International taxation and the organizational form of foreign direct investment

We investigate the relation between international taxation and the organizational form of foreign direct investment (FDI). Using micro-level data on inbound FDI relations in Germany, we find that a higher tax burden on income earned in a corporate subsidiary increases the probability that a multinational corporation (MNC) conducts foreign investment through a non-corporate flow-through. This effect is economically meaningful and varies with the relative importance of tax-motivated income shifting, a subsidiary’s non-tax benefits of limited liability and legal independence, and an MNC’s local knowledge. Moreover, we examine potential real effects of organizational form choices and document that affiliates established as flow-throughs exhibit a lower loss propensity and are less profitable than affiliates established as subsidiaries. Taken together, our findings inform policy makers about the potential response of MNCs to tax-law changes and suggest that the chosen organizational form can shape the future characteristics of investments abroad.


INTRODUCTION
Cross-border economic activities of multinational corporations (MNCs) have two sides.On the one hand, cross-border economic activities can stimulate economic growth and development (UNCTAD, 2017).On the other hand, MNCs can use the associated group structures to exploit differences between tax systems and to shift income to low-tax jurisdictions.Despite prior research showing that international taxation influences the location of foreign affiliates (Dyreng, Lindsey, Markle, & Shackelford, 2015;Lewellen & Robinson, 2013), we know little about how international taxation affects an MNC's organizational form choice for a newly established foreign affiliate.In general, an MNC can establish its affiliates as corporate subsidiaries or non-corporate flow-throughs.In this study, we examine the relationship between taxes levied on cross-border economic activities ('international taxation') and the organizational form that an MNC chooses for foreign direct investment ('FDI'). 1We also explore whether the future characteristics of a newly established foreign affiliate vary with the chosen organizational form.
Recent and planned changes to the international tax system, such as the Tax Cuts and Jobs Act of 2017 ('TCJA'), the OECD Action Plan on Base Erosion and Profit Shifting ('BEPS';OECD, 2013), and the OECD Two-Pillar Solution (OECD, 2021), could increase differences in the taxation of organizational forms.Thus, studying the impact of international taxation on organizational form choices provides insight into how MNCs might respond to such reforms and informs policy makers about the design of tax rules. 2 Moreover, organizational form choices are an integral element of crossborder investment decisions.Between 2005 and 2020, global net FDI inflows ranged between $1.6 and $3.1 trillion per year (World Bank, 2022), which suggests that MNCs regularly make large investments abroad, requiring the selection of an organizational form.A clear understanding of the drivers of these decisions and their potential implications for the future characteristics of new affiliates is therefore important for researchers and investors interested in MNCs and their cross-border economic activities.
Organizational forms differ in their tax and nontax costs and benefits and we expect the MNC to select the organizational form with the best costbenefit relation.The dividend-withholding tax is the main tax cost (benefit) of a subsidiary (flowthrough) because this tax applies only to dividend distributions of subsidiaries. 3Non-tax costs and benefits arise from differences in the MNC's liability exposure.Specifically, a subsidiary is a separate legal entity that provides the MNC with limited liability, shielding its operations from legal claims against the new affiliate and bankruptcy risk associated with the investment abroad (Bianco & Nicodano, 2006;Gordon & MacKie-Mason, 1994).
Transferring the ownership of a subsidiary is also less costly than transferring the ownership of a flow-through.In contrast, a subsidiary implies higher setup costs, compliance costs (Cerutti, Dell'Ariccia, & Martinez Peria, 2007;Demirguc-Kunt, Love, & Maksimovic, 2006), and coordination costs (Goolsbee & Maydew, 2002) than a flowthrough.Since international taxation affects the cost-benefit relation of organizational forms, we predict a positive relation between the tax cost of a subsidiary and the likelihood of establishing a new affiliate as a flow-through.Consistent with the MNC trading off the costs and benefits of organizational forms, we expect this relation to vary with the relative importance of other tax and non-tax factors, such as tax-motivated income shifting, the non-tax benefits of limited liability and legal independence, or local knowledge.
To address our research questions, we study organizational form choices of foreign investing entities establishing a new affiliate in Germany. 4 Germany provides an ideal setting because the country's extensive double tax treaty network implies country-pair-level variation in the tax burden difference between organizational forms, i.e., the tax burden difference varies conditional on the home country of the investing entity.At the same time, all organizational form choices occur in the same regulatory environment, holding constant the non-tax costs and benefits of organizational forms.Since most existing tax systems follow common principles (Barrios, Huizinga, Laeven, & Nicode `me, 2012), inferences regarding international taxation based on Germany should generalize to other FDI host countries.
We obtain data on organizational form choices from the Microdatabase Direct Investment (MiDi) of the Deutsche Bundesbank. 5 This dataset reports anonymized micro-level data on the stock of inbound FDI relations in Germany (Blank, Lipponer, Schild, & Scholz, 2020).As a key advantage, the MiDi database provides information on the organizational form of these FDI relations, which is unavailable in conventional datasets.For the sample period 2005 to 2013, we identify 2175 organizational form choices.The investing entities that make these choices are located in 59 home countries, which account for more than 99% of all inbound FDI relations recorded in the MiDi database.In total, 24.37% of the organizational form choices in our sample result in a non-corporate flow-through, underlining the relevance of this organizational form.
In the first set of tests, we investigate the determinants of organizational form choices.Our results suggest that the probability of establishing a new affiliate as a flow-through is positively associated with the tax cost of a subsidiary, consistent with MNCs being sensitive to international taxation in their organizational form choices.The tax effect is economically meaningful and comparable to the effect of non-tax determinants: a one standard deviation increase in the tax cost of a subsidiary, which is equivalent to an increase by 3.51 percentage points, is associated with a 4.06 percentage point higher probability of choosing a flow-through.
We next shed light on the tax and non-tax tradeoffs MNCs face in their organizational form choices. Specifically, we conduct several cross-sectional tests to identify conditions under which the effect of international taxation is likely to vary.We first exploit variation in tax-motivated income shifting and find that investing entities are less sensitive to the tax burden difference between organizational forms if the MNC can limit the affiliate's exposure to international taxation by shifting its income across borders.We also explore variation in non-tax factors and find that industrylevel differences in capital intensity and differences in regulatory quality between Germany and the home country of the MNC attenuate the tax sensitivity.These results suggest that the MNC trades off the tax cost of a subsidiary with its nontax benefits of limited liability and legal independence.Local knowledge of the FDI host country, in contrast, increases the tax sensitivity.
One concern regarding the determinants analysis is that unobserved MNC characteristics could drive our results.We address this concern in four ways.First, since the tax cost of a subsidiary varies at the country-pair level, our primary analysis exploits variation that is fairly exogenous to MNCs' organizational form choices. Second, we include a fixed effect for the MNC home country to control for common characteristics of MNCs from a given country.Third, we analyze changes in the tax cost of a subsidiary as opposed to levels.Fourth, we employ a quasi-natural experiment around a 2008 tax reform in Germany.This reform provides plausibly exogenous variation in the tax cost of a subsidiary for a subset of home countries, allowing a difference-in-differences analysis.Across all tests, we find evidence consistent with our baseline findings.
As a final analysis, we examine potential real effects of organizational form choices and test whether the chosen organizational form is associated with future characteristics of the new affiliate.To this end, we extend our initial sample to all observations available for a new affiliate in the MiDi database (sample period: 2005-2013).To address endogeneity concerns in this test, we leverage the tax cost of a subsidiary as an instrumental variable and implement a two-step estimation procedure, suggested by Wooldridge (2010).We document that flow-throughs exhibit a lower loss propensity and are less profitable than subsidiaries.These results suggest that the chosen organizational form affects future decisions at the affiliate level, shaping the future characteristics of the new affiliate.
Our study makes several contributions to the literature.First, our findings add to research on the effect of taxes on MNC group structures by showing that international taxation affects the organizational form of foreign affiliates.Prior research focuses on locational choices for foreign affiliates (Dyreng et al., 2015;Lewellen & Robinson, 2013) but does not examine the organizational form choices for these affiliates.More broadly, our findings speak to research on business-group affiliation.Prior literature shows that being affiliated with a business group (or an MNC) can affect innovation (Belenzon & Berkovitz, 2010), risk sharing (Khanna & Yafeh, 2005), the informativeness of earnings (Hong, Kalcheva, Kim, & Yi, 2017), and earnings management (Fan & Wong, 2002).Due to focusing either on the entire firm or individual corporate affiliates, these studies do not consider the implications of organizational form choices for other firm decisions.We inform this literature by documenting that MNCs frequently choose non-corporate flow-throughs and by showing that the organizational form selected for a foreign affiliate can affect subsequent decisions of the MNC.The latter result suggests that organizational form choices could moderate the economic effects of business-group affiliation.
Second, our study provides novel evidence for the effect of taxes on organizational form choices by identifying international taxation as an important determinant of these decisions when taken by MNCs.This result contrasts prior studies on domestic or standalone firms either finding that non-tax factors dominate organizational choices (Ayers, Cloyd, & Robinson, 1996;Gordon & MacKie-Mason, 1994;MacKie-Mason & Gordon, 1997) or documenting tax effects that are limited to industry-specific tax incentives (Hodder, McAnally, & Weaver, 2003;Petroni & Shackelford, 1995;Utke, 2019).Further, by showing that the effect of international taxation varies with the relative importance of tax-motivated income shifting, limited liability, legal independence, and local knowledge, our study sheds light on the trade-offs MNCs face in organizational form choices.We also find that new affiliates established as flow-throughs exhibit a lower loss propensity and are less profitable than those established as subsidiaries.In providing these results, we respond to a call for more research on the costs and benefits of holding ownership stakes in flow-throughs (Hanlon & Heitzman, 2010, p. 156), which has not yet received attention.
Our study has several implications for tax policy.By documenting an economically meaningful effect of international taxation on the organizational form choices of MNCs, our results suggest that these decisions can be particularly sensitive to tax-law changes. 6Moreover, recent proposals to reform the international tax system (Foss, Mudambi, & Murtinu, 2019;McGaughey & Raimondos, 2019), such as the OECD BEPS Action Plan or the Two-Pillar Solution, consider withholding taxes as one way to combat tax-base erosion (OECD, 2013(OECD, , 2021)).Our results indicate that, in order to be effective, such policies have to be implemented uniformly across organizational forms.
Our findings are also informative for researchers and investors.Since we find that MNCs frequently choose a flow-through organizational form for their cross-border investment, studies that focus on corporate subsidiaries are likely to neglect a substantial share of an MNC's cross-border economic activities.Moreover, our results suggest that the organizational form of foreign investment could signal differences in loss propensity and profitability across affiliates.Finally, our study informs future research on the economic effects of the TCJA.By transforming the U.S. worldwide tax system into a ''quasi'' territorial system, the reform might strengthen the impact of dividend-withholding taxes on the group structures of U.S. MNCs.

International Taxation and Organizational Forms
We study investing entities that select an organizational form for a newly established foreign affiliate.As depicted in Fig. 1, the investing entity could be located either in the same country as the parent of the MNC ('direct investment') or in a different country ('indirect investment').We group the organizational forms available to the investing entity into two categories: (i) subsidiaries and (ii) flow-throughs.Subsidiaries are corporate forms that are legally independent from the remaining operations of the MNC.Flow-throughs, in contrast, are non-corporate forms, such as partnerships or branches, that legally belong to the investing entity. 7 Income earned in a foreign subsidiary and in a foreign flow-through is subject to three layers of taxation that affect the after-tax return on the foreign investment.These layers include (i) the host country corporate income tax, (ii) the dividend-withholding tax on a subsidiary's dividend distributions, and (iii) the corporate income tax in the home country of the investing entity (Barrios et al., 2012).
If subsidiaries and flow-throughs are taxed differently along one of these dimensions, the tax burden on foreign income differs between the two organizational forms.Specifically, a subsidiary implies a higher tax burden, and therefore a lower after-tax return on the foreign investment, if the investing entity is located in a country that exempts foreign income from tax ('territorial tax system') or in a country that does not grant any relief from double taxation.This difference results from the dividend-withholding tax not being offset by a home-country tax credit.If the investing entity is located in a country that taxes foreign income but grants a tax credit for foreign taxes paid ('worldwide tax system'), the extent to which a subsidiary implies a higher tax burden depends on the host and home country corporate income tax rates as well as the extent of the tax credit (Kohlhase & Pierk, 2020).The dividend-withholding tax is again the main driver of the tax burden difference.We note that the dividend-withholding tax only applies whenever the income of a subsidiary is distributed to the investing entity.Hence, this tax has the character of a repatriation tax and the investing entity can reduce its effective burden by deferring the repatriation of foreign income (Foley, Hartzell, Titman, & Twite, 2007). 8 For each home-country-year in our sample, we calculate the tax burden difference between the two organizational forms and determine the tax cost of a subsidiary.We take statutory corporate income tax rates, dividend-withholding tax rates, the home country tax system, and double tax treaties into account.As a result, the tax burden difference depends on countries' international tax rules and varies conditional on the home country of the investing entity, inducing country-pair level variation in the tax cost of a subsidiary that is fairly exogenous to organizational form choices.We provide details on our approach in the Online Appendix.

Prior Research
The economic activities of MNCs often involve group structures with affiliates located in several countries (ICIJ, 2014) and prior research suggests that taxes can shape specific elements of these structures.On the one hand, MNCs strategically locate their foreign subsidiaries to minimize the withholding tax on intra-firm dividend distributions (Dyreng et al., 2015;Lewellen & Robinson, 2013).Similarly, Kohlhase and Pierk (2020) find that a worldwide tax system in the MNC home country reduces the incentives for tax management in foreign subsidiaries due to imposing additional home-country tax on cross-border dividend distributions.On the other hand, prior research finds that the tax-motivated presence of MNCs in tax havens reduces geographic transparency and could thus be inefficient (Foss et al., 2019).
In contrast to our study, prior studies do not consider the organizational form of foreign affiliates and evidence for the effect of taxes on organizational form choices is limited to domestic settings or standalone firms.Gordon and MacKie-Mason (1994), for instance, use industry-level data and find that the non-tax costs of non-corporate forms drive incorporation in the trade, service, and manufacturing sector.Similarly, MacKie-Mason and Gordon (1997) show that differences in the taxation of organizational forms affect the allocation of assets and taxable income between the corporate and non-corporate sector.Yet, the authors conclude that non-tax determinants dominate organizational form choices.More recent studies using firm-level data provide more nuanced results.Romanov (2006) studies closely held firms and finds that high personal income tax rates incentivize high-income individuals to shift income into lightly taxed corporate forms.Other studies focus on single industries and find that industry-specific tax incentives can outweigh the non-tax costs of organizational forms in narrow settings.For example, Hodder et al., (2003) find The investing entity selects an organizational form for a new affiliate in the host country (Germany).The investing entity is part of an MNC and located either in the same country as the parent (e.g., the United States: direct investment) or in a different country (e.g., Luxembourg: indirect investment).The three layers of international taxation determine the tax burden on income earned in a foreign subsidiary and a foreign flow-through, respectively.Layer 1 is the statutory corporate income tax in the host country (s host that the opportunity to avoid dividend taxes and alternative minimum taxes increases a bank's probability of converting from a taxable C-corporation to a non-taxable S-corporation. While the above studies examine the determinants of organizational form choices, evidence for the economic consequences of these decisions is scarce.In a cross-country setting, Demirguc-Kunt et al., (2006) find, on average, similar growth rates for corporate and non-corporate entities.Corporate forms, however, alleviate financing frictions and grow faster in countries with high-quality financial and legal institutions.The tax benefits of a particular organizational form can also affect subsequent decisions.Donohoe, Lisowsky, and Mayberry (2019) study the choice of U.S. banks to convert from a C-corporation to a tax-favored S-corporation and find that banks striving to be more competitive invest potential tax savings in advertising and higher deposit rates.Similarly, Utke (2019) investigates equity carve-outs into master limited partnerships (MLPs).His results suggest that tax-sensitive shareholders increase their ownership stake in the more lightly taxed MLP, whereas tax-exempt shareholders invest more heavily in the tax-disadvantaged parent firm.
Since we study the organizational form choice for foreign investment, our study also relates to research on MNCs' internal capital markets.These markets add value because they allow MNCs to raise external capital and to direct economic resources to their most efficient use (Stein, 2003).Consistent with these arguments, Desai, Foley, and Hines (2004) show that U.S. MNCs borrow less in countries with underdeveloped capital markets or weak creditor rights, which suggests that internal capital can substitute for costly external financing.Taxes can also influence internal capital markets and shape the allocation of capital.Specifically, several studies find that, due to the deductibility of interest payments, debt financing of a foreign affiliate is sensitive to the local corporate income tax rate and to tax-rate differences within the MNC (Desai et al., 2004).Egger, Keuschnigg, Merlo, and Wamser (2014) show that internal capital market frictions could mitigate or amplify these tax effects.

Hypotheses Development
An organizational form choice for a foreign affiliate becomes necessary if the firm prefers investing abroad over alternative modes to enter a foreign market, such as exporting, licensing, or forming a joint venture.A firm's foreign investment decision depends on firm-specific and country-specific advantages, complemented by industry-, nation-, and region-specific factors (Buckley & Casson, 1976).For example, an MNC needs sufficient nonlocation bound firm-specific advantages that can be exploited globally to benefit from investing abroad and to operate via a foreign affiliate.In contrast, if firm-specific advantages are location-bound, establishing an affiliate abroad may lead to unexpected costs for the MNC (Rugman & Verbeke, 1992).Establishing a foreign affiliate is preferable if the benefits of developing, deploying, and exploiting firm-specific advantages are higher than the benefits of exporting or forming joint ventures, as well as if the benefits of foreign investment exceed the costs of doing business abroad (Rugman & Verbeke, 2003).
Once the MNC decides to invest abroad and to establish a foreign affiliate, the investing entity selects the organizational form that maximizes the after-tax return on the foreign investment.As noted above, the dividend-withholding tax applies only to the dividend distributions of a subsidiary while a flow-through is not subject to this tax.Therefore, the dividend-withholding tax is the main tax cost (benefit) of a subsidiary (flowthrough) in cross-border settings.Additionally, an investing entity located in a country with a worldwide tax system can offset its taxable income with tax losses incurred in a foreign flow-through.Such a cross-border loss offset is typically not available for subsidiaries. 9 Aside from these tax aspects, several non-tax costs and benefits can influence organizational form choices in cross-border settings.For instance, a subsidiary is legally independent and offers limited liability, shielding the MNC from legal claims against the new affiliate (Gordon & MacKie-Mason, 1994).The difference in the investing entity's liability exposure matters because claims from creditors, employees, and customers against the new affiliate can expose the MNC and its operations to risk.Limited liability also limits the MNC's financial loss to the equity stake in the new affiliate, reducing bankruptcy risk associated with the foreign investment (Bianco & Nicodano, 2006).Moreover, the non-tax benefit of legal independence could be relevant for an MNC applying for a patent in Germany.If granted to a subsidiary, the asset would legally be attributed to the German affiliate.Otherwise, the patent would be attributed to the investing entity.Since Germany has ratified the European Patent Convention (EPC), an MNC holding a patent in a German subsidiary can benefit from patent projection in 38 countries (Hall & Helmers, 2019).
Another non-tax difference between subsidiaries and flow-throughs concerns the transfer of ownership in the new affiliate.While shares in a subsidiary can be transferred rather easily and at low cost, changing ownership in a flow-through requires the transfer of assets or the re-negotiation of the partnership agreement (Ayers et al., 1996).Minimum-capital and registration requirements, on the other hand, imply that setting up a subsidiary is more costly than setting up a flowthrough.Moreover, a subsidiary implies higher compliance and coordination costs due to stricter regulatory and financial-reporting requirements and the separation of ownership and control (Goolsbee & Maydew, 2002). 10 We expect the investing entity to consider these tax and non-tax costs and benefits and to select the organizational form with the best cost-benefit relation.Since taxes reduce the after-tax return on foreign investment, we expect the probability of choosing a flow-through to increase in the tax cost of a subsidiary.Based on these arguments, we posit the following baseline hypothesis: Hypothesis 1: The tax cost of a subsidiary is positively associated with the probability of establishing a flow-through.
The relation under Hypothesis 1 assumes that an MNC repatriates foreign income via dividends subject to international taxation.However, an MNC could limit the foreign affiliate's exposure to international taxation by shifting its income across borders (Dyreng & Markle, 2016).Common shifting strategies exploit discretion in setting intra-firm transfer prices (Klassen & Laplante, 2012) or make use of intra-firm interest or royalty payments as well as cost-sharing arrangements (De Simone, Mills, & Stomberg, 2019;Hopland, Lisowsky, Mardan, & Schindler, 2018).Since income shifting mitigates the tax cost of a subsidiary, we expect an MNC that shifts more income to be less sensitive to the tax burden difference between organizational forms. 11These arguments suggest the following cross-sectional hypothesis: Hypothesis 2: The association between the tax cost of a subsidiary and the probability of establishing a flow-through is weaker for MNCs that shift more income.
With respect to non-tax determinants, the benefit of limited liability seems particularly valuable in capital-intensive industries (Liu, 2014).Specifically, the likelihood and the extent of financial losses for the MNC could be higher in industries such as manufacturing or wholesale, potentially altering the cost-benefit relation of organizational forms.We therefore expect the tax burden difference to be less relevant for organizational form choices in capital-intensive industries, leading to the following cross-sectional hypothesis: Hypothesis 3a: The association between the tax cost of a subsidiary and the probability of establishing a flow-through is weaker for new affiliates in capital-intensive industries.
An MNC that engages in foreign investment is subject to several sets of regulation.Low regulatory quality in the parent home country, such as the inability of the government to implement and maintain stable regulation, could pose significant risk to the MNC and its cross-border investment (Dikova, Sahib, & van Witteloostuijn, 2010).While establishing a legally independent subsidiary prevents risk associated with low regulatory quality from spilling over to the new affiliate, a flowthrough legally belongs to the investing entity.As a result, low regulatory quality could impact the new affiliate and its operations.Since a subsidiary's legal independence shields cross-border investments from risk, we expect an MNC subject to low regulatory quality to be less sensitive to the tax burden difference.This leads to the following crosssectional hypothesis: Hypothesis 3b: The association between the tax cost of a subsidiary and the probability of establishing a flow-through is weaker for MNCs located in countries with low regulatory quality.
Establishing a new affiliate often coincides with a market entry in the FDI host country.An MNC that enters a foreign market for the first time has little knowledge about local market conditions and faces strategic decisions, such as how to finance the market entry or how much to invest in the host country.An MNC with prior host-country experience, in contrast, is knowledgeable about local market conditions and should have access to local tax knowledge (Feller & Schanz, 2017).Thus, we expect an investing entity with local knowledge to be more sensitive to the tax burden difference.
Based on these arguments, we state the following cross-sectional hypothesis: Hypothesis 3c: The association between the tax cost of a subsidiary and the probability of establishing a flow-through is stronger for MNCs with local knowledge.

MiDi Database and Supplementary Data Sources
We use the MiDi database of the Deutsche Bundesbank as our primary data source.The MiDi database is a below firm-level panel dataset that provides anonymized micro-level data on the stock of inand outbound FDI relations in Germany, starting in the year 1999.According to the German Foreign Trade and Payments Regulation (Außenwirtschaftsverordnung), an inbound FDI relation has to be reported to the Deutsche Bundesbank if an investing entity holds at least 10% of the shares or voting rights of a subsidiary or a partnership located in Germany with a balance sheet total of more than EUR 3 million.A German branch or a permanent establishment held by an investing entity has to be reported if the business assets amount to EUR 3 million and more (Blank et al., 2020).
The key advantage of the MiDi database is that we can identify the organizational form of an affiliate.This information is limited in conventional data sources, such as Orbis or Compustat.Data available for the German affiliate includes an identifier, industry affiliation, annual balance sheet positions, as well as annual turnover, after-tax profit, and the number of employees. 12Penalties and data appraisal techniques ensure high data quality (Friederich, Pham-Dao, Schild, Scholz & Schumacher, 2021).Being an FDI database, the dataset also provides the percentage of shares held by the investing entity and the share of the affiliate's assets and liabilities attributable to the investing entity or to other affiliates of the MNC (e.g., intra-firm debt).The main drawback of the MiDi database is the lack of income-statement information and that information on the investing entity and the MNC's parent is limited to identifiers for the investing entity and for the respective home countries.We are therefore unable to conduct indepth analysis on the parent and the MNC's operations or affiliates outside of Germany.
Although the MiDi database restricts our analysis to Germany, we have identified an ideal setting to test our predictions for several reasons.First, Germany has an extensive tax treaty network that implies country-pair-level variation in the tax burden difference between organizational forms (conditional on the home country of the investing entity).Since all organizational form choices in our sample are subject to identical host country regulation, we exploit this variation while holding constant the non-tax characteristics of organizational forms.Second, tax systems follow common principles so that inferences based on Germany should generalize to most FDI host countries (Barrios et al., 2012).Third, we observe organizational form choices for newly established affiliates.Since these decisions are unobservable in industry-level data (Gordon & MacKie-Mason, 1994;Liu, 2014;MacKie-Mason & Gordon, 1997) or when analyzing pre-existing firms (Ayers et al., 1996), our setting offers strong identification for the effect of international taxation on the organizational form choices of MNCs.Finally, Germany is a G7 country where MNCs invest with the intention to establish a long-term presence.Thus, MNCs are unlikely to establish German affiliates with the main purpose of lowering the group tax burden.
We supplement our data with information on dividend-withholding tax rates, home and host country corporate income tax rates, and the home country tax system from publicly available corporate tax guides (e.g., Ernst & Young, 2005-2013).The data for control variables stem from various sources, including the World Bank's regulatory quality indicators database (World Bank, 2005-2013), Gleditsch (2013), and Thomson Reuter's Datastream.We specify the respective data sources in the Online Appendix.

Sample Selection
To identify an organizational form choice, we focus on the first observation of an inbound FDI relation in Germany between 1999 and 2013 (18,265 observations). 13We first drop observations prior to 2005 because information to differentiate newly established affiliates from pre-existing affiliates is unavailable prior to 2005 (12,206 observations).Second, we drop observations where the investing entity holds less than 25% of the shares in the new affiliate (360 observations).Since strategic decisions under German corporate law require the consent of more than 75% of the shareholders, the 25% threshold ensures that the investing entity can actively influence the organizational form choice for the new affiliate. 14Third, we drop observations with insufficient data to calculate our regression variables (26 observations).Lastly, we exclude observations of pre-existing affiliates where the first observation in the database results from overshooting the reporting threshold (3498 observations).For the years 2005 to 2013, these restrictions result in a sample of 2,175 organizational form choices. Table 1 summarizes the sample selection.

Research Design
To model an organizational form choice, we estimate the following logistic regression: Pr Flow-Through i ð Þ is the probability that an investing entity establishes new affiliate i as a flow-through. 15The dependent variable, Flow-Through, is an indicator variable equal to one if affiliate i is established as a flow-through, and zero if the affiliate is established as a subsidiary.
Taxwedge is our main variable of interest and denotes the tax burden difference between organizational forms.As discussed above, we calculate Taxwedge for each home-country-year in our sample.Since Taxwedge in year t varies conditionally on the home country of the investing entity (country h), this variable captures variation in the tax cost of a subsidiary that is fairly exogenous to organizational form choices.We expect a positive coefficient on b 1 , consistent with the tax cost of a subsidiary being positively associated with the probability of establishing affiliate i as a flowthrough.
Vector X includes control variables for determinants of organizational form choices. First, we control for characteristics of affiliate i.We include LN(Employ) as the logarithm of employees and LN(Assets) as the logarithm of total assets to proxy for the size of the investment in affiliate i (Liu, 2014).Further, we include Roa as net profit over total assets and LossYear as an indicator variable equal to one if affiliate i reports a loss.While Roa controls for profitability, LossYear proxies for the propensity to incur a loss in affiliate i.A higher loss propensity increases the benefit of limited liability (Ayers et al., 1996). 16LossYear also captures potential tax benefits from a cross-border loss offset under a flow-through.We include Leverage to capture debt financing of the new affiliate and to control for debtholder demand for financial-statement information (Armstrong, Guay, & Weber, 2010).We add Brownfield as an indicator variable equal to one if the MNC establishes affiliate i through a merger and acquisition (M&A), and zero if affiliate i is newly founded.M&A is more likely to involve subsidiaries because their shares are easier to transfer.Since the MNC's repatriation strategy can influence the effective tax cost of a subsidiary (Foley et al., 2007), we include the indicator variable Distribution to control for profit distributions of affiliate i. 17 We also control for taxmotivated income shifting, which could affect the tax costs and benefits of organizational forms (De Simone et al., 2019;Hopland et al., 2018).Specifically, we add InternDebt as the ratio of intra-firm debt to total assets and Patents as an indicator variable equal to one if affiliate i operates in a patent-intensive industry (Hall, Helmers, Rogers, & Sena, 2014).These variables capture (i) intra-firm debt and (ii) patents as two key strategies MNCs use to relocate the income of their affiliates for tax purposes (Heckemeyer & Overesch, 2017).
Second, we control for expectations of the MNC that could influence the expected non-tax costs and benefits of organizational forms.Specifically, we include PE-Ratio as the median price-to-earnings (PE) ratio in year t, Industry-Roa as the median return on assets in year t, and SD-Industry-Roa as the standard deviation of the median return on assets over the 3-year period t to t+2.We follow Shroff, Verdi, and Yu (2014) and calculate all three variables based on data from publicly listed firms operating in the same industry as affiliate i. 18 These industry-level measures capture expectations at the time of the organizational form choice regarding growth opportunities (PE-Ratio), profitability (Industry-Roa), and risk (SD-Industry-Roa) for the investment in affiliate i. 19  Third, we control for characteristics of the investing entity.We add NumInv as the investing entity's total number of inbound FDI relations in Germany because the incentives to consider the tax cost of a subsidiary when establishing affiliate i could vary with the size of the operations in Germany.Moreover, a high number of FDI relations indicates local knowledge and suggests that risk is spread across multiple affiliates.We include the percentage of shares held by the investing entity in affiliate i (Holdings) as a control for the decision to cooperate with other investors.Cooperation requires commitment, for instance through unlimited liability.Finally, we add DirectFDI as an indicator variable equal to one for direct investment, and zero for indirect investment.An MNC might invest indirectly in Germany to exploit preferential dividend-withholding tax rates (Dyreng et al., 2015). 20 Fourth, we control for characteristics of the parent home country.We include LN(Dist) as the distance between Germany and the parent home country to capture coordination costs and crossborder frictions.While coordination costs are lower for flow-throughs (Goolsbee & Maydew, 2002), cross-border frictions increase the benefit of legal independence.We use World Bank's regulatory quality indicators (World Bank, 2005-2013) to construct DiffRegQuality as a control for countrylevel differences in regulatory quality.A subsidiary shields foreign investments from risk associated with low regulatory quality in the parent home country.To control for institutional differences between countries, we follow LaPorta, Lopez-de-Silanes, Shleifer, and Vishny (1998) and include indicator variables for the origin of the home country legal system (LegorUK, LegorFR, LegorSC).
Finally, we include Year and Industry fixed effects to capture year shocks and time-invariant industry characteristics. 21We cluster standard errors at the investing-entity level to account for serial correlation in the data (Petersen, 2009). 22To test our cross-sectional predictions, we add interactions to Eq. ( 1) that partition our sample into subsamples in which we expect the relation between Taxwedge and Flow-Through to vary.We discuss these tests in more detail below.

Sample Composition and Descriptive Statistics
The investing entities in our sample are located in 59 home countries, which account for more than 99% of all inbound FDI relations recorded in the MiDi database.Table 2, Panel A shows the mean of Taxwedge for the home countries in our sample. 23 Since the Parent-Subsidiary-Directive (Council Directive, 2011/96/EU) abolished withholding taxes on dividend distributions within the EU, Taxwedge has a mean of zero for most EU countries.In contrast, Taxwedge is relatively high for tax havens, such as Jersey.Germany does not sign double tax treaties with tax-haven countries so that the dividend-withholding tax becomes a final economic burden on a subsidiary's dividend distributions to these countries.Taxwedge ranges from 0 to 13.12% of pre-tax income earned in affiliate i, which suggests that the tax burden difference between organizational forms can be a substantial tax cost for the MNC. 24Taxwedge also varies over time as we record a total of 83 changes (45 increases and 38 decreases).Sixty-two of these changes occur around a 2008 tax reform in Germany.The remaining changes are scattered across time and result from changes in home country corporate income tax rates, home country tax systems, or dividendwithholding tax rates.
The 2175 organizational form choices in our sample result in 1645 subsidiaries and 530 flowthroughs, respectively.Hence, the unconditional probability of establishing a flow-through is 24.37% (= 530/2175).We observe the highest number of new affiliates for investing entities located in neighboring countries, such as Luxembourg, the Netherlands, Switzerland, and Austria, and major economies, such as the United States and the United Kingdom.Consistent with home-country variation in Taxwedge, the relative importance of organizational forms varies across countries. 25 Panel B presents organizational form choices by sample year.The number of new affiliates increases in the early sample years and then again after 2010.  3 presents the descriptive statistics for the full sample and separately for subsidiaries and flowthroughs.The average affiliate reports total assets of EUR 31.90 million, which suggests that the organizational form choices in our sample concern sizeable cross-border investments.Roa is negative on average, which is plausible given that we focus on affiliates in their first year of operation.We conduct t tests (Wilcoxon rank-sum tests) to assess differences in means (medians) between subsamples.The mean of Taxwedge is significantly larger for flowthroughs (p value = 0.005), consistent with investing entities being sensitive to tax burden differences in organizational form choices.The difference in medians is insignificant (p value = 0.556). 27Differences in means and medians for the remaining variables are consistent with our expectations.
Table 4 presents Pearson correlation coefficients.In line with the descriptive statistics in Table 3, the correlation between Taxwedge and Flow-Through is positive and significant (p value = 0.001).This result suggests that the tax cost of a subsidiary is positively associated with the probability of establishing a flow-through, providing preliminary support for Hypothesis 1. Correlations between the remaining variables are in line with the descriptive statistics.

Tests of Hypothesis 1
To test Hypothesis 1, we estimate Eq. ( 1) and present the results in Table 5.We standardize independent variables to have a mean of zero and a standard deviation of one prior to estimating regressions, which facilitates the comparison of (1)  estimates across variables and empirical specifications.We exclude year and industry fixed effects in column 1 and estimate the full model in column 3.
As predicted, we find positive and significant coefficients on Taxwedge in column 1 (p value = 0.001) and in column 3 (p value \ 0.001).These results suggest that the tax cost of a subsidiary is positively associated with the probability of establishing affiliate i as a flow-through.Results for the control variables are as expected. 28For instance, the probability of establishing a flow-through is negatively associated with affiliate size (LN(Employ) and loss propensity (LossYear) but positively associated with profit distributions (Distribution) and the number of FDI relations held by the investing entity (NumInv).
The area under the receiver operating characteristic ('ROC') curve suggests that our regression models exhibit reasonable predictive power (Hosmer, Lemeshow, & Sturdivant, 2013).For instance, the regression in column 3 predicts the correct organizational form for 79.1% of the observations in our sample.Comparing the area under the ROC curve for the regressions in columns 1 and 3, we note that year and industry fixed effects significantly increase the predictive power of our model (p value \ 0.001).This indicates that the determinants included in our model together with year and industry characteristics predict organizational form choices in our sample.
To gauge the economic significance of our results, we calculate marginal effects for all independent variables.In column 4, a one standard deviation increase in Taxwedge (i.e., an increase by 3.51 percentage points) is associated with a 4.06 percentage point higher probability of establishing a flow-through (p value \ 0.001, 95% confidence internal ('CI') between 1.76 and 6.37 percentage points).In comparison, a one standard deviation increase in Leverage implies a 3.73 percentage point lower probability of establishing a flow-through (p value = 0.012, 95% CI between -6.63 and -0.83 percentage points).The probability to establish a flow-through decreases by 7.54 percentage points for a one standard deviation increase in Roa (p value \ 0.001, 95% CI between -10.20 and -4.88 percentage points). 29These results suggest that the economic effect of Taxwedge on the organizational form choices of MNCs is sizeable and comparable to the effect of continuous non-tax determinants. 30 Overall, these results support Hypothesis 1: The probability of establishing a flow-through increases with the tax cost of a subsidiary.Hence, the tax burden difference between organizational forms is an important determinant of the organizational form MNCs select for foreign investment.

Tests of Hypothesis 2
Hypothesis 2 predicts that the tax burden difference is less relevant for organizational form choices if the MNC engages in more tax-motivated income shifting.To test this prediction, we estimate a modified version of Eq. ( 1) and present the results in Table 6.In column 1, we interact Taxwedge with HighIntDebt, which is an indicator variable equal to one if intra-firm debt provided to the new affiliate by foreign affiliates of the MNC is above the sample median.Since intra-firm interest payments allow an MNC to repatriate foreign income without This table presents results for the association between the tax burden difference and organizational form choices.The dependent variable, Flow-Through, is an indicator variable equal to one if the new affiliate is established as a flow-through, and zero otherwise.Columns 1 and 3 (2 and 4) report coefficients (marginal effects) for a logistic regression based on Eq. ( 1).We calculate marginal effects while holding continuous variables at their means.We standardize independent variables to have a mean of zero and a standard deviation of one prior to estimating regressions.The regression in column 1 (3) is estimated without (with) year and industry fixed effects.We winsorize continuous variables at the 1st and 99th percentile.We calculate heteroscedasticity-robust standard errors clustered at the investing-entity level.We define variables in the Online Appendix.Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations.
triggering the dividend-withholding tax, we expect a lower tax sensitivity for new affiliates with high intra-firm debt.Consistent with this expectation, the coefficient on Taxwedge*HighIntDebt is negative and significant (p value = 0.002).An F-test indicates that the joint effect of Taxwedge and HighIntDebt (b1+b3) is indistinguishable from zero (p value = 0.428), consistent with income shifting fully attenuating the effect of international taxation on organizational form choices.In column 3, we interact Taxwedge with Patents to proxy for income shifting via intra-firm royalty payments.Consistent with the previous test, we find a negative and significant coefficient on Taxwedge*Patents (column 3, p value = 0.056).The joint effect of Taxwedge and Patents (b1+b5) is again indistinguishable from zero (p value = 0.250).
In sum, these results support Hypothesis 2: The tax cost of a subsidiary is less relevant for organizational form choices of MNCs that engage in more tax-motivated income shifting.This repatriation strategy limits the extent to which an MNC's foreign income is exposed to international taxation, attenuating its effect on organizational form choices.

Tests of Hypotheses 3a-c
Hypotheses 3a-c predict that non-tax factors moderate the tax sensitivity of organizational form choices.To test the hypotheses, we again modify Eq. ( 1) and present the results in Table 7.In column 1, we test Hypothesis 3a and interact Taxwedge with HighIndRisk, which is an indicator variable equal to one if the new affiliate is established in the manufacturing or wholesale industry.Since the non-tax benefit of limited liability should be valuable in these capital-intensive industries, we expect Taxwedge to be less relevant.As expected, the coefficient on Taxwedge*HighIndRisk is negative and significant (p value = 0.084). 31The joint effect of Taxwedge and HighIndRisk (b1+b3) is indistinguishable from zero (p value = 0.397), which  1).We calculate marginal effects while holding continuous variables at their means.We standardize independent variables to have a mean of zero and a standard deviation of one prior to estimating regressions.All regressions are estimated with year and industry fixed effects.We winsorize continuous variables at the 1st and 99th percentile.We calculate heteroscedasticity-robust standard errors clustered at the investing-entity level.We define variables in the Online Appendix.Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations suggests that the benefit of limited liability fully attenuates the tax sensitivity of organizational form choices in capital-intensive industries.
In column 3, we test Hypothesis 3b and interact Taxwedge with HighRegDiff, which is an indicator variable equal to one if the difference in regulatory quality between Germany and the parent home country is above the sample median.Since a legally independent subsidiary shields cross-border investments from risk associated with low regulatory quality in the parent home country, we expect the tax cost of a subsidiary to become less relevant for organizational form choices.In line with this prediction, we find a negative and significant coefficient on Taxwedge*HighRegDiff (p value = 0.069); the joint effect of Taxwedge and HighRegDiff (b1+b5) is different from zero (p value = 0.039).These results suggest that low regulatory quality in the parent home country mitigates the effect of Taxwedge on organizational form choices.
In column 5, we test Hypothesis 3c and interact Taxwedge with HighExp, which is an indicator variable equal to one if the investing entity reports at least one additional inbound FDI relation in Germany at the time it establishes affiliate i.Such an MNC is familiar with local market conditions and has local tax knowledge.This knowledge facilitates the design of tax-efficient group Columns 1, 3, and 5 (2, 4, and 6) report coefficients (marginal effects) for a logistic regression based on Eq. (1).We calculate marginal effects while holding continuous variables at their means.We standardize independent variables to have a mean of zero and a standard deviation of one prior to estimating regressions.All regressions are estimated with year and industry fixed effects.We winsorize continuous variables at the 1st and 99th percentile.We calculate heteroscedasticity-robust standard errors clustered at the investing-entity level.We define variables in the Online Appendix.Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations structures, increasing the relevance of tax costs and benefits for organizational form choices.In support of this prediction, the coefficient on Taxwedge*-HighExp is positive and significant (p value = 0.007). 32These results suggest that local knowledge amplifies the effect of Taxwedge on organizational form choices.
Collectively, these results support Hypotheses 3a-c: Non-tax factors moderate the effect of the tax burden difference on organizational form choices.The results based on differences in capital intensity and regulatory quality indicate that MNCs highly value a subsidiary's non-tax benefits of limited liability and legal independence.Thus, in settings where these non-tax benefits are important, MNCs might respond weakly (or not at all) to tax-law changes in their organizational form choices.

Addressing Endogeneity Concerns
Although the tax cost of a subsidiary varies at the country-pair level and is therefore fairly exogenous to MNCs' organizational form choices, one concern with our analysis so far is that unobserved MNC characteristics could be jointly correlated with Taxwedge and the chosen organizational form.If not adequately controlled for, such variables could drive our results.We conduct three additional analyses to help address this concern.
First, we modify Eq. ( 1) and replace industryfixed effects with a fixed effect for the home country of the parent.This design controls for common characteristics of MNCs from a given country that could be correlated with both Taxwedge and the organizational form of affiliate i. 33  Consistent with our main results, the coefficient on Taxwedge remains positive and significant (p value = 0.07, untabulated).
Second, we re-estimate Eq. ( 1) using annual changes in Taxwedge.By focusing on changes in the tax cost of a subsidiary as opposed to levels, this test rules out that MNC characteristics correlated with the level of Taxwedge drive our results.The coefficient on DTaxwedge is positive and significant (p value = 0.072, untabulated).Thus, increases (decreases) in the tax cost of a subsidiary are associated with a higher (lower) probability of establishing affiliate i as a flow-through, corroborating our main results.
Third, we further tighten the changes analysis and employ a quasi-natural experiment around a 2008 tax reform in Germany.The reform was enacted in August 2007 and implemented in two steps.First, effective from January 1, 2008, the corporate income tax rate (including the local business tax; see the Online Appendix) was reduced by 9 percentage points (OECD, 2019).In a second step, the dividend-withholding tax was reduced by 10.6 percentage points, effective from January 1, 2009.While the lower corporate income tax rate led to a temporary increase in Taxwedge, the effect was more than offset by the reduction in the dividend-withholding tax.A strength of the setting is that the reform occurred in Germany -the FDI host country of our study.Moreover, although the reform changed two general elements of the German corporate tax system, it affected only a subset of home countries in our sample.Specifically, for investing entities located in countries with a nonzero Taxwedge prior the reform (44 countries), the reform led to an average decrease in Taxwedge by 1.47 percentage points.Yet, the reform did not alter the tax cost of a subsidiary for 15 home countries with a pre-reform Taxwedge of zero.As a result, the setting provides quasi random variation in the tax cost of a subsidiary that we can exploit in a difference-in-differences ('DiD') analysis.Taken together, these reform features make it unlikely that characteristics of foreign MNCs drive the observed changes in Taxwedge. 34 To implement a DiD analysis, we modify Eq. (1) as follows: Reform is an indicator variable equal to one for home countries of investing entities that experienced a change in Taxwedge through the reform (treatment group), and equal to zero if Taxwedge is zero during the entire sample period (control group).Post is an indicator variable equal to one for organizational form choices after the reform (2008 or later).The interaction Reform*Post assesses whether the reform altered the probability of establishing a flow-through for the treatment group relative to home countries unaffected by the reform.Since the reform led to a net decrease in Taxwedge, we expect a lower probability of establishing a flow-through after the reform, suggesting a negative coefficient on b 3 . 35 We present the results in Table 8.In column 1, the coefficients on Reform and Post are positive and significant (p values = 0.003 and 0.002).These results suggest that the probability of establishing a flow-through is generally higher for investing entities from countries affected by the reform and for observations in the post-reform period.Importantly, we find a negative and significant coefficient on Reform*Post (p value = 0.070).This result indicates that, relative to the control group, the probability of establishing a flow-through significantly decreased in the post-reform period for investing entities located in a country affected by the reform.In column 2, we replace Post with a set of year indicators to assess whether treatment and control group exhibit similar pre-reform trends in the probability of establishing a flow-through.Using 2007 as a reference year, we find insignificant coefficients on Reform*Year2005 and Reform*-Year2006 (p values = 0.375 and 0.616).These results indicate parallel pre-reform trends in the probability of establishing a flow-through, supporting the main identifying assumption of a DiD analysis.Coefficients on the remaining interactions are negative and mostly significant from 2010 onwards (p values = 0.069, 0.136, 0.075, and 0.095).This result is plausible because the overall effect of the staggered reform was fully effective only after 2009.
In sum, the analyses in this section indicate that our inferences hold across multiple empirical  2).We calculate marginal effects while holding continuous variables at their means.We standardize independent variables to have a mean of zero and a standard deviation of one prior to estimating regressions.All regressions are estimated with year and industry fixed effects.We winsorize continuous variables at the 1st and 99th percentile.We calculate heteroscedasticity-robust standard errors clustered at the investing-entity level.We define variables in the Online Appendix.Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations specifications and approaches, alleviating endogeneity concerns in the determinants analysis.We therefore conclude that the tax cost of a subsidiary has a likely causal effect on the organizational form choices of MNCs.and 6) of panel A (B) report marginal effects.We calculate marginal effects while holding continuous variables at their means.We standardize independent variables to have a mean of zero and a standard deviation of one prior to estimating regressions.All regressions are estimated with year and industry fixed effects.We winsorize continuous variables at the 1st and 99th percentile.We calculate heteroscedasticity-robust standard errors clustered at the investing-entity level.We define variables in the Online Appendix.Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations

Robustness Tests
To further assess the robustness of our main result, we report additional robustness tests in Table 9.
Panel A shows the results for tests using modified samples of organizational form choices. First, we assess whether a few large MNCs drive our results and drop observations of investing entities accounting for more than one new affiliate in our sample.Although this step reduces the sample to 1,659 observations, the coefficient on Taxwedge in column 1 remains positive and significant (p value = 0.016).The marginal effect in column 2 is similar to the baseline estimate (Table 5).Next, we address the concern that Taxwedge is relatively large for investing entities located in tax havens and therefore exclude these observations from our sample.In column 3, we continue to find a positive and significant coefficient on Taxwedge (p value = 0.019), which suggests that our main findings generalize beyond investing entities located in tax havens. 36 Panel B presents the results for additional robustness test.First, we control for the home country tax system (Worldwide) and the corporate income tax burden (CIT). 37Since withholding taxes are a final economic burden under a territorial tax system, MNCs head-quartered in a worldwide tax system might have weaker incentives to select a taxfavored organizational form for affiliate i. 38 Yet, the coefficient on Taxwedge in column 1 remains positive and significant (p value \ 0.001).Second, because Taxwedge is greater than zero primarily for home countries that do not have a double tax treaty with Germany or that are outside the scope of the Parent-Subsidiary Directive, these institutional features rather than the magnitude of Taxwedge could drive our results.To examine this possibility, we include our continuous measure Taxwedge together with an indicator variable taking the value of one if Taxwedge is greater than zero (PositiveTaxwedge).In column 3, the coefficient on Taxwedge remains positive and significant (p value \ 0.001) while the coefficient on PositiveTaxwedge (p value = 0.886) is insignificant.The marginal effect for Taxwedge in column 4 is very close to the baseline estimate (Table 5), consistent with the magnitude of the tax burden difference between organizational forms driving our results.Finally, in column 5, we include the two components of Taxwedge: (i) the dividend-withholding tax (Wht) and (ii) the corporate income tax levied or the tax credit granted in the home country (HomeTax).The coefficients are positive and significant for both components (p values = 0.001 and 0.045), which suggests that the tax sensitivity is not limited to the dividend-withholding tax and MNCs take the taxation of foreign income in the home country into account when selecting an organizational form for a new affiliate. 39

Theoretical Background
Our analysis so far has focused on the determinants of organizational form choices.In this section, we broaden our analysis by examining whether the chosen organizational form shapes future characteristics of the new affiliate.Subsidiaries and flowthroughs differ along several (tax and non-tax) dimensions so that the organizational form could affect subsequent decisions at the affiliate level (Demirguc-Kunt et al., 2006).For instance, limited liability and legal independence can protect the MNC from legal claims against the new affiliate or shield the MNC from bankruptcy risk associated with the investment abroad.An MNC considering these risks should be more likely to allocate risky investment projects, such as investment in research and development or the market entry of new products, to subsidiaries.As a result, compared to new affiliates established as flow-throughs, we expect subsidiaries to exhibit more investment (Coles, Daniel, & Naveen, 2006), higher risk-taking, and a higher propensity to incur losses (Acharya, Amihud, & Litov, 2011;John, Litov, & Yeung, 2008).Since a high-risk investment strategy should yield a higher expected return, we expect subsidiaries to be more profitable than affiliates established as a flow-through.Moreover, since limited liability provides an incentive to spread risky investment projects across several legally independent subsidiaries, new affiliates established as a subsidiary should be associated with a more complex group structure in the FDI host country than affiliates established as a flow-through.To sum up, we expect new affiliates established as flowthroughs to be associated with (i) less risk-taking, (ii) a lower loss propensity, (iii) less investment, (iv) a lower profitability, and (v) a less complex group structure in the FDI host country than affiliates established as subsidiaries.

Sample selection
To test our predictions and to study the future characteristics of newly established subsidiaries and flow-throughs, we extend our initial organizational form choice sample to any observation available for a new affiliate in the MiDi database (sample period: 2005 to 2013).Starting with the year of the organizational form choice, we obtain 6770 affiliate-year observations.The precise sample size varies across the empirical tests because some regression variables require several years of prior data.

Research design
One challenge in examining future characteristics of subsidiaries and flow-throughs is to rule out endogeneity concerns.For instance, the MNC might choose an organizational form anticipating future affiliate-level decisions about investment or risk-taking.Thus, the future characteristics of the new affiliate could be determined already at the time of the organizational form choice rather than resulting from the chosen organizational form.
Further, an unobserved variable could be correlated with the organizational form choice and the future characteristics of the new affiliate.For example, an MNC's repatriation strategy could affect both the choice between a subsidiary and a flow-through and future decisions at the affiliate level.
To address endogeneity concerns also in this setting and to test whether the chosen organizational form is related to future affiliate characteristics, we use a two-step estimation procedure suggested by Wooldridge (2010, p. 939). 40Specifically, we first re-estimate our determinants model, Eq. ( 1), on the organizational form choice sample using a probit regression.From this regression, we obtain fitted probabilities, i.e., the predicted value to establish affiliate i as a flow-through d Flow-Through .In a second step, we test for a relation between the chosen organizational form and the future characteristics of affiliate i using two-stage least squares (2SLS).As suggested by Wooldridge (2010), we instrument Flow-Through, i.e., the chosen organizational form, with d Flow-Through in this estimation.The second stage of the 2SLS-model is defined as follows: Depvar is a set of dependent variables and includes RiskTaking, LossPropensity, Investment, Roa, and Complexity, all measured in year t.We compute RiskTaking as the industry-year adjusted standard deviation of affiliate i's return on assets (Roa) over the 3-year period t to t+2 (Langenmayr & Lester, 2018).LossPropensity is an indicator variable equal to one if affiliate i reports a loss over the 3-year period t to t+2.This variable denotes the affiliate's long-term loss propensity and provides an alternative measure for the riskiness of operations.Investment, our proxy for annual investment of affiliate i, is the change in fixed and intangible assets from year t-1 to t, divided by lagged total assets.Roa is net profit over total assets and captures the profitability of affiliate i. Complexity, is the natural logarithm of the number of affiliates held by affiliate i in Germany and proxies for the complexity of the group structure in the FDI host country.
As noted above, Flow-Through is the organizational form of affiliate i that we instrument with its predicted values d Flow-Through in the 2SLSmodel.We expect negative coefficients on Flow-Through in all tests, which would suggest that establishing affiliate i as a flow-through rather than as a subsidiary is associated with less risk-taking, a lower loss propensity, less investment, a lower profitability, and a less complex group structure.
To yield consistent estimates, our empirical strategy requires an instrumental variable when estimating d Flow-Through (Wooldridge, 2010).A valid instrument must satisfy both the exclusion restriction and the relevance condition.Regarding the exclusion restriction, we note that Taxwedge captures the tax burden difference between organizational forms.Taxwedge is therefore unlikely to influence future characteristics of affiliate i other than through the chosen organizational form, satisfying the exclusion restriction.Second, the determinants analysis suggests that Taxwedge also satisfies the relevance condition because the variable is an important predictor of organizational form choices. Hence, Taxwedge appears to be a valid instrumental variable for studying the relation between the chosen organization form and the future characteristics of new affiliates.
Vector X is a set of control variables that follow from prior research on investment and risk-taking (Cummins, Hassett, & Hubbard, 1996;Langenmayr & Lester, 2018).LN(Employ), LN(Assets), Roa, Loss-Year, and Leverage control for affiliate size, profitability, the presence of a loss, and the availability of funds.Moreover, we include LN(Sales) as the natural logarithm of sales, Investment as the annual change in fixed and intangible assets over lagged total assets, Cash as cash over total assets, and Age as year t less the year in which affiliate i was established.We also add PE-Ratio, Industry-Roa, and SD-Industry-Roa to capture time-varying industrylevel differences in investment opportunities, profitability, and risk (Shroff et al., 2014).Following Shroff et al., (2014), we lag most control variables by 1 year.We again include year and industry fixed effects and cluster standard errors at the investingentity level.

Descriptive Statistics
Table 10 reports descriptive statistics and results for tests of differences in means and medians between affiliates established as subsidiaries and those established as flow-throughs.Flow-throughs exhibit lower risk-raking (RiskTaking, p value \ 0.001), a lower loss propensity (LossPropensity, p value \ 0.001), less investment (Investment, p value = 0.022), and a lower profitability (Roa, p value \ 0.001) than affiliates established as a subsidiary.Flow-throughs also hold fewer affiliates in Germany (Complexity, p value \ 0.001), consistent with a less complex group structure.Table 11 reports correlation coefficients between our dependent variables and the control variables included in Vector X. Correlations between our dependent variables are mostly weak or insignificant, which suggests that these variables capture distinct characteristics of new affiliates.

Regression Results
Table 12 presents the regression results.In panel A, we present the results of a probit regression based on Eq. (1) that we use to estimate predicted values of Flow-Through (step one).The results in this panel are consistent with Table 5. Panel B shows the results of 2SLS-regressions testing for a relation between the chosen organizational form and the future characteristics of affiliate i (step two).We estimate the 2SLS-model separately for all five dependent variables.Focusing on the second-stage results (columns 2, 4, 6, 8, and 10), we find that (2) (3)    3).Columns 1, 3, 5, 7, and 9 present the first-stage results.Columns 2, 4, 6, 8 and 10 present the second-stage results.In panel A, the dependent variable, Flow-Through, is an indicator variable equal to one if the new affiliate is established as a flow-through, and zero otherwise.In panel B, the dependent variables of the second-stage regressions are defined as follows: RiskTaking is the standard deviation of Roa of the new affiliate over the 3-year period t to t+2 and adjusted for the industry-year mean (based on one-digit NACE Rev. 2 industry codes).LossPropensity is an indicator equal to one if the new affiliate reports a loss over the 3-year period t to t+2.Investment is the annual change in fixed and intangible assets of the new affiliate dividend by lagged total assets.Roa is the net profit of the new affiliate and divided by total assets.Complexity is the natural logarithm of the number of inbound FDI relations of the new affiliate in Germany.The control variables for the second stage regressions LN(Employ), LN(Assets), Roa, Leverage, LN(Sales), Investment, and Cash are lagged by 1 year.We standardize independent variables to have a mean of zero and a standard deviation of one prior to estimating regressions.All regressions are estimated with year and industry fixed effects.All continuous variables are winsorized at the 1st and 99th percentile.We calculate heteroscedasticity-robust standard errors clustered at the investing-entity level.We define variables in the Online Appendix.Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations the tax burden difference between subsidiaries and flow-throughs is an economically important determinant of these decisions.The tax effect, which is comparable in magnitude to the effect of non-tax determinants, is weaker for MNCs that engage in more tax-motivated income shifting and MNCs that value a subsidiary's non-tax benefits of limited liability and legal independence.MNCs with local knowledge, in contrast, are more sensitive to the tax burden difference.Evidence from a 2008 tax reform in Germany alleviates endogeneity concerns and supports a causal interpretation of our findings.Finally, we examine potential real effects of organizational form choices and find that affiliates established as flow-throughs exhibit a lower loss propensity and are less profitable than affiliates established as a subsidiary.Despite these results and the various empirical approaches we apply to investigate our research questions, we cannot definitively rule out that additional aspects of an MNC's complex decision environment may influence organizational form choices in our setting.We therefore encourage readers to interpret our findings with this caveat in mind.
Our study makes several contributions to the literature.First, we extend research on the effect of taxes on the group structures of MNCs by showing that international taxation is associated with the organizational form of foreign affiliates.Second, we add to research on business-group affiliation by documenting that MNCs frequently choose a flowthrough organizational form and by showing that the organizational form of foreign investment could imply differences in loss propensity and profitability across affiliates.Third, we add to studies on the effect of taxes on organizational form choices.While prior research finds that nontax factors dominate organizational form choices in domestic settings, our findings suggest that taxes are an economically important determinant of the organizational form choices of MNCs.
Finally, our study has implications for tax policy.Since we find that organizational form choices of MNCs are sensitive to tax-law changes, our results suggest that measures against tax-base erosion (e.g., withholding taxes) have to be implemented uniformly across organizational forms in order to be effective.Moreover, our study informs future research on the economic effects of the TCJA.This reform, which transformed the U.S. worldwide tax system into a ''quasi'' territorial system, turns dividend-withholding taxes into a final economic burden and may strengthen the effect of these taxes on the group structures of U.S. MNCs.informative for two reasons.First, if implemented differently across organizational forms, recent policy proposals, such as the Subject-to-Tax-Rule of the OECD Pillar II proposal (OECD, 2021), could imply tax-rate differences.Second, prior research suggests that tax-rate differences are more salient to decision-makers than tax-base differences (Amberger, Eberhartinger, & Kasper, 2022).
3 Table 2 (panel A) suggests that the mean annual tax cost of a subsidiary in our sample amounts up to 13.12% of pre-tax income earned in the host country, underlining the economic significance if this tax factor.

4
The investing entity is located either in the same country as the parent of the MNC (direct investment) or in a different country (indirect investment).We analyze the organizational form choices of the investing entity because data on the entire group structure is not available.Since we focus on newly established foreign affiliates, our findings suggest that MNCs might respond to increases (decreases) in the tax burden difference between organizational forms by establishing more (fewer) new affiliates as flow-throughs.MNCs could react to tax-law changes also by adjusting the organizational form of existing affiliates.However, while most developed countries (including Germany) have tax rules in place that allow tax-neutral restructurings, the non-tax costs of changing an organizational form can be substantial (Tazhitdinova, 2020).

7
In the Online Appendix, we provide additional background on the organizational forms available in Germany.
Table 5, VIFs for Roa and LossYear amount to 1.57 and 1.40, respectively.These statistics suggest that including both variables does not pose multicollinearity concerns.In additional tests, we construct a long-term measure for general loss propensity by extending LossYear to 3 years (i.e., years t to t+2).Our inferences are unchanged (untabulated).
17 In untabulated tests, we interact Taxwedge with Distribution and find a positive coefficient on the interaction term (p value = 0.11).This result provides some evidence that profit distributions amplify the tax sensitivity of organizational form choices.The coefficient on Taxwedge remains positive and significant, consistent with organizational form choices for affiliates that do not distribute profits immediately being sensitive to the tax burden difference.This finding is plausible because MNCs should take into account potential future repatriations required to cover domestic financing needs (Dyreng & Markle, 2016) or to fund operations in other countries (Stein, 2003). 18We collect monthly data for publicly listed firms in Germany from Thomson Reuter's Datastream and take the median for each one-digit ICBindustry code to obtain annual values of PE-Ratio and Industry-Roa.For details and two recent applications of such measures, see Shroff et al., (2014) and Amberger, Markle, and Samuel (2021). 19We note that several control variables capture decisions of the MNC that might coincide with the organizational form choice, such as decisions regarding investment size (LN(Assets), LN(Employees)), financing (Leverage), the choice to spread risk across multiple affiliates (NumInv), M&A activity (Brownfield), profit distribution (Distribution), cooperation with other investors (Holdings), and direct versus indirect investment (DirectFDI).Since the coefficient on Taxwedge is estimated conditional on all variables included in Eq. ( 1), our research design controls for these decisions and for the MNC's expectations that may affect the organizational form choice. 20 In line with evidence in Lewellen and Robinson (2013), more than 75% of the organizational form choices in our sample stem from direct investment (Table 3).Although indirect investment offers potential tax savings, non-tax costs (e.g., set-up costs, administrative costs, agency costs, etc.) are higher than for direct investment (Amberger et al., 2021).Moreover, since a non-zero Taxwedge implies a higher tax burden for one of the organizational forms both in case of direct and indirect investment, we expect and find that the tax sensitivity of organizational form choices does not differ between direct and indirect investment (p value = 0.52, untabulated).Note that the relation between Taxwedge and Flow-Through remains positive and significant when limiting the sample to direct investment (p value \ 0.001, untabulated).
21 Since we focus on the first observation of a new affiliate established in Germany, our setting provides no within-affiliate variation required for including affiliate-level fixed effects.However, we employ several alternative approaches to address potential endogeneity concerns and present the results in section Addressing Endogeneity Concerns.
22 Our main inferences are unchanged when using alternative clusters (e.g., industry, year, the home country of the investing entity, or the parent home country).
23 In line with Deutsche Bundesbank's confidentiality rules as of March 1, 2021, we present home countries with a minimum of five observations per organizational form that result from five distinct parents.Our sample includes another 47 home countries that do not fulfill this requirement.
24 80.78% of the new affiliates report a Taxwedge of zero.We keep these observations in the sample because they provide information on the non-tax determinants of organizational form choices. Keeping these observations also increases the statistical power of our analysis.In untabulated tests, we reestimate our main tests (Table 5) and our crosssectional tests (Tables 6, 7) after dropping observations with a Taxwedge of zero, and find similar results.These tests alleviate concerns that observations with a Taxwedge of zero might drive our findings.
25 Despite a Taxwedge of zero, investing entities located in neighboring countries may still establish flow-throughs due to a low geographical distance diminishing coordination costs and similarities in legal systems reducing the risk associated with unlimited liability.Both aspects reduce the nontax costs of a flow-through, increasing this organizational form's relative attractiveness.
26 We aggregate observations based on one-digit NACE Rev. 2 codes to ensure a meaningful analysis. 27The relatively large share of observations with a Taxwedge of zero implies a similar central tendency of Taxwedge in both subsamples, rendering the difference in medians insignificant.
28 Given the large set of controls, we alleviate multicollinearity concerns by calculating VIFs.For the model in Table 5, column 3, the maximum VIF for the independent variables is 2.78 (Patents).Among year and industry fixed effects, the maximum VIF is 4.6 (indicator for ''Year 2006''), indicating a low risk of multicollinearity.
29 When calculating marginal effects for year and industry indicators (untabulated), we find that marginal effects for specific industries (except for the industry ''Energy Supply'') or specific years are in the range of the marginal effects obtained for categorical non-tax controls (e.g., LossYear, Brownfield, DirectFDI).This suggests that the variables in our regression model capture economically important determinants of organizational form choices. 30 To corroborate this conclusion, we conduct a likelihood ratio test and find that adding Taxwedge as an independent variable significantly improves the fit of our regression model (p value \ 0.001).
31 An alternative explanation for this result is that the investing entity might be less sensitive to Taxwedge if tax depreciation diminishes the taxable income of the new affiliate.To examine this argument, we interact Taxwedge with an indicator variable taking the value of one if the ratio of fixed and intangible assets to total assets of the new affiliate is above the sample median.We find a positive and significant coefficient on the interaction (p value = 0.002), which supports an explanation based on high capital intensity.
32 An F-test suggests that the joint effect of Taxwedge and HighExp (b1+b7) is larger than zero (p value \ 0.001). 33We present this analysis as supplementary rather than primary because we incur a slight loss in sample size.Specifically, home countries that contribute only one observation are dropped from the analysis. 34We note that the reform also did not change any non-tax costs and benefits of organizational forms.
35 Although the DiD analysis exploits an exogenous shock to the tax cost of a subsidiary, we present this analysis as supplementary because every affiliate contributes only one observation to our sample.Thus, rather than tracking affiliates over time, we test whether the reform led to a differential change in the probability of establishing a new affiliate as a flow-through. 36We use the tax-haven list by Gravelle (2009).From the home countries in Table 2 (panel A), we drop observations of investing entities located in Jersey. 37We use the statutory corporate income tax rate to proxy for the home country corporate income tax burden.
38 A country with a worldwide tax system taxes income earned abroad while granting a credit for foreign taxes paid.Under such a system, the MNC could pool dividend repatriations from various affiliates for purposes of home-country taxation (''cross-crediting'').Depending on the home-country tax burden, these features could weaken the incentive to select a tax-favored organizational form for affiliate i. 39 In additional tests (untabulated), we drop observations if the new affiliate is established in a regulated industry and if the parent is located in a tax haven (Gravelle, 2009), respectively.The coefficients on Taxwedge are positive and significant in both tests (p values = 0.003 and \ 0.001), further supporting the robustness of our results. 40In our setting, the endogenous explanatory variable (Flow-Through) is binary in nature.Wooldridge (2010) refers to such a regression model as a dummy endogenous variable model (p.938).
41 As expected, d Flow-Through exhibits a strong positive association with Flow-Through in the first stage of the 2SLS-regressions (columns 1, 3, 5, 7, and 9).
42 Limiting the analysis in Table 12 to observations with non-zero Taxwedge leads to qualitatively similar results (untabulated).Specifically, we still find a negative and significant coefficient on Flow-Through for LossPropensity (p value = 0.006).The coefficient on Flow-Through for Roa remains negative as expected but becomes statistically insignificant (p value = 0.369); the negative coefficient for Investment turns statistically significant (p value = 0.040).

Figure 1
Figure1International Taxation and Organizational Forms.Note: This figure outlines the three layers of international taxation.The investing entity selects an organizational form for a new affiliate in the host country (Germany).The investing entity is part of an MNC and located either in the same country as the parent (e.g., the United States: direct investment) or in a different country (e.g., Luxembourg: indirect investment).The three layers of international taxation determine the tax burden on income earned in a foreign subsidiary and a foreign flow-through, respectively.Layer 1 is the statutory corporate income tax in the host country (s host table presents Pearson correlation coefficients for the future characteristics of new affiliates.We winsorize continuous variables at the 1st and 99th percentile.We define variables in the Online Appendix.(Lag) denotes variables lagged by 1 year.Bold coefficients indicate a significance level of 0.01.Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations

Table 1
Sample selection (organizational form choice sample) This table presents the sample selection for the organizational form choice sample.The sample includes the first observation of a new affiliate in Germany (sample years 2005-2013).We define variables in the Online Appendix.Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations

Table 2
Sample compositionPanel A: Taxwedge and observations by home countryThis table presents information on the sample composition.Panel A presents the mean tax cost of a subsidiary (Taxwedge) and the number of observations by home country of the investing entity.In line with Deutsche Bundesbank's confidentiality rules, we present home countries with a minimum of five observations per organizational form that result from five distinct parents.Our sample includes another 47 home countries that do not fulfill this confidentiality requirement.These observations are summarized under Additional Observations.Panel B presents observations by sample year.Panel C presents observations by industry.Industry classification is based on one-digit NACE Rev. 2 industry codes.We define variables in the Online Appendix.
Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations

Table 3
Descriptive statistics (organizational form choice sample) This table presents descriptive statistics for the organizational form choice sample.We present means and standard deviations for dependent and independent variables.We present results for the full sample and separately for subsidiaries and flow-throughs.We conduct a two-sample t test assuming unequal variances (Wilcoxon rank-sum test) to compare means (medians) between subsamples.We winsorize continuous variables at the 1st and 99th percentile.We define variables in the Online Appendix *, **, and *** represent significance levels of 0.10, 0.05, and 0.01, respectively (two-tailed).Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations

Table 4
Correlation table (organizational form choice sample) This table presents Pearson correlation coefficients for the organizational form choice sample.We winsorize continuous variables at the 1st and 99th percentile.We define variables in the Online Appendix.Bold coefficients indicate a significance level of 0.01.Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Micro-database Direct Investment (MiDi) for the years 2005 to 2013, own calculations

Table 5
Baseline results: tax burden difference and organizational form choices

Table 6
Cross-sectional results: tax-motivated income shifting This table presents results for cross-sectional tests based on the extent of tax-motivated income shifting.The dependent variable, Flow-Through, is an indicator variable equal to one if the new affiliate is established as a flow-through, and zero otherwise.HighIntDebt is an indicator variable equal to one if the ratio of intra-firm debt provided by foreign affiliates to total assets of the new affiliate is above the sample median, and zero otherwise.Patents is an indicator variable equal to one if the new affiliate is established in a patent-asset intensive industry, and zero otherwise.Columns 1 and 3 (2 and 4) report coefficients (marginal effects) for a logistic regression based on Eq. (

Table 7
Cross-sectional results: non-tax factors This table presents results for cross-sectional tests based on non-tax factors.We drop regulated industries (industry classification D, J, and K based on one-digit NACE Rev. 2 industry codes) in columns 1 and 2. The dependent variable, Flow-Through, is an indicator variable equal to one if the new affiliate is established as a flow-through, and zero otherwise.HighIndRisk is an indicator variable equal to one if the new affiliate is established in the manufacturing or the wholesale industry, and zero otherwise.HighRegDiff is an indicator variable equal to one if the difference in regulatory quality between Germany and the parent home country is above the sample median, and zero otherwise.HighExp is an indicator variable equal to one if the investing entity holds at least one additional inbound FDI relation in Germany at the time of the organizational form choice, and zero otherwise.

Table 8
Difference-in-differences analysis: 2008 tax reform in Germany This table presents results for a DiD analysis based on the 2008 tax reform in Germany.The dependent variable, Flow-Through, is an indicator variable equal to one if the new affiliate is established as a flow-through, and zero otherwise.Columns 1 and 3 (2 and 4) report coefficients (marginal effects) for a logistic regression based on Eq. (

Table 9
Robustness testsThis table presents results for robustness tests.Panel A presents tests with modified samples.We drop observations if the investing entity establishes more than one new affiliate in Germany in column 1 and observations if the investing entity is located in a tax haven in column 3. Panel B presents results for tests with additional controls.In column 1 (3), we control for Worldwide and CIT (PositiveTaxwedge).In column 5, we replace Taxwedge with Wht and HomeTax.The dependent variable, Flow-Through, is an indicator variable equal to one if the new affiliate is established as a flow-through, and zero otherwise.Columns 1 and 3 (1, 3, and 5) of panel A (B) report coefficients for a logistic regression based on Eq. (1).Columns 2 and 4 (2, 4,

Table 10
Descriptive statistics (future characteristics of new affiliates sample)This table presents descriptive statistics for the future characteristics of new affiliates.In Panel A (B), we present means and standard deviations for our dependent (independent) variables.We conduct a two-sample t test assuming unequal variances (Wilcoxon rank-sum test) to compare means (medians) between subsamples.In Panel B, LN(Employ), LN(Assets), Roa, Leverage, LN(Sales), Investment, and Cash are lagged by 1 year.All continuous variables are winsorized at the 1st and 99th percentile.We define variables in the Online Appendix.*, **, and *** represent significance levels of 0.10, 0.05, and 0.01, respectively (two-tailed).Source: Research Data and Service Centre (RDSC) of the Deutsche Bundesbank, Microdatabase Direct Investment (MiDi) for the years 2005 to 2013, own calculations

Table 11
Correlation table (future characteristics of new affiliates sample)

Table 12
Future characteristics of new affiliates

Table 12 (
Continued)Panel B: 2SLS-regressions to examine future characteristics of new affiliates (step two)