Scholarly article on topic 'Location decisions of non-bank financial foreign direct investment: Firm-level evidence from Europe'

Location decisions of non-bank financial foreign direct investment: Firm-level evidence from Europe Academic research paper on "Social and economic geography"

0
0
Share paper
Academic journal
Rev International Economics
OECD Field of science
Keywords
{""}

Academic research paper on topic "Location decisions of non-bank financial foreign direct investment: Firm-level evidence from Europe"

ORIGINAL ARTICLE

Location decisions of non-bank financial foreign direct investment: Firm-level evidence from Europe

Ronald B. Davies1 I Neill Killeen2*

1 University College Dublin, Republic of Ireland

2 Central Bank of Ireland, Dublin, Republic of Ireland

Correspondence

Ronald B. Davies, University College Dublin, School of Economics, Newman Building, Belfield, Dublin, Republic of Ireland. Email: ronbdavies@gmail.com

Abstract

The non-bank financial sector in Europe has more than doubled in size between 2005 and 2015 reflecting the substantial growth in shadow banking activities. However, a large proportion of the non-bank financial sector that remains unmapped as granular balance sheet information is not available for over half of the sector. Motivated by these data gaps and employing firm-level data, this paper examines the location decisions of newly incorporated foreign affiliates in the non-bank financial sector across 27 European countries over the period 2004 to 2012. The probability of a country being chosen as the location for a new foreign affiliate is found to be negatively associated with higher corporate tax rates and geographic distance but increases with the size and financial development of the host country. The financial regulatory regime in the host country and gravity related controls such as the home and host country sharing a common legal system, language, border, and currency are also found to impact the likelihood of non-bank financial FDI.

1 I INTRODUCTION

The financial crisis highlighted the globalization of the financial sector including the substantial growth and complexity of corporate structures in the non-bank financial sector. A key feature of the globalization of the financial sector are the activities of multinational enterprises (MNEs). An active strand of the FDI literature has focussed on the effects of FDI in the financial sector (Goldberg, 2007), with empirical evidence suggesting that financial FDI may be less beneficial to the host country than general or greenfield FDI (Ostry, Ghosh, Habermeier, Chamon, Qureshi, & Reinhardt, 2010).

The financial sector comprises several types of entities, including banks and insurance companies. Of increasing importance, however, are the size and activities of non-bank financial institutions.

* Also affiliated to University College Dublin, Republic of Ireland. Currently on secondment to the European Systemic Risk Board (ESRB) Secretariat, Frankfurt am Main, Germany.

Rev IntEcon. 2017;l-26.

wileyonlinelibrary.com/joumal/roie

«3 2017 John Wiley & Sons Ltd I 1

Non-MMF Investment Funds

OFI Residual

Source: Grillet-Auberl et al. (2016)

FIGURE 1 Breakdown of European Union investment funds and OFIs by type, percent of total assets, Q4 2015 [Colour figure can be viewed at wileyonlinelibrary.com]

According to the European Systemic Risk Board (ESRB, 2016) and Grillet-Aubert, Haquin, Jackson, Killeen, and Weistroffer (2016), the non-bank financial sector in Europe has grown significantly, with total assets increasing from approximately £10 trillion in 2003 to €37 trillion in 2016. This in part reflects the growth of shadow banking activity in Europe as financial institutions adjust their activities and corporate structures in light of the increased regulatory requirements of the banking sector.1

Shadow banking can be broadly defined as credit intermediation outside the traditional banking system and therefore it is a catch-all term for non-bank financial institutions engaged in bank-like activities. Despite the substantial growth in this part of the financial system, a large number of non-bank financial institutions are to date unmapped, as reflected by the large residual of approximately 61 percent in the other financial intermediaries (OFI) sector in Europe (see Figure 1). The increasing importance and internationalization of non-bank financial institutions coupled with this large data gap raises a number of financial stability concerns. For example, many of the newly incorporated foreign affiliates in our sample are subject to lighter regulatory requirements than banks and could potentially be used as part of regulatory arbitrage strategies that can exacerbate vulnerabilities within the financial system. Non-bank financial institutions include financial corporations engaged in lending such as financial leasing (including aircraft leasing) or finance companies, specialized financial corporations such as factoring firms, special purpose vehicles (SPVs) including financial vehicle corporations (FVCs) engaged in securitization activity, financial holding companies, and securities and derivatives dealers. Despite the fact that some of these entities are engaged in credit intermediation, a large proportion of these institutions remain outside of the regulatory perimeter in many E.U. Member States.

This paper contributes to the existing literature by examining the determinants of location decisions of new non-bank financial foreign affiliates.2 Few studies in the empirical literature on firm location choice have examined sector-specific FDI in Europe while most exclude the financial sector from their analysis.3 Moreover, many of the papers that examine financial FDI focus on the activities of foreign banks (Grosse & Goldberg, 1991; Yamori, 1998; Buch & Lipponer, 2007; Claessens & Van Horen, 2014; Huizinga, Voget, & Wagner, 2014; Merz, Overesch, & Wamser, 2017). To the best of our knowledge no study examines the determinants of location decisions for FDI in the non-bank financial sector. This paper aims to fill this important gap.

Firms face a number of decisions when setting up their international activities and associated corporate structures. First, there is the choice of whether to serve foreign markets through the provision of cross-border services or a foreign affiliate. Second, if the firm chooses FDI, there is the question of where to locate. There is a large and growing number of studies that examines these FDI location decisions. One strand of the literature examines the wider determinants of location decisions (Head & Mayer, 2004; Basile, Castellani, & Zanfei, 2008) while a second strand of the literature focusses on the impact of taxation on FDI location decisions (Devereux & Griffith, 1998b; Hebous, Ruf, & Weichenrieder, 2011; Barrios, Huizinga, Laeven, & Nicodeme, 2012; Lawless, McCoy, Morgenroth, & O'Toole, 2015). A less developed third strand of the literature draws on more granular regional data to examine the determinants of sector-specific FDI location decisions (Siedschlag, Zhang, & Smith, 2013; Siedschlag, Smith, Turcu, & Zhang, 2013). Our paper is related to all three strands of the literature given that we focus on sector-specific FDI determinants, including corporate taxation. Our work extends this growing literature as many of the non-bank financial institutions included in our sample would not be captured within these studies.

In particular, we analyze several issues. First, we examine at a basic level the location of the financial activities of MNEs in the non-bank financial sector in Europe, finding substantial concentration. Second, we estimate the effect of key factors determining the location decisions of non-bank financial FDI using a conditional logit estimator. Third, we compare so-called "brass plate" entities (which despite large assets may have few employees) with other firms. We do this to examine how, in light of Goldberg (2007), non-bank financial FDI behaves differently to overall FDI. To do all of this, we construct a unique firm-level dataset of 8,724 newly incorporated foreign affiliates in the non-bank financial sector in the European Union over the 2004 to 2012 period.

Our results suggest that many traditional determinants of overall FDI, including market size, barriers between the home and host, taxes, and regulation have comparable effects in this sector. The first two of these results may be somewhat surprising given the perception that this industry is very footloose and not tied to a specific geographic location. That said, our results do mirror what is found in banking FDI. For example, the negative coefficient we obtain for distance is consistent with the findings of Buch (2005), and Claessens and Van Horen (2014). In addition, we find that the determinants of brass plate investment is comparable with that of larger firms alleviating concerns over differential responses between these two groups. Thus, our estimates suggest that non-bank financial FDI behaves much like FDI in the banking sector and overall FDI. In addition, although we do find an important effect from taxes, the significance of other determinants of FDI works to reduce concerns that FDI in this sector will merely flow to the lowest tax locations or those with less intrusive regulation.

The rest of the paper is structured as follows. Section 2 describes the data employed for our analysis. Section 3 discusses our empirical methodology and presents our results. Section 4 concludes.

2 I DATA

To investigate the factors driving non-bank financial FDI location decisions, we construct a unique dataset using firm-level data from Bureau van Dijk's Amadeus database. These data report financial and ownership information based on standardized financial statements for companies across Europe. We use data from unconsolidated accounts and collect information on the firm's date and location of incorporation, number of employees, the location of the foreign investor and the firm's sector classification. Using primary four digit European Classification of Economic Activities (NACE) Rev. 2 industrial classifications, we restrict our sample of firms to those classified in financial services, corresponding to NACE codes 6400 to 6630 inclusive (Table 1).4 This allows us to conduct a detailed sectoral analysis based on the firm's main activity with a specific focus on non-bank financial FDI.

TABLE 1 NACE Rev. 2 classification and number of non-bank financial foreign affiliates per NACE classification

% of total

NACE Rev. 2 Code/Description N sample

64 Financial services activities, except insurance and pension funding

6411 Central banking 0 0

6419 Other monetary intermediation 0 0

6420 Activities of holding companies" 5,987 68.5

6430 Trusts, funds and similar entitiesa 323 3.7

6491 Financial leasinga 213 2.4

6492 Other credit grantinga 0 0

6499 Other financial services activities, except insurance and pension fundinga 882 10.1

66 Activities auxiliary to financial services and insurance

6611 Administration of financial marketsb 74 0.8

6612 Security and commodity contracts brokerageb 202 2.3

6619 Other activities auxiliary to financial services except insurance and pension fundingb 689 7.9

6621 Risk and damage evaluation13 26 0.3

6622 Activities of insurance agents and brokersb 150 1.8

6629 Other activities auxiliary to insurance and pension fundingb 62 0.7

6630 Fund management activitiesb 116 1.3

Total 8,724 100

Note. Banks and insurance company accounts are not available on the Amadeus database. Therefore entities relating to NACE Rev. 2 65 "Insurance, reinsurance and pension funding, except compulsory social security" are not covered in our sample. aRefers to NACE Rev. 2 sectors that are classified as other financial intermediaries. Under ESA 2010, financial holding companies have been reclassified as captive financial institutions and money lenders. bRefers to NACE Rev. 2 sectors that are classified as financial auxiliaries.

Our dataset contains information on 8,724 new foreign affiliates across 27 European countries for the period 2004 to 2012.5 A firm is defined as foreign-owned if the firm has one foreign shareholder who holds at least 10 percent equity capital. A foreign affiliate is defined as new in its year of incorporation.

We merge these firm-level data with country-level data that account for a host of gravity related characteristics (e.g., cultural, macroeconomic, geographical, and institutional characteristics). These data are taken from a wide range of sources, including the World Bank, CEPII, the World Development Indicators (WDI), and the OECD. Drawing on the literature on location decisions we use a number of control variables that are described below. The source and definition of each of these variables are presented in Table 2. In addition, some essential data cleaning was performed prior to conducting the empirical analysis. A description of the data construction and cleaning process is presented in the

Appendix.

Before proceeding to the regression analysis, it is instructive to gain an overview of the broad patterns in the data. The number of foreign affiliates in each NACE Rev.2 sector classification considered for this analysis is shown in Table 1. Almost 70 percent of new foreign affiliates incorporated in Europe over our sample period relate to activities of financial holding companies. The second most prominent sector covers other financial services activities (except insurance and pension funding), which represents just over 10 percent of our sample of firms. Although 27 European countries are included in our final sample, Table 3 shows that a small number of countries, including the Netherlands, the United Kingdom, Germany, France, and Ireland represent almost 75 percent of our total sample. The Netherlands was the chosen host for 38.5 percent of foreign investments in the

TABLE 2 Variable definitions and data sources

Variable Description Data Source

Location Dummy variable equal to 1 if affiliate is located in a country and 0 otherwise Amadeus

Colony Dummy variable equal to 1 if home and host country ever shared a colonial relationship and 0 otherwise CEPII

Comcurr Dummy variable equal to 1 if home and host country share a common currency and 0 otherwise CEPII

Common legal Dummy variable equal to 1 if home and host country share a common legal system and 0 otherwise CEPII

Com. language Dummy variable equal to 1 if home and host country share a common language CEPII

Contiguity Dummy variable equal to 1 if home and host country share a common border and 0 otherwise CEPII

Corporation Tax Statutory corporation tax rate, % KPMG

Corp. Tax2 Squared term of statutory corporation tax rate, % KPMG

EATR Effective average tax rate as per Devereux and Griffith (1998a) CBT tax database

EATR2 Squared term effective average tax rate as per Griffith and Devereux (1998a) CBT tax database

Fin. Regulation Restriction index on financial conglomerates' activities. Higher values indicate more intensive restrictions Barth et al. (2001)

Fin. Development Private credit by deposit money banks and other financial institutions as a percentage of GDP GFD, World Bank

GDP growth Annual GDP growth, % WDI

Internet Fixed broadband internet subscribers (per 100 people) WDI

Distance Log of distance, measured by km between host and home country capital cities, weighted by population CEPII

GDP Log of GDP, constant 2005 prices in U.S. dollars WDI

OFFDIrestrict A measure of all discriminatory measures affecting foreign investors in the other financial sector. Varies from 0 (no impediments) to 100 (fully restricts) OECD

Per capita GDP GDP per capita in host country WDI

Unemployment rate Rate of country unemployment, % Datastream

non-bank financial sector with almost 95 percent of these investments in the Netherlands relating to activities of financial holding companies.6 Thus, FDI in this sector is remarkably concentrated.

The Amadeus database provides information on the Global Ultimate Owner (GUO) of the financial foreign affiliate, which we use to identify the home country of FDI. A GUO is defined as an investor who holds over 50 percent of the firm's equity capital. The GUO represents the home country for each

TABLE 3 Home and host countries of new non-bank financial foreign affiliates incorporated in the European Union (2004-2012)

Top 10 host countries N % Top 10 home countries N %

Netherlands 3,363 38.5 United States 1,854 21.2

United Kingdom 1,073 12.3 Luxembourg 842 9.7

Germany 1,047 12.0 United Kingdom 629 7.2

France 522 6.0 Germany 525 6.0

Ireland 505 5.8 Netherlands 372 4.3

Belgium 326 3.7 Switzerland 343 3.9

Austria 260 3.0 Cyprus 342 3.9

Italy 214 2.5 France 280 3.2

Denmark 202 2.3 Belgium 233 2.7

Spain 193 2.2 Austria 170 1.9

Total 8,724 Total 8,724

Source. Authors' calculations.

foreign investment while their location can be identified by their International Organization for Standardization (ISO) country code. We restrict our sample to firms who report their GUO ISO country code on Amadeus.7 Table 3 also provides information on the location of the GUO. The distribution is consistent with many of the related empirical studies (Barrios et al., 2012; Siedschlag, Zhang et al., 2013; Siedschlag, Smith et al., 2013; Lawless et al., 2015), and shows that OECD countries are the home country for the majority of GUOs. For instance, over 21 percent of the firms in our sample originate in the United States.

Given the large proportion of financial holding companies in our sample, Table 4 presents the number of firms per country in this NACE Rev. 2 classification over the analyzed period, 2004 to 2012. The first two columns show the top five host countries chosen as locations for financial holding

TABLE 4 The location of new financial holding companies and new non-bank financial foreign affiliates (excluding financial holding companies) incorporated in the European Union (2004-2012)

Top 5 host countries of financial holding companies N % Top 5 host countries (excluding financial holding companies) N %

Netherlands 3,194 53.4 United Kingdom 602 22.0

Germany 614 10.3 Ireland 505 18.5

United Kingdom 471 7.9 Germany 433 15.8

France 388 6.5 Netherlands 169 6.2

Belgium 281 4.7 France 134 4.9

Total 5,987 Total 2,737

Source. Authors' calculations.

US UK Luxembourg Germany Cyprus

All countries I__Netherlands

Source: Amadejs database, Bureau van Dijk FIGURE 2 Home country of top five foreign investors in financial holding companies incorporated in the European Union and the Netherlands (2004-2012) [Colour figure can be viewed at wileyonlinelibrary.com]

companies. The Netherlands dominates with over 53 percent followed by Germany (10.3 percent) and the United Kingdom (7.9 percent). Figure 2 displays the home country of the GUO for investments in financial holding companies in all countries as well as investors who set up financial holding companies in the Netherlands. It is noteworthy that the top home country for both samples is the United States. Table 4 also shows the top five host countries chosen as locations for all sectors excluding financial holding companies. Of the 2,737 foreign investments made excluding activities of financial holding companies, 22 percent chose to locate in the United Kingdom followed by Ireland with almost 18.5 percent and Germany with 15.8 percent.8

By focussing specifically on non-bank financial FDI we can explore the nature of foreign investments in our sample and examine whether these investments create real economic activity in their host country or whether the investments are brass plate in nature, such as the activities of SPVs. As of this writing, there is no single definition of SPVs. However, according to the Bank for International Settlements (BIS, 2009), they have a number of common features. First, they are usually created to perform a specific financial or business activity. Second, they are usually set up as bankruptcy remote vehicles, which ensures that the assets within the SPV are protected from the risk of bankruptcy of the originator, and therefore creditors of the originator do not have a claim on the assets of the SPV. Third, SPVs usually do not have any direct employees and instead use professional corporate service providers, directors, and trustees to perform the vehicle's duties. Fourth, SPVs may not have any physical presence or employees in their host country. Their only presence may be the address of their registered office. Fifth, in many European jurisdictions such as Ireland, the Netherlands, and the United Kingdom, the most common type of SPVs are orphan vehicles (BIS, 2009). Orphan vehicles are entities whose share capital is a nominal amount and which is held beneficially on trust for a charity. The holding of shares on trust for charitable purposes ensures that the SPV is not owned by the originator (BIS, 2009).

With this in mind, we use the number of employees as a proxy for "real" economic activity, with firms that have no or little employment being in line with the common features of SPVs as discussed in BIS (2009). Table 5 shows the top host countries for firms with between one and 10 employees. The Netherlands (with 47.9 percent) is the top host location and given the concentration of non-bank financial FDI in activities of holding companies in the Netherlands, it suggests that the establishment of holding companies does create some economic activity in the host country (as proxied by the

TABLE 5 The location of new non-bank financial foreign affiliates located in the European Union by number of employees (2004-2012)

Top 5 host countries of firms with 1-10 employees N % Top 5 host countries of firms with >10 employees N %

Netherlands 1,282 47.9 United Kingdom 82 21.8

Germany 467 17.4 Netherlands 72 19.1

United Kingdom 251 9.4 Germany 34 9.0

Ireland 99 3.7 Spain 26 6.9

Austria 92 3.4 Ireland 24 6.4

Total 2,679 Total 377

Source. Authors' calculations.

number of employees). Germany and the United Kingdom also attract financial investments that create some employment, followed by Ireland and Austria. Next, we explore whether the ranking of host countries changes when we adjust the number of employees' threshold to firms who report having greater than 10 employees. Table 5 shows that the United Kingdom is the top host country for this more employment intensive FDI followed by the Netherlands. It is worth noting that the share of foreign investments who chose to locate in the Netherlands falls significantly (to 19 percent) for this sub-sample of firms.

Drawing on the existing theoretical and empirical literature we gather data on a number of control variables that are likely to effect expected profits from locating in a particular country. Gravity variables describe market size and market access and have been used extensively in the empirical FDI literature with common patterns emerging (for an overview, see Blonigen & Piger, 2014). We proxy the size of the host market using GDP. GDP growth is also included as a further proxy for macroeconomic conditions in the host country. The literature generally finds that both attract FDI. To account for the wealth of the host country, we include Per capita GDP. The typical estimate for this can be positive or negative depending on whether FDI is horizontal (where wealthier consumers attract FDI) or vertical (where high incomes translate to high costs, reducing FDI). In a similar vein, we include the unemployment rate. As highlighted by Siedschlag, Zhang et al. (2013), high unemployment rates can indicate a pool of available labor but may also reflect a lack of flexibility in the labour market or poor labour quality. Therefore, the effect of the unemployment rate on the location probability is ambiguous.

In addition, it is standard to account for the cultural, geographical, and institutional characteristics of the host country. Distance, measured by kilometres between home and host capital cities, is used to proxy for factors that may hinder FDI between countries, such as information costs or time differences. We expect a negative effect on the probability of location choice.9 Similarly, we control for whether they share a common border (contiguity). To control for cultural barriers, we include a set of dummy variables describing whether the home and host countries share past colonial ties (colony), a common legal system (common legal), or a common language.10 We also include a dummy variable equal to one if they share a common currency (comcurr). In line with the existing literature, we expect all of these dummy variables to be positive. For example, Claessens and Van Horen (2014) include common language and legal system variables in their empirical analysis and find that they are important determinants of banking sector FDI. We anticipate that these factors are also important for non-bank financial FDI given the complexity of corporate structures and tax frameworks in this sector.

Beyond the traditional gravity variables, we expect that the regulatory policies of the host country have an impact on the location decision, particularly in this sector that is presumably rather mobile. The importance of tax on the location decisions of FDI has been the focus of much of the literature (Devereux & Griffith, 1998b; Barrios et al., 2012; Lawless et al., 2015) since corporation tax directly reduces the profits of firms. For our baseline, we use the statutory tax rate from KPMG. The advantage of this measure is its availability across countries. Its downside, however, is that relative to the effective average tax rate (EATR), it is less reflective of the actual taxes firms pay. Therefore, as a robustness check, we use the EATR measure collected by the Oxford University Centre for Business Taxation (which accounts for exemptions and other features of the tax code that affects the actual taxes paid). In any case, our two tax rate variables are highly correlated (0.949). Our a priori expectations is for a negative effect of taxation, with this marginal effect declining in the host tax rate (i.e., taxes matter more in low-tax locations).11

We control for the regulatory framework in place in the potential host country using OFFDIres-trict, the OECD's FDI restrictiveness index for the other financial services sector,12 which is described in Kalinova, Palerm, and Thomsen (2010). The index is a measure of all discriminatory measures affecting foreign investors and ranges from 0 (no impediments to FDI) to 100 (fully restricts FDI). The index covers four main areas including (i) foreign equity restrictions (ii) screening and prior approval requirements, (iii) rules for key personnel, and (iv) other restrictions on the operation of foreign enterprises. Our theoretical prior is for a negative effect of FDI restrictions on the location probability. In order to proxy for the quality of the financial regulatory regime, we include an index that measures restrictions on financial conglomerates' activities.13 It is important to note, however, that a large proportion of the institutions within our sample (e.g., financial holding companies, financial leasing companies, and SPVs) are outside of the regulatory perimeter in many E.U. Member States. Thus, it may have no impact or it may increase non-bank financial FDI if banking regulation is stringent in the host country.

Beyond these regulatory measures, we include two final variables relating to a country's attractiveness. First, we include private credit by deposit money banks and other financial institutions as a percentage of GDP as a measure of financial development.14 We do so as countries with deep financial development likely have characteristics that make them attractive to non-bank financial FDI.15 Second, we include the number of fixed broadband internet subscribers (per 100 people) as a proxy for infrastructure that may be particularly important for this services industry.

A number of our explanatory variables, namely unemployment, financial development, and OFFDIrestrict are included with a lag of one time period to mitigate potential endogeneity concerns from possible simultaneity of these variables. Standard errors are robust to heteroskedasticity and clustered at the firm level.16 Table 6 displays summary statistics for the explanatory variables used in our empirical analysis.17 We observe significant variation in country characteristics, most notably with respect to GDP growth and the unemployment rate. The variation on GDP growth reflects the impact of the 2008 financial crisis on macroeconomic conditions in Europe where a number of countries required external financial assistance programmes. Finally, for each of the time-varying controls, when an investment is made in year t, we use the t values (or t - 1 for those that are lagged).

3 | EMPIRICAL METHODOLOGY AND RESULTS

Our empirical methodology employs the conditional logit model as proposed by McFadden (1973). This model has been widely used in the empirical literature to examine the determinants of firms' location decisions (Head, Ries, & Swenson, 1995; Head & Mayer, 2004; Nefussi & Schwellnus, 2010; Barrios et al., 2012; Siedschlag, Zhang et al., 2013; Siedschlag, Smith et al., 2013; Lawless et al.,

TABLE 6 Summary statistics

Variable N Mean SD Min. Max.

Colony 191,160 0.1 0.2 0 1.0

Comcurr 191,160 0.2 0.4 0 1.0

Com language 191,160 0.1 0.3 0 1.0

Common legal 191,160 0.2 0.4 0 1.0

Contiguity 191,160 0.1 0.3 0 1.0

Corporation Tax 191,160 24.9 6.6 12.5 38.4

Corp. Tax2 191,160 660.3 327.3 156.3 1,471.5

EATR 166,414 22.6 5.5 11.1 34.0

EATR2 166,414 542.6 244.4 122.9 1,156.7

Fin. regulation 191,160 6.0 1.0 4.0 8.0

Fin. development 191,160 108.7 52.4 27.0 227.5

GDP growth 191,160 1.9 4.2 -18.0 12.2

Internet 191,160 20.6 9.0 0.5 39.8

Distance 191,160 7.8 1.1 5.1 9.9

GDP 191,160 26.2 1.5 23.3 28.8

Per capita GDP 191,160 10.1 0.6 8.7 11.4

OFFDIrestrict 191,160 3.8 4.9 0 20.5

Unemploy. rate 191,160 8.4 3.7 3.1 21.4

2015). The firm chooses its location to maximize profits, where Pj is the profit of firm i when locating in country j at time t:

% =Xijtb + Eijt. (1)

In this, Xijt is a vector of control variables including information about a potential host and the firm's home country, and Ejt is the error term. Although profits are unobserved directly, we do observe in which of the 27 European countries the firm locates.18 Therefore, our dependent variable, Location, is equal to one if the firm locates in a particular country j and zero otherwise. Assuming the error term Eji follows a type 1 extreme value independent and identically distributed (i.i.d) across all foreign affiliates and countries, the probability of choosing a particular country j can be written as follows:19

eXijt b

P(Y =j\1,..., K Xijt ) = . (2)

Ei51eXiktb

Table 7 presents the baseline estimates of the conditional logit model. Column (1) reports the estimates from our initial specification, which includes our full sample of firms. Of note is the statistically negative effect of the statutory corporation tax rate. This finding contrasts with other studies of location choice determinants (Basile et al., 2008; Siedschlag, Smith et al., 2013) who find an insignificant effect of the corporate tax rate on the location choice of MNEs. However, these studies do not consider

TABLE 7 Estimates of conditional logit model for non-bank financial firms' location decisions (baseline model)

(1) (2) (3) (4) (5)

Variable Full Full Sample Sample Sample

(expected sign) sample sample from (2) from (2) from (2)

GDP, (1) 0.706*** 0.743*** 0.758*** 0.954*** 1.101***

(0.012) (0.019) (0.011) (0.012) (0.022)

Per capita GDPt (1) 0.941*** -1.037*** 1.076*** 1.434*** -0.143*

(0.029) (0.057) (0.034) (0.043) (0.069)

GDP growth, (1) 0.032*** 0.122*** 0.031*** 0.010 0.048***

(0.007) (0.011) (0.007) (0.007) (0.010)

Distance (-) -1.574*** -1.318*** -1.657*** -1.583*** -1 172***

(0.030) (0.038) (0.031) (0.033) (0.040)

Corporation Tax, (-) -0.065*** -0.313*** -0.074***

(0.002) (0.018) (0.0019)

Corp. Tax,2 (1) 0.005***

(0.000)

Unemploy. Ratet-1 (?) -0.096*** -0.070***

(0.007) (0.007)

Fin. Developmentt-1(1) 0.012*** 0.010***

(0.000) (0.000)

OFFDIrestrict,_j (-) -0.030*** -0.098***

(0.004) (0.005)

Internet, (1) 0.083*** 0.033***

(0.004) (0.005)

Fin. Regulation (_) -0.198*** -0.285***

(0.018) (0.020)

EATRt (-) -0.116*** - 0.072**

(0.002) (0.025)

EATRt2 (1) -0.000

(0.001)

N 230,949 191,160 191,160 161,422 161,422

Firms 8,724 8,519 8,519 8,264 8,264

Log-pseudolikelihood -22,255.58 -18,630.90 -20,842.05 -18,940.80 -16,995.84

Note. The dependent variable Location equals 1 if a foreign affiliate is located in a country and 0 otherwise. Standard errors are robust to heteroskedasticity and clustered at the firm level. Standard errors in parentheses. *p<0.05; **p<0.01; ***p<0.001.

FDI determinants of the financial sector. Lawless et al. (2015), in contrast, do include financial sector firms and find that they are more responsive to changes in corporate tax rates than other sectors. Moreover, Merlo, Riedel, and Wamser (2016), and Merz et al. (2017) employ mixed logit models and find that corporate taxes have significant negative effects on the location of nonfinancial and financial FDI respectively. Of the other country controls included, we find the standard results from the gravity variables.

Column (2) includes a number of additional control variables, particularly with respect to regulation. Note that the inclusion of these additional controls reduces our sample by approximately 17 percent. Relative to the results in column (1), the primary change is that Per capita GDP is now significantly negative, suggesting that non-bank financial FDI avoids high income workers. In column (3), where we restrict the sample to those observations in (2) but do not include the additional controls, we see that this change is due to the new controls, not the change in sample. In line with the GDP growth variable, we find that higher unemployment reduces the probability of investment. This probability, however, increases with the size of the financial sector and internet usage. Regulation, be it with respect to FDI overall or to the financial sector, has a negative impact. This latter effect is notable as these firms generally fall outside the scope of these financial regulations. Finally, in line with Lawless et al. (2015), we see that the effect of the tax rate is nonlinear, with higher tax rates reducing FDI by more for low-tax countries.

As noted above, the corporate tax rate variable is limited as it does not account for other features of the tax system. With this in mind, columns (4) and (5) replaces this measure with the EATR. On the whole, our results are comparable, although we find no significant evidence for a nonlinear responsiveness to the tax rate. However, as can be seen, using this measure reduces our sample by nearly 30,000 observations. Owing to this data limitation, we proceed by using the statutory tax rate in our extended model specifications.

In Table 8 we arrive at our preferred specification by introducing the additional dummy variables proxying for barriers between the home and host countries. For comparison, column (1) repeats the specification from Table 7 (column 2). To this, we add our gravity related explanatory variables: sharing a common border, currency, colony, common legal system, and common language (column 2). Sharing a common language, border, currency, and legal system are found to increase the probability of firm location choice, with a shared colonial history having an unusual negative impact.20 Our other controls, however, remain as before. This then forms our preferred specification.

To this point we have examined the determinants of firm location decisions in the non-bank financial sector using our full sample of firms. However, as noted by the United Nations Economic Commission for Europe (UNECE, 2011), Claassen and Van den Dool (2013), and the OECD (2014) among others, FDI statistics can be distorted by SPVs and holding companies used to channel capital through countries. Such round tripping of capital may be part of MNEs tax planning strategies or a way of circumventing regulations imposed in certain jurisdictions. Given our focus on non-bank financial FDI, these issues are particularly pertinent for our sample of firms. As displayed in our descriptive statistics in Table 9, a large proportion of our sample relates to activities of brass plate entities that do not create sizeable economic activity in terms of employment. Therefore, in order to explore whether the determinants of location decisions are the same for FDI that creates employment and those that are brass plate in nature, we split our sample according to the number of employees reported by the firm. In column (3) of Table 8, we restrict our sample of firms to those who report having between one and 10 employ-ees.21 Note that with column (4) being larger employers, the number of observations falls significantly.

Comparing the results between columns (2) and (3) we find that, when looking at the low-employment firms, our results from the extended model are broadly unchanged. In column (4) however, we find much less significance. This may suggest that such firms are less sensitive to these

TABLE 8 Estimates of conditional logit model for non-bank financial firms' location decisions

(1) (2) (3) (4) (5) (6)

Variable (expected sign) Full sample Full sample Firms with 1-10 employees Firms with >10 employees Small firms Large firms

GDP, (1) 0.743*** 0.720*** 0.930*** 0.568*** 0.707*** 0.837***

(0.019) (0.019) (0.039) (0.080) (0.022) (0.039)

Per capita GDPt (1) -1.037*** -1.296*** -1.840*** -0.592* - 1.474*** -0.590***

(0.057) (0.060) (0.116) (0.275) (0.068) (0.136)

GDP growth, (1) 0.122*** 0.122*** 0.216*** 0.139** 0.134*** 0.068***

(0.011) (0.011) (0.021) (0.048) (0.013) (0.019)

Distance (-) -1.318*** -0.788*** -0.709*** -0.646*** -0.962*** -0.317***

(0.038) (0.048) (0.087) (0.224) (0.057) (0.081)

Corporation Taxt (-) -0.313*** -0.246*** -0.290*** -0.034 -0.229*** -0.196***

(0.018) (0.019) (0.043) (0.077) (0.022) (0.037)

Corp. Taxt2 (1) 0.005*** 0.004*** 0.005*** 0.000 0.004*** 0.003***

(0.000) (0.000) (0.001) (0.002) (0.000) (0.001)

Unemploy. ratet-1 (?) -0.096*** -0.105*** -0.176*** - 0.000 -0.109*** -0.096***

(0.007) (0.007) (0.016) (0.025) (0.008) (0.013)

Fin. developments (1) 0.012*** 0.013*** 0.014*** 0.013*** 0.013*** 0.013***

(0.000) (0.000) (0.001) (0.002) (0.000) (0.001)

OFFDIrestrictt-1 (-) -0.030*** -0.026*** -0.036*** 0.010 -0.004 -0.096***

(0.004) (0.004) (0.010) (0.018) (0.005) (0.008)

Internett (1 ) 0.083*** 0.093*** 0.113*** 0.043** 0.092*** 0.075***

(0.004) (0.004) (0.009) (0.016) (0.004) (0.008)

Fin. regulation (-) -0.198*** -0.187*** -0.238*** - 0.025 -0.108*** -0.409***

(0.018) (0.018) (0.033) (0.087) (0.022) (0.036)

Comcurrt (1 ) 0.599*** 1.078*** 0.401 0.553*** 0.690***

(0.054) (0.120) (0.254) (0.066) (0.094)

Colony (1) -0.224*** -0.468*** 0.007 -0.268*** -0.098

(0.045) (0.087) (0.199) (0.057) (0.075)

Common legal (1) 0.445*** 0.478*** 0.393* 0.468*** 0.407***

(0.038) (0.072) (0.194) (0.045) (0.070)

Com. language (1 ) 0.175*** 0.129 0.725*** 0.269*** 0.061

(0.049) (0.097) (0.257) (0.059) (0.091)

Contiguity (1 ) 0.467*** 0.687*** 0.005 0.457*** 0.440***

(0.057) (0.111) (0.265) (0.068) (0.104)

TABLE 8 (Continued)

(1) (2) (3) (4) (5) (6)

Variable Full Full Firms with Firms with Small Large

(expected sign) sample sample 1-10 employees >10 employees firms firms

N 191,160 191,160 58,915 7,840 122,662 68,498

Firms 8,519 8,519 2,621 352 5,464 3,055

Log-pseudolikelihood -18,630.90 -18,264.52 -4,991.97 -846.80 -11,931.74 -6,112.80

Note. The dependent variable Location equals 1 if a foreign affiliate is located in a country and 0 otherwise. Standard errors are robust to heteroskedasticity and clustered at the firm level. To be classified as a large company, a company must fulfill one of the following criteria (i) annual turnover must be €10 million or more, or (ii) total assets must be €20 million or more, or (iii) the number of employees is greater than 150. Small firms refer to those who do not meet these criteria. Standard errors in parentheses. *p<0.05; **p<0.01; ***p<0.001.

factors. However, it must be noted that the sample size in column (4) is much smaller than column (3). As such, particularly as the sign of the estimated impact is generally the same in columns (3) and (4), this difference may be due to the smaller sample size.

As an alternative approach to firm size, in columns (5) and (6) we split our sample into small and large firms. To be classified as a large company, a company must fulfill one of the following criteria (i) annual turnover must be €10 million or more, or (ii) total assets must be €20 million or more, or (iii) the number of employees is greater than 150 (with small firms failing at least one of these criteria). Note that in contrast to the employment division in columns (3) and (4), this division has a larger number of observations for the large firms. As can be seen, we find very similar results for the two groups, something that further supports the notion that the insignificances in column (4) may be driven by sample size rather than differences in behavior.

TABLE 9 Number of new financial foreign affiliates in the EU by year of incorporation (2004-2012)

All firms Firms with 1-10 employees Firms with >10 employees

Year N N N

2004 636 171 58

2005 802 232 48

2006 1,105 329 44

2007 1,356 425 52

2008 1,086 348 52

2009 893 275 44

2010 1,154 392 38

2011 1,038 351 21

2012 654 156 20

Total 8,724 2,679 377

Source. Authors' calculations.

TABLE 10 Average probability elasticities (APEs) of conditional logit model for non-bank financial firms' location decisions

(1) (2) (3) (4) (5) (6)

Variable (expected sign) Full sample Full sample Firms with 1-10 employees Firms with >10 employees Small firms Large firms

GDPt (1) 0.715*** 0.693*** 0.896*** 0.547*** 0.681*** 0.806***

(0.019) (0.019) (0.040) (0.080) (0.022) (0.039)

Per capita GDPt (1) —0.999*** -1.248*** 21 772*** -0.570* —1.419*** —0.568***

(0.057) (0.060) (0.116) (0.275) (0.068) (0.136)

GDP growth, (1) 0117*** 0.117*** 0.208*** 0.134** 0.129*** 0.066***

(0.011) (0.011) (0.021) (0.048) (0.013) (0.019)

Distance (-) -1.269*** -0.759*** -0.683*** -0.622*** -0.926*** —0.305***

(0.038) (0.048) (0.087) (0.224) (0.057) (0.081)

Corporation Taxt (-) -0.301*** -0.237*** -0.279*** -0.032 — 0.221*** — 0.189***

(0.018) (0.019) (0.043) (0.077) (0.022) (0.037)

Corp. Taxt2 (1) 0.005*** 0.004*** 0.004*** 0.000 0.003*** 0.003***

(0.000) (0.000) (0.001) (0.002) (0.000) (0.001)

Unemploy. ratet-1 (?) -0.093*** -0.101*** -0.169*** - 0.000 — 0.105*** — 0.092***

(0.007) (0.007) (0.016) (0.025) (0.008) (0.013)

Fin. developments (1) 0.011*** 0.012*** 0.013*** 0.012*** 0.012*** 0.013***

(0.000) (0.000) (0.001) (0.002) (0.000) (0.001)

OFFDIrestrict,-1 (-) -0.029*** -0.025*** -0.035*** 0.009 — 0.004 — 0.092***

(0.004) (0.004) (0.010) (0.018) (0.005) (0.009)

Internett (1) 0.080*** 0.089*** 0.109*** 0.042** 0.088*** 0.073***

(0.004) (0.004) (0.009) (0.016) (0.004) (0.008)

Fin. regulation (-) —0.191*** -0.180*** -0.229*** - 0.024 — 0.104*** — 0.394***

(0.018) (0.018) (0.033) (0.087) (0.022) (0.036)

Comcurrt (1) 0.577*** 1.038*** 0.386 0.533*** 0.664***

(0.054) (0.120) (0.254) (0.066) (0.094)

Colony (1) -0.216*** -0.451*** 0.007 —0.258*** — 0.094

(0.045) (0.087) (0.199) (0.057) (0.075)

Common legal (1) 0.429*** 0.460*** 0.378* 0.451*** 0.392***

(0.038) (0.072) (0.194) (0.045) (0.069)

Com. language (1) 0.169*** 0.124 0.698*** 0.208*** 0.059

(0.049) (0.097) (0.257) (0.059) (0.091)

TABLE 10 (Continued)

(1) (2) (3) (4) (5) (6)

Variable Full Full Firms with Firms with Small Large

(expected sign) sample sample 1-10 employees >10 employees firms firms

Contiguity (1) 0.450*** 0.662*** 0.005 0.440*** 0.424***

(0.057) (0.111) (0.265) (0.068) (0.104)

N 191,160 191,160 58,915 7,840 122,662 68,498

Firms 8,519 8,519 2,621 352 5,464 3,055

Log-pseudolikelihood -18,630.90 -18,264.52 -4,991.97 -846.80 -11,931.74 -6,112.80

Note. Figures shown are average probability elasticities (APEs) as described in Head and Mayer (2004). The APEs in the conditional logit model are calculated as follows: ex 5 bx(1 - (1 4 J)) where bx is the estimated parameter for x reported in Table 10 and J is the number of alternative countries in the choice set. The dependent variable Location equals 1 if a foreign affiliate is located in a country and 0 otherwise. Standard errors are robust to heteroskedasticity and clustered at the firm level. To be classified as a large company, a company must fulfil one of the following criteria (i) annual turnover must be €10 million or more, or (ii) total assets must be €20 million or more, or (iii) the number of employees is greater than 150. Small firms refer to those who do not meet these criteria. *p<0.05; **p<0.01; ***p<0.001.

As described in Greene (2012) the coefficients in the conditional logit model are not directly tied to marginal effects. We therefore follow Head et al. (1995) and Head and Mayer (2004) and calculate average probability elasticities (APEs) that have been applied in a number of related empirical studies (Head et al., 1995; Head & Mayer 2004; Nefussi & Schwellnus, 2010; Siedschlag, Zhang et al., 2013; Siedschlag, Smith et al., 2013).22 For the estimates from Table 8, these are reported in Table 10. As can be seen, these indicate that (setting aside the insignificance of coefficients when considering firms with more than 10 employees) the elasticities of the various factors is comparable across smaller and larger firms. Thus, combining the results we find that non-bank financial FDI behaves in ways comparable with other types of FDI: it is attracted to larger, fast-growing markets, is deterred by high labor costs, is restricted by the presence of geographic and cultural barriers, avoids regulation, and seeks to minimize tax payments.

In Table 11 we conduct a number of additional robustness checks on our results using subsamples of the data.23 In column (1) of Table 11, we show that our results are robust to the exclusion of firms whose home country for FDI is outside the OECD. In column (2), we instead limit our sample to firms whose home country is within the OECD. In addition, column (2) also excludes Ireland, Luxembourg, the Netherlands, and Switzerland. Our motivation for excluding these home countries relates to ownership structures and FDI pertinent to the financial sector. As noted by Claassen and Van den Dool (2013), these countries are prominent players in global headline FDI statistics despite having much smaller economies and financial centers compared with countries such as the United States and the United Kingdom and as such can act as a financial turntable for FDI transactions relating to SPVs. Therefore, headline FDI statistics for these countries are often distorted by these types of activity. Furthermore, some SPVs can be set up as orphan entities whose shares are held on trust for charitable purposes. These charitable trusts are often registered in jurisdictions such as the Netherlands and Ireland, which may potentially distort our ownership identification. We therefore control for such complex corporate structures used in these types of structured financial transactions by excluding these jurisdictions as home countries of FDI. Nevertheless, our main results are robust to their exclusion.

Furthermore, given the prominence of the United States as a source of FDI within our sample, we exclude the United States as a home country in column (3). The only significant difference is that now

TABLE 11 Estimates of conditional logit model for non-bank financial firms' location decisions on subsamples (robustness checks)

Variable (expected sign) (1) (2) (3) (4) (5) (6) (7)

Home country (OECD only) Home country (OECD excl. CH, IE, LU & NL) Home country (excl. USA) Home country (EU28 only) All NACE sectors (excl. fin. holding companies) Home country (excl. offshore fin. centers) Home country (excl. tax havens)

GDP, (+) 0.770*** 0.740*** 0.684*** 0.693*** 0.739*** 0.720*** 0.707***

(0.022) (0.025) (0.020) (0.024) (0.033) (0.020) (0.021)

Per capita GDP, (+) -1.131*** -1.256*** -1.266*** -1.201*** -0.059 -1.235*** —1.321***

(0.072) (0.082) (0.064) (0.080) (0.097) (0.063) (0.067)

GDP growth, (+) 0.105*** 0.108*** 0 J17*** 0.102*** 0.113*** 0i17*** 0.125***

(0.012) (0.014) (0.011) (0.013) (0.015) (0.011) (0.012)

Distance (-) -0.857*** -0.770*** -0.875*** -1.063*** -1.035*** -0.800*** -0.726***

(0.055) (0.075) (0.047) (0.050) (0.079) (0.050) (0.061)

Corporation Tax, (-) -0.183*** -0.189*** -0.206*** -0.122*** -0.345*** -0.232*** -0.247***

(0.021) (0.025) (0.020) (0.024) (0.025) (0.020) (0.021)

Corp. Tax,2 (+) 0.003*** 0.003*** 0.003*** 0.001** 0.005*** 0.004*** 0.004***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.001)

Un employ. rate,_i (?) -0.091*** -0.102*** -0.090*** -0.066*** 0.031*** -0.095*** -0.102***

(0.008) (0.009) (0.007) (0.008) (0.009) (0.007) (0.008)

Fin. development,_i(+) 0.012*** 0.012*** 0.012*** 0.010*** 0.005*** 0.012*** 0.012***

(0.000) (0.001) (0.000) (0.001) (0.001) (0.000) (0.000)

OFFDIrestrict,_i (-) -0.034*** -0.026*** -0.015** -0.010 0.048*** -0.024*** -0.015***

(0.005) (0.006) (0.005) (0.006) (0.005) (0.005) (0.005)

(Continues)

TABLE 11 (Continued)

(1) (2) (3) (4) (5) (6) (7)

Variable (expected sign) Home country (OECD only) Home country (OECD excl. CH, IE, LU & NL) Home country (excl. USA) Home country (EU28 only) All NACE sectors (excl. fin. holding companies) Home country (excl. offshore fin. centers) Home country (excl. tax havens)

Internet, (+) 0.082*** 0.090*** 0.089*** 0.075*** 0.040*** 0.092*** 0.099***

(0.004) (0.005) (0.004) (0.005) (0.005) (0.004) (0.005)

Fin. regulation, (-) -0.169*** -0.234*** -0.103*** -0.058* 0.115*** -0.168*** -0.195***

(0.022) (0.024) (0.019) (0.023) (0.029) (0.020) (0.021)

Comcurr, (+) 0.586*** 0.224** 0.687*** 0.815*** -0.036 0.589*** 0.423***

(0.055) (0.070) (0.053) (0.056) (0.085) (0.054) (0.061)

Colony (+) -0.305*** -0.759*** -0.079 -0.115 0.189** -0.190*** -0.361***

(0.049) (0.079) (0.053) (0.064) (0.069) (0.045) (0.065)

Common legal (+) 0.499*** 0.415*** 0.456*** 0.424*** 0.445*** 0.497*** 0.376***

(0.045) (0.060) (0.038) (0.048) (0.061) (0.040) (0.048)

Com. language (+) 0.094 0.564*** 0.338*** 0.274*** 0.629*** 0.0516 0.286***

(0.057) (0.081) (0.054) (0.070) (0.080) (0.053) (0.072)

Contiguity (+) 0.385*** 0.577*** 0.252*** 0.069 -0.053 0.485*** 0.641***

(0.062) (0.076) (0.059) (0.065) (0.092) (0.059) (0.068)

N 152,601 115,860 149,051 98,414 59,219 172,067 145,718

Finns 6,837 5,177 6,680 4,473 2,649 7,686 6,495

Log-pseudolikelihood -14,485.90 -10,957.21 -14,574.43 -9,710.09 -6,157.95 -16,500.04 -14,028.71

*Note. The dependent variable Location equals 1 if a foreign affiliate is located in a country and 0 otherwise. Standard errors are robust to heteroskedasticity and clustered at the firm level. Standard errors in parentheses. */?<0.05; **/?<0.01; ***/?<0.001.

TABLE 12 Estimates of mixed logit model for non-bank financial firms' location decisions

(1) (2) (3) (4) (5) (6)

Firms Firms

Variable Full Full with 1-10 with >10 Small Large

(expected sign) sample sample employees employees firms firms

GDP( (1) 0.843*** 0.810*** 1.098*** 0.745*** 0.764*** 0.967***

(0.0228) (0.0233) (0.0536) (0.102) (0.0277) (0.0414)

Per capita GDP, (1) — 1.074*** —1.295*** —1.703*** — 0.490 —1.478*** — 0.671***

(0.0567) (0.0610) (0.126) (0.309) (0.0707) (0.134)

GDP growth, (1) 0.121*** 0.114*** 0.181*** 0.145** 0.127*** 0.0719***

(0.0109) (0.0109) (0.0232) (0.0489) (0.0131) (0.0195)

Distance (-) — 1.391*** —0.784*** —0.683*** —0.581* —0.955*** — 0.311***

(0.0409) (0.0505) (0.0959) (0.233) (0.0598) (0.0863)

Corporation Taxi (-) —0.321*** —0.207*** 0.0107 0.0973 —0.153** —0.226***

(0.0181) (0.0351) (0.106) (0.144) (0.0481) (0.0385)

Corp. Tax,2 (1) 0.00525*** 0.00315*** — 0.000625 — 0.00213 0.00222* 0.00317***

(0.000344) (0.000675) (0.00206) (0.00274) (0.000932) (0.000717)

Unemploy. ratet-J (?) —0.0888*** — 0.103*** —0.191*** — 0.00752 —0.113*** —0.0787***

(0.00654) (0.00686) (0.0166) (0.0248) (0.00839) (0.0123)

Fin. developmentt-1(1) 0.0123*** 0.0129*** 0.0133*** 0.0131*** 0.0125*** 0.0140***

(0.000421) (0.000422) (0.000868) (0.00216) (0.000515) (0.000729)

OFFDIrestrictt-1 (-) —0.0677*** — 0.0755*** —0.218*** —0.0723* —0.0338*** — 0.151***

(0.00846) (0.00917) (0.0241) (0.0356) (0.00933) (0.0199)

Internet, (1) 0.0821*** 0.0874*** 0.0822*** 0.0223 0.0865*** 0.0818***

(0.00400) (0.00400) (0.00876) (0.0183) (0.00460) (0.00819)

Fin. regulation, (-) —0.247*** —0.227*** —0.242*** — 0.0974 —0.130*** —0.479***

(0.0180) (0.0194) (0.0421) (0.0961) (0.0231) (0.0366)

Comcurrt (1) 0.658*** 1.298*** 0.187 0.641*** 0.755***

(0.0522) (0.133) (0.267) (0.0617) (0.0973)

Colony (1) — 0.134** —0.328** 0.164 —0.196** — 0.0186

(0.0486) (0.109) (0.231) (0.0615) (0.0841)

Common legal (1) 0.445*** 0.542*** 0.465* 0.473*** 0.387***

(0.0401) (0.0828) (0.209) (0.0482) (0.0733)

Com language (1) 0.180*** 0.255* 0.845** 0.285*** 0.0352

(0.0519) (0.118) (0.291) (0.0625) (0.0954)

TABLE 12 (Continued)

(1) (2) (3) (4) (5) (6)

Firms Firms

Variable Full Full with 1-10 with >10 Small Large

(expected sign) sample sample employees employees firms firms

Contiguity (1) 0.561*** 0.933*** 0.115 (0.290) 0.543*** (0.0770) 0.551*** (0.107)

(0.0627) (0.136)

N 191,160 191,160 58,915 7,840 122,662 68,498

Firms 8,519 8,519 2,621 352 5,464 3,055

Log-likelihood -18,571.11 -18,187.46 -4,900.73 - 839.45 -11,891.64 - 6,076.38

*Note. The dependent variable Location equals 1 if a foreign affiliate is located in a country and 0 otherwise. The variables Corporation Tax, Corp. Tax2, Fin. Regulation, and OFFDIrestrict are specified as random. Standard errors are robust to heteroskedasticity and clustered at the firm level. To be classified as a large company, a company must fulfill one of the following criteria (i) annual turnover must be €10 million or more, or (ii) total assets must be €20 million or more, or (iii) the number of employees is greater than 150. Small firms refer to those who do not meet these criteria. Standard errors in parentheses. *p<0.05; **p<0.01; ***p<0.001.

the unusual negative coefficient on the colonial variable disappears, suggesting that this unexpected finding was driven by the United States-United Kingdom pair. Column (4) shows our results when we limit the sample to firms whose home country is within the EU28 (i.e., intra-E.U. FDI). As before, our main results hold, although contiguity and the overall FDI restrictiveness measure are now insignificant.

Next, we check our results by exploring heterogeneity across sectors within the non-bank financial sector. In column (5), the sample is restricted to financial NACE Rev. 2 sectors excluding the activities of financial holding companies. Given the large proportion of financial holding companies within our sample (almost 70 percent), the number of observations in this specification falls greatly resulting in a drop in significance of the coefficients. However, we now find that the financial regulation and FDI restriction coefficients are now significantly positive. This suggests that it is primarily the holding companies that seek to avoid such regulation, with other firms being attracted to such features. Alternatively, as these sectors are typically not subject to these financial regulations, this may be evidence of MNEs choosing the incorporation of new non-bank financial institutions over other forms of financial FDI to circumvent regulation. Outside of this change, however, we find a broadly similar story as in the baseline.

In columns (6) and (7) we exclude two sets of home countries that may be particularly sensitive to taxation. In column (6), combining the categorizations of Lane and Milesi-Ferretti (2011) and Claes-sens and Van Horen (2015), we exclude the following countries as offshore financial centers: Andorra, Antigua and Barbuda, Bahamas, Bahrain, Barbados, Bermuda, British Virgin Islands, Cayman Islands, Cyprus, Guernsey, Isle of Man, Jersey, Liechtenstein, Mauritius, Netherlands Antilles, Panama, Seychelles, and Singapore. Similarly, in column (7), following Davies, Martin, Parenti, and Toubal (in press), we exclude the Bahamas, Bermuda, Cayman Islands, Cyprus, Hong Kong, Ireland, Luxembourg, Malta, Singapore, and Switzerland as tax havens. In both cases, the pattern of coefficients is broadly the same in terms of significance and magnitude. This suggests that our results are not being driven by either group of countries.

As our last robustness check, we follow Merlo et al. (2016) and Merz et al. (2017) and use a mixed logit estimator for our baseline results and the firm size subsamples. These results are found in Table 12. As can be seen, the results are essentially equivalent to the conditional logit findings.24

4 | CONCLUSIONS

Despite the substantial growth and complexity of corporate structures in the non-bank financial sector and their large impact on headline FDI statistics in some European countries, empirical studies focussing on FDI location decisions in this sector are scarce. This is due to a general lack of granular data for this sector. Motivated by these data gaps, we build a unique firm-level dataset in order to overcome this problem. In this paper, we explore the determinants of location decisions for new foreign affiliates in the non-bank financial sector in Europe over the period 2004 to 2012.

Our results suggest that the probability of a country being chosen to host non-bank financial FDI is driven by many of the factors that affect FDI generally, that is, it increases with market size, falls with barriers between the host and the home countries, and declines with regulation and taxes. Given the large proportion of brass plate activity within the sector, we separately consider these firms, finding that they behave much like the sector as a whole. Finally, we conduct a series of robustness checks using different subsamples of firms to account for heterogeneity of determinants across different home countries of FDI, again finding consistency across groups.

While our results suggest that corporate taxes are a key factor in determining firms' location decisions, it also highlights the importance of traditional gravity controls such as sharing a common language and common legal system in determining non-bank financial FDI. Thus, concerns that this sector's investment will move en masse to low-tax jurisdictions may be somewhat assuaged. In addition, our analysis provides insight into the nature of non-bank financial FDI in Europe. Owing to a lack of official granular data for this sector, our sample suggests a high level of concentration within a few countries while non-bank FDI is also dominated by a few subsectors, most notably, the activities of financial holding companies. This implies that, by appropriately targeting data collection, a significant proportion of the data gaps in the non-bank financial sector can be filled.

ACKNOWLEDGMENT

The views expressed in this paper are those of the authors and do not necessarily reflect those of the Central Bank of Ireland, the European System of Central Banks (ESCB) or the ESRB. We thank the referee for valuable comments. We would also like to thank Daragh Clancy, Stefanie Haller, Hendrik Jungmann, Fergal McCann, Kitty Moloney, Gerard O'Reilly, Iulia Siedschlag, conference participants at the 2015 Annual Congress of the International Institute of Public Finance, 2015 European Trade Study Group Conference, 2015 Irish Economic Association Conference, and seminar participants at the Central Bank of Ireland for helpful comments on earlier drafts of this paper. Any remaining errors are our own.

1 The Financial Stability Board (FSB) define shadow banking as "credit intermediation involving entities and activities outside the regular banking system" or non-bank credit intermediation in short (FSB, 2013). However, there are a number of alternative definitions of shadow banking applied in the academic literature. For example, Claessens and Ratnovski (2014) define shadow banking as "all financial activities, except traditional banking, which require a private or public backstop to operate."

2 Banks and insurance companies' accounts are not available on the Amadeus database. Therefore, the majority of our sample consists of firms that would be classified within the OFI sector. Our sample includes multi-affiliate firms within the same corporate group. However, without consolidated information on these firms, these foreign affiliates cannot be identified or tracked.

3 An exception is Lawless et al. (2015) who include financial services firms in their sample of firms.

4 NACE is a statistical classification of economic activities developed in the European Community. NACE Rev. 2 was established by Regulation (EC) No. 1893/2006 of the European Parliament and of the Council on December 20, 2006.

5 While a number of the countries in our sample were not Member States of the European Union (Czech Republic, Estonia, Cyprus, Latvia, Lithuania, Hungary, Malta, Poland, Slovakia, and Slovenia joined the European Union in May 2004, while Romania and Bulgaria joined in January 2007) at the start of our sample period, the new accession countries represent only 7.84 percent of the final sample of host countries. When we exclude these countries, our results are qualitatively similar.

6 A number of factors may explain the large proportion of Dutch financial holding companies in our sample. As noted by the European Central Bank (ECB, 2014), 44 percent of the euro area shadow banking system remains unallocated to a broad and unspecified sector. They estimate that entities located in the Netherlands and Luxembourg account for approximately two-thirds of this residual. For the Netherlands, entities are most likely special financial institutions (SFIs) that comprise two-thirds of the Dutch shadow banking sector (ECB, 2014). As documented by Broos, Carlier, Kakes, and Klaaijsen (2012), SFIs are set up by mainly nonfinancial corporations (NFCs) for tax purposes, to attract external funding and to facilitate intragroup transactions. They suggest that most SFIs are classified as holding companies or group (finance) companies. Weyzig (2014) notes that there are approximately 12,000 of these entities located in the Netherlands that makes it the world's largest conduit country for FDI.

7 Imposing such a restriction is required in order to introduce our gravity related control variables that use bilateral home and host country information. However, this restriction reduces the sample of firms available by 69 percent owing to a large number of missing data for the GUO variable. Note that we use GUO data as of the data download date (2015).

8 Considering the home country of the GUO for foreign investments excluding the activities of financial holding companies shows a similar distribution with the United States again the main source of FDI followed by the United Kingdom.

9 The findings of Davies and Guillin (2014) validate the use of physical distance measures in empirical analyses on FDI in services.

10 Melitz and Toubal (2014) explore alternative measures of linguistic similarity beyond the common language dummy, finding similar results for the different measures. In order to best compare our results with the existing work on FDI, we, however, proceed with this measure.

11 In unreported results, we also included a variable measuring the time it takes for the average firm to prepare its taxes. Although statistically significant, the marginal effect was economically unimportant. As including this variable impacted the number of observations, we have excluded it here.

12 The "other finance" sector includes securities and commodities brokerage, fund management, and custodial services. As a robustness check, we also use the FDI restriction index for all industries and find similar results.

13 In alternative specifications, we replace this variable with a dummy variable that is equal to one if the financial supervisor has the power to place a troubled bank into insolvency. Our results are broadly similar using this alternative measure for the quality of the financial regulatory regime.

14 In alternative specifications, we include the number of commercial banks in the host country and find similar results.

15 This can also proxy for agglomeration effects as per Head and Mayer (2004). Barry, Gorg, and Strobl (2003) discuss these alternative interpretations for manufacturing FDI into Ireland.

16 We also compute heteroskedasticity-consistent standard errors and cluster at the country level that is in line with the methodology employed by Barrios et al. (2012). These estimates are not reported here, but are available, upon request, from the authors.

17 Note that these are for the sample used in the preferred baseline specification.

18 In some specifications the number of alternatives falls below 26 owing to missing data for some combinations of countries and years.

19 Inherent in this estimator, the i.i.d assumption imposes "Independence of Irrelevant Alternatives (IIA)." As is well known, this assumes that adding a new alternative cannot effect the relationship between a pair of existing alternatives.

20 As discussed below, this is driven by the United States and the United Kingdom.

21 We also run this specification on subsamples of firms who have no employees or who do not report employment data on Amadeus and obtain qualitatively similar results.

22 As noted by Nefussi and Schwellnus (2010), APEs are approximately equal to the estimated coefficients from the conditional logit model when there are a large number of choices in the choice set.

23 Additionally, we estimate a series of cross-sectional regressions for each year of data in our sample. Furthermore, we run our regressions on pre-crisis (pre-2008) and post-crisis year subsamples as a further modification of our data. In the majority of cases our results are unchanged. As a further robustness check, we estimate a nested logit model that relaxes the IIA assumption. While Hausman tests reject the assumption of IIA, the dissimilarity parameters are outside the normal range for some nests indicating inconsistency with profit maximization. That said, the results from the nested logit model are broadly similar to those obtained from our conditional logit model and as the vast majority of the location decisions' literature use the conditional logit model, we do so as well for comparability.

24 In line with Merz et al. (2017), we include the corporation tax (Corporation Tax, Corp. Tax2) and regulation (Fin. Regulation, OFFDIrestrict) variables as random in our mixed logit estimates in Table 12.

Ronald B. Davies ©http://orcid.org/0000-0002-9738-9256 REFERENCES

Bank for International Settlements (BIS). (2009). The Joint Forum: Report on special purpose entities. Basel: Basel Committee on Banking Supervision.

Barrios, S., Huizinga, H., Laeven, L., & Nicodeme, G. (2012). International taxation and multinational firm location decisions. Journal of Public Economics, 96(11-12), 946-958.

Barry, F., Gorg, H., & Strobl, E. (2003). Foreign direct investment, agglomerations, and demonstration effects: An empirical investigation. Review of World Economics, 139(4), 583-600.

Barth, J. R., Caprio, G., Jr, & Levine, R. (2001) The regulation and supervision of banks around the world—a new database (World Bank Policy Research Working Paper No. 2588). Washington, DC: World Bank.

Basile, R., Castellani, D., & Zanfei, A. (2008). Location choices of multinational firms in Europe: The role of EU cohesion policy. Journal of International Economics, 74(2), 328-340.

Blonigen, B. A., & Piger, J. (2014). Determinants of foreign direct investment. Canadian Journal of Economics, 47 (3), 775-812.

Broos, M., Carlier, K., Kakes, J., & Klaaijsen, E. (2012). Shadow banking: An exploratory study for the Netherlands (DNB Occasional Studies No. 1005). Amsterdam: Netherlands Central Bank, Research Department.

Buch, C. M. (2005). Distance and international banking. Review of International Economics, 13(4), 787-804.

Buch, C. M., & Lipponer, A. (2007). FDI versus exports: Evidence from German banks. Journal of Banking and Finance, 31(3), 805-826.

Claassen, P., & Van den Dool, G. (2013, October). The effects of including SPEs on BoP and FDI statistics. Paper prepared for the 26th Meeting of the IMF Committee on Balance of Payment Statistics, Oman.

Claessens, S., & Ratnovski, L. (2014). What is shadow banking? (IMF Working Papers no. 14/25). Washington, DC: International Monetary Fund.

Claessens, S., & Van Horen, N. (2014). Location decisions of foreign banks and competitor remoteness. Journal of Money, Credit and Banking, 46(1), 145-170.

Claessens, S., & Van Horen, N. (2015). The impact of the global financial crisis on banking globalization (DNB Working Papers No. 459). Amsterdam: Netherlands Central Bank, Research Department.

Davies, R. B., & Guillin, A. (2014). How far away is intangible? Services FDI and Distance. The World Economy, 37(12), 1731-1750.

Davies, R. B., Martin, J., Parenti, M., & Toubal, F. (in press). Knocking on tax haven's door: Multinational firms and transfer pricing. Review of Economics and Statistics.

Devereux, M. P., & R. Griffith, The taxation of discrete investment choices (IFS Working Paper No. W98/16). London: Institute for Fiscal Studies (1998a).

Devereux, M. P., & Griffith, R. (1998b). Taxes and the location of production: Evidence from a panel of US multinationals. Journal of Public Economics, 68(3), 335-367.

European Central Bank. (2014, October). Banking Structures Report. Frankfurt am Main: ECB.

European Systemic Risk Board. (2016, July) EU Shadow Banking Monitor, Issue 1. Frankfurt am Main: ESRB.

Financial Stability Board (FSB). (2013, August). Strengthening oversight and regulation of shadow banking: An overview of policy recommendations. Basel, Switzerland: FSB.

Goldberg, L. (2007). Financial sector FDI and host countries: New and old lessons. Economic Policy Review, 13(1), 1 -17.

Greene, W. (2012). Econometric analysis (7th ed.). New York: Pearson.

Grillet-Aubert, L., Haquin, J-B., Jackson, C., Killeen, N., & Weistroffer, C. (2016). Assessing shadow banking— non-bank financial intermediation in Europe (ESRB Occasional Paper No. 10). Frankfurt am Main: European Systemic Risk Board.

Grosse, R., & Goldberg, L. G. (1991). Foreign bank activity in the United States: An analysis by country of origin. Journal of Banking and Finance, 15(6), 1093-1112.

Head, K., & Mayer, T. (2004). Market potential and the location of Japanese investment in the European Union. The Review of Economics and Statistics, 86(4), 959-972.

Head, K., Ries, J., & Swenson, D. (1995). Agglomeration benefits and location choice: Evidence from Japanese manufacturing investments in the United States. Journal of International Economics, 38(3-4), 223-247.

Hebous, S., Ruf, M., & Weichenrieder, A. J. (2011). The effects of taxation on the location decision of multinational firms: M&A versus greenfield investments. National Tax Journal, 64(3), 817-838.

Huizinga, H., Voget, J., & Wagner, W. (2014). International taxation and cross-border banking. American Economic Journal: Economic Policy, 6(2), 94-125.

Kalinova, B., Palerm, A., & Thomsen, S. (2010). OECD's FDI restrictiveness index: 2010 (OECD Working Papers on International Investment No. 2010/3). Paris: OECD.

Lane, P. R., & Milesi-Ferretti, G. M. (2011). Cross-border investment in small international financial centres. International Finance, 14(2), 301-330.

Lawless, M., McCoy, D., Morgenroth, E., & O'Toole, C. (2015). Corporate tax and location choice of multinational firms (MPRA Paper No. 64769). Munich: Munich University Library.

McFadden, D. (1973). Conditional logit analysis and qualitative choice behaviour. In P. Zarembka (Ed.), Frontiers in econometrics, New York: Academic Press.

Melitz, J., & Toubal, F. (2014). Native language, spoken language, translation and trade. Journal of International Economics, 93(2), 351-363.

Merlo, V., Riedel, N., & Wamser, G. (2016). The impact of thin capitalization rules on the location of multinational firms' foreign affiliates (Research School of International Taxation Working Paper No. 1). Tubingen, Germany: Universität Tubingen.

Merz, J., Overesch, M., & Wamser, G. (2017). Tax vs. regulation policy and the location of financial sector FDI. Journal of Banking and Finance, 78, 14-26.

Nefussi, B., & Schwellnus, C. (2010). Does FDI in manufacturing cause FDI in business services? Evidence from French firm-level data. Canadian Journal of Economics, 43(1), 180-203.

OECD. (2014). FDI in Figures. Paris: OECD.

Ostry, J. D., Ghosh, A. R., Habermeier, K., Chamon, M., Qureshi, M. S., & Reinhardt, D. B. S. (2010). Capital inflows: The role of controls (IMF Staff Position Note No. 10/0). Washington, DC: International Monetary Fund.

Siedschlag, I., Zhang, X., & Smith, D. (2013a). What determines the location choice of multinational firms in the information and communication technologies sector? Economics of Innovation and New Technology, 22(6), 581-600.

Siedschlag, I., Smith, D., Turcu, C., & Zhang, X. (2013b). What determines the location choice of R&D activities

by multinational firms? Research Policy, 42(8), 1420-1430. UNECE. (2011). Impact of globalisation on national accounts (Note by UNECE Secretariat). Presented at 59th plenary session of the Conference of European Statisticians, Geneva, June 14-16, 2011.

Weyzig, F. (2014). The capital structure of large firms and the use of Dutch financing entities. Fiscal Studies, 35(2), 139-164.

Yamori, N. (1998). A note on the location choice of multinational banks: The case of Japanese financial institutions. Journal of Banking and Finance, 22(1), 109-120.

How to cite this article: Davies RB, Killeen N. Location decisions of non-bank financial foreign direct investment: Firm-level evidence from Europe. Rev Int Econ. 2017;00:1-26. https:// doi.org/10.1111/roie.12336

APPENDIX

A1 | Data cleaning

This section describes the data search criteria for obtaining our sample of firms on Amadeus. Prior to conducting our empirical analysis some essential data cleaning was performed. The search strategy was as follows:

• Firms classified as "active" on Amadeus;

• Firms corresponding to NACE Rev.2 64 (financial services activities, except insurance and pension funding), 65 (insurance, reinsurance, and pension funding, except compulsory social security) and 66 (activities auxiliary to financial services and insurance);

• Firms incorporated in EU28 countries (Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Germany, Denmark, Estonia, Spain, Finland, France, United Kingdom, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Poland, Portugal, Romania, Sweden, Slovenia, and Slovakia).

• Firms with at least one foreign shareholder (owning at least 10 percent equity) located in another country;

• Firms with unconsolidated financial accounts;

• Firms who were newly incorporated between 2004 and 2012 inclusive.

Based on the search criteria outlined above, Amadeus identifies just over 40,000 firms. However, we impose a number of essential data filters. First, we restrict our sample to firms whose GUO can be identified on Amadeus. Imposing such a restriction is required in order to introduce our gravity related control variables that use bilateral home and host country information. However, this restriction reduces the sample of firms available on Amadeus by 69 percent owing to a large number of missing data for the GUO variable. Second, we convert our data to long format that results in 27 observations per firm (26 zeros and a dummy variable equal to one if that country was the chosen location of the firm).

Third, we merge our firm-level data with country-specific controls (from various sources including the World Bank, Thomson Reuters Datastream, and KPMG) and gravity controls (taken from CEPII's Gravity dataset). Owing to missing data for some bilateral pairs in the CEPII Gravity dataset, 7,209 observations out of 329,859 observations were dropped representing just 2.2 percent

of our final sample. Fourth, we exclude Croatia from our analysis as a result of missing data for a number of our explanatory variables. Croatia represents a very small percentage of our final sample owing to the fact that only 16 firms were incorporated over our sample period once our data cleaning filters are applied.