Master thesis January 2015

(1)

MSc International Economics and Business MSc International Business and Management

Fire-Sale FDI During Crises: A Panel of Developed and

Developing Countries Between 1990-2010

Roderik Jongbloed

Supervisor: Dr. D.H.M. Akkermans Co-assessor: Dr. B.J.W. Pennink

Faculty of Economics and Business

(2)

1

ABSTRACT

This study investigates the FDI fire-sale hypothesis for 25 developed and 37 developing countries between 1990 and 2010. According to this hypothesis, financially distressed firms from countries affected by a crisis attract foreign buyers by selling their assets at discounts. We investigate the relationship between financial distress and inward brownfield FDI (M&A) by employing a panel data regression analysis. We find that FDI inflows do not increase when financial crises occur, but rather decrease as the investment climate deteriorates. No evidence for large-scale international fire-sales is found. The second part of this study investigates whether corporate assets in the U.K. and Malaysia sell at discounts during financial crises. We find that corporate assets of distressed firms in the U.K. indeed sell at discounts, and that discounts are greater during financial crises. This paper concludes with some firm management implications for dealing with fire-sales for both buyers and sellers.

Keywords:

(3)

2

TABLE OF CONTENT

1. INTRODUCTION……… 4

2. THEORY………... 6

2.1 Brownfield FDI in times of crises………...…..……... 6

2.2 Fire-sale FDI ………..……. 7

2.3 The role of credit……….. 9

2.4 Different crises, different impact?………...…. 10

3. DATA AND METHODS……….…. 12

3.1 Procedure………. 12

3.2 Sample………. 13

3.3 Variables and Measures………..…. 13

3.3.1 Inward Brownfield Foreign Direct Investment (dependent variable)……… 13

3.3.2 Financial crises (independent variables) ………..… 15

3.3.3 Credit supply………... 16

3.3.4 Control variables……….... 16

3.3.5 Tax havens………... 17

3.3.6 Limitations……….…. 19

3.4 The model……….... 20

3.4.1 Panel estimation method and assumption tests………..…… 22

4. RESULTS………..……..….. 23

4.1 Model fit………..…. 26

4.2 Regression results for hypothesis 2, the relationship between financial crises……… 26

and inward FDI 4.3 Regression results for hypothesis 3, the role of the supply of credit……….……. 27

4.4 Tax haven vs. non-tax havens & developing vs. developed economies………. 29

4.5 Control variables………..…… 30

(4)

3

5. ASSET PRICES……… 34

5.1 United Kingdom………... 35

5.1.1 Are U.K. bid premiums lower during financial crises? ………. 37

5.1.2 Are discounts for financially distressed firms greater than for ……….…. 38

non-distressed firms in the U.K.? 5.2 Malaysia……….….. 39

5.2.1 Are Malaysian bid premiums lower during financial crises? ……….….. 40

5.2.2 Are discounts for financially distressed firms greater than for ……….…. 41

non-distressed firms in Malaysia? 5.3 Limitations and alternatives……….…… 42

6. DISCUSSION……….………..…. 43

7. A MANAGERIAL PERSPECTIVE……….………….. 45

7.1 The acquiring firm – are fire-sales useful? ……….……… 45

7.1.1 Are fire-sales useful from a strategic perspective? ……… ……….. 46

7.2 The selling firm – How to deal with fire-sales? ……….. 47

7.2.1 How can firms prevent costly fire-sales? ……….. 47

7.2.2 If avoiding is no option anymore, how can target firms’ management …………..…….. 48

deal with fire-sales? 8. CONCLUDING REMARKS………..……. 49

REFERENCES………. 50

APPENDIX I: Country Samples……….. 56

APPENDIX II: Tax haven Economies……….…….... 57

APPENDIX III: Assumption Tests………... 58

APPENDIX IV: Correlation Matrix……….. 60

(5)

4

1. INTRODUCTION

In 2000, Krugman wrote an influential article about fire-sale FDI (Krugman, 2000). Until then, it was generally argued that investors pulled back from markets that were in crisis, thereby reducing inward foreign direct investment (hereafter FDI). Krugman argued in the opposite direction, by introducing the term ‘fire-sale FDI’ in relation to the Asian and the Latin-American crises. The fire-sale hypothesis entails that during periods of crises, corporate assets from crises-countries sell at discounts, attracting foreign investors – without liquidity problems – to buy these assets at prices below their fundamental values. As a result, the FDI flows into these crises-countries may increase instead of decrease.

In a recent study of Weitzel, Kling and Gerritsen (2014), this hypothesis was tested for European country-pairs during the period 1999-2012, including the current global financial crisis. The authors found no evidence for the fire-sale hypothesis in this context. The prices of corporate assets in European crisis-countries were not significantly lower during the crisis compared to more tranquil periods, and total inward FDI did not increase but actually decreased for all European countries. Also, Bogach and Noy (2012) found no evidence for the fire-sale hypothesis using a large panel of crises in developing countries.

These ambiguous results demand for more systematic research in understanding the relationship between financial crises and fire-sale FDI. Therefore, this study incorporates a panel of multiple financial crises that hit various developed and developing countries between 1990 and 2010, as identified by Reinhart and Rogoff (2009). We systematically assess the relation between these crises, credit supply, asset prices and the subsequent behavior of FDI. By combining the data of Reinhart and Rogoff (2009) with extensive World Bank and UNCTAD data about credit supply and FDI respectively, we investigate whether brownfield FDI increases in the wake of financial crises, and test whether Krugman’s hypothesis holds.

(6)

5 Our study contributes to this stream of literature in two important ways. First, it systematically assesses the fire-sale FDI hypothesis for developed countries over a longer time period containing multiple financial crises, which was still lacking (Bogach et al., 2014). Secondly, this study includes besides developed also developing countries, and thus we systematically asses for both groups how FDI flows have changed in response to crises. This allows for a proper comparison. Although not entirely new, we also investigate whether the supply of credit to firms, as important financing source (Campello, Graham, and Harvey, 2009), influences the effect of financial crises on these investment flows.

In addition, studying the fire-sale hypothesis requires to identify whether prices of corporate assets have dropped, and have been sold at discounts. This analysis is performed in the second part of our study for two countries in our panel, namely the U.K. and Malaysia. For the remaining countries, we follow Bogach et al. (2012) and study the impact of crises on FDI without specifically testing the intermediate process of declining prices of corporate assets. Lastly, a unique feature of this study is that it combines a broad, ‘generalizable’, cross-country analysis with an analysis of the prices of corporate assets within two countries, and ultimately combines the findings of both parts with some firm management implications.

The results of this study are relevant in several ways. Financial distress, and potential subsequent fire-sales, can result in the loss of value for firms and its shareholders, and can also cause considerable social cost when assets are sold below their fundamental values (e.g. Coval and Stafford, 2007; Schleifer and Vishney, 2011). Also, fire-sales can damage an entire economy when they evolve into a self-reinforcing process. This occurs when assets of other firms decline in value as well, also become financially distressed, and need to execute fire-sales (Schleifer et al., 2011). Lastly, countries may use expectations about brownfield investment inflows and outflows for making macroeconomic plans (Osei, Morrissey and Lensink, 2002). If policy makers know how a shock (like a crisis) affects the inflow of brownfield FDI, policy initiatives might be directed accordingly.

(7)

6

2. THEORY

2.1 Brownfield FDI in times of crises

Foreign direct investment can take the form of either greenfield investments or mergers and acquisitions (M&A), depending on whether a transaction involves creating mainly new assets or taking over existing ones respectively (Calderón, Loayza and Servén, 2002). Foreign M&A, henceforth denoted as brownfield FDI, involve the transfer of control from a local to a foreign firm. This type of investment is the central element of this study, as fire-sales concern the purchase of existing assets at discounts (Weitzel et al., 2014), rather than creating new ones.

Previous studies revealed that financial crises have different impact on FDI-flows compared to other capital flows (e.g. Krugman, 2000; Fernández-Arias and Hausman, 2001). For example, FDI flows tend to be less volatile to shocks than private debt or external debt flows (Osei et al., 2002). This is not surprising, since FDI is mainly driven by longer term strategic motivations, and is less driven by short-term capital motivations (Thangavelu, Yong and Chongvilaivan, 2009). For example, Thangavelu et al. (2009) found that FDI into Asia remained stable at the onset of the Asian crisis in 1997. Perhaps because of the strategic attractiveness of Asian markets and resources. Soliman (2005) observed that FDI acts as a stabilizing factor for currency crises, and that FDI-flows increase while other forms of capital flows are in decline.

However, the question about how specifically international brownfield investments behave during financial crises remains largely unanswered, perhaps because many studies do not clearly distinguish between brownfield and greenfield investments and rather study aggregated FDI-flows (e.g. Thangavelu et al., 2009; Osei et al., 2002). This distinction can be crucial, amongst others because the share of brownfield investment compared to total investments in developing countries has risen from 10% in 1980 to one third in 1990, and also cross-border brownfield FDI has increased (Calderon et al., 2002). Besides, brownfield investments (M&A) can usually be executed relatively quickly compared to greenfield investment (Bogach et al., 2012). Subsequently, Bogach et al (2012) argue that a short-lived financial crisis and the temporarily depreciation of assets can be accompanied by an M&A-boom while greenfield investment remains more stable.

(8)

7 reports that FDI has sharply declined, especially in developed countries, and in particular regarding cross-border M&A.

The second line of thought starts very similar as the first one described above. As the crisis ignites, the nation’s economy deteriorates, aggregated demand decreases, credit supply to firms contracts (Arteta et al., 2008), and the country’s firms may suffer as demand for their products and services decreases and as financing operations becomes more difficult. As from here, this second line of thought separates itself from the first one, because now foreign investors do not abstain from acquiring crisis-country targets, but instead invest in the crisis country’s firms. The rationale is that these firms are having difficulties in financing their operations and are therefore eager to sell off their assets, at discounts, to avoid financial disaster (Krugman, 2000) and to raise liquidity (e.g. Coval et al., 2007). If so, and if foreign investors are price-sensitive, foreign brownfield inward FDI may thus increase during crises. This process is often described as Krugman’s fire-sale hypothesis (Krugman, 2000). The media have often suggested the existence of this hypothesis (see Weitzel, 2014; and Baker et al., 2009), amongst others by claiming that Germany was buying cheap Greek assets during the global financial crisis. However, there is no convincing academic evidence in favor and there is some evidence against.

For example, some studies that investigated fire-sales found no effects, or even large decreases in FDI inflow during crisis, e.g. Weitzel et al. (2014) for FDI between European country-pairs during the ‘08 global financial crisis and Bogach et al. (2012) for multiple other crises in developing countries. Both found no evidence for the fire-sale hypothesis. These differences suggest that the effect of financial crises on foreign investments is context dependent. Therefore, it is for example inappropriate for FDI studies to mix rich and poor countries, as investors may have different kinds of motivations for each of them (Blonigen and Wang, 2004). For this reason, both developing and developed countries are included but are assessed separately to account for these differences. Below, we will further elaborate on the fire-sale hypothesis.

2.2 Fire-sale FDI

To the best of our knowledge, Krugman was the first to notice that although short-term capital fled rapidly during the Asian financial crisis, the flow of inward FDI increased. The same seemed to hold for the Latin American financial crisis in 1995, and especially for Mexico. For both incidences, the perception of large multinationals was that corporate assets in the Asian and Latin-American countries were available for large discounts. Krugman (2000) stated that this was widely spread across industries and argued that the overall financial situation thus played a major role.

(9)

8 becomes undermined by unnecessary panic (e.g. bank-runs) resulting in the disruption of productive activities (Krugman, 2000). For both situations, the result is a drop in asset prices, and insolvency of intermediaries, resulting in the termination of operations and further asset deflation. Coval et al. (2007) explain that capital providers may even force firms to immediately sell assets, resulting in asset prices dropping below their fundamental values. Although Krugman explained this sequence of effects for the Asian crisis, a similar process occurred during the current financial crisis, that started off with a credit-crunch and a troubled housing market (Brunnermeier, 2009), and is a common phenomenon in the wake of financial crisis in general (Reinhart et al., 2009).

H1: Prices of corporate assets decline during financial crises.

While these troubled firms are offering their assets at discounts, other local firms may not be inclined to buy these assets, because these firms are also suffering from the very same crisis. Therefore, the argument is that firms outside the crisis country can acquire these assets, since foreign firms are not liquidity constrained and are able to continue borrowing for regular interest rates, as opposed to local firms within the crisis country. Even though domestic firms are more efficient at running domestic investment projects, foreign firms can now outbid the domestic firms and takeover these assets because of their favorable liquidity conditions (Krugman, 2000). Hence, the fire-sale is born. This leads to the following hypothesis that will be tested in this study:

H2: Foreign Brownfield FDI (M&A) into crises countries increases during financial crises.

In a perfect world with perfect markets, where investors have all information and borrow at the same rates, financial arbitrage as in the form of fire-sale FDI would have been impossible. However, capital markets across countries are in fact not fully integrated and thus mispricing occurs. Baker, Foley and Wurgler (2009) argue that although previous literature did largely neglect, theoretical and empirical findings do indeed suggest that the slow-moving FDI-flows may reflect financial arbitrage by multinationals. It is thus possible that corporate assets are being sold under their fundamental values to foreign investors. Besides the fire-sale hypothesis as one form of arbitrage, FDI-flows can also result from another form of arbitrage resulting from cheap financial capital available to the investor (Baker et al., 2009). Weitzel et al. (2014) empirically confirm that acquirers come indeed from countries with easy access to capital.

2.3 The role of credit

(10)

9

Financial crises Inward Brownfield FDI

(foreign M&A)

Corporate asset prices

H1 H2 -+ Credit supply H3

-global financial crisis (Campello et al., 2009). As discussed before, especially those firms that are in immediate problems may sell these assets for discounts, resulting in fire-sales.

More studies confirm the importance of credit for financing investments. For example, Klein et al. (2000) found that it mattered for Japanese firms’ direct investments on which Japanese banks they were reliant during crises, as firms connected to the most troubled banks also showed the largest decreases in direct investments. Also, Di Giovanni (2002) found that firms with growth strategies often rely on external finance and that firms active in financially deep markets in terms of size and liquidity have better access to capital for financing their investments. Since firms may have enduring relationships with their domestic bank(s), it is likely that they are to a large extent dependent on these banks for financing their operations, which makes them vulnerable for shocks in their domestic financial sector.

So far, we have neglected that firms can also finance their operations by foreign financial institutions. Especially multinationals may use domestic as well as foreign sources, to diversify financing options, minimize taxes and to hedge against currency fluctuations (Sayek, 2009). Nevertheless, also foreign lenders may be declined to provide credit to firms from countries suffering from sovereign debt crises, as their perception of country risk deteriorates (e.g. Drudi and Giordano, 2000; Arteta and Hale, 2008). Interest rates rise and firms decrease their borrowing, but also future terms of borrowing may worsen. Arteta et al. (2008) found indeed significant evidence for a multi-year decline in foreign credit to the private sector in emerging markets after in this case, a sovereign debt crisis occurred.

Since credit is an important financing source for firms, we expect that contractions in the supply of credit will lead to (more) fire-sales, and thus more inward M&A. This leads to the following hypothesis:

H3: Especially when the supply of credit to firms contracts, inward brownfield FDI (M&A) increases during crises.

Figure 1 contains a visual representation of the conceptual model. Please note that the solid lines represent the relationships that will be tested empirically. The dashed lines are theoretical relationships which will not be tested empirically in this study. We will elaborate on our methods in chapter 3.

(11)

10

2.4 Different crises, different impact?

Definitions for financial crises differ largely in literature, for example, monetarists such as Friedman often link crises to banking panics while others, such as Kindleberger (1978) use a broader definition of financial crises (Mishkin, 1992). For example, sharp declines in the prices of assets, deflations, failures of firms, disruptions in foreign exchange markets or combinations of them, all having serious consequences for the overall economy (Mishkin, 1992). Financial crises result in inefficient financial markets which are not able to channel funds to the most productive investments, leading to contracting economic activity (e.g. Arteta and Hale, 2008).

According to Mishkin (1992), financial crises are caused by rising interest rates, stock market crashes, increasing uncertainty, bank panics, and sudden changes in aggregate price levels. Quite similarly, Reinhart and Rogoff (2009) identified six different types of financial crises and named them after their causes, including currency crises, banking crises, inflation crises, domestic and external sovereign debt crises and stock-market crashes. Now, the question rises whether these crises-distinctions matter for brownfield FDI. Or formulated differently, whether different types of financial crises have different impact on inward Brownfield FDI.

Mishkin (1992) argues that lenders, such as banks, do not have all the information for selecting creditworthy borrowers. Financial crises may lead to adverse selection, resulting in lenders not willing to supply loans, even if borrowers are willing to pay higher interest rates. This can happen because higher interest rates may signal that the lender is lending to someone with bad credit risk, even if the borrower has a promising investment opportunity. As depositors worry about their bank’s health, bank runs may occur, further decreasing the supply of loans to the private sector, resulting in a negative spiral. So the supply of loans can be largely affected, and also uncertainty increases dramatically in financial markets.

(12)

(13)

12

3. DATA AND METHODS

3.1 Procedure

The empirical procedure of this study consists of two parts. In the first part the relationship between the different crisis types and aggregated FDI flows will be estimated. It provides a good sense of how FDI flows respond to financial crises. In the second part, we focus on the prices of corporate assets as is an essential element of the fire-sale hypothesis (e.g. Ang et al., 2011) . Solely observing an increase in brownfield FDI may indicate the presence fire-sales but is evidently not enough to prove Krugman’s fire-sale hypothesis. In order to be more certain, the analysis should be shifted one level down from country-level to firm-level, to check whether prices of corporate assets have decreased, and preferably, whether they have dropped below their fundamental (theoretical) values. We do this by looking at the difference between the target firm’s share-price shortly before the acquisition and the share-price offered by the acquirer, called bid-premium (see for example Weitzel et al., 2014). Since this is a time-consuming process and the availability of detailed M&A-data is limited, we perform this test for one developed (U.K.) and one developing (Malaysia) country in our sample. This way, we combine the advantages of using a large macro-level sample with a micro-level look at the underlying mechanisms.

As from here, we discuss both stages separately, beginning with the relationship between financial crises and inward brownfield FDI, and the role of credit. The second stage will be discussed in chapter 5. Table 1 below provides an overview of the hypotheses and the associated methods.

Table 1: Hypotheses and Methods

H2: Foreign Brownfield FDI (M&A) into crisis countries

increases during financial crises.

1st stage of analysis:

by using panel-regression analysis

H3: Especially when the supply of credit to firms contracts,

inward brownfield FDI (M&A) increases during crises.

H1: Prices of corporate assets decline during financial crises.

(14)

13

3.2 Sample

Our original country-sample for which Reinhart and Rogoff (2009) crises-dates were available included 71 countries. Four countries are excluded from the panel due to missing FDI-data. These countries have less than five observations on inward brownfield FDI over a time-span of twenty years, and include the Democratic People’s Republic of Korea, Central African Republic and Myanmar. Taiwan is excluded because its FDI-data is consolidated with China. Next, we omit OPEC-member countries from our sample, as the concentration of FDI in natural resources may distort our results (Bogach et al., 2012). Following UNCTAD’s classification of developed and developing countries results in 25 developed countries and 37 developing countries (see appendix I for an overview). Russian Federation’s transition economy was added to the developing countries. We recognize that excluding countries with many missing values may result in sample selectivity bias as more developed countries may self-select into the sample. Specifically, the least developed part of the developing countries are likely to have more missing values due to lack of institutions and underreporting. Nevertheless, we have excluded only very few countries and the number of developing countries is still substantial, and remains larger than the number of developed countries (see appendix I).

The time span of our panel stretches from 1990 to 2010 since UNCTAD-data for brownfield FDI - and specifically the number and value of M&A – is available as from 1990 and crises-dates are identified up to and including 2010. Although we exclude interesting crisis-data of Reinhart and Rogoff (2009) between 1800 and 1990, our study still includes an extensive number of 444 crises as from 1990 for 62 countries. Of these 444 crises, 257 are banking crises and 187 are sovereign debt crises.

3.3 Variables and Measures

3.3.1. Inward Brownfield FDI (dependent variable)

Brownfield FDI is included as dependent variable in the regression models. FDI-data is obtained from UNCTAD, which uses definitions from the Balance of Payments Manual of the IMF and OECD. These institutions refer to FDI as ‘an investment made to acquire lasting interest in enterprises operating outside of the economy of the investor’ (UNCTAD, 2013). The direct investors are entities or groups of entities that have the purpose of gaining an effective voice in the management of the firm. This intention of having control over an enterprise is FDI’s unique feature and distinguishes it from portfolio investment. Having an effective voice often goes along with equity ownership. Therefore, both IMF and OECD use a 10 per cent threshold of equity ownership to qualify an investor as a foreign direct investor. There are two reasons why UNCTAD-data is used. First, they are widely used in other FDI studies and second, UNCTAD reports besides aggregated FDI-flows separately on brownfield FDI as is captured by the total value and total number of M&A.

(15)

14 negative sign are reverse flows, sometimes called divestments, and indicate that at least one of the components is negative and is not offset by positive amounts of the remaining ones. Data on FDI-flows are presented on a net basis (UNCTAD, 2014). It may be worthwhile to mention that FDI represents a financing flow, including transfers between headquarters and subsidiaries, and thus not only investments, as is often overlooked (Calderón, 2004).

Inward Brownfield FDI is included as our main dependent variable. It entails the acquisition of already existing facilities in the host country by a foreign acquirer (Johnson, 2005). For inward brownfield FDI we use two measures as provided by UNCTAD (2014), namely:

- The value of cross-border M&A by the economy of seller and; - The number of cross-border M&A by the economy of seller.

Both measures are included in the models since total M&A value may be sensitive to an incidental large deal. In that case, the number of M&A may remain stable. On the other hand, as the number of deals fluctuates it does not provide information about changes in total value, and thus the two complement each other perfectly. The value of M&A by the economy of seller is composed of the net sales by the economy of the acquired company, and can be calculated as follows: ‘Sales of companies in the host economy to foreign TNCs – Sales of foreign affiliates in the host economy’ (UNCTAD, 2014). Only deals that involved an acquisition of an equity stake of more than 10% are included in the data. Figure 2 and 3 below visually present the pattern of brownfield FDI over the years.

Figure 2: Average M&A value (million $) (data: UNCTAD)

(16)

15

3.3.2. Financial crises (independent variables)

For the identification of the two crises types and their dates, we use data of Reinhart and Rogoff (2009). Both banking crises and sovereign debt crises are captured by a dummy which indicates for all countries for each year between 1990 and 2010 whether that particular type of crisis has occurred. Below, the exact definitions and measures of both crises-types are briefly discussed.Table 2 provides an overview.

A banking crisis can be identified by two types of events (Reinhart and Rogoff, 2009). The first type, which is most severe and systemic, occurs when there are bank runs that lead to closure, merging, or take over by the public sector of one or more financial institutions. The second, less severe situation is when there are no bank runs, but merging, takeovers, or large-scale government assistance of an important financial institution(s), that marks the start of a sequence of similar outcomes for other financial institutions.

A sovereign debt crisis can either be external or domestic. Reinhart and Rogoff (2009) mark a sovereign default as the failure to meet a principal or interest payment on the due date, or when rescheduled debt is terminated in less favorable terms than originally agreed upon. Domestic debt crises also involve the freezing of bank deposits and/or forced conversions of deposits from dollars to local currency. This distinction is not necessary for our purposes, and therefore we simply add up these two crises-dummies during the regression analysis. Table 2 below provides an overview of the crises included and their definitions.

Table 2: Crises definitions

Crisis Type (independent, dummies) Definition / Measure (source: Reinhart and Rogoff, 2009)

1. Sovereign-debt crisis (external)

Failure to meet principle or interest payment on due date or within the set period, or when rescheduled debt is

extinguished in less favorable terms than the original ones.

2. Sovereign-debt crisis (domestic)

Same as above applies + the freezing of bank deposits and forcible conversions of deposits from dollars to local currencies.

3. Banking crisis

1. Systemic/severe: Bank runs leading to closure, merging or take-over by public sector of financial institution(s). 2. Financial distress/milder: No runs, but closure, merging

(17)

16

3.3.3. Credit supply

The supply of credit is included in our models as independent variable, as it is expected to affect the extent to which firms encounter liquidity shortages, and thus whether they are inclined to sell their corporate assets at discounts. It can be considered as a proxy for economy-wide financial liquidity. The exact measure included is the domestic credit to the private sector by banks, and is obtained from the World Bank, but originally composed by the International Monetary Fund (IMF) (World Bank, 2014a).

Domestic credit to the private sector by banks are ‘financial resources provided by depository corporations, through loans, purchases of non-equity securities, trade credits, and other accounts receivable that establish a claim for repayment’ (World Bank, 2014a). Central banks are not included as depository corporations. The variable is measured as percentage of GDP.

3.3.4. Control variables

Throughout the FDI-literature, a large number of different determinants are discussed. While many of them have significant effects on FDI there is no consensus on which ones to use (e.g. Blonigen, Piger, 2011). Therefore, we include control variables for robustness based on the studies most similar to ours (see Weitzel et al., 2014; Bogach et al., 2012). These and other related studies frequently include three main categories of control variables. First, economic variables capturing the overall economic and financial situation. Secondly, variables capturing the political and business environment, and third, gravity variables concerning distance, cultural differences and so forth. This study does not include gravity variables since we do not study bilateral flows (in contrast to Weitzel et al, 2014), and as we do not use one country as benchmark (e.g. the U.S. in the case of Bogach et al., 2012).

(18)

17 To capture unintended effects of differences in political situation, a measure for political stability as control for the aggregate political environment is included (see Bogach et al., 2012). Political risks can be an important determinant of FDI as Park, Pak and Lee (2006) found for South Korea. We use the measure ‘political instability and absence of violence’, which is an index from the World bank’s Worldwide Governance Indicators (World Bank, 2014f). This variable captures the perceptions of the chance that a government will be destabilized or overthrown and also includes politically-motivated violence and terrorism (Kaufmann, Kraay, Mastruzzi, 2010), and is constructed by the World Bank (2014f) based on a wide range of sources. Unfortunately, data for the control variable political stability is only available since 1996, and observations before 1996 will therefore not be included in the regression analysis.

Other variables such as the quality of institutions and socio-economic conditions as possible determinants of M&A (e.g. Rossi and Volpin, 2004) are not included as they presumably change in the longer-term only, and these variables are therefore not likely to explain differences in total M&A value or M&A number in a relatively short period around financial crises.

3.3.5. Tax havens

FDI flows into some countries are expected to be contaminated by FDI round tripping, which occurs when direct investors channel their funds abroad to special purpose entities (SPEs) and subsequently return these investments to the local economy in the form of FDI (IMF, 2004). As the terminology indicates, these investments are not channeled to ordinary firms, but roundtrip relatively quickly via these SPEs with specific or temporary objectives. These SPEs are often based in tax-haven economies with favorable tax regulations and convenient administrative support, incentivizing round tripping (IMF, 2004). Location decisions for greenfield investments are generally more sensitive to tax havens than brownfield investments, since tax-advantages are partly capitalized into the price of M&A (Hebous, Ruf and Weichenrieder, 2010). Nevertheless, Hebous et al. (2010) found that an increase in corporate tax rate of 10% reduces M&A investments with 3.6%. Therefore, we expect that our brownfield FDI-data might be biased by these tax haven effects.

We deal with this issue by creating a dummy variable for tax-haven countries. We prefer creating a dummy over omitting tax-haven countries since a great number of nations are identified as tax havens and deleting them may significantly reduce our sample-size. Moreover, using a dummy allows us to observe whether our results differ for tax haven and non-tax haven economies. We follow the Tax Justice Network for identifying tax haven economies and label an economy as tax haven when it is on the Tax Justice Network 2005 blacklist, beginning from the date it has history as secrecy jurisdiction (Tax Justice Network, 2014). Thus if the date by which a country became a secrecy jurisdiction is unknown, but it is on the blacklist, we mark that country as a

tax haven for the entire period (see appendix II for an overview). In the case countries have a history as

(19)

18 fairly sure that most commonly known tax havens are included. Table 2 below provides an overview of all variables and sources included.

Table 2: Variables and sources

Variable Source Code (if applicable) Date downloaded

Dependent:

M&A value (U.S.$) UNCTAD 12-09-14

M&A number UNCTAD 12-09-14

Crises dummies:

Banking crises Reinhart & Rogoff (‘09) 04-10-14

Sovereign debt crises Reinhart & Rogoff (‘09) 04-10-14

Credit supply:

Domestic credit by banks (% GDP) World Bank FD.AST.PRVT.GD.ZS 12-09-14

Controls:

GDP per capita (U.S.$) World Bank NY.GDP.PCAP.CD 12-09-14

GDP growth World Bank NY.GDP.MKTP.KD.ZG 12-09-14

Inflation World Bank FP.CPI.TOTL.ZG 12-09-14

Exchange rate World Bank PA.NUS.FCRF 23-11-14

Political stability World Bank - WGI 13-11-14

Dummy variables:

Tax haven dummy Tax Justice Network 11-11-14

Developing country dummy (DC) UNCTAD 12-09-14

(20)

19

3.3.6. Limitations

Although the dataset includes 62 countries and reflects a proper balance between developed and developing countries, missing values result in the loss of a relatively large number of observations that will be included in the panel-regression estimations. Since each variable discussed before has its theoretical foundations and plausibly contributes to the explanatory power of our model, it is preferred to include all these (control) variables. Nevertheless, especially the control variable political stability suffers from missing values. It appears that observations from 1990 until 1996 are missing for every country for this variable. Although this is an acknowledged limitation, it does not affect the sample regarding the selection of countries.

(21)

20

3.4 The model

During the first stage of this study, in which we test the relationship between financial crises and inward brownfield FDI (hypothesis 2), a panel estimation is executed. Panel analyses allow for testing multiple cross-sectional units (countries) over time and account for heterogeneity between these units (Hill, Griffiths, Lim, 2011). The following two models will be used to test our hypotheses:

(1) FDI valueit = β1 +β2 CRISP*DCit-1 + β3 CRED*DCit-1 + β4 DCi + β5 TAXit + β6 TAX*CRISP*DCit-1 + β7 CRED*CRISP*DCit-1 +β8 Xit-1 + ϵit (2) FDI numberit = β1 +β2 CRISP*DCit-1 + β3 CRED*DCit-1 + β4 DCi + β5 TAXit +

β6 TAX*CRISP*DCit-1 + β7 CRED*CRISP*DCit-1 +β8 Xit-1 + ϵit

- Where FDIit is a measure of brownfield FDI inflow (M&A-value/number) for country i in year t, - CRISPit is a binary crisis-indicator of Reinhart and Rogoff, for crisis type P in country i in year t, - CREDit is a measure for the supply of credit by banks,

- DCi is a dummy indicating a developing (DC) or developed country, - TAXit is the tax haven identifying dummy, and;

- Xit is a vector of control variables and ϵ is the error term.

Equations (1) and (2) above are the same except for the dependent variable, as we use both the value and the

number of brownfield FDI (M&A). Also, one can observe that all independent variables are systematically

interacted with the developing country dummy (DC), to avoid inappropriate pooling (Blonigen and Wang, 2004) as discussed before. It will provide separate coefficients for developing and developed countries. Furthermore, a few new things are introduced here. First, the equations contain two interaction terms. The first one, TAX*CRIS, will indicate whether inward brownfield FDI reacts differently to financial crises in tax haven economies. Or formulated differently, given there is a financial crisis in a country, if it matters for the relationship between financial crises and inward brownfield FDI whether this country is a tax haven economy. Secondly, we also interact the banking- and sovereign debt crises dummies with the credit supplied by banks since we expect that fire-sales, and thus increased inward brownfield FDI, especially occur when credit contracts (Weitzel et al., 2014). This is denoted by the interaction term CRED*CRIS.

(22)

21 changes in FDI flows. Formulated differently, the lags allow the crisis to have impacted the real economy via the contraction of aggregated demand and credit supply to firms, through which the troubles are transferred to the firms of which some may need to fire-sale their assets. Also the latter step may take some time as firms may be able to cushion the blow by using their liquid assets and reserves, but run out of liquidity after some time has elapsed. Since we are not aware of studies that estimated the time this process takes, we follow the study of Weitzel et al. (2014) and lag all independent variables by one period, which is in our case equivalent to one year. Although Weitzel et al. (2014) use lags of one quarter, argumentation for the length of this lag is lacking, and since our Reinhart and Rogoff crises data are only available on a yearly basis, we are bound to one-year lags. In the results section we will return to this issue by checking whether our results change when lags are removed as robustness test.

While equations (1) and (2) represent the structure of our full models, we also estimate two ‘reduced’ variants of the equations for each crisis (see table 3 for an overview). One model – or actually four in total (1, 4, 7, and 10) – include only the main effects, thus solely the crises-dummies and credit supply as independent variables. In the next models (2, 5, 8, and 11), we add the controls. This allows to observe whether including the controls has impact on the main effects. The full models (3, 6, 9, and 12) include the main effects, control variables, and the interaction terms.

Table 3: Overview of models

Dependent: M&A-value (equation 1)

Banking crises

Model 1 Main effects

Model 2 Main effects + controls

Model 3 Main effects + controls + interactions

Sovereign debt crises

Dependent: M&A-number

(equation 2)

Banking crises

Sovereign debt crises

(23)

22

3.4.1. Panel estimation method and assumption tests1

Multiple models are available for dealing with panel data, including a pooled model, a fixed effects model and a random effects model. The main disadvantage of the pooled model is that it ignores differences between countries that may lead to different coefficients, and is therefore put aside. The fixed effects model allows for unobserved individual heterogeneity, which is captured by the intercept, and the random effect model recognizes that the countries in our sample were randomly selected (Hill et al., 2011).

In choosing between random and fixed effects we employed an Hausman test (Hausman, 1978), and compared the coefficient estimates from random to those of fixed effects. The estimates for random and fixed effects differed significantly and therefore the fixed effects estimation is preferred over the random effects estimation2_{(see appendix III). The only exception is model 12, for which the test statistic points to using} random effects(Chi2 = 12.74; p > F = 0.753). The main drawback of using fixed effects (for the models 3, 6 and 9) is that interesting results are dropped during the regressions, as some variables are time invariant and will not be picked up when using fixed effects. Therefore, we follow Ortega-Argilés, Piva and Vivarelli (2011) and circumvent this problem by estimating all models both with fixed effects – as econometrically justified by the Hausman test – and random effects, which provide more complete results as it allows for the dummies and interactions.

Besides testing for inconsistent OLS-estimates, the fixed- and random-effects panel estimations also require the satisfaction of a few other important assumptions. First, we check for normality as non-normality may lead to misspecification (Jarque, Bera, 1987). The dependent variables for brownfield FDI, specifically the value of M&A and the number of M&A, are not normally distributed. An inspection of the residuals leads to the same conclusion, which is further confirmed by the results of a Jarque-Bera test we employed (see appendix III) (Jarque, Bera, 1987). This problem is resolved by transforming the variables into logarithms. Also, potential heteroskedasticity may bias the results, which occurs when observations have different variances (Hill et al., 2011). We employed a likelihood-ratio test, which indicated that all models suffer from heteroskedasticity. We circumvent this problem by using White’s heteroskedasticity-consistent standard errors for all models. Further, the data may be affected by autocorrelation. One can visually inspect the correlation matrix (appendix IV), but we also employed a Wooldrige test (Drukker, 2003), which indicated that models 9 and 12 suffer from autocorrelation3. Therefore, we use the cluster option which is robust for autocorrelation. Lastly, we have checked for multicollinearity by calculating variance inflation factors (VIF-values). All VIF-values were smaller than the commonly used threshold (10), which indicates that there are no serious multicollinearity problems (Adnan, Ahmad, Adnan, 2006). Appendix III provides an overview of the assumption tests and their outcomes.

1

Detailed results of the assumption tests can be found in appendix III.

2

P-values < 0.05, thus H0 rejected, indicating that the random effect model is affected by incorrect specification.

(24)

23

4. RESULTS

Table 4 presents the descriptive statistics. While most statistics are straightforward, please note that the domestic and external sovereign debt crises were sum up into one indicator for sovereign debt crises. Therefore, its maximum value is 2, which prevails when a country is hit by both a domestic- and external sovereign debt crisis in the same year.

Table 4: Descriptive statistics

Obs. Mean Std. Dev. Min. Max.

Dependent

M&A value (million $) 1114 6076 21613.030 -100577.600 252049.200

M&A number 1218 122 228.180 0 1967

Crises dummies

Banking crises 1302 0.197 0.398 0 1

Sovereign debt crises 1302 0.159 0.406 0 2

Credit supply

Domestic credit by banks (% GDP) 1279 62.319 46.572 3.657 319.461

Controls GDP per capita ($) 1302 13.231 15.401 0.223 95.190 GDP growth 1301 3.346 3.761 -17.669 15.240 Inflation 1241 25.937 248.341 -4.480 7481.664 Exchange rate 1295 0.259 1.094 2.96*10-8 10.390 Political stability 749 52.595 29.170 0 100 Tax haven 1302 0.239 0.427 0 1 Developing 1302 0.597 0.491 0 1

As discussed before, the dependent variable inward brownfield FDI is measured based on the total value of inward M&A and the total number of M&A conducted by foreign investors. Table 5 and 6 on the next pages present the results of the regression analyses for each of them respectively. The models 1-6 in table 5 are thus exactly similar to the models 7-12 in table 6, except for the dependent variable. Table 5 and 6 each include six models, three for banking crises, and three for sovereign debt crises.

(25)

24 Main effects RE FE RE FE RE FE RE FE RE FE RE FE Crisis*DC (0-1) 1-0 -0.673*** -0.678*** -0.248 -0.228 0.043 0.059 -1.029*** -1.025*** (0.173) (0.173) (0.305) (0.293) (0.670) (0.582) (0.041) (0.029) 1-1 -0.167 -0.150 0.413 0.300 0.307 0.133 -0.725 -0.681 0.530 1.449** 2.205** 2.522* (0.273) (0.280) (0.344) (0.324) (0.448) (0.489) (0.449) (0.487) (0.455) (0.502) (0.835) (1.179) 2-1 -1.968*** -2.067*** 0.117 -0.708 -1.567 -3.032*** (0.560) (0.543) (0.732) (0.851) (0.864) (0.793) Credit supply Credit by banks*DC(0) 0.015*** 0.016*** 0.004 0.006 0.006 0.010 0.014*** 0.014*** 0.002 0.006 0.002 0.006 (0.003) (0.003) (0.004) (0.004) (0.005) (0.005) (0.003) (0.003) (0.004) (0.004) (0.004) (0.004) DC(1) 0.025** 0.023* 0.015 -0.002 0.015 -0.003 0.023** 0.021* 0.019* -0.000 0.024** 0.003 (0.008) (0.010) (0.009) (0.012) (0.010) (0.012) (0.008) (0.009) (0.009) (0.013) (0.009) (0.012) Controls Developing (DC=1) -2.603*** -3.536*** -3.363** -2.377*** -3.940*** -4.036*** (0.586) (0.978) (1.027) -0.623 (1.076) (1.097) Tax haven (=1) -0.211 0.080 0.169 0.717 -0.176 0.333*** 0.152 0.828 (0.424) (0.065) (0.515) (0.639) (0.433) (0.001) (0.493) (0.587) GDP per capita*DC(0) 0.026* 0.017 0.021* 0.011 0.027* 0.015 0.029* 0.015 (0.011) (0.012) (0.011) (0.012) (0.011) (0.012) (0.011) (0.012) DC(1) 0.129*** 0.093 0.138*** 0.096 0.135** 0.095 0.144** 0.091 (0.039) (0.052) (0.040) (0.050) (0.042) (0.059) (0.044) (0.061) GDP growth*DC(0) 0.106*** 0.110*** 0.107*** 0.112*** 0.118*** 0.122*** 0.117*** 0.122*** (0.029) (0.029) (0.029) (0.029) (0.029) (0.026) (0.029) (0.026) DC(1) 0.074* 0.045 0.073* 0.045 0.079* 0.050 0.072 0.050 (0.037) (0.042) (0.037) (0.042) (0.037) (0.040) (0.041) (0.041) Inflation*DC(0) -0.022 -0.010 -0.022* -0.010 -0.030* -0.013 -0.031* -0.013 (0.013) (0.009) (0.011) (0.009) (0.015) (0.008) (0.016) (0.008) DC(1) -0.007*** -0.006*** -0.007*** -0.006*** -0.007*** -0.005*** -0.007*** -0.005*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Political stability*DC(0) -0.017* -0.006 -0.018* -0.007 -0.020* -0.007 -0.022** -0.007 (0.008) (0.008) (0.008) (0.008) (0.008) (0.009) (0.008) (0.009) DC(1) -0.030* -0.009 -0.024 -0.010 -0.033* -0.001 -0.027 -0.001 (0.015) (0.019) (0.015) (0.019) (0.014) (0.016) (0.014) (0.016) Exchange rate*DC(0) -1.049** -1.136*** -0.991** -1.070*** -1.047** -1.156*** -1.036* -1.156*** (0.360) (0.248) (0.363) (0.235) (0.396) (0.245) (0.434) (0.246) DC(1) 0.075 0.150 0.071 0.139 0.069 0.176* 0.048 0.154 (0.164) (0.075) (0.170) (0.077) (0.176) (0.082) (0.189) (0.078) Interactions Crisis*DC*Taxhaven 0-1-1 -1.627 -2.070* -2.195 (1.116) (0.955) (1.386) 1-0-1 0.696 0.783 (0.409) (0.396) 1-1-1 -2.080* -0.285 2.607 (0.981) (0.506) (1.568) Crisis*DC*Bankcredit 1-0 -0.005 -0.006 (0.004) (0.003) 1-1 0.003 0.004 -0.065* -0.040 (0.006) (0.006) (0.028) (0.032) 2-1 0.048 0.071*** (0.027) (0.015) Constant 6.542*** 5.363*** 8.221*** 6.298*** 7.973*** 6.325*** 6.548*** 5.415*** 8.533*** 6.100*** 8.427*** 6.245*** (0.379) (0.270) (0.834) (0.601) (0.869) (0.614) (0.376) (0.271) (0.890) (0.604) (0.900) (0.604) Observations 962 962 531 531 531 531 962 962 531 531 531 531 Wald Chi2 84.78*** 455.79*** 774.14*** 8394.61*** 665.46*** 3976.53*** F - 14.53*** 14.80*** - 18.62*** -Overall R-squared 0.301 0.127 0.445 0.373 0.463 0.316 0.301 0.156 0.449 0.084 0.486 0.157 Number of countries 62 62 60 60 60 60 62 62 60 60 60 60

Grey coloured rows present results for developing countries (DC=1) Robust standard errors in parentheses

All independent variables are lagged by 1 year *** p<0.001, ** p<0.01, * p<0.05

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

Table 5: Results fixed and random effects regression, M&A-value Dependent: Log M&A-value by economy of seller

(26)

25 Main effects RE FE RE FE RE FE RE FE RE FE RE FE Crisis*DC (0-1) 1-0 -0.341*** -0.341*** -0.304*** -0.306** -0.153 -0.148 -1.023*** -1.024*** (0.055) (0.054) (0.092) (0.095) (0.225) (0.228) (0.014) (0.012) 1-1 -0.108 -0.110 0.045 0.027 0.098 0.078 -0.577*** -0.567*** -0.398** -0.356** -0.031 -0.043 (0.112) (0.112) (0.142) (0.142) (0.182) (0.183) (0.141) (0.144) (0.129) (0.130) (0.243) (0.247) 2-1 -0.614*** -0.633*** -0.334 -0.393 0.002 -0.117 (0.165) (0.165) (0.215) (0.206) (0.396) (0.380) Credit supply Credit by banks*DC(0) 0.005*** 0.005*** 0.001 0.002* 0.002* 0.002** 0.004*** 0.004*** 0.001 0.001 0.001 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) DC(1) 0.016*** 0.016** 0.009* 0.007 0.010 0.008 0.015*** 0.015** 0.009* 0.007 0.010* 0.008* (0.005) (0.005) (0.004) (0.004) (0.005) (0.005) (0.005) (0.005) (0.004) (0.004) (0.004) (0.004) Controls Developing (DC=1) -2.495*** -2.606*** -2.349*** -2.353*** -2.651*** -2.429*** (0.425) (0.397) (0.453) (0.431) (0.427) (0.463) Tax haven (=1) 0.034 0.043* 0.005 0.569 0.092 0.170*** 0.010 0.557 (0.180) (0.020) (0.391) (0.488) (0.194) (0.000) (0.386) (0.489) GDP per capita*DC(0) 0.012*** 0.011*** 0.011*** 0.011*** 0.009* 0.008* 0.009* 0.008* (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) DC(1) 0.095* 0.095* 0.095* 0.093* 0.087* 0.087* 0.087* 0.085* (0.039) (0.042) (0.040) (0.043) (0.035) (0.039) (0.035) (0.038) GDP growth*DC(0) 0.014 0.015 0.014 0.015 0.030*** 0.031*** 0.030*** 0.031*** (0.010) (0.010) (0.010) (0.010) (0.008) (0.008) (0.008) (0.008) DC(1) 0.027** 0.025* 0.027** 0.024* 0.023 0.020 0.021 0.018 (0.010) (0.010) (0.010) (0.010) (0.012) (0.011) (0.011) (0.011) Inflation*DC(0) -0.004 -0.003 -0.006 -0.005 -0.008* -0.006 -0.008* -0.006 (0.003) (0.003) (0.005) (0.005) (0.004) (0.003) (0.004) (0.003) DC(1) -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Political stability*DC(0) -0.002 -0.000 -0.002 -0.001 -0.003 -0.001 -0.003 -0.001 (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) DC(1) -0.006 -0.003 -0.005 -0.003 -0.006 -0.003 -0.005 -0.004 (0.003) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Exchange rate*DC(0) 0.070 0.070* 0.082* 0.082* 0.045 0.046 0.045 0.046 (0.036) (0.028) (0.038) (0.032) (0.045) (0.033) (0.045) (0.033) DC(1) 0.138*** 0.140*** 0.138*** 0.143*** 0.152*** 0.159*** 0.150*** 0.158*** (0.030) (0.020) (0.030) (0.021) (0.027) (0.022) (0.033) (0.020) Interactions Crisis*DC*Taxhaven 0-1-1 -1.294 -0.052 -1.251 0.532 (0.784) (0.265) (0.746) (0.537) 1-0-1 -0.023 -0.026 (0.133) (0.137) 1-1-1 -1.259 -1.668* (0.706) (0.848) Crisis*DC*Bankcredit 1-0 -0.001 -0.001 (0.001) (0.001) 1-1 -0.001 -0.001 -0.011 -0.009 (0.003) (0.003) (0.010) (0.010) 2-1 -0.010 -0.008 (0.009) (0.008) Constant 4.353*** 2.975*** 4.554*** 3.026*** 4.349*** 3.011*** 4.346*** 3.028*** 4.722*** 3.153*** 4.523*** 3.072*** (0.273) (0.139) (0.273) (0.170) (0.322) (0.181) (0.284) (0.135) (0.305) (0.172) (0.343) (0.176) Observations 1,148 1,148 631 631 631 631 1,148 1,148 631 631 631 631 Wald Chi2 149.70*** 315.60*** 494.29*** 177452.89*** 194.29*** 209.08*** F - 24.90*** 27.27*** - 18.27*** 13.69*** Overall R-squared 0.415 0.005 0.484 0.031 0.507 0.03 0.428 0.040 0.499 0.026 0.525 0.007 Number of countries 61 61 59 59 59 59 61 61 59 59 59 59

All independent variables are lagged by 1 year *** p<0.001, ** p<0.01, * p<0.05

Model (12)

Model (7) Model (8) Model (9) Model (10) Model (11)

Table 6: Results fixed and random effects regression, M&A-number

Dependent: Log M&A-number by economy of seller

(27)

26

4.1 Model fit

The R-squared values for the full models for banking crises (models 3 and 9) are 31.6% and 3%, and models 6 and 12 for sovereign debt crises have R-squared values of 15.7% and 52.5% respectively (note that we look at the fixed effects columns for models 1,3 and 9, and at the random effects column for model 12). These differences indicate that the models vary in the extent to which they fit the data well, and that especially model 9 does a poor job in this regard. However, the F-values for the fixed effect models and Wald Chi2 values for the random effect models are all significant at the 1% level, indicating a good fit (see bottom of tables 5 and 6). Only for the fixed effects regression in model 6 for sovereign debt crises, we were not able to obtain an F-value. Therefore, especially the interpretation of models 9 and 6 requires some caution as their R-squared values are rather low.

4.2 Regression results for hypothesis 2, the relationship between financial crises and inward FDI

Now, we will discuss the findings from the regression analyses. The results are discussed per hypothesis, and we (re)present the relevant coefficients from the full models (3, 6, 9 and 12) in the tables below. We will refrain from discussing each individual regression coefficient from table 5 and 6 as there are too many, but we discuss the ones relevant for the hypotheses and those that are unexpected. In the tables, ‘DC’ stands for ‘developing country’. Thus for rows where DC = 0, the coefficient belongs to the developed countries, while rows with DC = 1 present the coefficients for the developing countries (grey colored rows). Note that the coefficients for sovereign debt crises in developed countries are missing (row 1-0). Sovereign debt crises occurred only 5 times in developed countries between 1990-2010, and these observations are omitted by the software in the full models due to missing values in the other (control) variables. The coefficients can only be obtained from the ‘main effects’ models 4 and 10. We start with hypothesis H2: Foreign Brownfield FDI

(M&A) into crises countries increases during financial crises. Table 7 below presents the relevant

coefficients.

Table 7: Regression results for hypothesis H2: ‘Inward M&A increases during crises’

Exp.

sign Dependent: Inward M&A value Dependent: Inward M&A number

Model 3 (Banking crises) Model 6 (Sovereign debt crises) Model 9 (Banking crises) Model 12 (sovereign debt crises) Crises*DC (0-1) 1-0 + 0.059 -0.148 (0.582) (0.228) 1-1 + 0.133 2.522* 0.078 -0.031 (0.489) (1.179) (0.183) (0.243) 2-1 + -3.032*** 0.002 (0.793) (0.396)

(28)

27 The coefficients in table 7 are ambivalent, and most are insignificant. Banking crises that occur in developed countries neither increase nor decrease M&A-value (0.059) and number (-0.148), as both coefficients are statistically insignificant. The positive signs of the coefficients for banking crises taking place in developing countries (see row 1-1) suggest an increase in the value (0.133) and number (0.078) of M&A, but these coefficients are also insignificant.

When we look at sovereign debt crises, the coefficient 2.522* is remarkable. It suggests that sovereign debt crises taking place in developing countries have a significant positive effect on inward M&A value, which is in accordance with our hypothesis that fire-sales may result in increased M&A-flows into crises-countries. However, the coefficient in model 12 for the number of M&A is not statistically significant, and is even negative. Therefore, we refrain from placing too much value on this individual significant value. Also, in cases where both an internal and external sovereign debt crisis occur (row 2-1), the coefficient is not significant anymore and turns even negative. This is strange, as one would expect that this positive relationship between a sovereign debt crises and inward M&A would even be stronger when two crises hit simultaneously.

Overall, the results are ambivalent and mostly insignificant. Also if we look at the other models in table 5 and 6 with only the main effects (models 1, 4, 7 and 10), most coefficients for both developed and developing countries are negative and some are (strongly) significant. When we add the control variables in the models 2, 5, 8 and 11, and later the interactions in the full models, these strong negative effects become weaker or disappear at all. The controls thus seem to explain much of the variation. Based on our results, we cannot conclude that financial crises increase inward Brownfield FDI. In fact, if at all, crises rather seem to decrease inward foreign FDI. This conclusion seems to hold for both developing and developed countries. The conclusion we can draw confidently is that we do not find evidence for a positive relationship between banking- or sovereign debt crises and inward brownfield FDI. Therefore, we reject hypothesis 2. This is in line with previous studies (e.g. Weitzel et al., 2014 & Bogach et al., 2012), which did not find evidence either and if at all, find that financial crises rather decrease inward Brownfield FDI across the board.

4.3 Regression results for hypothesis 3, the role of the supply of credit

We continue with the results related to hypothesis H3: Especially when the supply of credit to firms contracts,

inward brownfield FDI (M&A) increases during crises. The models in table 5 and 6 with only main effects (1,

(29)

28 a lower amount of credit supplied by banks increases inward M&A (via fire-sales). In fact, we find some evidence for a positive relationship between credit supplied to the private sector by banks and inward Brownfield FDI.

Although these coefficients are informative, we are mainly interested in the interaction between financial crises and credit (see lower part table 8). We want to know whether credit contractions increase inward Brownfield FDI during crises, as more fire-sales are expected when there is less credit available. The interaction effects are presented in the bottom rows of table 8 below. One coefficient is significant at the 0.1% level (0.071***), suggesting that more inward brownfield FDI is attracted when there is more credit supplied in cases when both a domestic and external sovereign debt crisis hit simultaneously. However, all other coefficients are insignificant, and the coefficient for situations where only one sovereign debt crisis occurs (-0.040) is even negative. All in all, the evidence supporting our third hypothesis is rather weak. Overall, we find no convincing evidence for the hypothesis that credit contractions increase inward Brownfield FDI during crises (through the occurrence of fire-sales). Weitzel et al. (2014) found no evidence either, and even found the opposite effect, namely that countries with higher credit levels attracted more cross-border acquisitions.

Table 8: Regression results for hypothesis H3, 'credit contractions further increase inward M&A'

Exp. sign

Dependent: Inward M&A value

Dependent: Inward M&A number Model 3 (Banking crises) Model 6 (Sovereign debt crises) Model 9 (Banking crises) Model 12 (sovereign debt crises) Credit by banks*DC(0) - 0.010 0.006 0.002** 0.001 (0.005) (0.004) (0.001) (0.001) DC(1) - -0.003 0.003 0.008 0.010* (0.012) (0.012) (0.005) (0.004) Interactions: Crises(0)*DC(0)*Bank-credit 1-0 - -0.006 -0.001 (0.003) (0.001) 1-1 - 0.004 -0.040 -0.001 -0.011 (0.006) (0.032) (0.003) (0.010) 2-1 - 0.071*** -0.010 (0.015) (0.009)

(30)

29

4.4 Tax haven vs. non-tax havens & developing vs. developed economies

As explained before, all independent variables were interacted with the developing country (DC) dummy to avoid inappropriate pooling of developing and developed countries, in accordance with Blonigen and Wang (2004). Further, a dummy indicating tax havens was included to check whether results differ for tax haven countries. Both dummies were also included individually, apart from the interactions for which these dummies were used. Their coefficients can be found in table 9 below. Remember that these are coefficients from our random effect models, as these time-invariant dummies are omitted in our fixed effect regressions. Note that, except for model 12, fixed effects should be used as indicated by the Hausman test, so coefficients should be interpreted with caution here.

Table 9: Regression results for the dummies ‘tax haven’ and ‘developing’ economies

Exp. sign

Dependent: Inward M&A number Model 3 (Banking crises) Model 6 (Sovereign debt crises) Model 9 (Banking crises) Model 12 (sovereign debt crises) Developing economy (DC) - -3.363** -4.036*** -2.349*** -2.429*** (1.027) (1.097) (0.453) (0.463) Tax haven + 0.717 0.082 0.569 0.557 (0.639) (0.587) (0.488) (0.489) Interaction: Crises*DC*Tax haven 0-1-1 -1.627 -2.070* -1.294 -1.251 (1.116) (0.955) (0.784) (0.746) 1-0-1 0.696 -0.023 (0.409) (0.133) 1-1-1 -2.080* 2.607 -1.259 -1.668* (0.981) (1.568) (0.706) (0.848)

*** p<0.001, ** p<0.01, * p<0.05

(31)

30 A remaining question is whether pooling of developed and developing countries would have been inappropriate in our study as Blonigen and Wang (2004) claimed. In other words, did it matter to systematically interact each independent variable with the DC dummy to avoid inappropriate pooling? While our main effects did not differ dramatically between developing and developed countries, the results for our controls (discussed in next subsection) sometimes did. For example, the effect of exchange rates on inward Brownfield FDI differs largely for developing and developed countries, as will be further discussed in the next section. Also, while most coefficients for the relationship between banking crises and inward Brownfield FDI were insignificant, sometimes their sign differed for developed and developing countries. These differences indicate that pooling developed and developing countries would indeed have been inappropriate. However, our results were not stronger (e.g. more significant) for either developing or developed countries.

Table 9 also presents the coefficients for the tax haven dummy, and the interaction between this dummy and the crises dummies. Tax haven economies seem to attract more inward Brownfield FDI as indicated by the positive coefficients, but all are statistically insignificant. When we look at the interaction of the two crises-types with the developing country and tax haven dummy (lower part table 9), we find that almost all coefficients are negative. The coefficient (-2.070*) in model 6 indicates that developing countries classified as tax havens attract less brownfield FDI. When these countries are hit by a banking crisis (see row 1-1-1), we observe that M&A value decreases significantly (-2.080*). When hit by a sovereign debt crises, we see that the number of M&A decreases significantly (-1.668*). At first glance it may seem strange that these interaction coefficients for tax haven economies are negative. However, it may simply mean that the ‘negative characteristic’ of being a developing country for attracting FDI outweighs the ‘positive characteristic’ of being a tax haven for attracting FDI. For developed countries we find no significant effects.

4.5 Control variables

(32)

31

Table 10: Regression results control variables

Exp. sign

Dependent: Inward M&A number Model 3 (Banking crises) Model 6 (Sovereign debt crises) Model 9 (Banking crises) Model 12 (sovereign debt crises)

GDP per capita*DC(0) ₊ _0.011 _0.015 _0.011*** _0.009* _(0.012) _(0.012) _(0.003) _(0.003) DC(1) 0.096 0.091 0.093* 0.087* (0.050) (0.061) (0.043) (0.035) GDP growth*DC(0) + 0.112*** 0.122*** 0.015 0.030*** _(0.029) _(0.026) _(0.010) _(0.008) DC(1) 0.045 0.050 0.024* 0.021 (0.042) (0.041) (0.010) (0.011) Inflation*DC(0) _- _-0.010 _-0.013 _-0.005 _-0.008* (0.009) (0.008) (0.005) (0.004) DC(1) -0.006*** -0.005*** -0.000 -0.000 (0.001) (0.001) (0.000) (0.000) Political stability*DC(0) + -0.007 -0.007 -0.001 -0.003 _(0.008) _(0.009) _(0.002) _(0.002) DC(1) -0.010 -0.001 -0.003 -0.005 (0.019) (0.016) (0.004) (0.004) Exchange rate*DC(0) - -1.070*** -1.156*** 0.082* 0.045 (0.235) (0.246) (0.032) (0.045) DC(1) 0.139 0.154 0.143*** 0.150*** _(0.077) _(0.078) _(0.021) _(0.033)

Grey rows present results for developing countries (DC=1) Robust standard errors in parentheses

*** p<0.001, ** p<0.01, * p<0.05