Multiple taxes and alternative forms of FDI: evidence from cross-border acquisitions

Abstract

This paper explores the role of tax instruments in affecting foreign direct investment (FDI), paying particular attention on their effect on two forms of FDI strategy, ‘horizontal’ and ‘vertical’. Applying a decomposition of FDI strategies to the universe of cross-border mergers (the dominant form of FDI) over the period 1999–2010, it emerges that taxes have a much more nuanced effect on FDI than frequently suggested; while corporate taxes affect FDI negatively, the tax elasticity varies depending on the FDI strategy (with vertical FDI being in general more responsive), the exact measure of taxation, and international tax considerations (double taxation, withholding taxes). Sales taxes also affect FDI, but only horizontally.

Appendices

Appendices

1.1 Appendix 1: Country coverage and summary statistics

1.1.1 Country coverage

The common sample covers the following countries. Wage data of UBS (various years) refer to the cities in parentheses:

As source: Australia (Sydney), Austria (Vienna), Belgium (Brussels), Brazil (Sao Paulo), Canada (Toronto), China (Shanghai), Czech Republic (Prague), Denmark (Copenhagen), Finland (Helsinki), France (Paris), Germany (Frankfurt), Greece (Athens), Hongkong (Hong- kong), Hungary (Budapest), Indonesia (Djakarta), Ireland (Dublin), Italy (Milan), Japan (Tokyo), Mexico (Mexico City), Netherlands (Amsterdam), Norway (Oslo), Poland (Warsaw), Portugal (Lisbon), Russia (Moscow), Singapore (Singapore), Slovakia (Bratislava), South Africa (Johannesburg), Spain (Madrid), Sweden (Stockholm), Switzerland (Zurich), UK (London), USA (Washington).

The common sample covers the following host countries. Wage data of UBS (various years) refer to the cities in parentheses:

As host: Argentina (Buenos Aires), Australia (Sydney), Austria (Vienna), Belgium (Brussels), Brazil (Sao Paulo), Canada (Toronto), Chile (Santiago de Chile), Denmark (Copenhagen), Finland (Helsinki), France (Paris), Germany (Frankfurt), Greece (Athens), India (Mumbai), Indonesia (Djakarta), Ireland (Dublin), Israel (Tel Aviv), Italy (Milan), Japan (Tokyo), Korea (Seoul), Mexico (Mexico City), Netherlands (Amsterdam), New Zealand (Auckland), Norway (Oslo), Portugal (Lisbon), South Africa (Johannesburg), Spain (Madrid), Sweden (Stockholm), Switzerland (Zurich), Turkey (Istanbul), UK (London), USA (Washington).

1.1.2 Summary statistics of the raw data

	CBA	GDP	Net wage	Distance	Border	Invest. freed.	Trade freed.	Labour freed.	Shareholder righ.	EU*EU
Mean	7.10	9.8e+11	2.16	6.81	0.05	67.90	78.45	65.76	3.03	0.23
Std	25.77	2.0e+12	3.88	5.09	0.23	14.64	8.26	16.44	1.38	0.42
Min	0	5.0e+10	0.01	0.06	0	30	24	37	0	0
Max	513	1.2e+13	84.74	19.84	1	95	90	100	5	1
	(1-EU) *EU	EURO* EURO	(1-EURO) *EURO	Exchange rate	Corporate tax			Sales tax	Labour tax
					Statutory	EATR	EMTR
Mean	0.23	0.11	0.24	1.05	30.90	27.50	18.39	16.49	24.63
Std	0.42	0.32	0.43	0.55	6.72	5.68	5.42	5.48	8.062
Min	0	0	0	0.16	10	8.63	4.78	5	8
Max	1	1	1	4.70	51.56	43.77	32.44	25	46

1.2 Appendix 2: Data description and additional results

Variable	Description	Source
Dependent variable:
$\hbox {n}_{ijt}$	Number of cross-border acquisition deals between the source country $i$ and host country $j$ during year $t$	Compiled
Tax variables
Corporate tax (statutory)	Statutory tax rate on corporate income in country $j$	KPMG, Corporate and Indirect Tax Survey
Corporate tax (EATR)	Effective average tax rate (EATR) on corporate income in country $j$	CBT Tax Database (2012)
Corporate tax (EMTR)	Effective marginal tax rate (EMTR) on corporate income in country $j$. This is calculated by the difference between the pre-tax and post-tax required rates of return	CBT Tax Database (2012)
Sales tax	Value-added tax (VAT) rate and other sales taxes	IMF, Tax Policy Division
Labour tax	Compulsory social security and income tax contributions in per cent of gross salaries as published in the Prices and Earnings survey of UBS. For the first part of our sample, the Prices and Earnings survey is only published triennially. Values of the missing years have been filled with the closest observation available. In particular, the values of the years 1999 and 2001 employ the 2000 data, the values for the years 2002 and 2004 employ the 2003 data, and the values for the years 2005 and 2007 employ the 2006 data. Since 2008 yearly updates of the Prices and Earnings survey are available	UBS, Prices and Earnings. See also Braconier et al. (2005)
Withholding tax	Withholding tax between countries assuming that profits are repatriated in form of dividends	KPMG, Corporate and Indirect Tax Survey. Deloitte International Tax Source.
Control variables:
Border	Common border between source and host country	Compiled
Distance	Great circular between the capital city of the source and host country	Compiled
$EU_{it}*EU_{jt}$	Variable indicating the EU membership of the source and host country	Compiled
$(1-EU_{it})*EU_{jt}$	Variable indicating the EU membership of the host (but not the source) country	Compiled
$EURO_{it}*EURO_{jt}$	Variable indicating that the source and host country share the Euro as common currency	Compiled
$(1-EURO_{it})*EURO_{jt}$	Variable indicating the EURO membership of the host (but not the source) country	Compiled
Exchange rate	Real (bilateral) exchange rate with US$	World Development Indicators
GDP	Real gross domestic product in US$ with base year 2000 of the host country $j$	World Development Indicators
Investment freedom	Index of freedom of investment referring to whether there is a foreign investment code that defines the country’s investment laws and procedures; whether the government encourages foreign investment through fair and equitable treatment of investors; whether there are restrictions on access to foreign exchange; whether foreign firms are treated the same as domestic firms under the law whether the government imposes restrictions on payments, transfers, and capital transactions; and whether specific industries are closed to foreign investment	Heritage Foundation
Labour freedom	Index of labour market freedom on a scale from 10 to 90 measuring dimension such as minimum wages, regulation against layoffs, regulatory burden on hirings etc.	Heritage Foundation
Net wage	Wage in the host country net of compulsory social security contributions as published in the Prices and Earnings survey of UBS. Wages are measured by an index referring to the hourly income of 13 comparable professions as paid in the capital city or the financial centre of a country. For the first part of our sample, the Prices and Earnings survey is only published triennially. Values of the missing years have been filled with the closest observation available. In particular, the values of the years 1999 and 2001 employ the 2000 data, the values for the years 2002 and 2004 employ the 2003 data, and the values for the years 2005 and 2007 employ the 2006 data. Since 2008 yearly updates of the Prices and Earnings survey are available	UBS, Prices and Earnings. See also Braconier et al. (2005)
Shareholder rights	Shareholder rights are measured by an anti-directors rights index reflecting (i) the possibility of shareholders to mail their proxy vote, (ii) whether shareholders are required to deposit their shares prior to the General Shareholders Meeting (iii) whether cumulative voting is allowed (iv) an oppressed minorities mechanism exists (5) whether the minimum stake allowing shareholders to call for an extraordinary shareholders meeting is more or less than 10 %. Higher values mean more power for shareholders	La Porta et al. (1998)
Trade freedom	Index of freedom of international trade (tariff and non-tariff barriers) on a scale from 10 to 90	Heritage Foundation

1.3 Appendix 3: On the choice of the Poisson regression

Econometric models that are capable to handle location choices include the conditional logit model, where $h_{ijt}^{d}$ is the dependent variable. The conditional logit models takes the joint distribution over all deals $d$, source countries $i$, host countries $j$, and the 11 years $t$ under consideration enter the log-likelihood function $\ln L_{cl} = \sum _{d=1}^{D}\sum _{i=1}^{N} \sum _{j=1}^{J} \sum _{t=1}^{T} \ln (P_{ijt}^{d})$ with $P_{ijt}^{d}$ defined by (9). Since $P_{ijt}^{d}=P_{ijt}$, the number $n_{ijt}$ of CBAs can be factored out, that is $L_{cl} = \sum _{i=1}^{I}\sum _{j=1}^{J} \sum _{t=1}^{T} n_{ijt} P_{ijt}$. Inserting (9) yields

$$\begin{aligned} \ln L_{cl}= & {} \sum _{i=1}^{I} \sum _{j=1}^{J}\sum _{t=1}^{T} n_{ijt} (\widetilde{x}_{ijt}\beta + \widetilde{\tau }_{ijt}\gamma +\delta _{j}) \nonumber \\&-\,\sum _{i=1}^{I}\sum _{j=1}^{J}\sum _{t=1}^{T} \bigg [n_{ijt}\ln \bigg (\sum _{i=1}^{I} \sum _{j=1}^{J}\sum _{t=1}^{T} \exp (\widetilde{x}_{ijt}\beta + \widetilde{\tau }_{ijt}\gamma +\delta _{j})\bigg )\bigg ], \end{aligned}$$

(10)

from which the coefficients $\beta$ and $\gamma$ can be estimated.

In practice, a caveat against the conditional logit model is that it can require massive amounts of data for estimation. To avoid this caveat, Guimarães et al. (2003) have proposed to turn to the Poisson regression for the coefficient estimation in location choice models. This assumes that $n_{ijt}$ is Poisson distributed, that is $Prob[n = n_{ijt}] = [\exp (-\lambda _{ijt})\lambda _{jt}^{n_{ijt}}]/n_{ijt}!$ whilst an exponential mean transformation connects the Poisson parameter $\lambda _{ijt}$ with the explanatory variables of (8), that is $E[n_{ijt}] = \lambda _{ijt} = \exp (\widetilde{x}_{ijt}\beta + \widetilde{\tau }_{ijt}\gamma + \delta _{i}+\delta _{j} + \delta _{t})= \alpha _{it}\exp (\widetilde{x}_{jt}\beta + \widetilde{\tau }_{ijt}\gamma +\delta _{j})$. For our case with panel data, $\alpha _{it}= \exp (\delta _{i}+\delta _{t})$ absorbs the heterogeneity from different source countries and years and is here treated as fixed effect. Guimarães et al. (2003) have shown that the concentrated log-likelihood function, which no longer depends on $\alpha _{it}$, equals

$$\begin{aligned} \ln L_{pc}= & {} \sum _{i=1}^{I} \sum _{j=1}^{J}\sum _{t=1}^{T} n_{ijt} (\widetilde{x}_{ijt}\beta + \widetilde{\tau }_{ijt}\gamma +\delta _{j}) \nonumber \\&-\,\sum _{i=1}^{I}\sum _{j=1}^{J}\sum _{t=1}^{T} \bigg [n_{ijt}\ln \bigg (\sum _{i=1}^{I} \sum _{j=1}^{J}\sum _{t=1}^{T} \exp (\widetilde{x}_{ijt}\beta + \widetilde{\tau }_{ijt}\gamma +\delta _{j})\bigg )\bigg ]+C. \end{aligned}$$

(11)

Since (11) differs from (10) only as regards the constant $C$, the estimates of $\beta$ and $\gamma$ of a Poisson regression and a conditional logit model are identical!

Owing to different asymptotic assumptions, in small samples, the standard deviations can differ between the logit model and the Poisson regression. However, Schmidheiny and Brülhart (2011, p. 219) show that clustering at the group level $\alpha _{it}$ yields asymptotically identical SE. For our case with thousands of count variables that reflect many more location choices embodied in CBA deals, these asymptotic properties are likely to hold as long as the SE are appropriately clustered.

As long as the variables are transformed into logarithms, the coefficients ($\beta$ and $\gamma$) of the Poisson regression have the interpretation of an elasticity with respect to the expected number of acquisitions $E[n_{ijt}]$. Hence, the (direct) tax elasticity $\eta$, given by

$$\begin{aligned} \eta = \frac{\partial E[n_{ijt}]}{\partial \tau _{ijt}}\frac{\tau _{ijt}}{E[n_{ijt}]}=\gamma , \end{aligned}$$

(12)

is constant.²⁷