The Impact of Terrorism on firm-level Mergers and
Acquisitions (M&As). Business-Related versus
Non-Business-Related Terrorism.
by
Babet Hogetoorn S2190001
b.hogetoorn@student.rug.nl
Master thesis for Msc. International Economics and Business Faculty of Economics and Business
University of Groningen
June 2017
2
Abstract
The increasing number of terrorist attacks and its related fear and anxiety has urged researchers to investigate the impact of terrorism on economic decisions, e.g. on foreign direct investments (FDI). This paper builds on the existing literature by realizing three improvements. First, the effect of different forms of terrorism, namely business-related terrorism (BT) and non-business-related terrorism (NBT), is examined. Second, as the level of analysis is firm-level instead of country-level, this paper controls for firm characteristics. Finally, bilateral (firm-level) merger and acquisition (M&A) data is used, instead of FDI data. Bilateral M&A data is preferred as it is based on where the decision is made, instead of where the investment flow comes from. Taken together, this paper investigates the impact of terrorism, BT, and NBT on cross-border firm-level M&As. To examine the effects, linear probability models and regressions models are applied using bilateral M&A data for 8412 firms from 120 countries over the period 2009-2015. It is found that terrorism and its two forms, BT and NBT, in a host country depress both the probability a foreign firm will perform a M&A investment in that host country, and lower the M&A investment value, conditional on entry. Furthermore, compared to NBT, terrorism targeted at firms has a more detrimental effect on the decision, but not on the M&A investment value, conditional on entry.
3
1. Introduction
9/11 Marks the anniversary of one of the worst terrorist attacks on American soil in U.S. history. On this date in 2001, terrorists associated with the Islamic extremist group Al-Qaeda, hijacked planes and deliberately flew them into major landmarks in America. The attack killed more than 2500 people, injured over 6000 others and caused billions of dollars in property and infrastructure damage.
Besides the financial damage and casualties, terrorist attacks induce fear and generate anxiety that make people more cautions and consequently might impact the decisions they make (Mirza & Verdier, 2008). The fear for more terrorist attacks has been legitimate as terrorism increased since 2004, with a steep rise from 2011 till 2014 (Figure 1).
Figure 1. Total number of Terrorist attacks in the World
Source: Global Terrorism Database (2016), https://www.start.umd.edu/gtd/
An increase in the number of terrorist attacks has urged researchers to investigate the effect of terrorism on the actions and decisions people make. In this paper, terrorism is defined as “the premeditated use or threat of use of extranormal violence or brutality by individuals or subnational groups to obtain a political, religious, social, or ideological objective through intimidation of a large audience, usually not directly involved with the decision making” (Llorca-Vivero, 2008; Rosendorff & Sandler, 2005; Enders & Sandler, 2000; Krueger & Malecková, 2003). One direction of interest of previous papers is the effect on foreign direct investment (FDI) as these investments have increased the last years and are an important source for economic growth (Di Giovanni, 2005; Bandyopadhyay, Sandler & Younas, 2014). Researchers found that (the risk of) terrorism reduces country-level FDI (Enders, Sachsida & Sandler, 2006; Bandyopadhyay et al., 2014; Shahzad, Zakaria, Rehman, Ahmed & Fida, 2016). For example, Abadie and Gardeazabal (2008) concluded that one standard deviation increase in terroristic risk reduces the net FDI position of the country by approximately 5% of gross domestic product (GDP). Besides, Enders and Sandler (1996) found that terrorism reduced annual (country-level) FDI by 13,5% and 11.9% in Spain and Greece, respectively. Although scholars have reached a consensus on the negative effect of terrorism on macro-level FDI, three main points for improvements can be identified. This paper implements the three improvements and thereby tries to build on the existing knowledge.
-2000 3000 8000 13000 18000 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
4 First, previous researchers overlooked the possible different effects of various forms of terrorism on FDI. Powers and Choi (2012) were the first in examining the different effects of business-related terrorism (BT) versus non-business-related terrorism (NBT). As it might suggest, BT attacks are targeted “on business or private citizens patronizing a business” (GTD, 2016), whereas NBT are aimed at all other types of targets or victims (see section 3.2). This paper follows the argument of Powers and Choi (2012) that not all forms of terrorism have the same effect on investment decisions. More specific, the former type of terrorism directly harms the multinational enterprise (MNE) by, for example, potentially assassinating employees, damaging assets, or disrupting the supply chain. Hence, MNEs might be more prudent regarding investing in countries where businesses are frequently targeted (Powers & Choi, 2012). On the other hand, MNEs might be less concerned with doing investments in countries prone to terrorist attacks that do not pose a direct threat to businesses (Powers & Choi, 2012). Therefore, in investigating the effect of terrorism on MNEs’ investment decisions, the distinction between BT and NBT is required and an interesting research direction. Powers and Choi (2012) concluded that transnational BT terrorism significantly decreases aggregate country-level FDI while transnational NBT terrorism does not. However, Powers and Choi (2012) used country-level FDI data and their paper is thus subjected to two limitation outlined below. So, their paper can be improved and their results should be confirmed.
Second, papers investigating FDI do not consider firm heterogeneity, e.g. firms’ interests in specific countries. More specific, previous papers looked at aggregate country-level FDI and do not consider the characteristics of firms performing these investments. For example, compared to Unilever, Shell might be less sensitive for terrorism in Kazakhstan as this country holds a large oil reserve. Shell might accept (increased) risk for this benefit and (keep) performing mergers and acquisitions (M&As). Not controlling for firm heterogeneity might therefore have impacted the results of previous researchers. To overcome this limitation, this paper investigates the impact of terrorism on aggregate firm-level investments and controls for firm heterogeneity.
Finally, most of these papers investigate the effect of terrorism on aggregate FDI flows (Powers & Choi, 2012; Shahzad et al., 2016; Enders & Sandler, 1996). This problem has already been identified and tackled by Ouyang and Rajan (2016). Decisions to perform FDI are usually made at the originating country of the firm, often the location of the headquarter. But, MNEs funnel a large proportion of FDI through offshore financial centers (OFCs) to obtain tax benefits (Hattari & Rajan, 2009; Ouyang & Rajan, 2016). Besides, enterprises establish special purpose vehicles (SPVs) to raise capital in countries with more favorable borrowing rates. As the decision to perform FDI is made in another country (headquarter country) than where the flow of FDI is coming from (OFC or location of SPV), looking at FDI data can be misleading in trying to understand linkages between countries (Ouyang & Rajan, 2016). Ouyang and Rajan (2016) propose to examine bilateral M&A data to overcome these data concerns. More specific, they explain that this data is “based on ownership of companies as opposed to flows of funds”. Following their advice, this paper considers bilateral (firm-level) M&A data as opposed to FDI data.
5 another source country into this specific host country. Following the conclusion of Powers and Choi (2012), the second research gap is whether BT has a more harming effect on firm-level cross-border M&As than NBT. Cross-border M&A is used as this paper is interested in FDI. Hence, based on the identified gaps, the research questions have been formulated as follows:
RQ 1: What is the effect of terrorism, BT, and NBT on firm-level cross-border M&As? RQ2: Does BT harm firm-level cross-border M&As more than NBT?
It is practically relevant to understand the effects of different forms of terrorism. Researchers showed that different forms of violence and conflicts have different effects on FDI (Blomberg & Hess, 2004; Nitsch & Schumacher, 2004; Blomberg & Mody, 2005). Following these conclusions, Oh and Oetzel (2011) investigated the impact of moderating governance characteristics on the relation between different disasters and strategic decisions of firms. They concluded that the importance of quality governance does vary given the type of disaster. So, governments should act differently on different types of disasters. Furthermore, Ouyang and Rajan (2016) found that good institutions negate the negative effects of terrorism. If this study shows that BT and NBT have different effects on M&A investments, this infers that government and/or policy makers should also respond differently to (the risk of) different forms of terrorism. For example, different counter terrorism policies should be implemented aimed at reducing the harming effect of BT and NBT on M&As. Furthermore, if the results show that BT has a more negative effect on M&A than NBT, this inference is even more important considering the fact that the number of BT attacks have more than doubled since 2009 (GTD). Responding in the incorrect way might even worsen the effect. However, these presumptions should be studied by future researchers (see further).
Besides the practical relevance, answering these research questions is also theoretically relevant. First, the conclusions add knowledge to the determinants of firm-level FDI (Belonigen, 2005) and cross-border M&As (Xie, Reddy & Liang, 2017). Second, this paper contributes to the literature examining the effect of terrorism, and its different forms, on investment decisions (Powers & Choi, 2012; Bandyopadhyay et al, 2014; Ouyang & Rajan, 2016). Finally, it leads to new further research ideas. For example, if results conclude different impacts of (N)BT, then it would be interesting to investigate the effectiveness of different anti-terrorism policies on the two forms of terrorism. Other ideas are presented in the conclusion, section 5.
To answer the research questions, linear probability models (LPMs) and ordinary least squares (OLS) regression models are applied using bilateral M&A data for 8412 firms from 120 countries over the period 2009-2015. The former type of models tests the effect of (N)(B)Terrorism on the extensive margin. That is, the effect of (N)(B)Terrorism on the probability a foreign firm decides (yes/no) to perform a M&A investment in a host country experiencing terrorism. The latter type of models tests the effect of (N)(B)Terrorism on the intensive margin. Or, the value of M&A investments, conditional on entry.
6 lower the M&A value invested, conditional on entry. Besides, compared to NBT, terrorism targeted at firms has a more detrimental effect on the decision, but not on the M&A value, conditional on entry. In other words, a firm is more prudent to invest in countries that experience BT. However, if a firm decides to enter such a country, BT has not a more negative impact on the value invested than NBT.
The remainder of the paper is organized as follows. Section 2 provides an overview of previous research examining the impact of terrorism. In this section, the hypotheses and their grounding are also presented. Section 3 outlines the methodology. The econometric model, the data sources, descriptive statistics and an explanation of the LPM and regression model are provided. Section 4 discusses the results. Finally, section 5 concludes the paper.
2. Literature review
Everyone reading the daily newspaper or watching the news on television, can understand we live in a disaster-prone world (Faulkner, 2001; Oh & Oetzel, 2011). Disasters can be split in three main categories; natural disasters, technological disasters and terrorist attacks (Oh & Oetzel, 2011). Some of the many examples are the tsunami in Southeast Asia in 2004, the Chernobyl disaster in 1986, and the 9/11 attack already detailed in the introduction of this paper, respectively. This paper focuses on the latter type of disaster, terrorism.
In research, it is suggested that disasters can both have negative and positive outcomes (Faulkner, 2001; Oh & Oetzel, 2011). Regarding the negative consequences, disasters bring both direct as well as indirect costs to organizations. Examples are lost or damaged assets and their replacing or repairing costs, loss of lives, loss of market opportunities, damaged infrastructure and negative impact on partners. The positive effects resulting from disasters can be attributable to the so called ‘stimulus effect’ and ‘productivity effect’ (Hallegatte & Przyluski, 2010). Concerning the stimulus effect, the damaging effects of disasters urge for reconstruction. As a result, this increases demand for (reconstruction) products, resources and production capacity. In this way, the disaster can act as a stimulus for economic growth. For example, in Turkey, the 1999 earthquake led to an increase in economic activities (Hallegatte & Przyluski, 2010). A condition for a disaster to bring stimulus advantages is that the economic resources of the country are not yet fully used (Hallegatte & Ghil, 2008). The need for reconstruction can also result in positive productivity effects. A disaster damages capital and assets which have to be repaired to ensure ongoing business. The assets can then be replaced with newer and more productive technologies. For example, new production technologies (e.g. new machines) and new infrastructure (Hallegatte & Przyluski, 2010). Nonetheless, different findings regarding productivity advantages are presented in literature. Albala-Bertrand (1993) studies the effect of 28 natural disasters on 26 countries in the period between 1960 and 1979. He concluded that GDP can increase after a disaster (partially) due to the replacement by more efficient capital. However, Hallegatte and Dumas (2009) stated that “disasters can influence the short-term growth rate, in the few years following each disaster, and the long-term production level, but cannot influence the long-term growth rate”.
7 related costs. As firms are willing to reduce the possibility of incurring these costs, they may opt to not invest in these countries. On the other hand, terrorism (disasters) may also present valuable opportunities for MNEs to take advantage of, as explained above. This would enhance new M&A investments in these more risky and uncertain countries (Oh & Oetzel, 2011; Li & Tong, 2007).
Besides the (possible) positive and negative effects of terrorism (disasters), terrorism also creates uncertainty regarding the possibility of future terroristic attacks (Abadie & Gardeazabal, 2008). This anxiety further shape investors’ economical decisions as the terroristic risks are included in their considerations (Mirza & Verdier, 2008). Besides, the fear associated with terrorism is often socially amplified (Oh & Oetzel, 2011). This means that, although the real risk of terrorism might be small, the risk is magnified and seen as a real concern in society. Furthermore, the perceived benefits and costs of (risk of) terrorism are (assumed) higher. As a result, firms react more dramatic than necessary with respect to the actual threat of terrorism to the firm (Kasperson, Kasperson, Pidgeon & Slovic, 2003).
8 Unfortunately, theory does not provide a clear direction whether the (perceived) costs or benefits outweigh. Moreover, this might even be firm- and terrorist-attack-specific. Research on the effect of disasters, and in specific terrorism, has been performed on different economic decisions. To examine the general (positive or negative) effect of terrorism, these papers will be reviewed in the next section. The conclusions and inferences from these previous papers will then be further used in hypotheses forming.
2.1 The effect of terrorism on economic decisions
Although terrorism is not a new phenomenon, since the 9/11 terrorist attack in New York there has been renewed interest in the impact of terrorism on different economic decisions (Mirza & Verdier, 2008). Examples are researchers investigating the impact of terrorism on economic activity (Abadie & Gardeazabal, 2008; Blomberg, Hess & Orphanides, 2004), on bilateral trade (Egger & Gassebner, 2015; Nitsch & Schumacher, 2004; Mirza & Verdier, 2008), international business (Czinkota, Knight, Liesch & Steen, 2010) and FDI (Shahzad et al., 2016; Abadie & Gardeazabal, 2008). Less exhaustive research has been done on the impact of terrorism on M&As (Ouyang & Rajan, 2016), and the effect of different forms of terrorism (Power & Choi, 2012).
As discussing the effect of terrorism on all the different economic decisions would be out of scope for this paper, this section will focus on the findings on the relationship between terrorism and bilateral trade, FDI and M&A. These topics are chosen because the first relation (terrorism-trade) has been thoroughly researched and thus provides a solid direction. The other relations (terrorism-FDI, terrorism-M&A) are the topics of interest of this paper. In the introduction, it was already explained why this paper adds new insights to these latter relations.
The papers examining the effect of terrorism on trade conclude a negative relationship. More specific, terroristic attacks detriments (bilateral) trade (Mirza & Verdier, 2008). For example, Nitsch and Schumacher (2004) found in their study that bilateral trade is reduced by 4% when the number of terrorist attacks are doubled. Egger and Gassebner (2015) also found a negative effect of terrorism on bilateral trade. However, they state that this effect is not strong, and almost negligible. Besides, this small effect is only observable in the medium run. Blomberg and Hess (2004) examined the effect of different forms of violence, of which terrorism is one. They found that terrorism has a negative effect on bilateral trade, but that the decline in trade due to a terrorist attack (-7.6%) is lower than other forms of violence, like external conflict (-20.8%) and inter-ethnic conflict (-17.8%).
Similar to trade, terrorism also limits (bilateral) FDI. This negative effect is found regardless of different measures of terrorism, e.g. by number (Enders et al., 2006) or intensity (Abadie & Gardeazabal, 2008), regardless of domestic or international terrorism (Bandyopadhyay et al, 2014), and regardless of the country of interest, e.g. Greece and Spain (Enders & Sandler, 1996) or Pakistan (Shahzad et al., 2016).
9 So, it can be inferred that terrorism negatively influences trade, FDI and M&A. Therefore, it seems that (risk of) terrorism increases the (perceived) costs of terrorism related to FDI more than the (perceived) benefits. Therefore, the first hypothesis for the first research question is:
H1: Terrorism has a negative impact on firm-level cross-border M&As
This paper is further interested in the effect of different forms of terrorist attacks on firm-level cross-border M&As. In the next section, conclusions from previous papers and the OLI-paradigm of Dunning (1998) are used in forming the hypotheses of the effect of BT and NBT on M&As.
2.2 The effect of BT and NBT of M&As
Previous research show that different forms of violence have different negative impacts on trade and FDI. Blomberg and Hess (2004) explored the effect of four different types of violence, namely terrorism, external conflict, revolutions and inter-ethnic fighting. Among others, two conclusions can be drawn from this paper. First, all different types have a negative effect on trade (although external conflict is not significant). Second, the coefficients show a different negative magnitude for the different types of violence. Nitsch and Schumacher (2004) also showed that different measures of internal and external conflict (assassinations, guerilla activities, purges, riots, and revolutions) show negative but different impacts on bilateral trade. Finally, Blomberg and Mody (2005) explored the effect of different forms of violence on bilateral FDI and also concluded the same. More specific, both terrorism and war and revolutions have a negative effect on FDI. In their paper, war and revolutions indicate a more harming effect than terrorism.
Powers and Choi (2012) were the first to show the different effects of terrorism aimed at different targets on FDI. In their paper, they investigated the different effects of BT and NBT on FDI. They made this distinction because they believed not all forms of terrorism have the same detrimental effect on economic activity. This idea is legitimate considering the findings above. Support for their idea was found. More specific, both BT and NBT showed negative coefficients. However, only BT has a significant negative effect on FDI.
Hence, two inferences can be drawn from the papers discussed. First, different forms of violence and terrorism have a negative impact on economic decisions, here trade and FDI. Therefore, the second hypothesis for the first research question is formulated as follows:
H2: Both BT and NBT have a negative effect on firm-level cross-border M&As.
10 2.3 The different effect of BT and NBT on M&As
Already touched upon in the first part of this literature review, the OLI-paradigm of Dunning (1998) is a useful tool in hypothesizing FDI (M&A) decisions. Here, it is used to provide examples and intuitions in favor of the idea that BT has a more harming effect on L- and I-advantages and thus on M&A investments. As the second hypothesis specified a negative relationship of BT and NBT on M&A, the focus in this section will be only on the possible harming effects.
Regarding L-advantages, it is expected that BT has a more harming effect on M&A investments than NBT. In other words, BT in a host country lowers the interests of investing firms more than NBT.
First, the hedonic market approach states that employees should be compensated for the distress of working in a region (highly) exposed to terrorism (Frey, Luechinger & Stutzer, 2007). In countries prone to BT, the employees and/or the firms they are working for, are the main target. Firms’ employees or the firms themselves are not a direct target of NBT. Logically, employees in countries often encountering BT attacks are exposed to higher risk and higher wages are thus demanded to compensate for that increased risk. As low wages are a form of L-advantages, BT lower these L-advantages more than NBT. Consequently, firms will be less inclined to merge with, or acquire, a firm located in a country threatened by BT than by NBT (Powers & Choi, 2012). Unfortunately, it is difficult to quantify the effect of terrorism on wages due to other impacting variables, like government reactions or inflation. Second, more BT means that future key value chain partners, suppliers and transportation providers (they are L-advantages why a firm might perform a M&A) are a more likely target. As a result, there are higher risks of interruptions in the supply chain and thus a higher risk of future costs (of doing business) (Czinkota et al., 2010). Again, this is less the case with NBT, as firms are not the main target. This higher risk concerned with BT might lower a firm’s interest in a specific location. With respect to the eclectic paradigm, this thus (further) reduces L-advantages (Powers & Choi, 2012).
Third, firms (or its managers) might be more aware of BT than of NBT in the host country because BT attacks are aimed towards ‘itself’ (or the business), while NBT has other targets. In other words, BT poses a higher perceived risk to the firm. Moreover, this perceived risk and the consequences of BT can be subject to social amplification and thus be overestimated even more. Taking together, this (overstated) risk might lead to the decision to not to invest or to invest in one of the many other countries of choice (Frey et al., 2007).
Countries prone to terrorism implement increased security measures which result in delays, higher costs, less reliable lead times and less reliable demand scenarios, thereby lowering the efficiency of the national transportation and logistical system (Czinkota et al., 2010; Sheffi, 2001). With respect to the eclectic paradigm, this makes a location less attractive (Powers & Choi, 2012). However, no argument can be provided why BT increases the costs related to anti-terrorism policies more compared to NBT.
Regarding I-advantages, it is also expected that BT has a stronger detrimental effect on M&A investments compared to NBT. More specific, BT lessens the willingness to own and control another firm in a specific country more than NBT.
11 that were hurt or even killed. This reduces operating profit. Besides the direct costs related to reconstruction, the time needed for reconstruction and the possibility to not be able to continue business for a while only further increases these costs (Hallegatte & Przyluski, 2010). The risk that the firm is damaged is larger with BT than with NBT because of the same reason stated over and over, namely that the firm is the main target in BT attacks. To prevent the possibility of incurring these costs, a MNE might decide not the merge or acquire a firm but enter the market via another strategy, like buying on the market. In this way, BT lowers I-advantaged more than NBT.
Second, affordable insurance for facilities, physical assets and even brand names reduced, and therefore increased the costs of doing business abroad (Gaibulloev & Sandler, 2011; Spich & Grosse, 2005). For example, after the 9/11 attack insurers eliminated terrorism from the ‘all risk’ policy. Firms now have to insure themselves for terrorism via a separate insurance, which entails much higher premiums (Wernick, 2006). In other words, an insurance coverage for terrorism-related activities is harder to obtain, and premiums have increased (Lenain, Bonturi & Koen, 2002). Lenain et al. (2002) conclude that resulting the 9/11 attack “it is estimated that commercial property and liability insurance rates have been raised by 30 per cent on average, with “target” structures such as […] office buildings seeing steeper increases”. So, the last part tells us that office buildings are real targets (BT) and that this results in even higher insurance costs. Besides, being at higher risk of being a target (BT) urges a firm to better insure themselves than when being not a target (NBT). This, as a result, raises costs (further). These higher costs lower the benefits of owning assets abroad and thus decrease I-advantages.
Third, besides the insurance costs, when at risk of being a target (BT) a firm might decide to also increase security expenditures to protect their facilities, buildings, factories, employees, and data (Gaibulloev & Sandler, 2011; Spich & Grosse, 2005). For example, BT may increase costs for security guards, data security and employee screening. There is less need for doing this when a firm is not at high risk of being a target (NBT). Again, because there are higher costs associated with BT than NBT, BT lowers I-advantages more than NBT.
Fourth, regardless of previous BT in the host country and the resulting risk, as being a MNE can provoke hostile attitudes and attacks on the firm (Li, Tallman & Ferreira, 2005). “MNEs seek to take advantage of a host countries’ resource endowments or compete more fiercely with local firms over the host country’s market, but without necessarily adding to local employment or skill bases” (Li et al., 2005). As a result, increased hostilities in host countries toward MNE’s have been observed. This increases the chance on BT in these host countries and the related costs. Other strategies, like outsourcing or trade, could then be preferred above M&A.
12 By combing the conclusion of Powers and Choi (2012) that only BT has a significant negative effect on FDI with the intuitions/arguments presented above, the hypothesis for research question two can be stated as follows:
H3: BT has a larger negative effect on firm-level cross-border M&As than NBT
In the next section, it will be explained how the three hypotheses will be tested. Furthermore, data resources will be provided.
3. Methodology
This section explains the econometric model. Furthermore, it provides the data sources for all the variables used in the model. Then, descriptive statistics on the main variables of interests are discussed. Finally, it is explained what different models are used to test the three
aforementioned hypotheses. 3.1 The econometric model
To test the three hypotheses provided in the literature section, a gravity model will be used. Although rooted in trade literature, among others, Di Giovanni (2005) and Ouyang and Rajan (2016) proved the gravity model is also useful in examining cross-border FDI and M&A flows, respectively.
As investigating the impact of (N)(B)Terrorism on cross-border firm-level M&A investments (MA) is the aim of this research, the main variables of interests in the gravity model will be the independent variables terrorism (T), NBT and BT.
To investigate the impact of T and its different forms (BT and NBT) on MA, LPMs and OLS regressions will be used. In the LPMs, the effect of (N)(B)T on the probability of a positive entry decision will be estimated. This is also called the extensive margin in this paper. The regression models examine the intensive margin. More specific, it estimates the effect of (N)(B)T on the value of M&As, conditional on entry. A simple correlation between MA and (N)(B)T would not suffice as this would only show the association between MA and (N)(B)T. Therefore, it gives no conclusion on the effect of (N)(B)T on MA. Besides, a correlation does not allow fixed effects and other (control) variables. This limits the possibility to isolate and thus better estimate the effect of (N)(B)T on MA. Although regressions cannot prove causality either, it helps to predict MA based on known values of (N)(B)T.
The LPM to estimate the extensive margin in this paper is:
D_MAzijt = a + b1((N)(B)Tjt-1) + b2 ln(GDPit) + b3 ln(GDPjt) + b3ln(Distijt) + b4 Xijt + b5 Fzt And the regression model to estimate the intensive margin in this paper is:
13 In both models, the main coefficient of interest is b1. This coefficient shows the effect of (N)(B)T on the extensive and intensive margin. Thereby, (tests on) the results for this coefficient can provide support for the theoretical expectations.
The main variables of interest, (N)(B)T, are proxied by the number of (N)(B)T attacks in the host country in the year preceding the M&A investment, divided by the population of that host country in the same year of the (N)(B)T attacks. More formally, Tjt-1 is the number of
terrorist attacks in host country j in year t-1 divided by the population of host country j in year t-1; BTjt-1 is the number of BT attacks in host country j in year t-1 divided by the population
of host country j in year t-1; NBTjt-1 is the number of NBT attacks in host country j in year t-1
divided by the population of host country j in year t-1. First, the lagged value of (N)(B)T attacks is preferred as this ensures that only terrorist attacks that might impact M&As are included. More specific, without taking the lagged value, it is assumed that a terrorist attack in Augustus 2010 can affect a M&A investment in February 2010. However, in practice, this is impossible. Taking the lagged value of attacks excludes the attacks that can have no impact in practical sense. Second, the number of attacks in the host country are scaled by the population of that host country. In this paper, it is assumed that one terrorist attack in Luxembourg will be of greater magnitude compared to an attack in Brazil because Luxembourg is a smaller country in terms of inhabitants. So, by scaling the number of attacks by population the measurement takes the magnitude of the attack into account. As the magnitude depends on the population of the host country, the effect of a (N)(B)T attack will differ per host country. The other variables can be defined as follows. D_MAzijt is the dummy MA variable, taking
value 1 if a M&A investment is made by firm z in source country i to host country j in year t, and 0 otherwise; ln(MAzijt) is the natural log of aggregate cross-border M&A value of firm z
in source country i to host country j in year t in millions US$; The vector Xijt contains all
control variables except the natural log of source and host country GDP (ln(GDPit) and
ln(GDPjt), respectively) and the natural log of the distance between source and host country
(ln(Distijt)). Finally, the model controls for firm-year fixed effects (Fzt). See appendix I for the
definitions of the variables.
14 When examining the intensive margin, the effect of (N)(B)T on the value of M&As, conditional on entry, will be tested by OLS estimation. More specific, the dependent variable in this research is the (natural log of the) aggregate cross-border M&A value of firm z in source country i to host country j in year t. This variable should lead to better results as explained in the introduction (limitation 1 and 2 of previous research). Moreover, it should lead to better results compared to the number of cross-border M&A deals because deal value consist of more information due to showing the magnitude of the deal. Deal value is measured in millions of US dollar. In this model, MA value is transformed to its natural log for three reasons. First, the econometric model will take a log-linear form, which is standard in gravity models (Brakman, Garretse, Marrewijk & Witteloostuijn, 2009). Second, the log transformation excludes the zero values, which is needed in estimating the intensive margin. Last, it increases the meaning of interpretation as the results can be interpreted in percentages. The percentage change gives a better idea about the magnitude of the effect. Like in the LPM, the coefficient of interest has to be scaled to the population of the host country. More specific, conditional on entry, one additional (N)(B)T attack in host country j in year t-1 leads to (approximately) a 100*(b/population host country t-1) percentage change in M&A investment value from firm z in source country i to host country j in year t (Hill, Griffiths & Lim, 2012, p. 142) (Appendix II). As intended, the effect of (N)(B)T differs per country. The same expectations regarding the signs of the coefficients of (N)(B)T are expected as stated in case of LPM (appendix I).
Previous literature showed that a high number of zero or missing values can be observed in gravity data (Di Giovanni, 2005; Ouyang & Rajan, 2016). When transforming the dependent variable to its natural log, zero and missing values are excluded from the data. As a result, the observations are not randomly selected from the population. This gives rise to concerns about sample selection bias. Previous papers used the Heckman Sample Selection method to solve this problem. However, due to the tight firm-year fixed effect structure (that results in many dummy variables), the Heckman model cannot be estimated. Therefore, this paper examines the intensive and extensive margin. This is another solution for the selection bias problem as the extensive margin measures the effect of (N)(B)T on M&A value, conditional on entry. Besides, it is a comparable method. However, the main difference is that the Heckman method implements the Mills-ratio, which is a measure of probability that a deal is observed or not, as a regressor. Importantly, the Heckman method “accounts for the fact that the inverse Mills ratio is itself an estimated value” (Hill et al., 2012, p. 623) and produces correct standard errors for the variables.
To limit the possibility of omitted variables affecting the results, control variables are implemented in the econometric model. These control variables are based on Giovanni (2005) and Ouyang & Rajan (2016) and include several gravity, macroeconomic, and financial variables. Each of them is expected to show different signs on their coefficients. Appendix I provides a clear overview of the measurement of the control variables and the expected signs of their coefficients.
15 investment opportunities. Also, Powers and Choi (2012) and Wang (2008) found a positive significant relationship between market size and FDI inflows. Furthermore, Ouyang and Rajan (2016) found that GDP of the host country positively influences cross-border M&A. Regarding the source country, Ouyang and Rajan (2016) found a significant positive relation between source country GDP and cross-border M&A value. Garita and Van Marrewijk (2007) explain this relationship by the effect that richer countries invest more in other countries. The distance between the source and host country is expected to show a negative sign. The intuition behind this negative relationship is the increased investment costs because of information asymmetries and the difficulty of managing the acquired assets (Di Giovanni, 2005). Erel, Liao and Weisbach (2012) also state that physical distance increases the costs of combining firms and found that the benefits of acquiring a firm in a nearby country are higher compared to an acquisition with a firm in a country far away. Finally, Ouyang and Rajan (2016) and Garita and Van Marrewijk (2007) also found a negative coefficient for the relationship between distance and cross-border M&A.
Common language, a common historical colonizer, common currency and a common religion are expected to show a positive sign. Opposed to distance, these factors are considered to reduce the cost of doing business (Di Giovanni, 2005; Ouyang and Rajan, 2016).
Exchange rate in the source country is expected to show a positive relationship with cross-border M&As, but a negative coefficient in the host country. As cross-cross-border transactions (often) requires domestic currencies to be converted into the currency of the host country, exchange rate has an impact on the value of the acquired assets. A currency appreciation in the source country reduces the costs of M&As, therefore stimulating cross-border M&As (Wang, 2008, Erel et al., 2012; Ouyang and Rajan, 2016). In contrast, a depreciation in the host country is hypothesized to lead to greater FDI (Di Giovanni, 2005; Erel at al., 2012; Ouyang and Rajan, 2016).
Regarding exchange rate volatility, the coefficients for both the source and host country can be positive and negative. Di Giovanni (2005) explains that depending on the correlation between the target firm’s exchange rate volatility and the overall acquiring firm’s exchange rate portfolio, high exchange rate volatility may have a negative or positive effect. The exchange rate volatility is calculated in line with Di Giovanni (2005). First, the log differences of the exchange rates of year t and year t-1 were taken. Then, the standard deviation is measured for 5 years prior to each period t.
Financial openness is expected to show a positive sign for both the source and host country, as it ought to facilitate cross-border capital flows, including M&A deals. These positive effects are found by Ouyang and Rajan (2016) and Garita and Van Marrewijk (2007). This measure takes a value ranging from 0 to 1, with one for the countries having most open and zero for the most closed financial markets.
16 is that they do not control for this. Second, a firm can show variability across different years that can have an impact on the choice to perform a M&A investment (yes/no) as well as the amount. For example, a firm might have lower interest in country A in year ’10 than in ’09. More specific, because it acquired a firm in country A in 2009, the investing firm might have less interest to perform another investment in country A in 2010 and might be more interested in doing a M&A in another country. Another example might be that the same firm has a lower availability of investment funds in two different years. As the firm has done a M&A in 2009, the funds available in 2010 might not suffice to do an investment, or might lower the amount of investment. By comparing observations from the same firm in the same year, the differences (in levels of interests) between firms and over years can no longer explain the results. So, with these fixed effects, the models can better isolate and thus predict the effect of (N)(B)T on the choice to invest (LPM) and the level of investment (regression).
3.2 Data sources
This section describes the data sources used for the dependent and the independent variables. Appendix I provides a clear table specifying the variables and their data sources.
The data used for the dependent variables is taken from the Zephyr database. As the focus of this research is on M&As, institutional buy-outs, capital increases, joint-ventures, management buy-outs, management buy-ins, demergers, minority stakes, and share buy backs have been excluded. Besides, only cross-border deals are included for the period of 2009 until 2015. This period is opted to limit the effect of the global financial crisis on cross-border M&A deals. Finally, only completed deals are included with a minimum (sum)deal value of $1 million.
17 not needed in this paper and this disadvantage can thus be disregarded. GTD defines a terrorist attack as “the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious, or social goal through fear, coercion, or intimidation” (GTD, 2016). For an incident to be included in the GTD dataset, the following three criteria must be satisfied; (1) The incident must be intentional; (2) the incident must entail some level of violence or immediate threat of violence; (3) the perpetrators of the incidents must be sub-national actors. GTD (2016) analyzes for each incident the target/victim. GTD identifies 22 types of victims, one of them characterized as ‘Business’ (Appendix III). As such, it is possible to distinguish between BT and NBT. More specific, in this paper terroristic events with a ‘Business’-target will be assigned to BT, while attacks with other categories of victims will be assigned to NBT. The World Development Indicators (WDI) database of the World Bank provided population data to scale the magnitude of the (N)(B)T attack to the population of the host country.
Different databases are used for the control variables in the gravity model. Data on market size (GDP) is drawn from the WDI database. Data for the gravity control variables, which are distance, common language, common colonizer, common currency and common religion, are taken from CEPII. This is a French research center in international economics. The Chinn-Ito index generated by Chinn and Ito (2007) is used as a proxy for financial market openness for both the source and host country. This index takes a value ranging from 0 to 1, with one for the countries having the most open financial markets and zero for the most closed financial markets. Finally, data on exchange rates are form Darvas’ (2012) REER database. The exchange rate volatility is calculated in line with Di Giovanni (2005) and by using the data from Darvas’ (2012) REER database.
3.3 Descriptive statistics
A description of selected variables is provided in this part. From this description, various observations can be made.
The dataset consists of 120 countries and 8412 different firms. Appendix IV contains a list of countries included in the sample. Using these 8412 firms, the dataset ends up with more than 7 million observations. Of these 7 million observations, the largest part contains zero values (as seen in previous papers and as discussed above). In this case, only 10.069 observations show a M&A value. Of the 10.069 observations, 9374 are unique firm-year observations. The regression model with firm-year fixed effects compares different observations from the same firm in the same year. This means that observations are only taken into account for firms that have invested at least twice in the same year. By excluding the observations of firms that only invest once in the same year, only 1276 observations are left. This means that the regression will only have 1276 or less observations (dependent on the number of observations of the dependent variables). Counting the number of unique values reveals that only 581 firms have invested twice or more in the same year.
18 M&A deals have declined since 2011 (as opposed to the Worldwide trend), while Southeast Asia shows an increase over the period. Central Asia and Central America & Caribbean show fluctuations in the number of M&A, while Western Europe show a stable growth. Besides, the number and the value of M&As do not always follow the same trend. For example, South Asia shows an increase in the number since 2012, but a decrease in value since 2011. The same holds for Eastern Europe. This region shows an increase in the number of M&As in the period from 2011 until 2014, but a decrease in value in the same period. On the contrary, in Central Asia and North America, the number and value of M&As do follow the same trend. The descriptive statistics in table 1 provides more information regarding M&As, conditional on entry. It shows that the lowest firm-host country M&A investment is 1 million US$ (due to excluding deals with (sum)deal value of lower than 1 million) and that the highest firm-host country M&A investment is 70,5 billion US$. The sum value of all the deals together is 3012 billion US$. The mean firm-host country M&A investment is 299 million US$ with a standard deviation of 1,5 billion. This indicates quite a variation between different firm host-country pairs.
Figure 3. Number of M&A deals in the world Figure 4. Aggregate value M&Adeals in the world
source: Zephyr database. source: Zephyr database.
https://www.bvdinfo.com/en-gb/our-products/economic-and- https://www.bvdinfo.com/en-gb/our-products/economic-and
m-a/m-a-data/zephyr m-a/m-a-data/zephyr
Table 1. Descriptive statistics
Variable Observations Mean Standard
deviation
Min Max Sum
M&A value in millions of US$ 10069 299 1503 1 70500 3012777
Number of Terrorist attacks in host country
840 54 284.85 0 3925 45420
Number of BT attacks in host country
840 5 19.60 0 228 4130
Number of NBT attacks in host country 840 49 230.43 0 3697 41290 2000 2200 2400 2600 2800 3000 3200 3400 2009 2010 2011 2012 2013 2014 2015
Number M&A deals World
200000 400000 600000 800000 1000000 1200000 2009 2010 2011 2012 2013 2014 2015
Aggregate value M&A deals World
19 With respect to terrorism, the number of terrorist-, BT- and NBT- attacks have increased (Appendix VI). Two things stand out. First, the number of BT attacks are far lower than the number of NBT attacks (mind that the different figures have different scales). Second, while the number of NBT attacks declined since 2014, the number of BT attacks increased. A more thorough look into the number of (N)(B)T attacks in different regions of the world provides some observations (Appendix VI). First, large differences can be seen in the total number of attacks across countries. Australasia & Oceania, Central America & Caribbean, and Central Asia have experienced less than 50 attacks a year from 2009 until 2015, contrary to Middle East & North Africa, South Asia, and Sub-Saharan Africa that experienced thousands of attacks per year. Second, almost all regions experienced an increased number of terrorist attacks, except Central Asia. Finally, the trend of BT and NBT follows almost in all regions the same trend as Terrorism. Only North and South America are exceptions. In North America, terrorism increased during the period 2010 until 2014, but BT decreased in that period. The same observation can be inferred for South America for the period 2011 until 2013.
Although the variables of interest are scaled to population of host country, table 1 provides descriptive statistics on the number of (N)(B)T. The descriptive statistics for the scaled variables are harder to interpret, less easy understandable and less informative. For example, the mean terrorist attack per inhabitant is 0,00000163. Therefore, statistics on the number of (N)(B)T are provided instead. Over the years, there were 45420 terrorist attacks of which 4130 had a business-related target. Thus, terrorist attacks comprise only a small number of BT attacks. This was already indicated above. The remaining 41290 attacks were aimed at non-business related victims. The mean number of terrorist attacks, BT attacks and NBT attacks were 54, 5, and 49, respectively. Large standard deviations indicate great variance. This might lead to the estimates not being significantly different from zero and wide interval estimates. Do not be distracted by the number of observations. As the data set comprises of data on 120 countries over 7 year, the number of observations needed to calculate the total amount of (N)(B)T is (120 * 7=) 840. Calculating the total number of (N)(B)T attacks on the 7 million observations will result in erroneous sum statistics, as the number of (N)(B)T attacks in the same country in the same year will be summed multiple times.
When considering correlation levels of 0.6 as high, the variables T, BT and NBT are highly correlated (appendix VII). This is not surprisingly following the observation that they move together in systematic ways. This was already indicated above. Correlation levels of other variables are relatively low (appendix VII). High correlation levels between variables might indicate multicollinearity. Although this does not violate the OLS, the standard errors tend to be inflated. Besides, it might produce parameters signs that seem theoretically questionable (O’Brien, 2007). Finally, the test has a higher probability that it will conclude that the coefficient estimates are not significantly different from zero. In other words, “the correlating variables do not provide enough information to estimate their separate effects, even though theory may indicate their importance in the relationship” (Hill et al., 2012, p. 240).
20 model (Craney & Surles, 2002). Because VIFi = 1/(1-r2i), VIFs between 3.52 and 3.79 imply
that r2i = 0,72 and 0,74, or between 72% and 74% of the variability in BT and NBT is
explained by the remainder of the independent variables in the model. When including BT or NBT in the LPM, the VIFs of BT and NBT become 1.03 and 1.04, respectively. Excluding BT or NBT from the regression models results in a VIF of 2.08 for BT and 2.04 for NBT. The variance of BT is thus mainly increased due to correlation with NBT, and vice versa. Although the VIFs are not extremely high when implementing both BT and NBT in the same model (they are lower than 10), one might become concerned with VIFs higher than 2.5. Following this indication, it is therefore less satisfying to use NBT and BT together as dependent variables in the same model.
In both the extensive and intensive margin models, heteroskedastic and cluster robust errors are used. First, a limitation of the LPM is that the error terms are heteroskedastic (Hill et al., 2012, p. 588). Besides, a graphical detection by plotting the residuals against the independent variables further indicates that heteroskedasticity is also present in the data for the intensive margin regression models. This means that the variance of the error term varies from one observation to the other. Heteroskedasticity is a violation of the OLS assumptions. Although the OLS estimators remains unbiased, the estimated standard errors are wrong (Hill et al., 2012, p. 588). Second, terrorism data is non-independent for host-country years. More specific, the value of (N)(B)T does not vary within a host country within a year. In other words, for all observations of host country j in year t the value for, for example, BT is x. Assume there is an observation where a firm Q from source country U invest in host country J in year t. Host country J experiences a number of attacks n and has a population of p in year t-1. Value for BT in year t-1 is then x=n/p. Now, another firm W from source country R also performs a M&A investment in host country J in year t. The value for BT for this observation is also x=n/p, because firm W invests in the same host country in the same year. So, the number of observations is increased by one. However, the amount of information on BT is in reality not increased by this extra observation. Neglecting this non-independence results in standard errors that would be too small as too many observations are included in the calculation. To prevent this, errors are clustered by host country-year. So, heteroskedastic and cluster robust standard errors are corrected for heteroskedasticity and non-dependence of data, and are thus correct.
Finally, Li and Schaub (2004) reject the reverse causality hypothesis. One might argue that higher capital inflows results in higher growth and likely decrease terrorist attacks in the host country. With a sample of 112 countries from the period of 1975 to 1997, they show that trade FDI and portfolio investment do not have a direct positive effect on terrorism within countries.
3.4 Baseline models and tests explained
To test the three hypotheses, seven different models will be used. In all models, the control variables GDPit, GDPjt, Distijt, and Xijt are present. The differences between the models results
21 (decision and value). However, simply observing that the estimates show negative signs do not provide scientific proof. Therefore, left-tailed tests will be performed to determine whether there is convincing statistical evidence for the hypotheses (one and two) that (N)(B)T have a negative effect on M&As.
The fourth model includes both BT and NBT. In this model, the relative impact of BT and NBT is examined by letting both variables compete in the same model. Thereby it is possible to investigate which of the two forms of terrorism has a stronger relation with M&A investments. Beside the left-tailed tests, a Wald test will be performed to see if the coefficients of BT and NBT show statistically significant differences. This test can provide support for the theoretical expectation that BT is more harmful than NBT.
The fifth, sixth and seventh model includes standardized regression (or beta) coefficients of BT, NBT, and BT and NBT respectively. For this, variables are transformed to standard scores, or z-scores. In this case, each variable has been rescaled to have a mean of zero and a standard deviation of one. Appendix VIII provides the formula for this transformation. As beta coefficients are measured in standard deviations and are thus ‘metric-free’, it is possible to compare coefficients from two different models. Besides it provides a better comparison of coefficients in the same model. In this way, it is possible to provide a more solid conclusion regarding the expectation that BT has a greater effect on MA (decision and value) than NBT. The interpretation is as follows; a one standard deviation change in (N)(B)T would yield a b1 standard deviation increase in MA. As the probability of rejection does not change when standardizing variables, left-tailed tests and Wald tests on unstandardized variables will conclude the same as on standardized variables.
To overcome the possible multicollinearity problem in model 4 and 7, J-tests are performed. Indicated above, implementing both BT and NBT might induce multicollinearity problems. As multicollinearity inflate the standard errors, it can impact the left-tailed tests and Wald test and therefore the conclusions. The J-test compares two models with the same dependent variable, but with different independent variables. Subsequently, it decides which model is better or, in other words, is the ‘appropriate model’. The procedure for the J-test is as follows. The predicted (fitted) value of model 2 is implemented as a regressor in model 3. If the coefficient of the added regressor is significantly different from zero then the model of the added regressor (model 2) has extra explanatory power above the other model (model 3). If this is not the case, then the other model (model 3) is a better predictor of the dependent variable, in this case MA. Then, the process is reversed. Comparing the conclusions might indicate one model to be the appropriate model. It might also be the case that both models do not contain the correct set of regressors (Davidson & MacKinnon, 1981). The J-test will report the same results on the unstandardized and standardized results.
To check the robustness of the baseline results, different robustness checks are performed (section 4.3).
4. Results
22 4.1 Baseline results on the extensive margin
Table 2 presents the results of the main variables of interest in the seven different LPMs. Appendix IX provides a table where the results on the control variables are also shown. In these models the effect of (N)(B)T on the extensive margin is tested. In other, more technical words, table 2 (and appendix IX) shows the effect of (N)(B)T in host country j in year t-1 on the probability that a firm z in country i performs a M&A investment in country j in year t. Models 1-3 show that terrorism and its different forms, NBT and BT, all have a significant negative impact on the probability that a foreign firm will perform a M&A investment in a host country at the one percent significance level. Regarding the extensive margin, left-tailed tests provide convincing (statistical) evidence for the hypotheses (one and two) that T, BT and NBT have a negative effect on the probability of a positive M&A decision. As explained before, the effect of one (N)(B)T attack cannot be inferred from the coefficient at glance because the measurement of (N)(B)T is scaled by the population of the host country. In specific, the slopes of the variables (N)(B)T can be interpreted as follows; one extra (N)(B)T attack in host country j in year t-1 changes the probability that a firm z from source country i will perform a M&A investment in year t by 100*(b/population host country t-1) percentage point. The effect of a terrorist attack thus differs per country, as stated in section 3.1. For example, the country with the smallest population in the dataset of this paper is Belize in 2008. In that year, Belize had a population of 306165. So, if Belize experienced ten extra T-, BT-, or NBT attacks in 2008, the probability that a foreign firm performed a M&A investment in 2009 declined with (10*100*(b/306165)=) 0,16 percentage point (T), 2,25 percentage point (BT), and 0,16 percentage point (NBT) ceteris paribus, respectively. The country with the largest population in the dataset is China in 2014. China had a population of 1364270000. Then, if China experienced ten extra T-, BT-, or NBT attacks in 2014, the probability that a foreign firm performed a M&A in 2015 declined with 0.000036 percentage point (T), 0,0005 percentage point (BT), and 0,000038 percentage point ceteris paribus, respectively. The other countries in the dataset have a larger population than Belize in 2008 and a smaller population than China in 2014. Therefore, the effects experienced by other countries in the same or different years lie between the effects of Belize and China. In practice, the effects of (N)(B)T are very small, and maybe even neglectable.
23 the exact interpretations of model 1-3 are more informative. Therefore, no exact interpretation is provided here.
The J-test concludes that the models with BT (model 2 and model 5) are the appropriate models to predict the probability a foreign firm will do a M&A investment in a host country. More specific, when implementing the fitted value of a model including BT (model 2/model 5) in a model with NBT (model 3/model 6) results in a significant added regressor. Thus, model 2 (model 5) has extra explanatory power above model 3 (model 6). When reversed, the added regressor is not significant.
The control variables concerning the source country are omitted because of collinearity. This is the result of the firm-year fixed effect structure implemented in the models. In this fixed effect structure, observations of the same firm in the same year are compared. As a result, no variance in source country variables is present. Assume firm Z in country I has performed M&A investments in host country J, L and M in year t. In year t, the GDP of source country I was 10. So, when comparing the observations of firm z in year t, GDP is 10 in all three observations and thus shows no variability. As a result, the variables are fully collinear with the fixed effects and are therefore omitted.
The other control variables have their expected signs, except for the variable “common currency”, and are all strongly significant. Firms invest in host countries with a higher GDP, indicating that these host countries have larger economies with more investment opportunities. Furthermore, there is a higher probability that a firm invests in a foreign country that is located closer, has a common official language, a common (historical) colonizer and the same religion. Contrary to the idea in this paper, a firm is less likely to invest in a country with a common currency. One reason for this negative finding might be that the M&A data is dominated by observations of firm-host countries without the same currencies. Of the 7 million observations, only 214000 observations are of source-host countries with a common currency. An appreciation in the host country currency decreases the probability of a M&A investment, whereas the volatility of the exchange rate increases this probability. The former effect was hypothesized, why the latter effect could have shown a positive and a negative sign. As explained in the section 3.1, the positive or negative sign depends on the correlation between the target firms exchange rate volatilities and the overall acquiring firms exchange rate portfolios (Di Giovanni, 2005). Finally, financial openness in the host country facilitates M&A investments and thereby significantly increases the probability of M&A investments.
24
Table 2. Baseline results: The impact of (N)(B)T on the probability of a positive M&A decision
Dependent variable: Dummy MA, (D_MAzijt)
(1) (2) (3) (4) (5) (6) (7) Variables standardized T -49.34*** (12.26) BT -681.3*** -671.6*** -0.0107*** -0.0105*** (125.9) (200.7) (0.00197) (0.00314) NBT -51.83*** -1.011 -0.00998*** -0.000195 (12.90) (19.58) (0.00248) (0.00377)
Controls Yes Yes Yes Yes Yes Yes Yes
Firm-year fixed effects Yes Yes Yes Yes Yes Yes Yes
Heteroskedastic and cluster robust errors Yes Yes Yes Yes Yes Yes Yes
N 5837016 5837016 5837016 5837016 5837016 5837016 5837016
Notes: * p<0.10, ** p<0.05, *** p<0.01. Heteroskedastic and cluster robust standard errors are in parentheses. (N)(B)T is measured by the number of (N)(B)T attacks in host country j divided by the population of host country j (Appendix I).
Table 3. Baseline results: The impact of (N)(B)T on the M&A investment value, conditional on entry
Dependent variable: natural log of aggregate cross-border M&A investments (Ln(MAzijt))
(1) (2) (3) (4) (5) (6) (7) Variables standardized T -87814.1*** (24969.2) BT -505074.2** -49626.8 -0.155** -0.0152 (236585.3) (260422.8) (0.0727) (0.0800) NBT -95562.1*** -91378.6** -0.361*** -0.345** (27375.0) (35945.5) (0.103) (0.136)
Controls Yes Yes Yes Yes Yes Yes Yes
Firm-year fixed effects Yes Yes Yes Yes Yes Yes Yes
Heteroskedastic and cluster robust errors Yes Yes Yes Yes Yes Yes Yes
N 1079 1079 1079 1079 1079 1079 1079
25 more consistently to BT. So, the third hypothesis is supported. This conclusion is furthermore in line with the findings of Powers and Choi (2012).
4.2 Baseline result on the intensive margin
Table 3 shows the results of the seven different regression models. Appendix X provides a table with the same models, however this table also presents the results on the control variables. In these models, the effect of (N)(B)T on the intensive margin is tested. In other words, table 3 (and appendix X) shows the effect of (N)(B)T in host country j in year t-1 on the M&A investment value of firm z in country i to country j in year t, conditional on entry. A first thing to mention is the low number of observations in all seven models (N=1079). However, this can be explained. First, as the models investigate the effect of (N)(B)T on M&A value, conditional on entry, all the zero value observations are excluded. Section 3.3 of this paper already showed that only 10,069 observations are non-zero. All other observations are thus dropped. Second, by including firm-year fixed effect the models compare observations of the same firm (z) in the same year (t). Therefore, observations of firms that have invested only in one country in a specific year are excluded. Section 3.3 stated that only 1276 unique firm-year observations exist in the dataset. Some more observations are dropped as not all these 1279 observations have data on all dependent variables.
Models 1 until 3 show that T, BT and NBT all three have a significant negative effect on the M&A value, conditional on entry. T and NBT at the one percent level and BT on the five percent level. Regarding the intensive margin, left-tailed tests furthermore proof that that T, BT and NBT have a negative effect on M&A investment value. Again, the exact interpretation of the effect of one (N)(B)T attack cannot be inferred from the coefficient at glance, as it differs per country. More specific, conditional on entry, one additional (N)(B)T attack in host country j in year t-1 leads to (approximately) a 100 *(b/population host country t-1) percentage change in M&A investment value from firm z in source country i to host country j in year t. For example, Greece had a population of 11104899 in 2011. An additional T-, BT-, or NBT- attack in Greece in 2011 decreases aggregate firm-level M&A value invested by a foreign firm in Greece in 2012 by (100*(*(b/11104899)=) 0,79 (T), 4,5% (BT), and 0.86% (NBT) ceteris paribus, respectively. Because the average amount of M&A investment in the dataset is approximately 299 million US$, an extra (N)(B)T attack results in a loss of approximately $2.4 million for every T attack, $13.5 million for every BT attack, and $2.6 million for every NBT attack in Greece in 2012. The population of Greece is the most close to the median population. Explained differently, half of the countries have a population smaller than Greece and half have a population larger than Greece. Therefore, half of the countries experience effects larger than Greece, while the other half experience effects smaller than Greece.
