Reducing Profit Shifting: An Evaluation of the Norwegian Thin Capitalization Rule.
Frida Kvamme
Thesis submitted for the degree of Master of Philosophy in Economics
Department of Economics Faculty of Social Sciences
UNIVERSITY OF OSLO
May 2020
Reducing Profit Shifting: An Evaluation of the Norwegian Thin Capitalization Rule
Frida Kvamme
Preface
This thesis marks the end of my years as a student at the University of Oslo. Although I feel a sense of relief as I write my concluding sentences, I realize I have spent the last year dreading the master thesis process for little reason. Even though it’s been challenging and frustrating at times, writing a master thesis has also been surprisingly fun, and of course very educational. This process has been eased through academic and mental support from several people. I first want to thank my supervisor, Martin Eckhoff Andresen at Statistics Norway, for excellent supervision. Our many discussions have been inspiring. Thank you for always keeping the door open and for answering a limitless amount of questions.
This thesis has also benefited greatly from comments by participants at a department sem- inar at the Norwegian School of Economics in March. Thank you to Jarle Møen for inviting me to present my results, this was a very valuable experience. Thanks also to Morten Bø˚as at NUPI and Se´an Kennedy at the OECD for proofreading and for very helpful comments regarding the structure and language.
A special thank you goes to my roommates whom I have lived with throughout the mas- ter process. As the majority of the thesis has been written at home due to the ongoing coronavirus-pandemic, I feel very grateful for your humorous and thoughtful personalities.
Your daily support and motivating efforts have been highly appreciated. I’m also thankful for the master thesis scholarship awarded to me by Oslo Fiscal Studies and for the data provided by Statistics Norway.
Finally, I would like to thank my friends and family for always cheering me on. A big hug awaits you all once the threat of the pandemic is reduced. Thank you for always being by biggest supporters.
Frida Kvamme May 2020
Abstract
Profit shifting by multinational enterprises (MNEs) erodes government tax bases and dis- torts market competition. Many countries have implemented laws to curb profit shifting including through interest expenses on debt. Rules that aim to deter this profit shift- ing are often called thin capitalization rules (TC rules). Much of the research literature finds that companies respond to TC rule by reducing internal debt, but the reductions are surprisingly small and the reasons for them remain disputed. Using over 2.5 million observations of company tax records in Norway between 2005 and 2017, I employ a gen- eralized difference-in-difference model to evaluate the effects of the Norwegian TC rule on profit shifting. Five main findings are established. Firstly, foreign MNEs respond to the rule through a reduction in the transfer pricing mechanism, indicating that these compa- nies reduce internal interest expenses while maintaining levels of internal debt. This could explain why previous research has found little response in debt levels. Secondly, I show that domestic MNEs behave differently from foreign MNEs by reducing both their debt levels and interest expenses. This difference in the response to rule between foreign and domestic MNEs is striking and does not appear to be a significant part of the discussion in the research literature. Thirdly, foreign MNEs pay more tax per NOK in response to the rule than domestic MNEs in part because domestic MNEs can substitute their internal interest expenses to Norwegian affiliates with intragroup contributions. Fourthly, I find no evidence of substitution effects towards alternative channels of profit shifting or to external debt or interest expenses. Finally, a simulation analysis suggests that the rule produces in the region of 2 billion NOK (1.25% of corporate income tax revenue in 2017) in additional tax revenue annually, with little apparent cost in either reduced investment or profitability among the affected companies.
The analysis was carried out using Stata. All remaining errors are my own.
Contents
1 Introduction 1
2 Literature review 3
2.1 Capital structure of multinational companies . . . 3
2.2 Thin capitalization rules . . . 4
2.3 Contribution of this thesis . . . 6
3 Institutional background 7 3.1 Interest limitation rules . . . 7
3.2 The Norwegian thin capitalization rule . . . 8
4 Theory 9 4.1 Optimal capital structure in firms . . . 9
4.2 Financing a multinational company . . . 10
4.3 Thin capitalization rules . . . 13
4.4 Substitution between profit shifting alternatives . . . 15
5 Data and descriptive statistics 18 5.1 Data and sample restrictions . . . 18
5.2 Central variables . . . 19
5.3 Descriptive statistics . . . 20
6 Empirical approach 23 6.1 The Generalized Difference-in-Difference Model . . . 24
6.2 Threats to identification . . . 25
7 Results and discussion 27 7.1 Main results . . . 27
7.2 Differential firm behavioral responses to TC rules . . . 31
7.2.1 Foreign and domestic multinationals . . . 32
7.3 Tax revenue analysis . . . 34
8 Conclusion 36
Bibliography 39
A Robustness check — industry controls 42
B Robustness check — attrition 43
List of Figures
1 Estimated and actual disallowed variables from RF-1315 . . . 20
2 Estimated and actual disallowed interest (2005-2017) . . . 22
3 Aggregated RF-1315 data by destination and source countries (2014-2017) . 23 4 Regression results — complete sample . . . 28
5 Balance sheet variables — complete sample . . . 29
6 Assets and transactions — complete sample . . . 30
7 Regression results — FMNE DMNE . . . 32
8 Balance sheet variables — FMNE DMNE . . . 33
9 Assets and transactions — FMNE DMNE . . . 34
10 Extra tax revenue (2014-2017) . . . 35
11 Regressions with industry controls . . . 42
12 Regressions, balanced —whole sample . . . 43
List of Tables
1 Sample characteristics for selected variables (2014-2017) . . . 212 Sample characteristics for subsamples (2014-2017) . . . 21
1 Introduction
Stronger integration and globalization in recent decades have facilitated increased capital mobility. Many multinational enterprises (MNEs) are in a position to benefit from this shift and minimize the tax they pay, by exploiting differences in legislation and tax rules prac- tices between countries. The Organisation for Economic Co-operation and Development (OECD) estimates that these practices cost countries USD 100-240 billion in lost revenue annually, which equals around 4-10% of the global corporate income tax revenue(OECD, 2015a). This erodes the tax bases of countries, constrains government resources, while distorting competition in favor of MNEs relative to domestic firms. It was against this background in 2013 that the OECD launched its work plan for Base Erosion and Profit Shifting project (BEPS).
One mechanism through which MNEs can shift profits is through interest expenses on debt. By lending money from a related affiliate in a low-tax jurisdiction, a multinational company effectively shifts parts of their profits to a jurisdiction with lower tax rates, saving the tax difference. Consequently, BEPS Action number 4 called for rules to ”limit base erosion via interest deductions and other financial payments” (OECD, 2013). Rules that aim to hinder base erosion via interest deductions are often called thin capitalization rules (TC rules henceforth). Understanding the effect of TC rules is important both to ensure that the rules actually hinder profit shifting, and to uncover and understand any possible unintended consequences. This thesis will evaluate the effects of the Norwegian TC rule implemented in 2014, and will especially focus on the rule’s effectiveness in hindering profit shifting.
Previous research has established that the location of MNE’s debt are influenced by tax rates (Desai et al., 2004; Huizinga et al., 2008; Mintz and Weichenrieder, 2010; Møen et al., 2019). Moreover, Blouin et al. (2014) and Buettner et al. (2012) finds that TC rules are effective in reducing debt levels. However, the reductions have been surprisingly small and the reasons why remain disputed. Some point to substitution effects towards external debt and external interest expenses (Blouin et al., 2014; Dreßler and Scheuering, 2015).
Others argue that there are loopholes in the rules that firms exploit (Weichenrieder and Windischbauer, 2008; Saunders-Scott, 2015), or that changing the capital structure takes time (Hopland et al., 2018). Finally, some argue that previous research has mostly looked at operating companies and thus overlook important holding companies(Ruf, 2011).
The Norwegian TC rule has so far only been studied by master theses and Andresen and Kvamme (2019). Finnanger and Leland (2017) finds that the firms subject to the rule became more profitable after implementation, indicating that the rule reduced firms’
manipulation of transfer prices. Skjæveland and Viung (2016) finds a reduction in both internal and total debt, and that this the reduction is bigger for national firms than foreign
firms. In contrast, Ahmed (2017) finds that MNEs convert external debt into internal debt and then reduces the internal interest expenses.
This thesis builds on the analysis in Andresen and Kvamme (2019)1 which finds that firms reduce their interest and pay more tax in response to the rule. They find no evidence of substitution effects towards external interest expenses. The contribution of this thesis is twofold. First, I provide a new explanation as to why previous research has found surprisingly low reductions in debt. By exploiting rich Norwegian register data I’m able to look at the effects of the rule and it’s mechanisms in a more thorough way than previous research has. I later argue why exploring both the debt levels and interest expenses is essential in order to capture the full effect of implementing a TC rule, and why this approach can be explanatory for the low elasticities found in previous research.
Secondly, this thesis contributes to the discussion on TC rules and it’s effects on profit shifting. The microdata allows the sample to be split into subsamples, which allows for isolating the different behavioral responses across firms. More specifically, I divide the firms into foreign MNEs and domestic MNEs. This is important as earlier research has suggested that multinational firms face opportunities of mitigating the effects of the rule, possibly through alternative channels of profit shifting (Saunders-Scott, 2015). By looking at the multinational firms by groups, I’m able to isolate the different reaction patterns and investigate whether they are able to avoid the rule.
In order to study the effects of the Norwegian TC rule I will first set up a theoretic frame- work for firms’ capital structure based on previous work by Møen et al. (2019) and Schindler and Schjelderup (2016). I then try to extend this model to include the Norwegian TC rule and predict some possible behavioral responses for firms. Next I will take these predictions to the data where I use a generalized difference-in-difference model to test the predictions.
Using a number of years before and after treatment, this method produces treatment coef- ficients by year, enabling a visual inspection of possible treatment trends.
There are five main findings from this research. First, foreign multinationals respond to the rule through reduction of the transfer pricing mechanism, implying that these companies reduce internal interest expenses while maintaining levels of internal debt. This could explain why previous research has found little response in debt levels. Second, I show that domestic multinationals behave differently from foreign multinationals by reducing both their debt levels and interest expenses. Thirdly, foreign multinationals enterprises pay more tax per NOK in response to the rule than domestic multinationals because domestic multinationals can substitute their internal interest expenses to Norwegian affiliates with intragroup contributions. Forth, and contrary to previous research, I find no evidence of
1Parts of this thesis were developed for the Statistics Norway report nr. 2019/41, evaluating the effects of the Norwegian TC rule. The report was commissioned by the Norwegian Ministry of Finance.
substitution effects towards alternative channels of profit shifting nor to external debt or interest expenses. Finally, a simulation analysis suggests that the TC rule produces in the region of 2 billion NOK in additional tax revenue annually, with no apparent cost of reduced investment or profitability in the affected companies.
This thesis is structured in the following way: Section 2 presents related literature and my contributions to this literature. Section 3 provides the institutional background for TC rules and section 4 presents a theoretical framework for the analysis. In section 5 I describe the data sources and present descriptive statistics. In section 6 I explain the empirical approach and discuss possible threats to the analysis. Section 7 presents and discusses the results. Section 8 concludes.
2 Literature review
In the following section I will outline what I consider to be the most thesis relevant research conducted on capital structure of MNEs. I have divided the literature review into two subsections, one on capital structure in an international setting, and the second on existing research on thin capitalization rules. I have chosen not to divide the research further into sections of theoretical and empirical papers as the papers that presents a theoretic model also tests their model empirically. The final subsection discusses limitations of previous research and presents my contributions.
2.1 Capital structure of multinational companies
Several studies show that multinational firms leverage higher debt levels in countries with high tax rates.
Using data on US multinationals, Desai et al. (2004) highlights how companies have higher debt-to-equity ratios in countries with higher statutory tax rates than in countries with lower tax rates. The levels of internal debt were particularly sensitive to tax rates. They also investigated the relationship between strong creditor rights and debt levels. Here they found that strong creditor rights incentivized companies to increase external borrowing and decrease internal borrowing.
The model developed by Huizinga et al. (2008), predicts that profit shifting done by a multinational in a country depends on the national tax rate and a weighted average of international tax rate differences. Testing their theory empirically on a novel dataset of Eu- ropean firms in 23 countries, the results show that increased tax rates incentivize increased debt levels, consistent with Desai et al. (2004). Using a dataset on German outbound
FDI, Mintz and Weichenrieder (2010) similarly find a positive correlation between the debt levels and the host country’s tax rate.
The model of Huizinga et al. (2008) was extended by Møen et al. (2019) to include internal debt. The model predicts that it is always optimal for a MNE to use a mix of both external and internal debt and that the gain from interest deductions from internal debt are maximized if the creditor is the group affiliate placed in the country with the lowest tax. When testing their model on German micro data they found that internal and external debt were of about equal importance for debt shifting purposes.
Egger et al. (2010) questioned how foreign ownership affected the debt levels of firms. Using a dataset which covered 27 European countries, they found that foreign-owned firms had significantly larger debt ratios than their counterparts in their host country. Additionally, they demonstrated how this was increasing in tax rates. A limitation to their study is that they did not have data to investigate the internal debt levels on firm level, only the total.
Similarly, Taylor and Richardson (2013) finds that variables pertaining to multinationality and connections to tax havens are significantly and positively associated with firm’s debt ratio.
A quantitative review of research on multinational companies’ profit shifting behavior can be found in Heckemeyer and Overesch (2017). They use 27 research articles investigating different forms of profit shifting and the response to tax rate differentials. They find a tax semi-elasticity of pre-tax profit of 0.8, which means that reported profits decrease by about 0.8% if the international tax rate differential increases by one percentage point. The results also suggest that transfer pricing (price manipulation of intragroup transactions) and licensing costs are the main channels of profit shifting.
2.2 Thin capitalization rules
The research literature on thin capitalization rules has so far been few in number and the results are varied. On balance, most of the research literature tends to find a reduced use of internal debt as a response to thin capitalization rules. However, the results regarding the size of the response and substitution effects are mixed.
Buettner et al. (2012) analyzed the effects of thin capitalization rules using a dataset on all foreign affiliates of German companies which covers 36 countries (28 of these were European). They found that German affiliates abroad affected by the rule reduce their internal debt, though substitute parts of it by increasing their external debt. Wamser (2014) investigates the effect of the German TC rule on affiliates in Germany owned by foreign multinationals. Using propensity score matching, he finds similarly to Buettner et.
al. that firms affected by the rule reduced their internal debt, but substitute internal debt with external debt. However, other studies find different results. For example, de Mooij and Hebous (2017) find that there is no substitution to external debt if the TC rule only applies to related parties.
Using data on US multinationals in 54 countries, Blouin et al. (2014) also finds thin cap- italization rules to decrease firm’s debt ratio. Moreover, they found that the companies change their financing promptly to the rule and that the firms’ market value were reduced.
They find no effects on investment levels.
Buslei and Simmler (2012) finds that the introduction of the German thin capitalization rule led German firms to either reduce their debt ratio or split their assets to exploit the exemption clause. Dreßler and Scheuering (2015) use the same data and reform as Buslei and Simmler (2012) and extends their analysis by using 3 years before and after the reform, as well as separating firms into groups. Contrary to Buslei and Simmer, Dreßler and Scheuering find the thin capitalization rules to decrease external debt, not internal, and that the effect were mostly present for national companies. Multinational firms did not decrease their debt levels in the same matter as national firms.
Though the research conducted so far on TC rules is consistent in that they all show a negative response in debt levels to the rules, Ruf (2011) points out that the elasticities are surprisingly low. He presents three possible explanations for this, the first being that the studies have only looked at operating subsidiaries and thus have overlooked the role of holding companies. Secondly, he theorizes that transfer pricing guidelines could decentivize debt shifting. An elaboration of the relationship between transfer pricing and debt shifting is provided in the next subsection. Thirdly, debt shifting could be a tool only used by multinationals. He finds evidence for the two first hypotheses, but not the third.
Hopland et al. (2018) shows that when companies unexpectedly incur losses and wishes to shift money in to a country, debt is less flexible than changing transfer prices. One explanation is that changing internal financing is a longer-term decision and thus offers less flexibility. It seems likely that the same explanation should hold for shifting profits out of a country.
Research done by Egger et al. (2014) provides much larger elasticises than other studies.
They control for the possibilities of access to external borrowing in the host country. This is in line with the results presented by Desai et al. (2004) and suggests that firms have other reasons of using internal borrowing other than profit shifting.
The most relevant master thesis to include in the literature section is the thesis by Skjæve- land and Viung (2016). They evaluate the effects of the Norwegian earnings stripping rule, and focus on debt levels as the dependent variables. They find a reduction in both internal
and total debt, and that this reduction is bigger for national firms than foreign firms. They propose that this could be due to possible alternative profit shifting mechanisms.
2.3 Contribution of this thesis
Common for the majority of the research discussed in the literature section is the research design based on exploiting cross-country differences in tax rates, thin capitalization rules and debt ratios. One pronounced challenge is that the credibility of the causal claim is lower in a cross country setting due to unobserved factors. Even though country fixed effects are included, hence controlling for constant country specific factors such as geography, governance and size, it is close to impossible to control for all important factors. Local policy changes, for example regarding the access to capital, could also impact the debt ratios. In contrast, the Norwegian implementation of the earnings stripping rule provides a setting for a natural experiment. This is arguably more favorable than a cross country setting seeing as a more credibly control group is easier to establish.
Furthermore, as the design of thin capitalization rules differs across countries, previous research has struggled to precisely determine firms’ treatment status. In contrast to this, the rich register data from Statistics Norway enables close to perfect determination of the treatment group.
Due to what seems like data limitation, most research done on thin capitalization rules only investigates the changes to firms’ debt levels and debt-to-equity ratios. Arguably, focusing on balance sheet variables is sufficient when investigating thin capitalization rules effects on capital structure. However, when looking at the effects on profit shifting, the role of interest expenses should not be overlooked. If firms are able to artificially set the internal interest rate, they could react to the implementation of a thin capitalization rule by reducing the interest rate instead of debt levels. This could explain why previous research has found little response in debt levels.
How this impacts the estimates of profit shifting depends on the rule’s design. If the purpose is to estimate the effect of a safe harbor rule (a set debt-to-equity ratio) on profit shifting, not looking at the actual interest expenses could overestimate the effects on profit shifting. This is because firms can adjust their debt levels to the rule, but simultaneously increase the internal interest rate to maintain the same level of profit shifting. If the rule in question is an earnings stripping rule(a set interest-to-earnings ratio), not looking at interest expenses would underestimate the rule’s profit shifting effect. The debt reduction could be low, giving the impression of little response to the rule, whereas the firms have actually reduced their internal interest expenses, and consequently their profit shifting.
The lack of research looking at both debt levels and interest expenses seems to be explained
by limited access to data on firms interest expenses. In contrast, the data provided by Statistics Norway enables me to look at both debt levels as well as the actual interest expenses paid on the debt. As discussed, this is essential in order to capture the full effect of implementing a thin capitalization rule.
In sum, my contributions to the literature are facilitated by the rich Norwegian register data. I’m able to look at the effects of the rule and it’s mechanisms in a more thorough way than previous research. I’m also able to split the sample into different subsamples, thus isolating the differential reaction patterns between different types of firms. The data also includes detailed tax data suitable for estimations of tax responses, a type of analysis very few papers have undertaken. Finally, the tax data allows for predictions concerning the potential extra tax revenue from reducing profit shifting.
3 Institutional background
3.1 Interest limitation rules
The incentives for using debt financing and debt shifting is rooted in the way the corporate tax system is structured. The corporate tax in Norway is designed so that financing costs for debt are deductible, whereas financing costs for equity financing are not. One argument for this are, inter alia, that interest expense on debt is a payable cost to the firm. In contrast, costs associated with equity financing, viewed as the return the same capital could have had elsewhere, is neither payable nor easy to estimate. Considering the fact that only debt is deductible, the incentive to prefer debt as a method of financing is thus strong.
The benefit of using debt as financing is even greater for multinational enterprises than for purely domestic firms. MNEs can minimize their total global tax burden by financing the companies in high-tax countries with debt, and placing the equity-financed parent company in a low-tax country. By using loans from the parent company to the subsidiary in the high-tax country, the subsidiary can move the profits to a low-tax country through internal interest costs and at the same time deduct the expenses to minimize taxable profits in the high-tax country.
In addition to reducing government revenue, this practice distorts the competition in strong disfavour of domestic firms without foreign affiliates. Relative to MNEs that shift profits, the domestic firms will face a higher effective tax rate. In addition, the erosion of the domestic tax base and the competitive disadvantage of domestic firms affect the integrity of the tax system. Against this background, countries started to implement thin capitalization
rules targeting this practice.
The types of TC rules to hinder this is usually divided into two groups, rules based on a ratio of balance sheet variables such as set debt-to-equity ratio (”safe harbor”) , and rules based on a ratio of interest expenses and economic performance, such as interest expenses/earnings (”earnings stripping”). While safe harbor rules long was seen as the method of choice, earnings stripping rule are now the best practice approach recommended by the OECD (OECD, 2015b). They argue that safe harbor rules come with several dis- advantages. The main disadvantage refers back to the discussion of manipulation of the interest rate, a debt-to-equity rule does nothing to hinder this. They also point to the fact that it seems easy for a MNE to manipulate the rule by transferring equity to a specific entity located in a country with a safe harbor rule.
In contrast to this, rules setting the limit to a percent interest deductibility of earnings is a better tool to combat profit shifting. Linking the interest deductions to economic activity and taxable income makes the rule robust to other types of profit shifting, such a transfer mispricing, which would have reduced the earnings and consequently the amount of allowed interest deductions. The measure of earnings is usually a firm’s earnings before interest, tax, depreciation and amortization (EBITDA). A disadvantage with using earnings based rules is that it creates an unstability in the predictions of how large the interest deductibles will be at the end of the year. The best practice approach thus includes the opportunity of carrying forward disallowed interest which can be offset against future unused allowed interest.
3.2 The Norwegian thin capitalization rule
As part of the effort to minimize companies’ profit shifting out of Norway, an earnings stripping rule was implemented in 2014. The rule restricts firms’ deductions for interest expenses according to a standard limitation, cf. § 6-41 of the Tax Act. If the company’s interest expenses exceed the interest deduction limit of 30% (25% from 2016) of the com- pany’s EBITDA, the company’s internal interest expenses will be limited by the amount exceeding the limit. The disallowed amount will then be returned to the company’s tax base.
The rule limits interest expenses on internal debt to related companies (direct ownership or control of at least 50 per cent), as well as certain types of external debt that were regarded as internal debt under the rule2. This included, among other things, debt to external lenders
2At its introduction in 2014, it was pointed out that the transfer of profits through interest de- ductions could just as easily happen through interest external paid to independent lenders, through so-called ”external debt relief”, cf. Meld. St 4. In 2019, the rule was therefore extended to include disallowance of external interest costs at group level.
where related affiliated had provided collateral. Only the companies with net interest expenses over NOK 5 million are subject to the rule. The rule does not apply to personal taxpayers, financial corporations or companies that are taxed only by the petroleum tax office.
In accordance with the principle of non-discrimination in the EEA Agreement, the state cannot impose rules that favor local loan conditions. This means that the TC rule applies equally to both Norwegian, partly Norwegian companies, and foreign companies with Nor- wegian branches. The accuracy of the rule, with regard to the purpose, has therefore been criticized. Purely domestic companies and groups are at risk of having interest expenses disallowed, which will not be relevant for the purpose of the rule - limiting the shifting of profits abroad. In principle, Norwegian companies can guard against this in the form of substituting domestic interest expenses with intragroup contributions. In reality, how- ever, it can be challenging to predict which companies in the business group that will have disallowed interest and needs to substitute the interest expenses with contributions. The degree to which they are able to substitute the interest expenses with contributions will be analyzed in section 7.
4 Theory
In order to understand the incentives the earnings strippings rule introduce, it’s useful to set up a model. I will first introduce a model of optimal tax-efficient capital structure of a MNE and it’s affiliates. This is a model developed by Møen et al. (2019) and builds on previous work by Huizinga et al. (2008). I will then attempt to extend their model by introducing the Norwegian earning strippings rule. Moreover, I extend the model to include both debt shifting and transfer pricing opportunities and a substitution effect between the two. This extension is based on a model developed by Schindler and Schjelderup (2016).
Finally, I will discuss the implications of several profit shifting alternatives on the incentives of the thin capitalization rule.
4.1 Optimal capital structure in firms
The theoretic foundation of optimal capital structure in firms is perceived to origin with Modigliani and Miller and their 1958 paper ”The Cost of Capital”. In their model they showed how in a world with perfect capital markets and no taxes, capital structure was irrelevant for the value of a firm. Even though interest expense on debt is deductible from the tax base whereas the cost of equity is not, increasing debt comes with a bankruptcy cost, effectively equalizing the use of equity and debt. Expanding their theory in 1963 to
include corporate taxes, they showed, contrary to their 1958 article, that a firm’s value does in fact increase together with the increased use of debt. In a world with corporate taxes, the value of debt deduction, often called the debt shield, is rising in tax rates. According to Modigliani and Miller (1963), this incentivizes the firm to finance only by debt.
There are, however, several reasons why firms are financed by both equity and debt.
Modigliani and Miller (1963) point out that there are costs associated with debt and that lenders impose restrictions on the amount of debt a firm can incur. Berk and Demarzo (2014) lists agent costs, transactions costs, bankruptcy costs, costs associated with infor- mation asymmetries as additional reasons as to why a firm will never be financed purely by debt. According to the trade off theory3, a firm will in optimum balance the tax gains of the debt shield against the costs of debt.
4.2 Financing a multinational company
The model developed by Møen et al. (2019) extends previous theory on capital structure by showing that there are three ways a MNE can save tax: In addition to the standard debt tax shield, MNEs can exploit debt tax shields from both internal and external debt capital markets to shift profit. I have devoted the most space to internal debt shifting as the Norwegian earning strippings rule targets internal interest expenses4.
A multinational enterprise (MNE) has a number of affiliates i= 1,2...ninncountries that each have tax rate ti. MNE structure and location is exogenous and they only have one company in each countryi. The head company of the MNE is denotedp. For simplification, it is assumed that each affiliateionly has has fixed assetsKiwhich they use to produce one homogenous good with production function yi = f(Ki). Price of the good is normalized to 1. This is the same in all countries.
CapitalKi can be financed either by equityEi, external debtDiE or internal debtDIi from related parties. This gives affiliateithe balance sheet Ki=Ei+DiE+DIi and the MNE’s balance sheet will be P
i6=pEi = Ep +DpE +DIp. This means that the equity (Ei) in all affiliates is distributed from the MNE head companyp, and consists of equity and external and internal debt.
The internal debt-to-asset ratio in affiliate i is given by bIi = DKIi
i and external debt is bEi = DKEi
i. Total debt ratio is bi= DEiK+DIi
i . The cost function for internal debt and external
3The origins of trade off theory — the theory of optimal financial leverage for firms — is often credited to Kraus and Litzenberger (1973), Jensen and Meckling (1976) and Myers (1984).
4As of 2019 the Norwegian earnings strippings rule is extended to include external interest expenses on group level. This is, however, not part of this thesis as the time period since the change is too short.
debt is given by CI(bIi) and CE(bEi ), respectively. The assumptions common for both cost functions are that they are additively separable, convex in the debt-to-asset ratio and proportional to capital employed. Yet, the actual specifications for the cost functions for external debt and internal debt differ. Internal debt comes with costs relating to engineering expenses such as lawyers and accountants, and establishment of internal banks. On the other hand, internal debt is not associated with costs relating to bankruptcy, nor does it affect the information asymmetry between investors and management. The properties of the cost function for internal debt are expressed as follows:
CI(bIi)>0, CI0(bIi)>0, CI00(bIi)>0 if bIi >0 CI(bIi) = 0 if bIi ≤0
External debt, as aforementioned, comes with costs such as bankruptcy risk costs and agency costs. Following the trade-off theory, firms balance the risks and gains by setting an optimal level of debt-to- asset ratio of b∗. When the ratio of external debt-to-asset, bEi is less than b∗, the gains of increasing debt exceeds the costs. Atb∗ the marginal costs of one increased unit of debt is equal to the marginal gain. This gives the following properties for the cost function:
CE(bEi )>0, CE0(bEi )>0, CE00(bEi )>0 if bEi > b∗
CE0(bEi )<0 CE00(bEi )>0 if bEi < b∗
In addition, the MNE faces bankruptcy costs at the level of the head company because it guarantees for some of it’s affiliates’ external debt. This is given by Cf(bf) which equals the MNE’s total debt ratio
P
iDEi P
iKi. The costs of bankruptcy is increasing in the total debt ratio, which gives the properties Cf0(bf)>0 andCf00(bf)>0.
The true profit (πei) and taxable profit (πti) of firmiis given by
πie=yi−rKi−CE(bEi )−CI(bIi) and πti =yi−r(DiE−DIi)
True profit (πie) for firmiequals sales minus the interest expenses of debt and equity (Ki), minus the costs associated with the two types of debt. For taxable profit (πti) I assume that neither equity costs nor costs associated with debt are deductible, only interest expenses, which is in line with Schindler and Schjelderup (2016) and Møen et al. (2019). This is strong assumption which might be only partly correct. Møen et al. (2019) however, argue that this is a necessary simplification so that the model yields well-defined structural equations.
The same assumption is also applied in Huizinga et al. (2008).
Firm i’s after-tax profit (πi ) is thus is equal to πie−tiπit. Inserting for true profit and
taxable profit gives the following after profit equation:
πi= (1−ti)yi−rKi+tir(DEi +DIi)−CE(bEi )−CI(bIi)
From the after profit equation it is clear that the value of the firm is increasing in the tax debt shield tir(DEi +DiI) for both internal and external debt. Costs associated with taxation, equity and debt decreases firm value. For a profit maximizing MNE it is necessary to weigh the tax benefits of debts against the costs of debt.
The value of the MNE is given by the sum of after-tax profit of all affiliatesiplus the MNE’s cost of bankruptcyCf(bf). The MNE wishes to maximize their use of debt, but takes into account the extra costs associated with this. They also take into account that the sum of internal interest expense and income needs to sum out to zero across the affiliates. Total capitalKi is a considered a fixed variable and not part of the first order conditions. This is because MNE provides the affiliates with the equity necessary to reach both a tax-efficient financing structure and the optimal level of real capital. If debt is increased, Ei is reduced to maintain the sameKi The head of the MNE’s profit maximization problem becomes as follows:
maxDE
i,DiI
Y
p
=X
i
(1−ti)yi−rKi+tir(DEi +DiI)−CE(bEi )−CI(bIi) −Cf(bf) st. X
i
rDIi = 0 (λ)
Differentiating with respect toDEi andDIi gives the following first order conditions:
DEi : tir= ∂CiE(bEi )
∂bEi ∗ ∂bEi
∂DiE +∂Cf(bf)
∂bf ∗ ∂bf
∂DEi DIi : r(ti−λ) = ∂CiI(bIi)
∂bIi ∗ ∂bIi
∂DIi
The first order conditions show that in optimum, marginal costs of increasing debt by one unit equals the marginal tax savings gain. For external debt, the amount is increasing in tax rate of affiliate i. This is because a higher debt tax shield allows for higher marginal costs of external debt. The amount of internal debt is given by the tax rate differential of in the country of firm i and the lowest tax rate the MNE faces globally. This is given by the lagrange multiplier (λ), which is the shadow price of receiving one unit of internal interest income. In optimum, this is minimized by λ=miniti, meaning that the marginal costs of one unit internal interest income is minimized when it’s the affiliate in the country with lowest tax rate who functions as the MNE’s bank. By numbering the country with
the lowest tax rate to 1 i.e., miniti =t1 and inserting this in the first order condition for internal debt, I get the following equation:
r(ti−t1) = ∂CiI(bIi)
∂bIi ∗ ∂bIi
∂DIi
When i = 1, the left hand side equals to zero. This means that the affiliate located in country 1 will have no internal debt, but solely be the internal debt lender. It’s also possible to see that the higher the tax rate differentialti−t1, the more the gain from internal debt in country i will be. In theory, several affiliates could be creditors and receiving interest expenses, however, the tax-efficient structure implies that it’s the affiliated in t1 who will function as the MNE’s bank.
4.3 Thin capitalization rules
As described in section 2, the Norwegian interest limitation rule sets a threshold to a limited percentage interest deduction of a company’s EBITDA. The rule is strictly binding, meaning that firms are unable to have exceeding amount of interest expenses without this being disallowed. To simplify the model, EBITDA is here assumed to equal sales (yi). The rule only applies to firms above a certain threshold and in Norway this level is set to NOK 5 million. For the following modelling it is assumed that all firms are above this NOK 5M threshold and that firms are unable to manipulate their interest rate r (transfer pricing).
The rule does not apply if the firm has zero net internal interest expenses. The earnings strippings rule is given by the indicator function:
1min(rDiI, rDIi +rDEi −ηyi)
| {z }
Disallowed interest
1=
if rDiI+rDiE > ηyi
0 otherwise
The indicator function (1) equals to one if firmi0stotal interest expenses exceed the allowed percentage of EBITDA,ηyi . The amount disallowed and added to the tax base will be the minimum amount of total internal debt expenses and the amount exceeding the percentage.
Moreover, the disallowed interest will be taxed with countryi0stax rate, and after-tax profit (πi) will consequently decrease with the same amount. The new profit function for firm i, with exceeding interest expenses, is now given by:
πi= (1−ti)yi−rKi+tir(DiE+DIi)
−CE(bEi )−CI(bIi)−ti1imin(rDIi, rDIi +rDEi −ηyi)
As before the MNE wants to maximize the sum of after-tax profits taking into account the zero sum of internal debt. Now they additionally have to take into account any potential extra costs of disallowed interest. The new maximization problem for the MNE becomes as follows:
maxDE i ,DIi
Y
p
=X
i
(1−ti)yi−rKi+tir(DiI+DEi )−CE(bEi )−CI(bIi)
−ti1imin(rDiI, r(DIi +DiE)−ηyi) −Cf(bf)
st. X
i
rDIi = 0 (λ)
The indicator function and minimization function in the maximization problem gives rise to 3 different after-tax profit functions for firm i:
1.
(1−ti)yi−rKi+tir(DIi +DEi )−CE(bEi )−CI(bIi) if1i= 0
In the scenario where the indicator function is equal to zero,1i = 0, there is no extra cost added to the firm’s maximization problem. The profit functions is consequently equivalent to the results from Møen et al. (2019), derived in section 4.2. Firmi does not have excess interest expenses and can fully exploit the tax debt shields.
2.
(1−ti)yi−rKi+tirDiE−CE(bEi )−CI(bIi) if1i= 1 and rDiI≤r(DiI+DEi )−ηyi
In scenario 2 firm i have excess interest above the threshold and the amount of internal interest expenses is lower than the sum of interest expenses exceeding the rule. The min- imization function in the rule states that if rDiI ≤ rDIi +rDiE −ηyi, rDIi would be the disallowed interest and added to the profit function. This gives the extra cost of tirDiI which effectively cancels out the tax debt shield for internal debt, tirDiI from beforehand.
3.
(1−(1−η)ti)yi−rKi−CE(bEi )−CI(bIi) if1i= 1 and rDiI > r(DIi +DEi )−ηyi
In the scenario where rDiI+rDiE −ηyi < rDIi, each marginal unit debt of both internal and external debt is disallowed and−(ti(r(DiI+DEi −ηyi)) is added to the tax base. This cancels out the previous tax debt shield of tir(DIi +DiE), but note that tiηyi is still left in the equation and functions as a tax rebate. This symbols the fact that firms can still exploit the tax shield up until the level of ηyi.
4.4 Substitution between profit shifting alternatives
I have so far assumed that the interest cost r is non-manipulative for the modelling of debt financing and thin capitalization. In this part I will extend the model to include the possibility of transfer pricing through manipulation of the interest rate ˜ri. I will first introduce firm i0s maximization problem in a situation with substitution between transfer pricing and debt shifting, but without an interest limitation rule. The model is developed by Schindler and Schjelderup (2016). The scenario with both substitution opportunities and the implementation of an interest limitation rule will be discussed in the next subsection.
Total interest cost for internal debt is given byr+ ˜ri and the amount of profit shifted away through artificial interest rates ˜riis given byPi = ˜riDi >0. The cost functions are assumed to inhibit the same properties as before: convex in both internal and external debt, as the risk of bankruptcy increases with increased debt. The total debt cost function is additively separable, (CD(bEi , bIi, Pi) = CE(bEi ) +CI(bIi, Pi)), but the cost function for internal debt (CiI) is now dependent on both the internal debt-to-asset ratio (bIi) and the amount of profit shifted through transfer pricing (Pi). The intuition behind this is that profit shifting and/or high debt-to-asset ration can come with a cost in the form of tax administrations suspicion, which again can potentially lead to a costly audit. Thus low profits due to profit shifting through transfer pricing can make it costly to use more internal debt.
As before the MNE maximizes after-tax profit given the condition of zero-sum internal debt lending and borrowing across affiliates. Moreover, the same condition now has to apply for the sum of profit shifting. Firm i can deduct the artificial level of interest expenses (˜riDi), something that increased the tax debt shield (ti(rDiE +riDiI+ ˜riDi)). The new profit maximization function for the MNE is now:
maxDE
i,DiI,˜r
Y
p
=X
i
(1−ti)yi−rKi−˜rDi+ti(rDEi +rDiI+ ˜rDi)
−CE(bEi )−CI(bIi, Pi) −Cf(bf)
st. X
i
rDiI= 0 (λ) & X
i
˜
riDIi = 0 (γ)
In addition to the first order conditions derived in 4.3, I get the following first order con- dition by maximizing wrt. ˜ri:
˜
ri : DIi(ti−1)−∂CiI(bIi, Pi)
∂Pi
∗ ∂Pi
∂˜r +γDIi = 0
∂Pi
∂r˜ =DIi Rearranging and dividing by DIi gives:
˜
ri: µ+ (ti−1) = ∂CiI(bIi, Pi)
∂Pi
The new Lagrange multiplierγis the shadow value of an additional unit shifted. According to Schindler and Schjelderup (2016), it can be shown to be equal to maxi = (1−ti). By doing as before and numbering country 1 as the country with the lowest tax rate, it is clear that the value of transfer mispricing is maximized when the affiliate receiving the transfers is located in the country with the lowest tax rate:
ti−t1 = ∂CiI(bIi, Pi)
∂Pi
The right hand side is the marginal benefit of overpricing the interest rate which in optimum equals the marginal cost. The overpricing benefit is given by the tax rate differential of countryiand the country with the lowest tax rate, t1.
The interplay between debt shifting and transfer pricing through interest rates is dependent on the cross-derivatives of the concealment costs of profit shifting, given through the cost functionCI(bIi, Pi). Theory provides ambigious conclusions on the directions of these cross- derivatives. There is cost substitutability if there is an increased use of one type of profit shifting if the cost of the other channel increases, ∂P∂2CI
i∂bIi > 0. This is the direction of the cross-derivations which is intuitively more likely. However, there is also a case for cost complementarity (∂P∂2CI
i∂bIi < 0) if there are economies of scale related to increased concealment costs of one channel. If a MNE acquires skills in tax engineering through transfer pricing, these may be transferable to profit shifting through debt shifting as well, and vice versa. In line with Schindler and Schjelderup (2016) and Eggert and Goerdt (2020) the following discussion will assume cost substitution.
I can use the results from Schindler and Schjelderup (2016) to present possible outcome scenarios for firm capital structure when a country implements an earning strippings rule.
With substitution between debt shifting and interest rate manipulation, the modelling of the earnings strippings rule needs slight modifying:
1min((r+ ˜ri)DIi,(r+ ˜ri)DIi +rDEi −ηyi)
| {z }
Disallowed interest
1=
if (r+ ˜ri)DIi +rDEi > ηyi
0 otherwise
Recall that the earnings stripping rule is based on a percentage of EBITDA, (earnings before interest, taxes, depreciation and amortization), which I defined as a share of sales(ηyi). Of course in real life EBITDA includes all other costs than interest and taxes, such as operating costs, financial costs, etc.. Using transfer pricing to shift profits reduces the before-tax profit and the threshold amount, which is now disadvantageous for a firm with internal debt. Thus the earnings stripping rule effectively targets debt shifting by increasing the cost of all types of transfer pricing, not just interest rate manipulation.
Theory work on thin capitalization has so far only provided ambiguous proposals of ex- pected firm behavior under concealment cost substitutability. An earnings strippings rule which raises the marginal costs of both channels, may, under certain circumstances, lead to either more debt shifting or more abusive interest rates. However, based on the evidence from research discussed earlier, this results seems unlikely. Additionally, the model for the earnings strippings rule, provided a clear conclusion in that the rule is strictly binding and that firms would want to avoid the extra cost of disallowed interest. With this in mind, I propose the following alternative scenarios:
(1) As the earnings stripping rule targets both types of profit shifting, firm i will reduce both their artificial interest rates and levels of internal debt. Eggert and Goerdt (2020) argue that firms will still substitute one method in favor of the other, but that this is challenging to see when the use of both methods is reduced.
(2) Firms can decide to either reduce interest rates or debt levels, depending on where the marginal cost of reducing profit shifting is lowest. Schindler and Schjelderup (2016) claim that the marginal cost increase is lower for transfer pricing than debt shifting, something that is in line with other research that provides evidence for higher flexibility in transfer pricing (Buettner et al., 2012; Møen et al., 2019). It then seems probable that a firm will choose to reduce interest rate ˜r than rearrange their financing to get under the threshold.
(3) In the case where the MNE has additional mechanisms for profit shifting, such as over- pricing royalty payments or payments for expertise services, there could be a substitution effect between increased concealment costs for debt (both ˜r and rDIi) and this other type of profit shifting. One argument against this is that increased profit shifting would reduce EBITDA and consequently reduce the threshold determining disallowed interest. However,
if firm i is solely using internal debt for the purpose of profit shifting, it shouldn’t be anything hindering them from reducing internal debt expenses to zero and substitute this form of shifting towards a type of profit shifting that is not targeted by a rule i.e royalty payments.
5 Data and descriptive statistics
5.1 Data and sample restrictions
Based on the theory above, I take the following hypothesis to the data to investigate the effects of the TC introduction in 2014: Firms that are exposed to the rule in 2013 should reduce their interest expenses in order to not have parts of their interest expenses disallowed.
I also expect to see a reduction in internal debt levels. To do this I use rich register data provided by Statistics Norway spanning the period 2005-2017. By combining several datasets, the final dataset covers large areas of firms’ accounting numbers and tax records.
I follow Andresen and Kvamme (2019) and use as a starting point a dataset containing registration data for all firms in Norway5. All firms are listed with their organizational number, something which enables perfect matching of data sources. Furthermore, I have access to both the firms’ accounting data as well as data based on their tax records. The accounting data comes from source ”Income Statement 2”, a form all firms, except small firms or financial firms, have to turn in. Income Statement 2 includes all variables of firms income and sales, as well as all posts on their balance sheets. Of special interest are the variables for interest expenses, as well as debt and equity.
The tax data comes from the form ”Tax return for private limited companies”. In addition to tax liability, this form provides information on intragroup contributions. Moreover, I have access to the form ”Limitation of deduction for interest between related entities”
(RF-1315), a data source which has previously only been used by Andresen and Kvamme (2019). The form was introduced in 2014 as the TC rule was implemented. All firms with net interest expenses above 5 million NOK form have to turn it in. The form contains detailed information on all transactions regarding interest expenses and interest income, including source and/or destination country. Besides using it for descriptive statistics, this form was used as a blueprint for determining firms’ levels of exposure to the rule in 2013.
The earnings stripping rule does not apply to firms in the financial industry (i.e banks, insurance companies etc.), tax exempt industries (non-profit), or firms which are subject to petroleum tax. Consequently, these firms are dropped from the baseline dataset. Addi- tionally, shipping companies are also excluded, as these companies follow a separate interest
5Virksomhets- og foretaksregisteret (VOF)
model. The sample is also restricted to only include firms that exist in both Income State- ment 2 and Tax return form. This is because both forms are needed to construct the variables determining rule threshold and disallowed interest. The final baseline sample contains 2 751 108 observations.
In addition to this, I added data from several sources in order to divide the firms into subsamples. To identify firms as foreign MNEs (FMNEs and foreign MNEs is used in- terchangeably), I used the Norwegian shareholder register which lists all private limited companies, shareholders and the amount of shares they hold. A firm is categorized as a FMNE if a foreign shareholder holds at least 50% of total shares. Additionally, the forms ”Investments abroad” and ”Controlled transactions and accounts outstanding” re- quire firms to list their mother company and mother company’s country of domestication.
This was used as a complementary source to identify FMNEs, and as the main source to identify DMNEs. One downside of the last two data sources is that they do not cover the universe of firms, only a sample. The form ”Investments abroad” has to be answered by all Norwegian shareholders with more than 20% direct ownership abroad. Firms with transactions to related affiliates above a total of NOK 10 million are required to turn in
”Controlled transactions and accounts outstanding”.
5.2 Central variables
In order to determine the treatment status and level of exposure to the rule in 2013, pre- cise calculation of the variables EBITDA, net interest expenses and net internal interest expenses is needed. EBITDA is an earnings definitions previously not used in any report- ing to the tax administration, but was established in the form ”Limitation of deduction for interest between related entities” (RF-1315). To calculate EBITDA for all years, I use information gathered from the tax return and from form RF-1084 on Depreciations and mortizations. To compute EBITDA I start with the variable for net income (Income state- ment 2), deduct intragroup contributions (Tax return) then add depreciations (RF-1084) and net interest expenses (Income statement 2).
For net intragroup interest expenses, the form RF-1315 uses a somewhat wider definition of interest expenses than Income statement 2. For example, RF-1315 requires firms to include costs associated with providing collateral for related affiliates’ external debt. This is not considered to be a large problem, but might explain some of the discrepancies in reported internal interest expenses and total net interest in Income statement 2 and RF-1315.
The degree to which I was able to calculate the variables needed for treatment status is given by the correlations in figure 1. The correlations are between the estimated variables using other data sources and the blueprint variables from RF-1315 for the non-random
sample of firms who filled this form. All correlations are fairly high with coefficients over 0.9, except for coefficient for the amount of disallowed interest carried forward. One would primarily assume here that firms, if in a position to do so, would monetize the disallowed interest carried forward. A somewhat puzzling discovery was that not all firms did this, but chose to carry this further forward. This might be explanatory for the somewhat lower correlation for this variable.
Figure 1: Estimated and actual disallowed variables from RF-1315
Note: The figure displays the plotted correlation coefficients between estimated and actual numbers, for variables that are central to the determination of treatment status.
5.3 Descriptive statistics
Presented below is a table containing information on mean differences between firms subject to the rule and firms not subject to rule. All observations are given by observation-year, meaning that a firm can be included several times. Column (1) contains all firms from 2014-2017. Column (2) is all firms with net interest expenses above 5 millions NOK. Recall that these firms have to submit the interest limitation form, even though they might be below the threshold for having disallowed interest. Column (3) is all the firms that don’t turn in the form. Column (4) contains the firms that actually did have part of their interest disallowed. It’s clear from the table that the firms submitting RF-1315 are different than