subsidiaries is positive and significant at the 1% level, while the corresponding coefficient is insignificant for low-tax subsidiaries.²⁵ Lastly, the insignificant treatment effect on ETRs at the subsidiary level indicates that the increase in tax costs at the consolidated level was driven by a reallocation of profits among subsidiaries, not an increase of tax rates within subsidiaries (see Column (5) of Table 7). In sum, the results on the effects of CbCR at the subsidiary level support the interpretation that CbCR leads to a reduction in tax avoidance of companies with reporting obligation.

Fig. 17

Treatment effect on the tax sensitivity of subsidiary RoA. Notes: Coefficients on the term $\sum_{{\text{T}}=2010}^{2018}{\upbeta }_{{\text{T}}}\left({\text{t}}={\text{T}}\right){{\text{CbCR}}}_{{\text{i}}}$ from Eq. (3). Outcomes are normalized to zero in 2012 in the middle of the pre-reform period. The dependent variable is the unconsolidated return on assets at the subsidiary level. Ultimate parent entities are not included. The dashed vertical lines indicate the 95%-confidence interval for the coefficient estimates with standard errors clustered at the company level

Table 7 Effects of CbCR on tax avoidance at the subsidiary level