Mergers & Acquisitions in Banking: The effect of the Economic Business Cycle Student name: Lucy Hazen Master student Finance at Tilburg University Administration number: 507779 E-mail address: 1st Supervisor: Prof. Dr. S.C.W. Eijffinger President Tilburg University Society 2nd Supervisor: M.L. Kobielarz MSc PhD-student at Tilburg University Date of submission final draft: 2 October, 2016 Abstract Literature argues that the business cycle of a country plays a role in overall M&A activity, and thus also for bank specifically. In previous studies different forms of GDP have been used for indicating the swings of the business cycle. However, there are more factors that show the state of the business cycle. In this study I investigate whether indeed the business cycle, measured by more than one variable, is an important driver. This will be analysed on the number and value of bank M&A activity in EU15 countries during the period from January 1985 to December 2014. Furthermore, this study will also examine if differences in M&A characteristics, such as if domestic and cross-border deals are also linked to the business cycle. The results suggest there does exist a relationship between the business cycle proxies and bank M&A activity. However, not all business cycle measures appear to have an influence.
1. Introduction In the past three decades a lot has changed in the banking industry, where most of the Mergers and Acquisitions (M&As) have occurred in the US as well as in Europe. The number of M&As mostly increased in the 1990s compared to the 1980s in Europe. Theory argues that there are several reason for banks to participate in a M&A. Some of these underlying motives can be value (i.e. achieve economies of scale and scope, increase market power, decreasing risk by expanding their products and outside borders, etc.) and non-value based (i.e. based on managerial motives). However, deciding to agree in a M&A as a bank might be influenced by exogenous factors such as the economic environment of a country. This has been widely analysed and discussed, and it is confirmed that there is an important relation between the economic development of a country and M&A activity. Previous literature on M&A determinants have focused a lot on US bank, while there are relatively a few that focused on European (EU) countries. Studies on EU countries in this field are upcoming. The determinants of overall M&A activity have been widely investigated, where sector specific M&As, such as the banking sector, somewhat less. Also, these studies mostly looked at the number of M&As rather than deal value. Nonetheless, there are studies that focussed on bank M&As, where it is still important to use the knowledge from early research about overall M&As. There have been many attempts in previous literature to reach a better understanding of the influence macroeconomic fundamentals on M&A activity. In a lot of studies about different M&As they argued that macroeconomic variables play an important role (e.g. Jovanovic and Rousseau, 2002; Resende, 2008; Albertazzi and Gambacorta, 2002; Choi and Jeon, 2011). They investigated the drivers of bank M&As and came to the conclusion that the business cycle plays an important role. Jovanovic and Rousseau (2002) argued specifically that the business cycle is an important explanatory variable. In theory there are more proxies for the business cycle than only GDP, which were almost never used. Although, the business cycle based on GDP (per capita) has an effect, it can be measured with several other variables, such as unemployment rate and the output gap. These other measures are also important to include if we want to conclude that the business cycle has an significant effect on M&A activity in every aspect. In broader terms, this research contributes to the understanding of bank M&A activity and if it is determined by the state of the business cycle. Since M&As and in this case bank M&As can contribute in a countries economy, it remains important to understand the determinants of M&A activity. In the past 30 years a lot happened economically and by including the most recent financial crisis, it would be interesting to see what kind of effect these economic swings have for bank M&As. Beltratti and Paladino (2013) studied the abnormal return during the recent financial crisis and concluded that the behaviour of M&A activity in the banking sector was indeed different than if M&As were carried out in normal times. With the financial crisis practically passed but still leaving its marks, it is interesting to see whether the business cycle still has an influence, where most research on the determinants of M&A activity looked at the period before the crisis occurred. This results in the following main hypothesis, whether the state of the business cycle still has an effect on M&A activity when the output gap and 1
unemployment rate are included. As the results indicate, the business cycle proxies have different influences on M&A activity. The output gap does not have a strong effect on the number of M&As, where GDP per capita growth only seems to matter when looking at the delayed effect and the unemployment rate has a significant effect in case of both the immediate and delayed effect. However, the business cycle does appear to have more explanatory power when looking at the delayed effect. Each M&A is different and is based on different characteristics such as domestic or cross-border deals and deal attitude. In previous literature they already tried to find what drives these differences. 1 In most of these studies they used GDP (per capita) growth as explanatory variable. However, as argued before there are more business cycle measures, hence this study will also look if these differences in M&A characteristics are indeed linked to the business cycle. As for deal attitude, hostile deals only occurred a few times in the data used in this study, which is in line with the study of Rossi and Volpin (2004) where they argued that the frequency of hostile takeovers is very small. Therefore, this will not be further analysed. There are also other characteristics that are important such as bank characteristics. However, this study will not focus on these characteristics, since bank s will not be studied in detail. The main focus lies more on domestic and cross-border M&As and use country characteristics such as the strength of legal rights, bank concentration and other macroeconomic factors as control variables. This results in the following second hypothesis, whether the business cycle still has an influence on domestic and cross-border deals when the output gap and unemployment rate are included. However, the results came out very different than expected. As previous literature argued that macroeconomic variables have somewhat the same influence on both domestic and cross-border deals, which will be argued in the next section, this was not confirmed in this analysis. Furthermore, there is also no consistency seen regarding which business cycle measure is influencing either domestic or cross-border bank M&As. As for being a target in a cross-border deal, it appears the business cycle does not play a role. The difference with this study and previous studies is that we use a different model, set of control variables and a larger period sample. In this paper I analyse a sample of completed bank M&As announced within the period January 1, 1985 and December 12, 2014. The sample of this study comprises bank M&A deals of the EU15 countries and shows the influence of the business cycle on the pattern of M&A activity. 2 My review of this subject proceeds as follows. In Section 2, I will review previous empirical literature related to M&A activity and the business cycle to understand the phenomenon that will be investigated in this study. In this section the most important findings and conclusions will be compared of these studies next to what theory predicts. In Section 3 I present a description of the data and its recourses and discuss the empirical specification of the model used in my estimation. In the next section sections, Section 4 and 5, the empirical results will be presented and discussed. In the 6 th and final section I will briefly summarize 1 In Section 2 Literature review an overview of these findings will be presented. 2 The official definition of EU15 countries includes Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Spain, Sweden and the United Kingdom. 2
my major results from previous sections and formulate my conclusions. If possible, I will propose some recommendations for future research. 2. Literature review In theory bank M&As are driven by enhancing their efficiency and profitability, which of course does not only apply for banks. Bank M&As may be set up to exploit their economies of scale and/or scope or to increase market power. This has been widely analysed but there are much more other determinants. In the past a lot of research has been done regarding M&A activity and what drives it. M&A activity in general has puzzled researchers widely and there have been several attempts to explain the wave pattern of M&As. Most of these attempts were based on the US and some focussed on Europe; mostly country or region specific. Besides this, most studies were based on overall M&As, where only a few were specifically focussed on the banking sector. It is important to understand what drives bank M&A activity since it is a crucial industry of our global economy, especially after the latest financial crisis. In one of the most earliest about M&A activity, Beckenstein (1979) concluded that the financial environment is a significant force in explaining the behaviour of mergers during the period beginning at 1949 up to 1970. A more recent study by Resende (2008) looks at M&A waves in the UK and confirms, with alternative empirical analysis, the role for macroeconomic variables in determining M&A activity. This is also in line with Choi and Jeon (2011) who investigated the dynamic impact of the macroeconomic environment on M&A activity in the US. They argued that macroeconomic factors play an important role in determining the trend of aggregate M&A activity in the US. Besides the fact Resende (2008) and Choi and Jeon (2011) used different models, looked at the overall M&A activity of different countries, where they did analyse a larger period. In this way, it is better to capture economic development and analyse its relationship with M&A activity. Resende (2008) looked at a large period sample starting at 1969 up to 2004, where Choi and Jeon (2011) covered the period January 1980 to December 2004. Therefore, this study will also use a larger period sample on specifically bank M&A activity. 2.1 Business cycle It is important to understand the effect of the business cycle on the banking industry in order to analyse the sensibility of this industry. In bad economic times the conditions of the banks can worsen its qualities and profits. Jovanovic and Rousseau (2002) argued that different stages of a business cycle do have a significant impact on overall M&A activity. However, since this study focusses on bank specific M&A deals it is important to know whether the business cycle has the same effect on this sector. As found by U. Albertazzi and L. Gambacorta (2009) bank profits pro-cyclical and are affected by macroeconomic variables. Hence, in this sense it could be possible that a M&A occurs when a bank falls and is being rescued by another bank. However, Becketti (1986) argues that M&A activity typically 3
occurs increases during expansions and decreases during recessions. His statement makes sense due to the fact that during an expansion, where GDP is high, banks are more profitable and willingly to invest. Therefore, expanding would be more attractive. His study is most closely related to my paper besides that he looks at overall M&A activity. Becketti (1986) looks at the amount and value of M&As and only finds a significant positive relation between GNP and the value of M&As on the short run. He takes the variable GNP as a proxy for the business cycle and uses OLS regressions to analyse the relationship between financial activity and mergers and vice versa, where his sample begin at 1960 up to 1980. However, his methodology looks rather simple since he uses no control variables and does not account for any time independent effects for each variable that could possibly be correlated with the regressors. The business cycle can be measured by several factors, but the commonly used variable in this topic is GDP (per capita) (growth). As discussed above, Choi and Jeon (2011) argued that macroeconomic variables have an influence on M&A activity. In their analysis they used a dummy variable of the business cycle (1 is for ascending periods and 0 for descending periods based on realized marginal rate of substitution) and real GDP as one of their main variables, where only real GDP has a positive significant effect. This indicates that when the economy is expanding it allows firms to make M&As decisions due to a more favourable environment. However, real GDP can be difficult to compare the economic conditions among different countries and thus this variable will not be used as business cycle proxy. Rossi and Volpin (2004) look at cross-country determinants of M&As looking at a 12 year period; starting at 1990 to 2002. Their dependent variable is the percentage of traded firms that are targets of successful M&A, using a Tobit model. They used GDP growth as control variable, which had a significant negative effect. This indicates that it is more likely being a target when GDP growth is negative, hence this mean the state economy is going down. However, the model used would not fit this study since we look at the number and value of M&As per country per year, where no censoring is needed. Their findings on GDP growth are in line with what Claessens and van Horen (2007) and Hernando et al. (2009) found on bank specific M&As. Claessens and van Horen (2007) used the change in the number of foreign banks over the period 1995 until 2006 and also included the change in GDP per capita, which had a positive significant effect. Hernando et al. (2009) analysed the probability of a bank being acquired using a Logit model over a period starting at 1997 until 2004. The methodology used in the study of Hernando et al. (2009) does not fit into this study since they look at the probability, where this study looks at the number and value of M&As. All these studies have in common that their period sample is too short to capture any business cycle effect. M&A characteristics such as domestic and cross-border deals may depend on different events to occur, where being a target or an acquirer in a cross-border deal also could matter. Focarelli and Pozzolo (2001) analyse the pattern of cross-border banking M&As relative to non-financial M&As and argued that cross-border deals are rarer in the banking sector than other sectors. They looked at 29 OECD countries between 1990 and 1999, where they used panel data and a Probit model. For the panel data 4
they controlled for common shock and by introducing year and country dummies also for idiosyncratic factors, which is most suitable for the data in this study. Focarelli and Pozzolo (2001) found that total and per capita GDP have a positive effect on M&A activity, where larger and richer countries are more likely to have the size required to expand abroad. In a later study by the same researchers they found that total and per capita GDP have no significant effect anymore on the overall number of cross-border M&As in the whole financial sector and in particular for banks (Focarelli and Pozzolo, 2008). However, Focarelli and Pozzolo (2008) did use another empirical model, namely a negative binomial regression model. However, they did find that financial institutions in countries that have a lower GDP per capita are more likely to be targets of a cross-border. These results concerning the negative effect of GDP per capita are consistent with Lanine and Vander Vennet (2007). Focarelli and Pozzolo (2005) share the same results and added to these finding that they are associated with higher expected economic growth in the future. Buch and DeLong (2004) come to the same conclusion with using a broader time period starting at 1985 up to 2001 and using another empirical method namely a Tobit model. The findings of these papers indicate that banks are more willingly to acquirer another bank when GDP per capita is positive in their own country and is negative in the targets country when economic growth is expected in the future. Caiazza et al. (2012) analysed whether targets in cross-border bank M&As are different from target banks in domestic M&A deals between 1992 and 2006 using a Probit model. They find that crossborder deals are more likely to be driven by diversification motives, but that the probability is on average much lower that a bank is the target of a cross-border M&A than the probability of being the target of a domestic operation. Hence, from this we can conclude there will be more domestic than cross-border bank deals on average as also argued by Focarelli and Pozzolo (2008). Caiazza et al. (2012) found that GDP has a negative significant effect for domestic deals in EU15 countries. In a later study by Caiazza et al. (2014) they found different results between 1992 and 2007 using the same model with different control variables, where a wide variety of countries has been used in the sample. They argued that the sign of the impact of these characteristics on domestic and cross-border deals is the same, but its magnitude often differs. They found that it is more likely that a bank bids when GDP is high. There are several things most of these have in common. First, is that they only look at a form of GDP to interpret what kind of influence the state of the economy has on M&A activity. There is a difference in effect on M&A activity by either using GDP or GDP per capita. This makes sense since a country that has a high GDP but has an intensely large population will end up with a low GDP per capita. A low GDP per capita means that each citizen will get a small amount when a nation s wealth is equally distributed, indicating a not so favourable living standard. Therefore, a high GDP per capita shows a more efficient economy. This measure would be a more reliable business cycle proxy for determining and comparing the economic state of each country on an individual level. However, this variables cannot be the only explanation for the business cycle. Hence, this study does not limit the business cycle proxy to one variable but combines two or three measures that are considered to be a good measure for the 5
business cycle. Besides the most commonly used variable in this field, GDP per capita growth, the unemployment rate and the output gap will be used for the business cycle. 3 Second, most studies have analysed the period before the recent financial crisis. After the financial crisis banks in Europe suffered a lot and had to deal with sharpened regulations, which could have had an effect on their M&A activity. Therefore this period is included in the sample of this research. Third, the period most researchers look at are mostly around 10-15 years. However, as argued before this is too short to capture any business cycle effect. Hence, a large sample will be analysed of approximately 30 years. By extending the period sample, the risk of missing data for some variables is higher and thus some of the control variables cannot be included in this study. 2.2 Control variables The control variables that will be used for the analysis are based on the most commonly used variables in this topic, which showed to have a significant effect on M&A deals. The three variables that will be used as control variables and will be briefly discussed are: inflation, private credit to GDP and concentration of the banking sector per country per year. Inflation is not the most commonly used variable on this topic, where I think it could be interesting to add since it does have influence on several economic variables and is an important macroeconomic variable. Focarelli and Pozzolo (2001) used inflation in their analysis but not as main explanatory variable and show that is does have a negative significant effect on bank M&As in OECD countries. This indicates that less bank M&As will occur when the inflation rate is high. However, when they removed some countries from the sample, due to lack of information on regulatory restrictions, inflation did not have a significant effect anymore, which could be a result of a small sample of only 10 years. The ratio private sector credit to GDP is a widely used measure for banking sector development, which does show mixed results among researchers. This can be due to different investor interests. An important reason can be that an investor wants the opportunity to expand the banking sector of a country, where it may not be possible in the country they are located in since the banking system is already well developed. Therefore, these banks want to enter a market which is less financially developed (Buch and DeLong, 2004; Focarelli and Pozzolo, 2005; Focarelli and Pozzolo, 2008; Correa, 2009; Caiazza et al.,2014). Other investors want to expand to a market where large profit opportunities are more likely and thus want to enter a banking system that is more developed (Caiazza et al., 2014). Since they all look at different countries, factors such as distance, language and culture can have an influence, which will not be considered in this study. Concentration is also a common used measure in determining M&A activity. Acquiring a bank across the border with a highly concentrated market may offer higher profit opportunities, where 3 Based on http://www.imf.org/external/pubs/ft/fandd/2013/09/basics.htm 6
domestic deals with high concentrated market could may be less likely since authorities want to maintain the competition in the banking market. Hernando et al. (2009) found that it is less likely being acquired by a domestic or foreign bank when bank concentration is high, which is contradicting with Caiazza et al. (2014) findings. Caiazza et al. (2014) and Kohler (2009) find that market concentration has a positive and significant effect on both domestic and cross-border targets, where Correa (2009) only argues that cross-border deals are more likely with concentrated banking systems. However, Wheelock and Wilson (2004) and Pasiouras and Zopounidis (2008) find a negative relation. Pasiouras and Gaganis (2009) also tried to analyse this relationship but their findings were mixed. However, they all look at different regions around the world and using different models and periods, which could result in different effects. For this paper other control variables that had a significant effect on M&A activity were considered, such as the interest rate spread and the strength of legal right index. Supervisory authorities and regulation affect a banks incentive to engage in a M&A for either being a target or acquiring bank (Buch and DeLong, 200; Pasiouras et al., 2011; Rossi and Volpin, 2004). Caiazza et al. (2012) show that the strength of legal rights in EU15 countries is a good measure. However, the strength of legal right index is only available for the last 10 years of my sample. In this way, the sample will be reduced and we cannot speak of a business cycle effect regarding the main variables. This is an issue that was found on many potential control variables. 3. Data 3.1 Data description and resources For this study the sample contains all completed bank M&As for EU 15 countries, where the announcement date was within the period January 1, 1985 and December 31, 2014, and reported by Security Data Corporation (SDC) Platinum, a database from Thomson Financial. Because I wish to study the fraction of bank M&As, I selected M&A deals between financial institutions where at least the target or the acquirer is a commercial bank or bank holding company. Furthermore, I collected information from the same database on the transaction value of the deals in dollar values, where the transaction value is not known for all bank M&As. Figure 1 presents a plot of M&A activity per country showing the total numbers and total values within the sample period divided by the number of Monetary Financial Institutions (MFIs). 4 The ratio used in this figure gives a better understanding of the M&A activity in relation to the size of a countries banking industry. 5 However, the graph is somewhat biased since there was no data available on the number of MFIs for the whole sample period. The graphs shows that Belgium, Greece and Spain had twice as much M&A deals in 30-years period time than banks existing. Furthermore, the value of a deal 4 The number of MFIs in the (non-) participating member states in Europe are presented as of July 2016. This refers to the position on the last working day of the month. 5 This ratio will not be used in further regressions in this study. 7
Fig. 1. M&A Activity EU-15 countries Total M&A Activity Total Value M&A Activity Austria Belgium Denmark Finland France Germany Greece Ireland Italy Luxembourg Netherlands Portugal Spain Sweden United Kingdom 0.5 1 1.5 2 M&A activity to total banks ratio 0 500 1,000 1,500 M&A value to total banks ratio ($ million) Data source: SDC Platinum Mergers and Acquisition produced by Thomson Reuters and the European Central Bank. is not known for every single M&A and thus M&A activity based on total value in this figure is not commensurate with total M&A activity in number of deals. 6 In this way, the graph indicates that the M&A deals in the Netherlands where worth a lot more than in other countries. This is a biased conclusions since it could be that another country had more M&A deals with higher values, but where just not known and registered. When looking at the value of the M&As it reduces the initial sample of 5699 M&A deals to 2492 M&A deals on a country level, which is a huge reduction of the sample. Nonetheless, the known deal values will still be used in this study to see if the business cycle has some effect. For this study data on different macroeconomic variables are gathered for each EU-15 country from different databases. GDP per capita growth, unemployment rate and the output gap are used as a measurement for the business cycle taken from World Bank s Global Development Indicators database and the IMF World Economic, respectively. Table 1 present some more detailed descriptive statistics for the business cycle data used in the empirical analysis. In this sample, the mean of GDP per capita growth ranges from 0.77% in Greece to 3.72% in Ireland, where its maximum is 10,07%. It is still remarkable in this sense that the standard deviation for Ireland (3.90%) is a little higher than for Greece (3.6%). However, one explanation can be that the high swings for both countries are due to the fact they both suffered a lot from the financial crisis. The mean of the unemployment rate ranges from 3.27% in Luxembourg to 17.35% in Spain. It is therefore not remarkable that the lowest minimum also belongs to Luxembourg (1.50%) and the highest 6 For more detailed results on this see Table 2. 8
Table 1. Summary statistics for country-specific dependent variables Country GDP per capita growth Unemployment rate Output Gap Mean Min Max Sd Mean Min Max Sd Mean Min Max Sd Austria 1.72-4.05 3.55 1.59 4.42 3.10 5.80 0.81 0.37-2.68 3.72 1.64 Belgium 1.58-3.07 4.39 1.61 8.53 6.20 11.90 1.50-0.20-2.15 2.24 1.33 Denmark 1.34-5.5 5.17 2.18 6.44 3.40 10.70 1.85 0.11-4.06 4.00 2.07 Finland 1.79-8.71 5.94 3.45 9.28 2.50 18.60 4.29-0.25-7.88 6.09 3.48 France 1.33-3.44 4.09 1.49 10.00 7.40 12.70 1.44-0.70-2.70 1.86 1.50 Germany 1.67-5.38 4.35 2.09 7.70 4.90 11.20 1.74-0.49-4.03 2.80 1.72 Greece 0.77-9.00 5.54 3.67 10.58 7.00 27.30 4.65 0.04-8.89 10.65 4.55 Ireland 3.72-6.59 10.07 3.90 11.19 3.70 18.10 5.20-0.77-7.20 9.99 4.21 Italy 1.04-5.91 4.14 2.20 9.77 6.10 12.10 1.72-1.14-4.07 2.69 1.65 Luxembourg 3.01-7.11 9.49 3.70 3.27 1.50 5.80 1.29 0.97-4.06 9.58 3.56 Netherlands 1.76-4.26 4.35 1.87 6.01 2.10 14.20 2.97-2.77-5.19 1.00 1.60 Portugal 1.80-3.64 7.60 2.88 7.26 3.80 16.20 3.14-1.03-6.27 4.28 2.81 Spain 1.71-4.42 5.29 2.32 17.35 8.20 26.10 5.22 0.16-6.60 5.95 3.36 Sweden 1.69-5.99 5.09 2.52 6.15 1.60 10.40 2.70-0.45-6.79 4.56 2.79 United Kingdom 1.96-4.91 5.70 2.08 7.58 4.60 11.50 2.178-0.16-3.18 4.86 2.08 Total 1.79-9.00 10.07 2.68 8.37 1.50 27.30 4.43-0.24-8.9 81.45 4.78 Notes: The values of these country characteristics are presented in percentages. Specific information about the variables and their meaning see Appendix A. Data on GDP per capita growth and Unemployment rate are from the World Bank s Global Development Indicators database; Output Gap is from the IMF World Economic Outlook. maximum to Spain (26.10%), since their means have the same pattern. Something that can also be taken from the table is, that the countries that have been hit the most during the recent crisis have the highest unemployment rate mean and the highest standard deviation. The means of the output gap are mostly negative and ranges from -2.77% in the Netherlands to 0.97% in Luxembourg. The output gap of the Netherlands was mostly negative during these 30 years, which can be due to low investment growth and slow economic recovery. Looking at Greece over a 30 year period its output gap was also mainly negative or slightly positive. Prior to the financial crisis Greece was performing really well and had a peak up to 10.65% but during the crisis the output gap fell drastically, which explains why Greece has the lowest minimum and highest maximum and standard deviation of all countries. The tables also clearly shows a pattern where most countries that have a high unemployment rate also have a negative output gap when their potential GDP exceeds the actual GDP. The countries France, Greece Ireland, Italy and Spain show this pattern, which are also the countries that suffered most from the recent financial crisis. However, GDP per capita growth does not comply in this pattern since the size of each country s population is different. When looking at Ireland and Luxembourg they have a relatively high GDP per capita than all the other EU15 countries, which is due to the fact that they have less residents compared to the other countries. However, the GDP of Ireland has experienced a 9
much higher growth over a 30 year period than Luxembourg did, which is the same for the total population. 7 The same summary statistics has been produced for the control variables (see Appendix B). The first control variable is the inflation rate. The average inflation rate is the lowest in Germany (1.67%) and the highest in Greece (8.31%), where the highest maximum lies at 24.68% in Portugal and the lowest maximum lies at 4.95% in Austria. Here we can see big differences between the EU-15 countries. The second variable is domestic private credit as a proportion GDP, which is a customary variable used in the literature and shows the development of the banking sector. A low ratio indicates that there is less development possible since financial institutions are not able to provide a lot of financial resources to the private sector. The table shows some significant cross-country differences ranging from 54.68% in Belgium and 55.11% in Greece to 124.64% in the UK and 105.33% in Spain. It would seem strange that Belgium has such a low ratio and is even slightly below Greece. However, Belgium always struggled with a large budget deficit, which was only decreased to the social security system in the early 1990s, and their taxes are high compared to the other EU15 countries. These factors could have had an influenced on private sector credit, and therefore could explain the low ratio. Data on all the control variables discussed so far are from World Bank s Global Development Indicators database. Concentration of the banking sector in each country also has an influence on M&A activity. Therefore, the share of the Five Large Banks is used as a measure of concentration. Based on the average of these measures the banking industry of Finland and the Netherlands are highly concentrated, whereas Germany has the lowest concentrated banking industry, which seems logic based on the number and size of banks per country. Data on banks concentrations comes from Bankscope. In this study I will also look at M&A characteristics as mentioned before, and thus data on crossborder and domestic deals are collected from SDC Platinum. I define domestic M&As as those where the nationalities of the acquirer and the target banks are identical; I define cross-border M&As as those where the nationalities differ. 8 My initial data set comprises over 5699 M&A deals on a country level, where 3927 are domestic and 1772 cross-border. Clearly, domestic deals are more frequent, which is in line with facts reported by Focarelli and Pozzolo (2001), also highlighted by S. Caiazza et al. (2012), that domestic deals are more likely to occur than cross-border deals. Table 2 shows a full summary of the deal characteristics per country. When a country participates in a cross-border deal it means two countries have a share in M&A activity, and thus one M&A deal occurs in both countries as M&A activity. By having a more detailed look, it is remarkable to see from the table that Belgium, Ireland, Luxembourg, and the Netherlands participated in more cross-border than domestic deals. A reason could be since they are one of the smallest countries of the sample based on their environs and banking sector, where their banks need to expand across border at some point to gain more profit opportunities. Even 7 Information on GDP and population comes from http://data.worldbank.org/. 8 In this study I do not distinguish between mergers and acquisitions, which is typical for research in this field. 10
Table 2. Summary statistics for M&A-specific variables Country All M&As Cross-border Domestic Austria Total number 163 52 111 Total value 24883.61 17119.89 7763.724 Total number (value based) 48 19 29 Belgium Total number 235 133 102 Total value 111853 46885.18 64967.8 Total number (value based) 103 63 40 Denmark Total number 195 49 146 Total value 30318.49 13425.45 16893.05 Total number (value based) 77 21 56 Finland Total number 180 26 154 Total value 32503.46 19838.78 12664.68 Total number (value based) 54 14 40 France Total number 883 292 591 Total value 255767.8 89749.92 166017.9 Total number (value based) 365 145 220 Germany Total number 837 233 604 Total value 193318.7 80347.44 112971.2 Total number (value based) 205 94 111 Greece Total number 116 22 94 Total value 54395.5 4982.182 49413.31 Total number (value based) 58 12 46 Ireland Total number 74 46 28 Total value 29751.79 12891.17 16860.62 Total number (value based) 41 24 17 Italy Total number 1150 197 953 Total value 344324.3 64836.69 279487.6 Total number (value based) 594 105 489 Luxembourg Total number 114 97 17 Total value 11989.22 10547.55 1441.665 Total number (value based) 37 35 2 Netherlands Total number 219 121 98 Total value 149485.5 32543.92 116941.5 Total number (value based) 68 48 20 11
Table 2. Summary statistics for M&A-specific variables (continued) Country All M&As Cross-border Domestic Portugal Total number 194 77 117 Total value 30119.39 9492.389 20627 Total number (value based) 120 43 77 Spain Total number 643 161 482 Total value 226678.6 44292.45 182386.1 Total number (value based) 357 108 249 Sweden Total number 163 60 103 Total value 30891.94 14453.92 16438.01 Total number (value based) 72 33 39 United Kingdom Total number 533 206 327 Total value 222725.8 59264.13 163461.6 Total number (value based) 293 112 181 Notes: Total number and value (total value is in $ million) of bank M&A-specific variables per country that took place between 1985 and 2014. The data comes from SDC Platinum Mergers and Acquisition database. when looking at the activity where deal values are known this stays the same. This is the contradicting for the fact that the known deal values of Belgium, Ireland, and the Netherlands are higher for domestic than for cross-border deals. As for Finland, it did participated more in domestic (40) than in cross-border deals (14), but the value of the deals that are known are higher for cross-border ($19838.78 million) than for domestic deals ($12664.68 million). Drawing a conclusion based on this is difficult because this study will not look into each bank deal within a country specifically, and thus will not be discussed any further. 3.2 Methodology In paragraph 3.1 I presented descriptive statistics that showed that there appeared to be differences between countries in the amount and total value of bank M&A activity and among the business cycle proxies and control variables. Furthermore, also some differences emerged on M&A characteristics. I begin by using a fixed effect regression to estimate whether M&A activity among countries within the period sample are due to changes in the business cycle, which will not only be measured by the commonly used GDP per capita growth. By indexing countries with j and years with t, the econometric analysis is carried out using the following benchmark model: Y jt = α j + β 1 Business Cycle N 1jt + β 2 Control variables N 1jt + γ 1 Country j + γ 2 Time t + ε jt Y jt is the number or value of M&As in a country j at time t. Business Cycle jt exists of three proxies for the business cycle, namely GDP per capita growth, the unemployment rate and the output gap per country j at time t. Some control variables are added in order to control for other possible 12
macroeconomic effects causing M&A activity and time and country effects. 9 Finally, α j is the unknown intercept for each entity (i.e. n country-specific intercepts) and ε jt is the error term of the fixed effects regression. Since the variables are measured either by taking the average of the whole year or just by taking the variable known at the end of the year, all the variables will be lagged one period. One of the main reason for using lagged variables is to control for reverse causality. Becketti (1986) also used lagged macroeconomic variables in his study to be certain that the estimated effect of the M&A variables isolated the independent influence of M&A activity. As this study follows Becketti (1986) in some way, this will also be used in the analysis. Besides this, all standard errors are robust and clustered at country level since not all variables are considered to be independent of each other. It could take some time to see the effect of the business cycle on M&A activity. Therefore, using variables lagged one extra period (N 2 jt ) could be important for understanding the influence of the economy and how fast the banking sector is reacting to the economic environment. This is also referred to as the delayed effect and will only be performed on the business cycle variables. 4. Business cycle effect I estimate a fixed effects model, where the dependent variable is either based on the number or the value of M&A activity per country per year. In some years not every country had bank s that participated in M&As. This will be taken into account during the regressions since this can also be due to several factors such as macroeconomic changes. In all the regression I control for time and country fixed effects to avoid overspecifying the equation. Besides this, I allow a 10% significance level in all the regressions. In this section I wanted to find out if the business cycle has explanatory power using two or three measures at once. If the business proxies are used separately and being the only main variable, it shows they have a highly significant effect on the number and value of bank M&As for both the immediate and delayed effect. 10 These results can be found in the tables of Appendix C and D for both the immediate and delayed effect on the number and value of M&A activity. 4.1 Number of M&A deals Table 3 presents the results of the estimates of the business cycle on the number of bank M&A deals in the EU-15 countries. In the table there a two different parts, Panel A reports the results based on the immediate effect, and Panel B shows the results based on the delayed effect. Each panel is also divided into four columns, where it takes two business cycle variables in the first three columns and all three variables in the last column. 9 For all variables and specific definitions see Appendix A. 10 The business cycle variables are not highly correlated with each other. By adding them in pairs of two or all at once, it could improve the model. 13
Table 3. M&A activity on country level Business cycle Panel A: Immediate effect Panel B: Delayed effect (1) (2) (3) (4) (5) (6) (7) (8) Output Gap -0.78 *** -0.19-0.16-0.45-0.03 0.07 (-3.549) (-0.810) (-0.702) (-1.758) (-0.135) (0.260) GDP per capita growth -0.24-0.25-0.23-0.65 * -0.63 * -0.64 * (-1.036) (-0.912) (-0.848) (-1.912) (-2.115) (-1.965) Unemployment rate 1.02 *** 0.94 *** 0.94 *** 0.76 *** 0.80 *** 0.80 *** Control variables (4.558) (3.233) (3.234) (4.455) (3.535) (3.887) Inflation rate -0.43 0.10 0.02 0.07-0.35-0.04-0.28-0.02 (-0.827) (0.310) (0.064) (0.214) (-0.575) (-0.079) (-0.583) (-0.035) Private Credit to GDP 0.08 * 0.07 * 0.08 * 0.07 * 0.09 ** 0.08 ** 0.10 ** 0.08 ** (2.119) (1.877) (2.082) (1.871) (2.336) (2.176) (2.302) (2.146) Share Five Large Banks -0.35 ** -0.43 *** -0.43 *** -0.44 *** -0.33 ** -0.39 ** -0.36 ** -0.39 ** (-2.550) (-3.094) (-3.023) (-3.102) (-2.329) (-2.902) (-2.601) (-2.894) Year FE Yes Yes Yes Yes Yes Yes Yes Yes Country FE Yes Yes Yes Yes Yes Yes Yes Yes Observations 227 227 227 227 227 227 227 227 R-square 0.375 0.420 0.419 0.421 0.364 0.397 0.381 0.398 Adjusted R-square 0.311 0.361 0.360 0.359 0.299 0.336 0.318 0.333 RMSE 5.659 5.449 5.454 5.457 5.706 5.556 5.629 5.569 Notes: This table shows the influence of all business cycle proxies and other macroeconomic control variables on M&A activity on a country level. The dependent variable is the number of M&A activity per year per country. Panel A shows the immediate effect of the business cycle on M&A activity and Panel B shows the delayed effect of the business cycle on M&A activity (lagged one period). For further variable information see Table 1 and 2 and Appendix A and B. All the regressions are adjusted for robust standard errors. *** Significance level of 1 % or less ** Significance level between 1% and 5% * Significance level between 5% and 10% Column (1) Panel A shows that an M&A announcement is more likely to happen when the actual output is less than a countries economy could yield at full capacity and with a negative GDP per capita growth. Where the negative sign of GDP per capita growth is not as expected based on most studies, it makes sense in comparison with the negative sign of the output gap. However, the coefficient is statistically insignificant. In column (2) the sign of GDP per capita is negative but still not significant, where the unemployment rate is positive and highly statistically significant. The unemployment rate remains to have a significant effect in column (3), where the output gap does not anymore. When adding all three business cycle measures at once, only the unemployment rate still has an immediate positive effect on bank M&As. Hence, once added more business cycle proxies the unemployment rate seems to have more influence, where GDP per capita growth does not. Indicating that a higher unemployment rate on a very short notice can result in M&A activity. A higher unemployment rate normally means an underperforming economy, and thus could increase the urge to expand to get better profit opportunities. 14
In the columns of Panel B the same patterns can somewhat be seen as in Panel A. However, when looking at the delayed effect GDP per capita growth appears to have a statistically significant effect on bank M&As. The output gap has does not appear to have any influence on M&A activity when considering the delayed effect. GDP per capita growth was expected to have a positive and significant effect based on previous literature, which in Table 3 does only appear to have a negative influence on M&As. Correa (2009) did found a negative GDP per capita growth but used a very small sample of 10 years. However, the unemployment rate is highly statistically significant in both immediate and delayed effect indicating that a higher unemployment rate, and thus an underperforming economy increases the likelihood of a bank M&A to occur around approximately a year. The overall pattern that can be seen in Table 3 is that bank M&A activity in EU-15 countries is more likely to happen in a period of recession. 4.2 Value of M&A deals Table 4 shows the results of the estimates of the business cycle on the value of bank M&A deals in the EU-15 countries. Also this table is divided into two different parts, Panel A presents the results based on the immediate effect, and Panel B reports the results based on the delayed effect, where each panel takes two business cycle proxies in the first three columns and all three proxies in the last column. Column (1) of Panel A shows that a in country with negative output gap and GDP per growth it is more likely to increase the value of a bank M&A. However, the output gap does not appear to have a statistically significant effect throughout the whole first Panel. In all the columns (1)-(4) GDP per capita growth and the unemployment rate show they keep having an immediate significant effect within a year on the transaction value of M&A activity. Hence, taken from the table this indicates that a negative GDP per capita growth and a high unemployment rate, and thus a downturn in the economy will increase the value of M&A activity. In Panel B all three the business cycle proxies appear to have a highly significant relation with the value of bank M&As regarding a delayed effect. It is however remarkable that the output gap has a positive sign, where it in the previous regressions always had a negative sign. The coefficients GDP per capita growth and the unemployment rate indicate a the value of a M&A is more likely increase in a recession. However, the output gap has a positive effect meaning that when a countries GDP still yields above potential GDP it raises the transaction value. It seems contradicting with the other two business cycle proxies and especially with GDP per capita growth. Nonetheless, GDP per capita growth is based on the number of residents in each country and gives a different view than only GDP, where they always have opposite signs as argued before in the literature review. When comparing Panel A and B somewhat the same pattern can be see regarding GDP per capita growth and the unemployment rate. Therefore, we can say that no matter the swings in a short period these business cycle proxies remain to have a significant influence. Only the output gap had a negative insignificant sign when it comes to the immediate effect, but turns into a positive significant sign when looking at the delayed effect, which in a short time is contradicting. 15
Table 4. Value of M&A activity on country level Panel A: Immediate effect Panel B: Delayed effect (1) (2) (3) (4) (5) (6) (7) (8) Business cycle Output Gap -669.04-223.98-8.47 33.89 766.89 ** 889.67 *** (-1.559) (-0.638) (-0.028) (0.076) (2.506) (3.080) GDP per capita growth -1103.79 * -1098.47 ** -1096.80 ** -1254.15 ** -1167.14 ** -1251.05 *** (-2.034) (-2.488) (-2.439) (-2.830) (-2.807) (-3.270) Unemployment rate 1010.78 *** 1012.29 ** 1006.64 ** 807.87 ** 1236.34 *** 1234.08 *** (3.038) (2.680) (2.735) (2.700) (3.849) (4.061) Control variables Inflation rate -1319.66-648.61-885.99-650.45-1128.24-701.21-1100.58-465.74 (-1.735) (-0.957) (-1.312) (-0.926) (-1.274) (-0.960) (-1.507) (-0.597) Private Credit to GDP -33.11-45.03-19.95-45.03-19.64-20.43 3.59-23.79 (-0.966) (-1.205) (-0.524) (-1.199) (-0.575) (-0.537) (0.096) (-0.608) Share Five Large Banks -616.49 ** -696.49 ** -665.95 ** -696.63 ** -559.54 * -636.92 ** -578.68 ** -633.30 ** (-2.233) (-2.634) (-2.433) (-2.598) (-2.050) (-2.452) (-2.168) (-2.471) Year FE Yes Yes Yes Yes Yes Yes Yes Yes Country FE Yes Yes Yes Yes Yes Yes Yes Yes Observations 192 192 192 192 192 192 192 192 R-square 0.212 0.226 0.213 0.226 0.201 0.215 0.207 0.222 Adjusted R-square 0.114 0.131 0.116 0.125 0.102 0.118 0.109 0.121 RMSE 12113.038 12000.601 12099.707 12036.047 12192.464 12088.278 12150.117 12065.059 Notes: This table shows the influence of all the business cycle proxies and other macroeconomic control variables on M&A activity on a country level. The dependent variable is the total value of M&A activity per year per country ($ millions). Panel A shows the immediate effect of the business cycle on M&A activity and Panel B shows the delayed effect of the business cycle on M&A activity (lagged one period). For further variable information see Table 1 and 2 and Appendix A and B. All the regressions are adjusted for robust standard errors. *** Significance level of 1 % or less ** Significance level between 1% and 5% * Significance level between 5% and 10% The results show us that there is a relationship between the business cycle and the value of a bank M&A, where it does matter if you look at different time effects. However, it is still difficult to formulate hard conclusions since not all values for every M&A are known. 5. M&A deal characteristics For estimating the influence of the business cycle on deal characteristic, the same fixed effects model and assumptions are used as in Section 4. However, the dependent variable is now based on the number of cross-border or domestic M&A deals per country per year, where still a distinction will be made between immediate and delayed effect. The only difference made is that the value of M&As will not be considered in the regressions since it is hard to provide an appropriate conclusion. 16