Mergers and acquisitions in the financial sector, does the crisis have an influence?

(1)

UNIVERSITEIT VAN AMSTERDAM

Mergers and acquisitions

in the financial sector,

does the crisis have an

influence?

Finance

Ivo de Meij 5978106 6/29/2015

This paper tries to look if the deal value influence the stock price reactions during a merger or acquisition of an financial institution during the financial crisis. An event study with cross sectional analyses showed no significant influence. Therefore concluding that the abnormal returns behave the same as before the crisis.

(2)

2

1. Introduction

Organizations worldwide have the goal to create value for their shareholders. And with this growth is natural phenomenon. But this growth can progress naturally, but it is also possible to grow by participating in mergers and acquisitions activity. This also holds for financial institutions but the has the recent crisis had an influence?

The question that this paper will try to answer is: Does the deal value influence the

stock price reactions during a merger or acquisition of an financial institution during the financial crisis?

Eichengreen & Rourke (2009)argue that latest crisis was worse than the great depression of the nineteen twenties. Banks were not too big to fail anymore, looking at the Leman Brothers debacle. During the crisis financial institutions seemed to be hit the hardest. So have the

expectations on banks and their performance changed and can this been seen in the stock price reaction during a merger or acquisition?

With this paper I wish to shed more light on merger and acquisitions during the financial crisis with respect to financial institutions. Although the research on this subject of M&A is substantial, it is mostly does not look at external influences as a crisis. They look at more quantifiable data to look for abnormal returns during the announcement of M&A deal. Furthermore the majority of the studies use the CAPM, but literature from Fama & French, (1992) shows that CAPM is not the best model to be used for estimation share return. therefore this paper will use the Fama French three factor model. This should lead to better estimates and in term lead to better conclusion.

This paper will be build up in 6 sections: first the introduction. secondly a literary review will be presented with an overview of previous academic papers on the subject of merger and acquisitions. In the third part the methodology is explained (how the empirical design is build up). In the fourth part of this paper the data will be discussed, where it originates from and what are the basic statistics. In the fifth part will show the results and with these results the null hypotheses will be check. finishing with section 6 which is the conclusion.

(4)

4

2. Literature review

The amount of literature available on mergers and acquisitions is substantial and therefore I will try to create a good overview on this subject. This overview will start with an overview literature on M&A with an emphasis on literature related to M&A of financial institutions, after that an explanation is given on why this paper uses the Fama French three factor model and finishing with a brief overview on performance of financial institutions during the crisis.

The general view on the subject of merger and acquisitions is that the target firm on announcement has a positive significant short term return. On the other hand the acquiring company has either a negative reaction or a small plus in the period surrounding the

announcement date. This assumption holds for M&As that take place in the financial

industry (Becher, 2000; Crouzille, Lepetit, & Bautista, 2008; de Young, Evanoff, & Molyneux, 2009; Delong, 2001; Tourani Rad & Van Beek, 1999). So on the short run only shareholders of the target really benefit.

2.1 Mergers & Acquisitions

Observing the merger activities over time made scientist conclude that there are six major waves during which there was a significantly higher number of mergers.

These periods are: beginning of the 20th century, the 1920, sixties, eighties, nineties and the beginning of the 21st century. An interesting fact that all of these merger waves ended

abruptly thanks to a collapse of the stock markets. (Alexandridis, Mavrovitis, & Travlos, 2011; Gugler, Mueller, Yurtoglu, & Zulehner, 2003; Martynova & Renneboog, 2008).

Companies partake in merger and acquisition actives for several reasons. The vast majority of these mergers reasons are: economies of scale, economies of scope and size. Economies of scale are generally regarded as the most important. M&As create advantages in scale because companies can lower the cost that is needed to operate for example they can cut down in costs of buildings, there is no longer a need to have two corporate

headquarters. furthermore companies can also reduce the costs in personnel costs, by combining divisions within the organization. The paper of Becher (2000) gives evidence that a merger that is driven by synergy generate a higher premium compared to empire building. Although later in this literary review I will show that empire building is an important driver none the less. But this is not based on shareholder returns but on survivability. Houston & Ryngaert 1994 also find that banks with a higher overlap and therefore a higher potential of

(5)

5

synergy cost saving tended to perform better then where this is not the case. This is

observation is also backed by the research of Delong (2001), where the bank mergers where divided in four different types. The dividing line was based on activity, the geographic focus and diversification (Delong, 2001, p. 250). Delong showed that the market does 'recognize' the type merger that took place and that the focusing mergers (with higher synergy

potential performed better than non focusing mergers.

Although the studies show that mergers with focusing nature tend to perform better on announcement compared to diversifying merger announcements. So why do financial institutions increase their scope with mergers when it is not in the best interest of their shareholders? The paper of Milbourn, Boot, & Thakor, (1999) gives an explanation Milbourn argues that running a larger financial institution requires more skill than running a smaller firm, so this will improve the status of the CEO. And this can could lead to overconfident management. On the other hand Milbourn tries to give an explanation with regards of future uncertainty with respect to skill. Anticipating future changes in regulatory rulings and

anticipating on this in advance to get the required skills in house beforehand. These finding are also shared in the literary study of Martynova & Renneboog (2008).

Size can be an important driver for mergers and acquisitions, certainly for financial institutions where banks can be driven to create an organization that is too big to fail. Which in a way cements their position in the market. Because failure of such a entity would create a lot of problems within the financial sector and the economy in general This is backed by Molyneux, Schaeck, & Zhou (2014) who researched bank mergers in Europe between 1997 and 2007 and if the increased size increased the chance of getting a safety net subsidy. And concluded that these larger firms indeed had a higher chance of being saved. The literary overview from de Young et al. (2009) come to the conclusion, that too big to fail was an imported driver in the most recent M&A wave in the financial industry. However

developments during the recent crisis with the collapse of Lehman Brothers showed that this is not an insurance in itself. Seeing that Lehman brothers was one of the oldest companies on the New York Stock Exchange and turned out to be the largest firm to ever file for bankruptcy in the United States (Johnson & Mamun, 2012). Furthermore acquisitions of larger banks that took place in later part of the nineteen eighties banks in the United States showed that there is no apparent combined positive return (Houston & Ryngaert, 1994).

(6)

6

Furthermore takeovers could be hostile by nature, although the amount of scientific papers is substantial on this specific subject. The real number of these hostile takeovers is relativity small in 1986 it were just 40 of the 3300 takeovers. Although these mergers have an higher return exceeding 30 percent (Jensen, 1988). In recent years there are not as many of these hostile takeovers as in the eighties. As (Martynova & Renneboog, 2008) show that this partly is the reason because of the introduction of an anti takeover law in US. Therefore making it more difficult for firms to just take over each other. Another reason is the increase in cooperate governance tools available to managers. The study of Schwert & Schwert (2000) looked at mergers that took place in the United State between 1975 and 1996 and showed that the deals described as hostile did not react different economically then non hostile takeovers.

However the above being said the premiums that are being observer by multiple studies are not just down on the types of mergers. There are more things that are important during these announcement. For example the amount of cash paid for the target. The acquirer has many possibilities to finance an acquisition just cash, just stock or a

combination of the two. And this way of payment tend to have a impact on which premium is observed on the market. Previous study of Travlos (1987) show that bidding organizations have different stock premiums on announcement. Companies that financed their merger with stock tended to have a negative return during the event itself. However if cash was used to finance the merger the abnormal returns where normal. During the sixth merger wave M&A's tend to be financed more with cash then equity, which led to be more in line with the neoclassical definitions of mergers (Alexandridis et al., 2011).

Research also shows that the premium reacts to other financial numbers like the return on equity (ROE). ROE basically is the net income divided by the total amount of equity, this measure is a useful measure in measuring the profitability of companies that are active in the same industry.

Furthermore multiple studies show that locations are of influence on premiums as well. These locations are defined differently throughout the academic literature. There are several studies on US financial institutions that show that the state in which a financial institution has its company headquarters and the headquarters of the target influence the premium paid. Delong (2001) shows that premiums paid for financial institutions in the same state tend to generate a higher abnormal return on announcement then M&A's that take

(7)

7

place between financial institutions in different states. This finding is further backed by the research of Houston & Ryngaert (1994)

That being said the research of (Asimakopoulos & Athanasoglou, 2013) show that cross border M&A activity in the European financial market (mergers that take place between financial institutions of different countries) does not create value for the acquirer on

announcement. This would suggest that the shareholders prefer 'local' mergers more. This is in line with the research that has been done on M&A's in the United States. One of the explanations of this phenomena is that the papers in question look at markets with a relative high protection of the shareholder. The research of Bris & Cabolis ( 2008) show that markets with a high level of shareholders interest protection create higher M&A premiums and that in term will lead to a lower level of abnormal returns on the announcement date.

2.2 Fama French Three factor model

As stated in the introduction this paper will use the Fama French Three factor model, this is different from the majority of the papers that are written on the subject of mergers and acquisitions. In order to show why this model is preferable to the other models, this part will focus on a brief history of the CAPM and the Fama French three factor model.

The capital asset pricing model was first introduced in 1965 which was build upon previous work of Tobin. And was based around two assumptions namely the market assumption that assumed that every investor has the possibility to invest any part of their capital in a risk free asset. This can be seen as a simple savings account on which you gain a percentage of interest or for example country bonds (Lintner, 1965). The later of these is generally used as the measure of the risk free rate in the literature as well as in the study books. The second pillar on which the CAPM is based is the investors assumption. This

assumes that every investor has its own idea of how a stock performs and that if the investor has two shares or portfolios with the same return he will chose the one with less risk.

However the model was not without its problems, as Reinganum (1981) show that high beta portfolios are not significantly different to low beta ones. Which makes these beta's less important to traders. Fama and French saw the problem too that and look at ways of improving the CAPM. They did this by adding two factor small minus big and high minus low in their paper Fama and French showed that just using the two factors SMB and HML it did not help to explain the variance in a portfolio. However if it was used together with market

(8)

8

return it did provide them better fit of the model. Fama and French showed that CAPM only had an R2 of 69% and that the lowest fitting three factor model had an R2 of 83%, but overall the model showed an R2 of above 90%. This is a gain of at least 14% but for majority proved to be an improvement of over 25% (Fama & French, 1992, 1993).

2.3 Crisis

The most recent crisis showed the vulnerability of the economic system, but academic papers also found that some banks did perform better than others. Beltratti & Stulz (2012) found that banks that are financed with more short term assets where significantly more fragile compared to banks that where financed with more long term funds. Which makes sense, because short term funding is much more intertwined with the market and therefore much more vulnerable to market changes. And thus where more exposed to an abrupt market change like the crisis. Beltratti & Stulz (2012) furthermore conclude that banks with management that was more geared towards the shareholders performed worse that banks that did not do this. These findings are also backed by (Berger & Bouwman, 2013) who looks at the capital and performance of banks during a crisis. They found that capital generally helps large banks during a crisis. Furthermore Crouzille et al. ( 2008) looked at the Asian financial crisis of the nineties with respect to mergers and acquisitions on financial institutions they found that the crisis did have a negative effect on the abnormal returns. During this crisis the market did not react different towards target and bidder.

3. Methodology

3.1 Hypotheses

In order to test if main question raised in this paper indeed has an influence on the abnormal returns of a company. The main regression equation that will be used in this paper is as follows:

equation 1

(9)

9

With this regression formula I will introduce two hypotheses in order to test the main research question. These hypotheses will be formulated as the H0, the H1 will then be the

opposite which would be equal to the sign ≠. 1. H0: Dummy crisis = 0

With in testing if the recent financial crisis has an influence on the abnormal returns that can be observed surrounding the time of the announcement. By using this notation the influence can either be negative or positive. With respect to research of (Crouzille et al., 2008) I will suspect a negative reaction.

2. H0: Sizedeal*dummy crisis = 0

In essence the most import of the hypotheses test because this is the closed to answering the main research question. Arguing that size as well as the crisis has an influence on the observed abnormal returns.

3.2 event study

The purpose of this study is to conclude if the M&A deals create shareholder value. In order to test if this is indeed the case, the standard method of an event study is used. Event studies can deliver useful results if one's to use one of the classical rules in economics. That the market and it's participants behave rational. Which in turn should lead to a fast reaction to the stock price given a certain event would occur, in this case a merger announcement. (Asimakopoulos & Athanasoglou, 2013; MacKinlay, 1997). Furthermore the paper of (Brown & Warner, 1985) shows that an event study can be carried out with daily stock data

compared to the monthly data what was being used in older literature. Having said the majority of papers written in the last fifteen years all use daily stock data. The estimation window which will be used for this study will start 300 days (calendar days) and will stop 60 (calendar) days before the event. This is in line with majority of the academic papers available. (Asimakopoulos & Athanasoglou, 2013; Brown & Warner, 1985; Delong, 2001) The reason for this length is that it will give a sufficient number of observations prior to the event (generally around 220). And because with the inclusion of the crisis period having a longer time span prior to the event can lead to a biased estimator and therefore a less accurate model

(10)

10

The majority of the literature on M&A uses the CAPM model or the market model that being said more recent studies have showed that the use the Fama-French three factor model should lead to better estimated results as is suggested in the literary review.

equation 2

Where ret is the expected return of the stock for the given time period, (Km-Rf) the market

portfolio return minus the risk free interest rate. SMB stand for Small Minus Big and HML High Minus Low these variables are used to calculate the historic excess returns (Fama & French, 1993)

equation 3

Abnormal returns (AR) are generated by subtracting the expected returns from the actual returns for the given event window observations.

equation 4

Here the cumulative abnormal returns are calculated.

These CAR's are calculated for the duration of the event window. Because in relevant literature there is not one standard for the time period in question, I will use multiple event windows ranging from t -20 as a start date till t=0. And for the end t=0 till t=5.

3.3 analysis of abnormal returns

While CAR's will show the actual accumulated difference in stock price returns compared to estimated values, they cannot give information on the underlying causes of these anomalies. For instance there is a possibility that the recent crisis has an influence on the returns

generated during the event window. In order to test this paper will use a cross section analysis. This method is not a new thing in literature and has been done a number of studies on mergers and acquisitions for example Asimakopoulos & Athanasoglou (2013), Becher (2000) and Delong (2001).

(11)

11

4. Data

This section will be build up in two sections, the first part is used to explain what data is used and where it originates from. In the second part the emphasis will lie on the methodology

4.1 Data collection.

4.1.1 Merger and acquisition data

The first part of the data is acquired using the Zephyr database for the announcement dates of the mergers and acquisitions. These events are order chronological starting from1st of January 2004 till 31st December 2013. In this period I searched for Zephyr classification: banking, Insurance & financial services and NACE rev.2 (primary codes) 6419 other monetary intermediation. With a deal type acquisition, merger on all available stock markets.

With the added information that the acquiring firm and target firm are both based in the United States. Furthermore the deal has to be completed. This resulted in total of 355 mergers and acquisitions. However with the introduction of restrictions with respect to ISIN company codes and completeness of deal information on how the deals cash/equity is divided made the size of the available M&A's drop to 201 observations. But within this data there was still a problem of overlapping data with respect to the estimation window data. In order to try and not create a bias in the estimators for the abnormal returns all events of the same company that has multiple mergers in a same time frame have been deleted. This decreased the number of events to 186.

The following two tables will show the number of M&As over time and the average deal size of the complete number of observations (355). These graphs show that the number of deals as well as the value of the deals plummeted during the start of the crisis. And that they remain under the average over the majority of the remaining time.

(12)

12

figure 1

figure 2

4.1.2 Stock price information

Stock price information is gathered via the DataStream software package. This data will contain the daily share price returns for the designated time period 300 days before event till five days after (calendar days).

4.1.3 Fama French three factor information

In order to test for abnormal returns this paper will use the Fama French three factor model as described earlier. This information is gathered from the website form Ken French

(http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html) , which contains the daily values of SMB (Small Minus Big) and HML (High Minus Low), risk free rate

0 10 20 30 40 50 60 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Number of M&As

Number of M&A's average M&A's

0 500000000 1000000000 1500000000 2000000000 2500000000 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Deal Value

(13)

13

(1-month T-Bill) and the market return. These daily values range from July 1 1926 till April 30 2015.

4.2 control variables

This section will contain the various control variables and describe them in sort and in depth. In order to give a better understanding of the values and their importance for the model which will be tested in section 5.

Table 1

 lndealsize, the market can react differently to the size of the takeover as stated in the literary review larger deals have different impact on shareholder than smaller deals. The natural logarithm is used in order to make the OLS estimations better this is also a normal procedure when taking other papers in account.

 lnsize, natural logarithm of total assets is widely used as an control variable for M&A premiums in the academic literature.

 Dummy_Crisis The crisis had a major impact on the economy as a whole and there should not be left out of this regression as stated in the literary review that every M&A wave ended in the 'collapse' of the stock market the date used in this study is

Variable name: Description

lndealsize The Natural logarithm of the deal size

lnsize Natural logarithm of total assets

Dummy_crisis Dummy for crisis 1 if deal takes place after August 2007

ROE Return on equity

EPS Earnings per share

Cash_ Percentage of Cash paid in de deal compared to

shares

Large deal Dummy: 1 if deal is in 4th quartile

Employees Natural logarithm of the number of employees

that work for specific company

Crisisdeal Dummy_crisis * lndealsize

(14)

14

August 2007 so every observation that occurs in August 2007 or after will be given the value of 1. That being said (Molyneux et al., 2014) put the financial crisis in-between 2008 and 2009. However during time period of the second half of 2007 till the present day (2015) there was no just a single crisis. The most obvious was the subprime mortgage crisis. But there were also two souvenir debt crises that took place round that same time period. And an argument can be made if the underlying problems that created these crises have been dealt with properly therefore in this paper I will use the crisis dummy till the end of the data set which is available to me.

 ROE, return on equity can be seen as an important control variable with respect to financial institutions as stated in the research of (Asimakopoulos & Athanasoglou, 2013).

 EPS or earning per share can been seen as one of the most import variables in determining the share price. so in that respect it should be helpful in the regression on explaining the abnormal returns. These earnings per share are the values that were presented the in the latest quarterly statement before the merger

announcement date.

 Cash_ will give the percentage of cash used for the deal. previous literature shows that the amount of cash paid makes a significant influence on the price premium (Martynova & Renneboog, 2008). These values are calculated via the following formula:

equation 5

 Large deal, Larger deals tend to have a different expose because shareholders expect different things from larger mergers then smaller ones. Because of the implication that are inherent on larger deals.

 Employees, the size of a firm can be measured in different ways you can look at het the total assets but a larger firm in general has a higher number of employed staff.

 Crisisdeal is a dummy variable combining the dummy variable for crisis times the lndeal size. As shown in the figures the value of the deal has decreased by a huge

(15)

15

amount and therefore I will include this variable in order to test if the deal size in the crisis is significantly different from the deals that took place before the crisis.

 Crisissize is a dummy variable that combines the dummy for crisis together with the control variable lnsize. This is to check if there is a difference in the size of the acquirer during the crisis.

4.3 Basic statistics

The next two tables will show the basic statics of the variables that are used in this paper.

Table 2

Summery statisics

Variable name obs mean median std dev

cash_ 136 0.33 0.3 0.338 ROE 136 10.071 9.96 7.403 EPS 135 3.316 1.38 15.230 employees 133 6.797 6.531 1.601 dummy_crisis 137 0.416 0 0.495 lndealsize 137 18.330 18.150 1.577 lnsize 137 15.155 14.850 1.669 crisisdealsize 137 6.144 0 7.346 crisissize 137 7.456 0 8.912 largedeal 137 0.255 0 0.438

table 2 shows the basic statistic of the control variables like the mean and median. analyzing the results shows that the variables Mean and median are all close to each other apart from the dummy variables. This in turn indicates that the variables are near symmetrical which makes these variables useable in a regression. However looking at the standard deviation shows that the variables ROE and EPS have large standard errors indicate that these

variables have fat tails. The other high standard deviations are those of interaction variables that either are zero or the original value and these original values have relativity small standard deviations. So these high standard deviations are to be expected.

(16)

16 Table 3 Correlation matrix Number of observations 131 largede a crisissiz e crisisdea l lnsize lndealsiz e dummy_

c emp EPS ROE

largedeal 1 crisissize -0.1402 1 crisisdeal -0.1220 0.9961 1 lnsize 0.5607 -0.1367 -0.1545 1 Lndealsiz 0.7388 -0.1775 -0.1563 0.7379 1 dummy_c -0.1600 0.9949 0.9950 -0.1825 -0.2134 1 emp 0.5626 -0.1976 -0.2144 0.9728 0.7178 -0.2432 1 EPS -0.0517 -0.0946 -0.0956 0.0278 0.0352 -0.0963 0.0425 1 ROE 0.2012 -0.3723 -0.3567 0.2815 0.3341 -0.3873 0.3452 0.1458 1

Table 3 shows the correlation table of the all the variables that are going to be used in the cross sectional regression. The rule of thumb with respect to correlation is that if the value is higher than +/-0.70 gives an indication of high correlation between the variables in the equation. Looking at the table shows that the variable employees with lnsize (0.9728) and lndealsize (0.7178) have a high correlation. So it is best to avoid combining these variables the regressions. That said lnsize and employees do basically measure the same thing the size of the organization although with different forms of measurement. The same can be said of the correlation between dummy_crisis and crisissize (0.9949) and crisisdealsize (0.9950). These correlation can be explained in the same way as was done with the summary statistics of these variable, because they are interaction terms and there for show a high correlation with the crisis dummy.

5. Results

This section will discuss the results from the regressions. The first part will describe the CAR's the second part will be divided up in two parts. Part a will discuss the main findings with respect to the cross section analysis while the second part will discuss the robustness check.

5.1 Cumulative abnormal returns

The results show in table 4 are the cumulative abnormal returns, these results are in

conformation with the general literature on mergers and acquisitions on the acquirer site. It shows that indeed the acquirers stock return is negative round the time of an

(17)

17

announcement. The standard deviations calculate in the next two tables are calculated from the car returns and not the more advanced method suggested by McKinley (McKinley variance). Furthermore looking at the minimum and maximum returns that are observed during the event windows show the extremes are quite far apart. This can lead to the regression result that are shown in section 5.2 have bias to accommodate these outliers in the regression.

Table 4

Cumulative abnormal returns of acquirers

Acquirers (N = 137) in % number of deals event window car

mean car median std

dev. positive car negative car Min Max

t -20 +5 -1.322 -2.073 9.355 54 83 -26.754 23.437

t -5 +5 -1.551 -1.967 6.955 52 85 -25.368 31.664

t -3 +3 -1.243 -1.521 6.401 50 87 -23.49 31.619

t -1 +1 -0.803 -0.85 5.223 56 81 -19.393 27.629

t 0 -0.233 -1.967 4.457 60 77 -13.38 30.845

Looking at the figures supplied in appendix one, which show the scatter plots of the CAR's of the various event windows. This gives a better indication on how the CAR's are spaced and it indeed shows that there are outliers. To try and correct for these outliers the cross sectional analyses uses CAR's that are winsorized. Winsorizing is replacing the extreme outliers with a value closer to the median. This is an academic approved way of dealing with outliers and was first introduced by (Tukey & Mclaughlin Donald H, 1963). The top one percent of values are replaced with the value at the 99th percentile and the bottom one percent is replaced with the value at the 1st percentile. Going further with winsorizing for example five percent would induce to many artificial results and thus cloud the results which is academically not preferable.

(18)

18

Table 5

Cumulative abnormal returns of acquirers (winsorized)

Acquirers (N = 137) in % number of deals event window car mean car median std

dev. positive car

negative

car Min Max

t -20 +5 -1.268 -2.073 9.130 54 83 -26.7541 23.437

t -5 +5 -1.614 -1.967 6.331 52 85 -17.636 15.324

t -3 +3 -1.336 -1.521 5.904 50 87 -21.601 16.931

t -1 +1 -0.813 -0.850 4.856 56 81 -14.099 20.964

t 0 -0.285 -0.145 3.914 60 77 -10.146 20.347

Table 5 shows the winsorized data and here it can been seen that the all minimum and maximum have indeed been lowered and as a result also lowered the standard deviation of all event windows. Thanks to the use of this technique the number of observations positive or negative stay the same.

5.2 Cross sectional analyses

5.2.1 Main results

In this section the main findings will be discussed per time window starting at the event window minus twenty, plus five. all results are rounded after three decimals apart from the R2. All the regressions will us robust standard errors. Robust standard errors are in general a bit larger the standard errors in the normal regression model but robust standard errors help with potential heteroskedasticity problems within the data. Furthermore its usage is widely accepted in the economic literature.

(19)

19

Table 6

OLS Regression results for window t -20 +5

depended variable CAR (winsorized)

Name Reg 1 Reg 2 Reg 3 Reg 4 Reg 5 Reg 6 Constant 4.452 -0.885 -1.232 -0.671 -0.663 8.656 (6.595) (1.026) (0.801) (1.664) (0.880) (7.101) lnsize -0.377 -0.586 (0.418) (0.424) cash_ -0.923 -0.979 (2.195) (2.444) EPS 0.040* 0.034* (0.011) (0.012) dummy_crisis -1.435 0.079 (0.878) (24.239) crisisdealsize -0.081 -0.083 (0.089) (1.330) R2 0.0048 0.0012 0.0045 0.0060 0.0063 0.0177 obs 137 136 135 137 137 134

note: *, **, *** denotes 1%, 5%, 10% significance respectively

the first value is the regression coefficient

the values in () denote the standard error

this regression is done with robust standard errors

The first table shows the results of the regression on the cumulative abnormal returns for the time window t= -20 +5. For this 26 day time window a regression is done on all individual variables and the whole model. The variable cash_ should have a positive reaction to

abnormal returns if we look to the literature (Travlos, 1987). However this variable has an negative influence (-0.979). So cash paid mergers in this window have a negative reaction on the abnormal returns generated. Furthermore the size of the organization has an negative influence on the price as well larger firms will have a lower abnormal return. The interaction variable crisissize shows a negative return (-0.083)which indicates that larger deals during the crisis generated lower abnormal returns compared to deals that took place before the crisis.

Analyzing the results shows that the dummy variable of the crisis in the individual regression has an negative bias to the results. However changing to the complete model the coefficient changes for negative to positive so with extra explanatory variables the crisis no longer has an negative influence on the price (-1.435) but has a rather small plus (0.079). This is not in line with the results form Crouzille et al., (2008) which showed a negative reaction. That

(20)

20

being said apart from the earning per share (which is significant in both their individual regression as in the combined model) None of the variables show a significant relation between the cumulative abnormal returns and the variables used in the regression.

Therefore the null hypotheses show earlier on in section 3 cannot be rejected. Which mean there is that the variables dummy_crisis and dealcrisis do not significantly differ from zero. And thus their influence can be seen as limited.

Table 7

OLS Regression results for window t -5 +5 depended variable CAR (winsorized)

Name Reg 1 Reg 2 Reg 3 Reg 4 Reg 5 Reg 6 Constant 3.299 -1.510*** -1.668* -1.307*** -1.281*** 5.048 (4.935) (0.786) (0.558) (0.667) (0.666) (5.660) lnsize -0.324 -0.411 (0.321) (0.349) cash_ -0.100 -0.309 (1.472) (1.600) EPS 0.042* 0.040* (0.007) (0.008) dummy_crisis -0.735 1.175 (1.125) (14.112) crisisdealsize -0.045 -0.106 (0.063) (0.792) R2 0.0073 0.0000 0.0102 0.0033 0.0039 0.0227 obs 137 136 135 137 137 134

Moving to table 7 shows the regression results for the time window t +5 -5 in this shorter time span the variables behave in relatively the same way as show in table 6. However the result that are generated show that the constant factors (β0) of the individual regressions

are more significant compared to table 6. Apart from the first regression on lnsize and the regression on the complete model. In the complete model the variables behave in the same fashion as in the previous time window. However the variables in which are important for testing the hypotheses remain not significant and therefore again it is not possible to reject the two null hypotheses. Again showing that there is no significant relation between the

(21)

21

crisis and the observed abnormal returns generated by the announcement of a merger or acquisition.

Table 8

Name Reg 1 Reg 2 Reg 3 Reg 4 Reg 5 Reg 6 Constant 6.466 -1.264 -1.380** -1.319** -1.284** 6.965 (4.964) (0.757) (0.520) (0.574) (0.574) (5.527) lnsize -0.515 -0.540 (0.328) (0.354) cash_ 0.058 -0.142 (1.421) (1.421) EPS 0.032* 0.032 (0.008) (0.009) dummy_crisis -0.042 0.018 (1.077) (12.638) crisisdealsize -0.007 -0.003 (0.061) (0.715) R2 0.0212 0.0000 0.0066 0.0000 0.0001 0.0296 obs 137 136 135 137 137 134

Continuing with table 8 with the time window -3 +3 again shows that the individual regressions show that the constant is significant and that only earning per share have a statistically significantly influence on the observed cumulative abnormal returns. The

variables still maintain their positive/negative influence on the cumulative abnormal returns, but as can been seen the standard errors decreased with the move to the shorter time span. The main variables that are important for this study namely crisisdealsize and the dummy for crisis maintain insignificant and therefore the hypotheses given in section 3 can't be reject. so there is no statistically significant influence on the observed cumulative abnormal returns.

(22)

22

Table 9

Name Reg 1 Reg 2 Reg 3 Reg 4 Reg 5 Reg 6 Constant 8.748** -0.962 -0.908** -1.043** -0.953** 6.167 (3.562) (0.636) (0.420) (0.454) (0.454) (4.026) lnsize -0.630* -0.442** (0.231) (0.251) cash_ 0.747 0.183 (1.059) (1.021) EPS 0.035* 0.039* (0.013) (0.015) dummy_crisis 0.553 12.412 (0.893) (9.505) crisisdealsize 0.0188 -0.657 (0.052) (0.538) R2 0.0470 0.0028 0.0124 0.0032 0.0012 0.0771 obs 137 136 135 137 137 134

In table 9 an even shorter event is introduced namely -1 +1. And here things change cash now has an positive influence on the abnormal returns (0.183), which is in line with previous academic papers. However these values are not significant. Another change in the complete model is that no longer earnings per share is the only statistically significant variable. Lnsize now has an significant influence on the CAR's. The rest of the model keeps behaving in the same manner as it did with the previous event windows. The crisis still has an positive effect on the observed CAR's and larger deals during the crisis still generate less returns. But as with the other regressions these results stay insignificant and there for the null hypotheses cannot be reject. Which in term means that in the smaller event window the crisis and crisisdealsize do not statically differ from zero. So their influence can be seen as statistically insignificant.

(23)

23

Table 10

OLS Regression results for window t 0 depended variable CAR (winsorized)

Name Reg 1 Reg 2 Reg 3 Reg 4 Reg 5 Reg 6 Constant 10.056* -0.305 -0.325 -0.422 -0.310 8.252* (2.899) (0.543) (0.338) (0.352) (0.354) (3.053) lnsize -0.682* -0.543* (0.187) (0.189) cash_ 0.279 -0.952 (0.821) (0.808) EPS 0.014 0.014 (0.011) (0.012) dummy_crisis 0.328 20.541* (0. 727) (7.281) crisisdealsize 0.003 -1.140* (0.041) (0.402) R2 0.0846 0.0006 0.0030 0.0017 0.0001 0.1455 obs 137 136 135 137 137 134

Looking at the last event window of t 0 which are the abnormal returns generated on the day of the announcement show a different result compared to the other event windows in that the two variables dummy_crisis and crisisdealsize now show to have an significant influence on the results given by this regression. Although the crisis dummy does have a big positive effect on the abnormal returns. Which is the complete opposite of what the work of Crouzille et al. (2008) suggest. Furthermore the variable crisisdealsize has is negative which means that the larger the size of the merger or acquisitions the lower the abnormal returns are. One possible explanation is that the deal values during this crisis period are lower than the period before. And smaller deals are met with different expectations from the

shareholder compared to the larger deals. Continuing with the rest of the model shows the constant and lnsize have a significant influence on the regression results as well. However the variable that has been significant in all previous regressions namely EPS is not

statistically significant anymore. And the percentage of cash paid has no longer a positive relation on the abnormal returns. With this in mind both null hypothesis given in section 3 can be reject . With a significance level of α 1% the crisis does have an significant influence

(24)

24

on the abnormal returns during the event window of t 0. And crisisdealsize also is significant at α 1% so this variable statistically significantly differs from zero for the time window t 0.

Summarizing the result show that for all the time windows apart from the last one (t 0) the hypothesis H0: Dummy crisis = 0 and H0: dealsize*dummy crisis = 0 are reject and therefore

their influence to the model and the cumulative abnormal returns to statistically insignificant.

5.2.2 Robustness check

For the robustness checks I will use the data from the regression from the time window t +1 -1. The reason for this is that this window has the best fitting R2 of all the models apart from t =0 but this is a check on abnormal returns from the day that announcement was made. This cannot be called an event window but just an event.

Non robust standard errors

This section of the robustness checks will justify the use of the robust standard errors in the main model by comparing the regression done with robust standard errors (shown in table 11 regression one) with standard, standard errors (shown in table 11 regression two). Table 11

depended variable

CAR

(winsorized)

Name Regression 1 Regression 2

Constant 6.167 6.167 (4.026) (4.236) lnsize -0.442** -0.442** (0.251) (0.264) cash_ 0.183 0.183 (1.021) (1.291) EPS 0.039* 0.039 (0.015) (0.026) dummy_crisis 12.412 12.412 (9.505) (8.980) crisisdealsize -0.657 -0.657 (0.538) (0.498) R2 0.0771 0.0771 obs 134 134

note: *, **, *** denotes 1%, 5%, 10% significance respectively the values in () denote the standard error

(25)

25

As table 11 shows the estimators stay the same with the use of robust standard errors but the errors themselves change. As a result the EPS went from insignificant to significant at α 1% (0.015) with robust standard errors to not being significant with normal standard errors. This indicates that there is a problem of heteroskedasticity in the data. Doing a Breusch-Pagan test for

heteroskedasticity with a null hypotheses that the variance is constant is rejected with an alpha of 10% thus justifying the use of robust standard errors for the main regression.

Separate regression on the crisis.

Table 12 will show two regressions done on different time periods the first will displays the regression of before the crisis, while the second regression will show the crisis regression.

Table 12

depended variable

CAR

(winsorized)

Constant 4.692 20.43** (4.290) (7.903) lnsize -0.421 -1.493* (0.271) (0.519) cash_ 1.784*** -1.831 (1.002) (1.722) EPS 0.031* 1.274* (0 .005) (0 .217) R2 0.0874 0.3284 obs 79 55

Looking at the output of table 12 shows that the crisis has an effect on the returns. Where regression one shows that the pre crisis regression is much more conform the existing literature, with cash having a positive and significant influence on the abnormal returns. However the regression on the crisis time period shows that cash paid has an negative influence on the cumulative abnormal returns. Furthermore the size has an higher negative reaction (-1.493 compared to -0.421 pre crisis) that is significant at α 1% which suggest that larger financial institutions have lower abnormal returns that before the crisis.

This shows that the crisis has an influence on the abnormal returns during mergers and acquisitions. Although these results can't be compared in that sense because it are two

(26)

26

individual regressions with their own residuals and standard errors. Therefore it is better to do one regression on the complete dataset and include a dummy for the crisis.

Using employees as company size

The next table will show the regression results of using a different variable for the size of the organization regression one will show the regression with lnsize and the second will use employees as the size variable

Table 13

depended variable

CAR

(winsorized)

Constant 6.167 2.700 (4.026) (2.142) lnsize -0.442** (0.251) employees -0.547** (0.275) cash_ 0.183 -0.034 (1.021) (1.027) EPS 0.039* 0.039 (0.015) (0.015) dummy_crisis 12.412 13.445 (9.505) (9.902) crisisdealsize -0.657 -0.725 (0.538) (0.563) R2 0.0771 0.0805 obs 134 130

In table 13 the results for using an alternative size variable show that both are relatively close together value wise and standard error wise both are significant at α 5%. With the use of employees the fit of the model is even a bit better compared to using the natural

logarithm of total assets. However the use of employees as a measure of size is not common in the academic literature on this subject and therefore it is more logical to use the more generally accepted measure for size namely the natural logarithm of total assets.

(27)

27

6. Conclusion and Discussion

This paper tried to show if the deal value influence the stock price reactions during a merger or acquisition of an financial institution during the financial crisis. It did so by comparing the cumulative abnormal returns that are generated during the event of a merger and

acquisition announcement from before and during the crisis. This was done by using a cross section analysis of the CAR's for five different time windows. The observed CAR's where on average negative for the acquiring company which is in line with all the related literature. With each time window the two hypothesis's H0: Dummy crisis = 0 and H0: dealsize*dummy

crisis = 0 were tested. And resulted in that for all the event windows apart from the smallest t 0 the crisis had a positive effect on the abnormal returns which was not in line with the previous literature, but it did not significantly differ from zero and thus the null hypothesis that the crisis has no influence could not be reject. Only on the day of the announcement (t 0) the null hypotheses could be rejected and thus showing that the crisis has and an

statistically significant influence. The hypotheses that the size of the deal has an influence on the premiums were rejected by al time windows apart from the last window (t 0)

Therefore I come to the conclusion that the crisis did not have influence on the

observed cumulative abnormal returns during an announcement of a merger or acquisition as does the deal size. Which means that during the crisis the premiums generate are same.

Although that the prices of shares in question might have dropped the percentage premium on such events have stayed unchanged.

One possible explanation is given in the paper of Bris & Cabolis, (2008) that markets like the US have a high level of share holder interest protection. Which will lead to high premiums and thus lowering the cumulative abnormal returns.

Although the robustness check on the different time spans showed that during the crisis the observed CAR's have a different reaction to the pre crisis model, its influence on the whole model showed insignificant. Reasons for this could be the number of acquisitions during the crisis period where smaller and the overall price of the acquisitions was lower as well. Because shareholders expectations of smaller deals is different to that of larger deals.

One of the problems with this paper was unavailability of data on the side of the target therefore it was not possible to get enough data for the target this problem was mainly down to DataStream and therefore it was not possible to look at the CAR's of the target or the combined results which could have led to different conclusions than the one

(28)

28

that came out of this research. Furthermore the data showed that crisis premiums have different reaction but the inability of proving this could also lie in use of the wrong control variables. But again for more advanced control variable you need the data of the target. Future research could improve on this by testing the influence of the target on CAR's of the acquirer. It could also be possible to use a different estimation window which build up its history on a longer time before the event. although that can also lead to biased results because the crisis had a impact on the stock market.

7. References

Alexandridis, G., Mavrovitis, C. F., & Travlos, N. G. (2011). How have M&As changed? Evidence from the sixth merger wave, (June 2015), 37–41. http://doi.org/10.1080/1351847X.2011.628401

Asimakopoulos, I., & Athanasoglou, P. P. (2013). Revisiting the merger and acquisition performance of European banks. International Review of Financial Analysis, 29, 237–249.

http://doi.org/10.1016/j.irfa.2012.08.010

Becher, D. a. (2000). The valuation effects of bank mergers. Journal of Corporate Finance, 6(2), 189– 214. http://doi.org/10.1016/S0929-1199(00)00013-4

Beltratti, A., & Stulz, R. M. (2012). The credit crisis around the globe: Why did some banks perform better? Journal of Financial Economics, 105(1), 1–17.

http://doi.org/10.1016/j.jfineco.2011.12.005

Berger, A. N., & Bouwman, C. H. S. (2013). How does capital affect bank performance during financial crises?? Journal of Financial Economics, 109(1), 146–176.

http://doi.org/10.1016/j.jfineco.2013.02.008

Bris, A., & Cabolis, C. (2008). The value of investor protection: Firm evidence from cross-border mergers. Review of Financial Studies, 21(2), 605–648. http://doi.org/10.1093/rfs/hhm089

Brown, S. J., & Warner, J. B. (1985). Using daily stock returns. Journal of Financial Economics, 14(1), 3–31. http://doi.org/10.1016/0304-405X(85)90042-X

Crouzille, C., Lepetit, L., & Bautista, C. (2008). How Did the Asian Stock Markets React To Bank Mergers After the 1997 Financial Crisis? Pacific Economic Review, 13(2), 171–182. http://doi.org/10.1111/j.1468-0106.2008.00395.x

De Young, R., Evanoff, D. D., & Molyneux, P. (2009). Mergers and acquisitions of financial institutions: A review of the post-2000 literature. Journal of Financial Services Research, 36(2), 87–110. http://doi.org/10.1007/s10693-009-0066-7

Delong, G. L. (2001). Stockholder gains from focusing versus diversifying bank mergers. Journal of Financial Economics, 59(2), 221–252. http://doi.org/10.1016/S0304-405X(00)00086-6

(29)

29

Fama, E. F., & French, K. R. (1992). The Cross-Section of Expected Stock Returns. Journal of Finance, 47, 427–465. http://doi.org/10.2307/2329112

Fama, E. F., & French, K. R. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics.

Gugler, K., Mueller, D. C., Yurtoglu, B. B., & Zulehner, C. (2003). The effects of mergers: An international comparison. International Journal of Industrial Organization, 21(5), 625–653. http://doi.org/10.1016/S0167-7187(02)00107-8

Houston, J. F., & Ryngaert, M. D. (1994). The overall gains from large bank mergers. Journal of Banking & Finance, 18(6), 1155–1176. http://doi.org/10.1016/0378-4266(94)00065-4

Jensen, M. C. (1988). Takeovers: Their Causes and Consequences. Journal of Economic Perspectives, 2(1), 21–48. http://doi.org/10.1257/jep.2.1.21

Johnson, M. A., & Mamun, A. (2012). The failure of Lehman Brothers and its impact on other financial institutions. Applied Financial Economics, 22(5), 375–385.

http://doi.org/10.1080/09603107.2011.613762

Lintner, J. (1965). The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets. The Review of Economics and Statistics, 47(1), 13–37. http://doi.org/10.2307/1924119

MacKinlay, A. C. (1997). Event Studies in Economics and Finance. Journal of Economic Literature, 35(1), pp. 13–39.

Martynova, M., & Renneboog, L. (2008). A century of corporate takeovers: What have we learned and where do we stand? Journal of Banking and Finance, 32(10), 2148–2177.

http://doi.org/10.1016/j.jbankfin.2007.12.038

Milbourn, T. T., Boot, A. W. a., & Thakor, A. V. (1999). Megamergers and expanded scope: Theories of bank size and activity diversity. Journal of Banking & Finance, 23(2-4), 195–214.

http://doi.org/10.1016/S0378-4266(98)00079-X

Molyneux, P., Schaeck, K., & Zhou, T. M. (2014). “Too systemically important to fail” in banking - Evidence from bank mergers and acquisitions. Journal of International Money and Finance, 49, 258–282. http://doi.org/10.1016/j.jimonfin.2014.03.006

Reinganum, M. R. (1981). A New Empirical Perspective on the CAPM.pdf. Journal of Financial and Quantitative Analysis, 16(4), 439–462. http://doi.org/10.2307/2330365

Schwert, G. W., & Schwert, G. W. (2000). Hostility in Takeovers: In the Eyes of the Beholder? Journal of Finance, 55(6), 2599 – 2640.

http://doi.org/http://www.blackwellpublishing.com/journal.asp?ref=0022-1082

Tourani Rad, A., & Van Beek, L. (1999). Market valuation of European bank mergers. European Management Journal, 17(5), 532–540. http://doi.org/10.1016/S0263-2373(99)00042-0 Travlos, N. G. (1987). Corporate Takeover Bids, Methods of Payment, and Bidding Firms’ Stock

(30)

30

Tukey, J. W., & Mclaughlin Donald H. (1963). Less Vulnerable Confidence and Significance Procedures for Location Based on a Single Sample: Trimming/Winsorization 1, 25(3), pp. 331–352.

Appendix

Appendix 1

the following graphs are of scatter plots with the car on y axis and the event dates on x axis graph 1 is for time window (-20 +5), graph 2 (-5 +5), graph 3 (-3 +3), graph 4 (-1 +1) and graph 5 (0)

appendix 1 a -4 0 -2 0 0 20 cu mu la ti ve _ a b n o rm a l_ re tu rn 1/1/2002 1/1/2004 1/1/2006 1/1/2008 1/1/2010 1/1/2012 1/1/2014 event_date

(31)

31 appendix 1 b appendix 1 c -4 0 -2 0 0 20 40 cu mu la ti ve _ a b n o rm a l_ re tu rn 1/1/2002 1/1/2004 1/1/2006 1/1/2008 1/1/2010 1/1/2012 1/1/2014 event_date -2 0 -1 0 0 10 20 30 cu mu la ti ve _ a b n o rm a l_ re tu rn 1/1/2002 1/1/2004 1/1/2006 1/1/2008 1/1/2010 1/1/2012 1/1/2014 event_date

(32)

32 appendix 1 d appendix 1 e -2 0 -1 0 0 10 20 30 cu mu la ti ve _ a b n o rm a l_ re tu rn 1/1/2002 1/1/2004 1/1/2006 1/1/2008 1/1/2010 1/1/2012 1/1/2014 event_date -1 0 0 10 20 30 a b n o rm a l_ re tu rn 1/1/2002 1/1/2004 1/1/2006 1/1/2008 1/1/2010 1/1/2012 1/1/2014 event_date

Mergers and acquisitions in the financial sector, does the crisis have an influence?