Ownership structure and market efficiency in the context of hazard rates in merger deals

(1)

Universiteit Van Amsterdam Amsterdam Business School MSc Business Economics, Finance

Master Thesis

To obtain the academic degree

Master of Science Business Economics, Finance

Ownership Structure and Market Efficiency in the

Context of Hazard Rates in Merger Deals

Name: Funda E. Toptaner

Address: Jahnstraße 18, 65185 Wiesbaden Submitted to: Florian S. Peters, Ph.D.

(2)

Statement of Originality

This document is written by Student Funda E. Toptaner who declares to take full responsibility for the content of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the content.

(3)

Abstract

This thesis finds that the passage of time after merger announcement contains information and that financial structure, deal attitude, acquirer leverage, acquirer leverage in combination with financial structure, merger arbitrage spread, and termination fee have an effect on the hazard rate of completion after merger announcement. The main finding is that despite inefficiencies occurring and semi-parametric and graphical analysis hinting at the fact that ownership structure has an effect on inefficiencies in the estimation of hazard rates in pricing, the hypothesis that non-institutional investors increase the failure to price hazard rates of completion into the market correctly could not be shown.

(4)

List of Abbreviations

OLS Ordinary-Least-Squares

SEC Securities and Exchange Commssion

STATA Data Analysis and Statistical Software for Professionals U.S. United States

(6)

List of Figures and Tables

Figure 1 Frequency of Withdrawn and Completed Deals... 22

Figure 2 Hazard Rate of Completion of All Deals (Smoothed) ... 29

Figure 3 Hazard Rate of Completion of All Deals (Minor Smoothing) ... 29

Figure 4 Graphical Analysis of Heterogeneity Issues ... 33

Figure 5 Actual versus Market Hazard Rate of Completion ... 44

Figure 6 Frequency of Industry Category in Dataset Based on Industry SIC Codes ... 62

Table 1 Summary Statistics: Deal Characteristics ... 23

Table 2 Summary Statistics: Company Characteristics ... 26

Table 3 Log-Rank Test of Survival Function ... 40

Table 4 OLS Regression of Ownership Structure on the Difference Between the Actual and Market Hazard Rates ... 48

Table 5 Cox-Model Regression of Ownership Structure on The Actual Hazard Rate ... 54

Table 6 OLS Regression of Ownership Structure on Market Hazard Rates ... 56

Table 7 OLS Regression of Ownership Structure on the Difference Between Actual and Market Hazard Rates Using Different Independent Variables ... 59

Table 8 Industry SIC Codes Description ... 62

Table 9 OLS Regression of Ownership Structure on the Difference Between Actual and Market Hazard Rates with Different Datasets ... 63

(7)

1 Introduction

Ownership of shares of non-institutional investors fell from 15.1% in 2010 to 13.8% in 2014, while the value of direct and indirect holdings of corporate equities increased (pg. 3, Federal Reserve, 2014). The effect of investors on the overall economy is not negligible. Explaining actions of different types of investors is increasingly important due to the decrease in barriers to entry and the decrease in transaction costs, also due to technological developments. In his book from 1994, Roe already pointed out that policies shape the American economy and the American political system systematically discourages large investors. Roe (1994) explains that some company types, such as banks, insurance companies, mutual and pension funds are increasingly aiming to have less and less of an influence on corporations, concluding that this reduces efficiency of the economic system.

The discussion of market efficiency is an ongoing debate; making Fama (1970) one of the most important papers of traditional finance theory. Fama (1970) introduced the concept that a share’s price equals its fundamental value and that the market is efficient in the sense that it incorporates all available information in the pricing of the securities. In contrast, many examples exist of why this is actually not the case. Questions such as why investors participate in the market in the first place if no money is earnable, or why above-average returns are achievable, stand against the market efficiency hypothesis. The efficient market hypothesis, however, does not argue that each individual person is inefficient or involved in “wrong” investment decisions. It further does not argue that price irregularities occur or even persist in the short-run.

Fama (1998) defends the position of efficient market hypothesis by finding shortcomings in the model set-ups of other researchers claiming inefficiencies of the market. These models oftentimes arise from the broader research area “behavioural finance”. Research in this area is mainly concerned with explaining irrationality observed in the market, usually by using psychological or sociological phenomena as

(8)

support. Generally, most irrationalities are observed as under- and overreaction. Disagreement persists between market efficiency supporters and their opponents of whether under- and overreaction offset each other (, whereas latter hold the position that they offset each other.)

Most major price changes analyzed follow from news. News, for example concerning management, earnings announcements, changes in strategy or mergers and acquisitions. The definition of news already includes the fact that it contains random and unpredictable information, making this absurdity even more interesting. In the analysis of merger news, statements about managers strategically announcing changes or plans have been made by researchers (for example, Sun and Louis (2015) or Ahern and Sosyura (2014)). The strategy behind this evidently supports the importance of news for the financial markets. Or reversed, information revealed by analyst recommendations is even said to influence the outcomes of a proposed merger (Becher, Cohn and Juergens, 2014).

The outcome of merger announcement has not been highlighted thoroughly in research. Many topics explain returns before and after merger announcements, or find relationships between company and deal properties and the earnings from a merger deal or similar, however, facts about the probability that the merger actually completes or withdraws has not been the focus of analysis of a wide range of researchers. Likewise, whereas affects of news have been studied extensively, the information contained in not receiving any news is very interesting as well for multiple reasons. Giglio and Shue (2014) introduced this type of study and their analysis builds the foundation of this thesis in the sense that their models are used. With models, the hazard rate model for the completion of merger events, as well as their general conception that analyzing the time after mergers is deceived as the information that the simple passage of time unfolds, is meant. More precisely, many economists who believe in market efficiency view the market as successful devices for reflecting new

(9)

information (Malkiel, 2003). The question then arises, what happens when no new information is revealed. This must also be incorporated in the price of the security, meaning that the price must reflect the probability that a merger completes conditional on the probability that it has not completed or withdrawn up to that date. If this probability changes over time, then time actually contains information on merger completion or withdrawal. Giglio and Shue (2014) show that their model agrees with the behavioural model of underreaction, in this case, to the passage of time. Underreaction in this context refers to agents “underreacting”/”not reacting (enough)” to information contained in the passage of time (Giglio and Shue, 2014).

A deeper look into the analysis of behavioural economics, but also other areas, reveals that not all investors act alike. For example, differences between smart money and feedback traders (for example, De Long, Shleifer, Summers and Waldmann, 1990; Goetzmann and Massa, 1999…), informed versus uninformed investors (Carter and Manaster, 1990; Glosten and Milgrom, 1985…), and between institutional and individual investors (weekend effect of Lakonishok and Maberly, 1990; attention and news by Barber and Odean, 2008; herding and feedback trading by Nofsinger and Sias, 1999; returns by Kaniel, Saar and Titman, 2008…) exist.

First of all, the different types of investors, representing the ownership structure of a company, can influence the actions of management (a focused area of governance research topics). For example, institutional investors might want to encourage short-term earnings goals, or in general influence management or achieve specific company goals. Secondly, ownership structure can have a direct effect on the companies’ performance. For example provide liquidity in the markets or in general determining the financing structure of a company. Thirdly and in this context most important, the stock performance on the market plays a big role for the company but also the investors in the securities market. As described, the market underreacts to the probability of merger completion (Giglio and Shue, 2014). This novel idea that

(10)

individual investors compared to institutional investors underreact to the passage of time in merger deals will be examined.

Individual investors are said to underperform benchmarks (e.g. Coval, Hirshleifer and Shumway, 2005), make wrong investment decisions (buy and sell the wrong investments – “disposition effect” – e.g Frazzini, 2006), have limited attention (e.g. Hirshleifer and Teoh, 2003), and many other inefficient decisions. But also simply: individual investors may invest when they have a personal relationship to the product or service being sold, or a family tie to the company. Individual investors probably are not as efficient in their research also based on the lack of access to all data and thus might make less rational investments. Also, individuals might have a tendency to show interest in shares which are discussed in the news, hoping to be able to read between the lines.

Answering this research question does not only contribute to classical finance theory and behavioural finance literature, but also reveals useful information for many different areas: believing in the efficient market theories may lead to misinterpretations of major stock events and bubbles. Furthermore, mergers are major and often one of the most significant events in a company’s lifetime and the value revelation of the mergers reveals important information for investors, companies and researchers. Moreover, the promotion of different types of investors is done by policymakers; thus, more information on specific investor groups would be informative for them, i.e. stimulate major institutional changes where they need to be made. Apart from the beliefs described by Roe (1994), the OECD (Celik & Isaksson, 2014) is committed to understanding the character and degree of ownership engagement as regulations in OECD countries exist, e.g. some institutions in the US are to some extent obligated to vote all shares under management. The UK has a Stewardship Code, which essentially encourages shareholder voting. On the other hand, in Sweden for example, a pension fund is explicitly prohibited from voting their shares in any Swedish companies, by law.

(11)

Similarly, in Turkey some mutual funds are not allowed to participate in the governance of the investee companies. Further, companies can introduce “voting caps” to limit the number of voting shares; this is allowed in Belgium, Denmark, France, Norway, Spain, the UK and the US (OECD (Celik & Isaksson, 2014)). In some countries, it is also common practice to address disclosure of voting policies in the company’s rules and codes. Thus, shareholder engagement is not internationally agreed upon, suggesting that arguments for different types of engagements from different institutions exist.

Specifically examining merger failure allows to separate value implication of the takeover itself (Malmendier, Opp & Saidi, 2015). The arbitrage opportunity found in underreaction is relevant for investors and companies. But all in all, understanding the markets and the media’s role in passing information are generally acknowledged as interesting for policy makers, media, society, investors, and companies. As Fama (1970) also points out, the capital market prices should reveal information about the production-investment possibilities of firms, and investment decisions of investors. With more information on the source of underreaction, managers and investors could incorporate this information in evaluating their opportunities. Perhaps, information which is found in this analysis may help prevent a merger withdrawal (or completion if that is the goal).

Further, the information retrieved can give hints about possible desired effects. Grullon, Kanatas, and Weston (2004) show that companies with higher advertising expenditures also have a larger number of individual and institutional investors. All in all, market inefficiencies and the role of investors in the merger context reveals important information on many levels.

The remainder of this thesis is structured as follows. Section 2 first summarizes the literature on the efficient market hypothesis and its challenges leading to second, behavioural finance research focusing on underreaction and news, and then third the smallest level introducing literature comparing institutional to non-institutional

(12)

investors. Section 3 provides explanations on the set-up of the data, the meanings and formulas behind the necessary variables and ultimately the analysis of the variables. Section 4 first provides the derivation of the actual hazard rate of completion, then examines heterogeneity issues in the dataset, before comparing the hazard rate seen in the dataset and the market observed hazard rate, and then ultimately analyzing the effect of non-institutional investors on the difference between the actual and market hazard rates. Section 5 then explores possible shortcomings of the analysis: perhaps the effect of ownership structure on the actual hazard rate and the market hazard rate calculated separately reveal errors of the main analysis, or changing the continuous independent variables into dummy variables as in the section analyzing heterogeneity reveals different findings, or lastly the separation of the dataset according to industries might disclose instable findings.

(13)

2 Literature Review

This thesis contributes to several areas of research. The starting point is the efficient market hypothesis in the field of traditional finance theory. An introduction of this research area and the opposing theories in the area of behavioral finance build the foundation of this thesis. A third area of research which is not categorized to one specific field of research but can touch many is the differentiation between institutional and non-institutional investors and their characteristics. This thesis moves along these three areas.

2.1 Shift From Traditional Finance Theory

Before the 21st century, most researchers in finance believed that markets are efficient

(Fama, 1970). The “efficiency of markets” was first introduced by Eugene Fama and describes that information is incorporated into the securities market prices as soon as it is revealed (Fama, 1970). This means that the security’s price equals its fundamental value. The supporters of those theories, neoclassical economists, argue that behavioural biases are irrelevant in the market because so-called “rational investors” would correct for this irrationality by exploiting arbitrage opportunities (Friedman, 1953). In contrast, now researchers developed rational explanations of why abnormalities are not exploited. Some of the main explanations are transaction costs (Friedman, 1953), search costs and tax effects (Basu, 1977), noise trader risk (Shleifer & Vishny, 1997), and fundamental risk (Merton, 1987). Starting the 21st century, however, an even greater

change in thought arose: economists emphasized psychological and behavioural elements in stock prices.

It is generally acknowledged that anomalies in the market give rise to possible market inefficiencies, which can be defined as markets allowing traders to earn above-average returns without having to accept an above-above-average risk (Malkiel, 2003). In other words this statement implies the predictability of stock prices. Research on

(14)

possible predictability of markets can be split into two categories: predictability through “fundamental valuation metrics” and predictability on the basis of past information.

Multiple patterns based on fundamental valuation metrics have been examined and have not yet been discarded: the size effect examined by Fama and French (1993) is the effect that, over the long-run, smaller firm stocks generate larger returns than larger firm stocks. Another metric is the price per earnings multiple or price to book ratio: Basu (1977) was among the first to show that value stocks, meaning stocks with low price per earnings ratio, generate higher returns than growth stocks. This was later also supported by Fama and French (1993). Mehra and Prescott (1988) introduced the equity risk premium puzzle, i.e. the irrationally high US historical equity risk premiums. Campbell and Shiller (1988) found that depending on the investment horizon, a significant portion of future returns could be predicted by dividend yields. Fama and Schwert (1977) and Campbell (1987) relate short-term interest rates to future stock returns.

Research from the category “predictability of future prices based on the passed” generally analyzes past events and predicts repetition in the future. For example, “long-term return anomalies” were first introduced by DeBondt and Thaler (1985) who show that investors give too much weight to past information of firms by displaying that past winners tend to be future loser and vice-versa. DeBondt and Thaler (1985) link this investment behaviour of waves of optimism followed by waves of pessimism to the behavioural decision theory of Kahneman & Tversky (1982) which among others states that investors are systematically overconfident in their ability to process information and forecast future stock prices or future earnings of companies.

Rationalizations of these anomalies are not decided on. Especially other arguments, which are not regarded as exploitable but still hint at examples of market inefficiency, are found when examining the Market Crash of October 1987 or the Internet Bubble in the late 1990s. Explanations for these and other events must be

(15)

related to more psychological than rational valuations of stock prices. For example, Malkiel (2003) explains how researchers argue that some investors are demonstrably less rational than others, and thus pricing irregularities and even predictable patterns in stock returns exist and can persist over short periods.

Overall, Fama (1998) claims that market inefficiency has not been proven, and explains that anomalies are chance results, the empirical analysis are not robust, and that efficient markets allow individual and irrational reactions, however, they will be split equally. For example, post-event continuation of pre-event abnormal return is about as frequent as post-even reversal and the number of underreaction and overreaction should be approximately the same. Malkiel (2003) points out that this view still exists among a group of “market efficiency believers”. Nevertheless, no proficient empirical or theoretical evidence is found, especially on the last statement made from Fama’s (1998) research. Interestingly, Schwert (2001) finds that some strategies which have been pointed out in the finance literature, such as the “January Effect” have diminished after publication. Accordingly, Grossman and Stiglitz (1980) further argue that there would be no incentives for investors to uncover information that is reflected in the markets if the markets immediately captured the value.

2.2 Behavioral Finance: Underreaction and News

The research area “behavioural finance” mainly emerged in the 1980s to explore the anomalies mentioned in the previous section. Especially the anomalies which are not evidently explainable by traditional finance theories or are not agreed upon in the financial markets are of great interest. Whereas traditional finance theory as explained above is based on rational markets where the price of a security equals its actual “fundamental value”, behavioural finance seeks to explain irrationality driven from market participants, oftentimes supported by psychological theory or a sociological perspective. Barberis and Thaler (2003) give two definitions for rationality: updating beliefs with new information as described by Bayes’ law and “making choices that are

(16)

normatively acceptable, in the sense that they are consistent with Savage’s notion of Subjective Expected Utility” (p. 2).

The ongoing debate about specific “market inefficiencies” include, bounded rationality, representative bias, contagion effect, conservatism, availability bias, judgement bias, overconfidence, self-attribution, and optimism and wishful thinking. Bounded rationality was introduced by Simon (1957) who argues that market participants are restricted in their capacities and hence are only reduced capable of rationality. Barberis, Shleifer and Vishny (1998) point out a first cognitive psychological explanation as the representative bias of Kahneman and Tversky (1982): investors giving too much weight to recent patterns and too little to the data basics with properties of the population. The contagion effect is described by Shiller (2000) as investors buying stocks with increasing prices as this caused attention – this is consistent with the short-run momentum and leads to irrational exuberance. Conservatism is based on Kahneman and Tversky (1974) and explains that when people form estimates they start off with a possible estimate and adjust away from it. According to Barberis & Thaler (2003), conservatism describes that people overweigh data because some data is representative of specific underlying models. Conservatism outlined by Edwards (1968) is described as updates being only slowly included in the model. Kahneman and Tversky (1974) also introduce the availability bias: when judging, people use their memory which usually can only invoke information partially. Daniel, Hirshleifer and Subrahmanyam (1998) describe a model with informed investors and uninformed investors. The latter not being subject to judgment bias. Informed investors determine the price; however, they are subject to two biases: overconfidence (exaggerate precision of information) and self-attribution (down-weight public signals, especially if they contradict private signals). The overconfidence theory states that investors’ confidence leads to biased valuation of information and thus investors trade too much because they are too optimistic or too pessimistic (Odean

(17)

1998, Alexandros V Benos 1998, in line with Fischhoff, Slovic, Lichtenstein, 1977). The psychological concept of optimism and wishful thinking was introduced by Buehler, Griffin and Ross (1994). They show that people think “they finish their tasks sooner” and Weinstein (1980), who display that people have rosy views of abilities and prospects.

All these explanations may be used to explain why underreaction and overreaction to information occur. Public news or events are the most obvious information that the market participants receive. Underreaction to events was first introduced by Ball and Brown (1968) who analyzed the stock prices response to earnings for approximately a year after announcement. Kothari, Lewellen and Warner (2006) study the stock market’s reaction to aggregate earning news. They find that returns are unrelated to past earnings, which suggests that there is neither underreaction nor overreaction, but actually the discount rate shocks explain a significant fraction of aggregate stock returns. This is contrary to other findings. Apart from earnings announcements, the concept of underreaction is analyzed in other event types, for example in merger deals. In this research area there are also contradictory findings of underreaction and true price reflection. Sun and Louis (2015) observe that managers of inflated earnings firms purposely announce mergers on a Friday where limited attention of investors is present, leading to underreaction of the investors. Ahern and Sosyura (2014) also show how that the timing of merger announcements and prior negotiations is systematically organized to increase favourable valuation. Frazzini (2006) analyzes whether underreaction to news is due to the disposition effect (the disposition effect was labelled by Shefrin and Statman (1985)) and finds that bad news travels slowly among stocks trading at large capital losses, in turn leading to a negative price drift, whereas good news travels slowly among stocks trading at large capital gains, in turn leading to a positive price drift. Similarly, Da, Gurun, and Warachka (2014) find that markets generally underreact to slow release of news. On the contrary, important evidence on the non-existence of underreaction is presented by Agrawal et

(18)

al. (1992). Further, Fama (1998) often observes that patterns on investor reactions are not consistent and sometime disappear when accounting for further variables.

Apart from public news and events, there is also information flow not directly observed. Whereas reaction to news has been studied extensively in different areas, the effect of no news is more difficult to examine. Huberman and Regev (2001) find that stock prices probably move to no new news, and the movements might be concentrated in stocks that have attributes in common which are not necessarily company fundamentals. Shiller (1981) even suggests that investors overreact to unobserved stimuli. Chan (2003) compares stock returns that exhibit momentum with or without accompanying news. Chan (2003) finds evidence that investors overreact to spurious price movements even without news. Giglio and Shue (2014) believe to be the first to investigate underreaction to simply the passing of time. Passage of time is analyzed by observing changes in prices and returns of stocks after merger announcements. Giglio and Shue (2014) find that the time after a merger announcement actually contains information, and that agents underreact. They test how the market reacts to “no new news” in the context of mergers and claim that if markets are rational, prices should contain the information of the hazard rate, “the probability that the merger will complete in event week t conditional on it not completing or withdrawing prior to week t” (p. 3390-3391).

So, the discussion of whether market participants underreact to news is already discussed quite extensively, the question of underreaction to no new news is remains very unexamined. Especially a setting where time to an event can be examined, meaning, that an announcement occurs, time passes, the event occurs or withdraws is representative for passage of time in well-defined manner. There is a set starting and ending point, the information is available and the interim period is suitable. A more detailed analysis of which factors influence the passage time also reveals information on the completion or withdrawal of merger deals, which is barely discussed in

(19)

research. Further, the behavioural models can help explain the failure to incorporate the probability of merger completion and are thereby represented in a different context. As behavioural models actually examine the investors in detail, the question arises whether the type of investor plays a major role in inefficiencies occurring in the markets.

2.3 Irrational and Rational Investors

Following the debate of market efficiency supporters and non-supporters, coupled with explanations from behavioural finance for market inefficiencies, some researchers focus on analyzing differences between rational and irrational investors. Despite the efficient market hypothesis stating that rational investors undo the damage done by irrational investors, finance theory does not necessarily imply that these fully offset each other (Shiller, 2003).

Stein (2009) supports this view by stating that a large number of institutional traders or information intermediaries do not increase market efficiency. Sinha (2015) even finds that underreaction to news suggests behavioural biases and is more common among large stocks and stocks with high institutional holdings. Barberis and Thaler (2003) explain that limited arbitrage shows that if irrational traders cause deviations from the fundamental value, rational traders cannot reverse this. Hirshleifer, Subrahmanyam and Titman (2006) even developed a model from which they concluded that irrational trades influence underlying cash flows, by which irrational market participants can earn abnormal profits that can exceed the abnormal profits of rational informed investors. Whereas Odean (1999) states that with more trading more losses occur, still hinting at inefficiencies coupled with individual investors.

The question which arises is that perhaps the behavioural biases are attributable to ownership structure. Examining this in the context provided by Giglio and Shue (2014) has multiple advantages. With this analysis, multiple discrepancies in literature are addressed: efficient market hypothesis, behavioural finance, and research

(20)

on individual versus institutional investors. Further, characteristics which affect merger completion and failure are directly examined.

(21)

3 Data and Descriptive Statistics

To answer the research question “how does underestimation to the passage of time between merger announcement and merger completion change with changes in non-institutional versus non-institutional ownership” data on merger deals, target and acquiring company details, stock prices and ownership structure is necessary. As most analysis in the underreaction literature, the maximum time length as possible is used: 1985 – 2014, as other papers have found negligible differences in this time frame.

3.1 Data Selection

The focus is on all publicly available merger deals with U.S. target companies. U.S. is the nation of analysis due to the higher comparability with other findings in the traditional finance and underreaction literature. Necessary conditions for a merger event to be included are: valid announcement date, valid completion date or failure date within 5 to 250 trading days after the announcement, whereas the upper bound is set to avoid capturing information that is unrelated to the merge as reasoned in the methodology section and stated by Malmendier, Saidi & Opp, 2016. Further, in their comparable dataset, Giglio and Shue (2014) state a median time to completion of 88 days and a mean of 103.5. This confirms the meaningfulness of the maximum and minimum days to completion. Mergers that fail or complete before announcement are excluded as time after announcement of a “potential” merger is analyzed. This may also hint to errors in the database. Moreover, information on institutional holdings and exchange ratio of the deal are inevitable. In addition, hybrid forms of financing, meaning a mix of cash and stock financing, are excluded as the relevance of the financing information has been proven by many researchers (general: Demodaran (aswath), Malmendier, Saidi & Opp, 2016, news focused: Ahren and Sosyura, 2014) and the separations of hybrid into cash and stock to include the observed differences is problematic. Giglio and Shue (2014) point out the importance of excluding tender

(22)

offers with expiration date as the passage of time has a different interpretation if an end date is known, thus these deals are also excluded. Deals where the target announced bids within the past three years will be excluded as the company hints at a special case. Deals with undisclosed M&A value (cannot calculate the premium of the deal), and special deal types such as privatizations and LBOs will be disregarded as well. Deals between companies from all industries are included in the analysis. This is meaningful as the general underreaction and market efficiency of all deals in the market are of interest.

The Thomson ONE M&A database contains information on merger deals including relevant information on the merger and target companies: ID, industry codes, nation, industry SIC code, and public status. Deal information includes date of announcement, date of (unconditional) completion, withdrawal date, deal type/attitude (hostile, friendly or other), value of transaction, (initial) price per share, number of shares acquired, whether a collar agreement was set, financing structure (“consideration structure”), and termination fee.

To obtain other relevant information about the companies, the COMPUSTAT database was used. It is available through the Wharton-WDRS website and among other contains yearly information on the total assets of the company (at), total liabilities (lt), and total common/ordinary equity (ceq) which are relevant for this study. This data was retrieved for the target and acquirer companies of each deal for the past fiscal year before merger announcement. The “current” (including the announcement date) fiscal year already incorporates information on the potential, completed, or withdrawn merger and thus the valuation or changes that came with the merger would already have been incorporated in these variables.

The database CRSP contains valuable information on stocks. Specifically, data on stock prices and average returns for the target and acquirer are necessary for two business days before the deal announcement for the calculation of the takeover

(23)

premium for cash financed deals and the size of the target and acquirer, and the price of the target is downloaded up until 250 business days after announcement for the calculation of the hazard rate. Further, the number of shares outstanding for the target and acquirer two days before the announcement are retrieved for the calculation of the size of the target and acquirer.

The remaining data is retrieved from the Thomson Reuters Institutional Holdings (13F) database. Holdings data, including the typecodes, shares outstanding, manager id, number of shares owned, as reported on Form 13F filed with the SEC, are necessary for the target company in the most frequent fashion possible. In this case this means the quartile before the date of announcement of the merger. The database lists the share ownerships of each manager. This extensive detail is not necessary for the purpose of this analysis, therefore, the data is collapsed to merely deliver summed information on one deal, i.e. total shares outstanding, total shares owned by type1 – type 5 institutional investors, total shares held by all types.

3.2 Variable Preparation

The basis of survival analysis is time. For this thesis, in line with the calculations of Giglio and Shue (2014), weeks after announcement are examined. The reasoning behind this is that the deals are naturally likely to complete within the next year and examining daily (in contrast to stock prices) probabilities of completion is too detailed as rarely “open announcements”, i.e. announcements without a deadline, complete within the first days, and many generally take a few months to actually complete or withdraw, as will be seen in section 3.3. So, first, the time to completion and time to withdrawal after merger announcement are calculated simply by subtracting completion or withdrawal date, depending on which actually occurred, from the date of the announcement. The number is divided by 7 and rounded down to the nearest integer to result in the average “weeks after announcement”. For setting the data as survival time data in STATA it is necessary to merge the information for the deals as

(24)

“studytime”, i.e. a variable which states the time after announcement until event (whether completion or withdrawal). To be able to differentiate the event-type a dummy variable indicating whether the deal completed (1) or withdrew (0) is generated.

The deal characteristics which are also analyzed by Giglio and Shure (2014) are defined and created quite similar to their study for the purpose of this thesis. The deal type is split into three groups: friendly, hostile, and other. A variable deal attitude for the heterogeneity analysis is created which represents 0 if the deal is hostile, 1 if it is friendly and 2 if other, whereas other includes unknown and unsolicited deals. For the regression analysis, two dummy variables are created: hostdummy, which is 1 if the deal is hostile, and 0 otherwise, as well as otherdummy, which is 1 if the deal is from the attitude category other and 0 otherwise. Calendar time is the actual time period of the deal. This information is examined as otherwise calendar time information is not included in the survival time analysis. For this variable, the time-span 1985 to 2014 is split into centuries as this seems like a reasonable split of changes in the cultural, business and political environment, for example, the opening of the west and east, or the opening of China and Russia to the rest of the world starting from the 90s, as well as the start of the internet era in the 1999s, and the advancements in globalization over the past 15 years. A variable calendar time is created whereby 0 represents before 1990 (1985 – 1990), 1 between 1990 and 2000, 2 deals from 2000 to 2010 and the final group 3 represents the deals that occurred between 2010 and 2014. For the regression analysis, 3 dummy variables are created. The dummy before90 is 1 if the deal took place before 1990, and 0 otherwise. The dummy in90 and in00 are 1 if the deal took place in the 1900s and 2000s, respectively and 0 otherwise. In order to actually observe effects in the transaction environment, the merger deals are split into whether the deal occurred during a merger wave or not. A closer look at the spread of the deals according to years reveals that in general before the 1999s (as mentioned the internet era/leading to the dotcom bubble) the amount of transactions was not comparable to

(25)

the later dates. Thus, a dummy variable is generated whereas the deals before the 1999s are split into 1 if during a merger wave and 0 if not. The deals after the 1998s are represented in this dummy variable accordingly. Target and acquirer size (the second was not created by Giglio and Shue (2014)) are calculated by multiplying shares outstanding of the target/acquiring company by the price of their shares two days before the announcement. For the heterogeneity analysis this variable is grouped into large (1) versus small (0). A dummy variable named financing is created for the variable financial structure (“consideration structure”): 0 for equity and 1 for cash deals. A further dummy is generated indicating whether the deal was industry diversifying or not. This was simply created by assigning the dummy variable diversifying a “0” if the SIC codes of the acquirer and target were in one industry, i.e. 1-9 agriculture, 10-14 mining, 15-17 construction, and so on, and a “1” if not (industry diversifying merger). The merger arbitrage spread is according to Giglio and Shue (2014) defined as the relative difference between the effective offer price and the target price two days after the merger announcement, whereas a large merger arbitrage spread would hint at the fact that the merger is less likely to complete. For the heterogeneity analysis the merger arbitrage spread was also split into equally sized groups referring to 0 for relatively lower and 1 for relatively higher merger arbitrage spreads. The deal premium for cash-financed deals is defined according to the definition of Giglio and Shue (2014) as the initial offer price at merger announcement divided by the price of the target two days before the merger announcement. For equity-financed deals the premium is calculated as the number of shares the target shareholders receive of the acquiring companies’ shares upon deal completion times the price of the acquirer two days before deal announcement divided by the price of the target two days prior to deal announcement. Based on this variable a dummy variable is created: 0/1 for relatively lower/higher deal premia. The target price is already created when merging the main database (Thomson One) with the CRSP data – for this already the variable price was split for

(26)

price of the target and price of the acquirer. For the heterogeneity analysis, the deals were split evenly into group 0 and 1, meaning relatively low and relatively high target prices respectively (as Giglio and Shue (2014) did). The termination fee (given in million USD) was directly retrieved from the Thomson One database. Both variables in USD, termination fee and size are scaled to represent million USD.

Other variables are defined and generated as the following: acquirer leverage represents the acquirer’s debt-to-equity ratio. The total liabilities of the acquirer in the fiscal year before the announcement are divided by the capital market value, whereas this is total assets minus the book equity value plus the market equity value. A dummy variable categorizing this information into acquirer with higher (1) versus lower (0) leverage is generated. Relative deal size is defined as the value of the transaction divided by the size of the acquirer in the fiscal year before the merger announcement. Again, a dummy is created: higher leverage deal size is indicated with 1, and lower leverage deal size with a 0. Ownership structure is derived by adding the number of shares held by the different types of investors for each target company for the quartile before the announcement and dividing this by the total number of shares outstanding. This group is split into two equally sized groups creating the dummy ownership structure used in the heterogeneity analysis with 0 and 1 for low and high institutional ownership, respectively. For the regression analysis the variable non-institutional ownership is created by simply calculating 1 minus the variable ownership structure. This gives the ratio of shares held by non-institutional and total shares in the target company.

For the calculation of the market hazard rates, as described in section 4.3, the prices and returns from the stocks downloaded from the CRSP database will be averaged like the variable “time after announcement until completion/withdrawal in weeks” , so that the prices and returns for the weeks (1 to 50) are calculated. This could mean that a “week” actually starts in the middle or end of the week, i.e. it does not refer

(27)

to calendar weeks.

At last, time dummies for each time t are generated to represent the average curve. This means that a panel data set up is necessary when running the main regression. The “time dummy” for each t is 1 if the specific time is represented, 0 otherwise.

3.3 Variable Analysis

The generated dataset includes 3,036 merger deals which are described in table 1. All target companies and 94% of the acquiring companies are U.S. based, as explained in subsection 3.1. Panel A reports figures separating the dataset into characteristics according to completed, which are 77%, and withdrawn, which are the remaining number of deals. Many deals contain missing deal attitudes, but it is safe to assume that actually only a few mergers were hostile takeovers, or attempted hostile takeovers. This is more thoroughly described in the upcoming section since this characteristic might have an effect on hazard rates. Further, as expected (better infrastructure, lower transaction costs, easier flow of information and alike), the number of mergers increases over the decades, whereas the relative percentage of deals withdrawn increases as well: in the first decade only 14% of the deals completed (small total number of deals though), and approximately 70%, 80%, and 85% of the announced deals were completed in the 90s, 2000s, and 2010-2014., respectively. A high percentage of 85% of the deals are cash deals, whereby the relation of completed (81%) and withdrawn (19%) is comparable to the other characteristics. The acquirer signals that equity is undervalued which therefore has positive reactions on the acquirer’s stock price. A similar case is observable with industry diversifying deals (86% and 14%), which represent 31% of the total dataset.

Additional, but more financial, characteristics of a merger deal are shown in Panel B. The transaction volume ranges from small deal sizes of 130,000 up to 6.2bn USD and the standard deviation and mean hint that the transaction volumes are spread quite

(28)

extensively. The dataset did not exclude any specific deal sizes and therefore this represents all deal sizes available after the sorting out of characteristics described in section 3.1. The initial offer price per share and the termination fee also have a wide range. The merger arbitrage spread has a wide range, however, the figures hint at the spread averaging around 0.76 with a standard deviation of 5.23. The offer price more often exceeds the target price two days after merger announcement. Arbitrage opportunities are thus existent, but not extremely high.

Panel C shows that the average deal completes or withdraws in 105 business days/21 weeks after merger announcement. As described, the minimum and maximum business days are set at 5 and 250, respectively. Both, the time to completion and time to withdrawal, are observed throughout this entire range. In comparison, merger deals which withdraw occur earlier than deals which complete on average, with a difference of about 20 business days/4 weeks after merger announcement. Figure 1 displays the

frequency of merger completion and withdrawal over the examined time. Whereas not many deals complete or withdraw after week 40, most deals withdraw quickly after announcement and most deals which complete do so around week 20.

Apart from deal characteristics, a closer look at the companies involved in the

(29)

Table 1 Summary Statistics: Deal Characteristics

Summary Statistics: Deal Characteristics Panel A

Completed Withdrawn N

Absolute Relative Absolute Relative Absolute Relative

Number of Deals 2,341 77% 695 23% 3,036 1% Attitude: Friendly 991 66% 505 34% 1,496 49% Attitude: Hostile 10 19% 44 81% 54 18% Attitude: Other/Missing 1,340 90% 146 10% 1,486 49% Announced in the 85-90s 13 14% 82 86% 95 3% Announced in the 90s 415 70% 177 30% 592 19% Announced in the 2000s 1,332 80% 333 20% 1,665 55% Announced in the 2010-14s 581 85% 103 15% 684 23% % Cash Deal 2,083 81% 489 19% 2,572 85% % Industry Diversifying 801 86% 131 14% 932 31%

% Acquirer from U,S. 2,391 84% 455 16% 2,846 94%

Panel B

Mean Std. Dev, Min Max N

Transaction Value ($mil) 889 3,332 0.13 6,211 3,036

Initial Price per Share 23 59 0.01 2,930 3,036

Termination Fee ($mil) 45 119 0.1 2,300 2,300

Merger Arbitrage Spread 0.76 5.23 -43 126.50 3,036

Deal Premium 1.19 0.37 0.00 9.09 3,036

(30)

Table 1 - Continued

Panel C

Mean Std. Dev, Min Max

Time to Event (days) 105 55 5 250

Time to Completion (days) 108 53 5 250

(31)

merger is reasonable as their characteristics might affect the hazard rate of completion. These are presented in table 2.

Panel A of table 2 shows that the acquirer prices on average are higher than the share prices of the target company. The returns for the acquirer are quite comparable, whereas the 25th percentile and the mean of the target returns hint at the fact that the

target share returns are more skewed towards the negative numbers than the acquirer share returns. The average size of the target company is larger than the average size of the acquirer, however, the higher median and 75th percentile support that most acquirer

companies are larger than their respective targets. The high differences in leverage among the acquirers is possibly due to the fact that all kind of industries are present in this dataset, whereas the top 25th percentile is difficult to explain. Panel C examines one

of the most important variables in the dataset, namely the percentage of institutional investors. The data is said to be not adjusted by the databank as it is directly retrieved from the SEC filings and no analysis of whether it may be accurate is applicable to these figures. As expected after having examined the characteristics of the deal and the involved companies, the percentage of institutional investors is widely spread, whereas the median and mean center around 50%. On average, insurance companies and investment companies with their managers only make up a small amount of investors of the total company, which makes sense as the majority of these companies makes rather safe investments, usually in very large companies, i.e. few percent could make up a significant high amount. Banks and independent investment advisors are more highly invested in companies overall which is probably explainable by their business set-up (amount of capital available and interest in different company sized). All other investors (type 5) in the data have a large stake, a mean and median of 67 and 72, respectively.

(32)

Table 2 Summary Statistics: Company Characteristics

Summary Statistics: Company Characteristics

Mean 25th

Percentile Median

75th

Percentile Panel A: Two Days Before Announcement Date (Daily Shares)

Target Price 20 6 14 26

Target Returns (%) 0.2 -0.9 0 1

Acquirer Price 26 10 20 37

Acquirer Returns (%) 0.1 -0.1 0 1.2

Panel B: Fiscal Year Before Announcement Date

Target Size ($bn) 1.8 0.07 0.2 0.8

Acquirer Size ($bn) 0.7 0.3 0.6 1.5

Acquirer Leverage (%) 2 0.02 0.06 27

Panel C: Quartile Before Announcement Date

% Institutional Investors* 51 27 52 75

% Type 1 Institutional Investors 11 4 7 13

% Type 2 Institutional Investors 3 0.4 1 3

% Type 3 Institutional Investors 3 0.8 1 2

% Type 4 Institutional Investors 21 8 15 27

% Type 5 Institutional Investors* 67 58 72 82

Type 1 institutional investors refers to banks, 2 to insurance companies, 3 to investment companies and their mangers, 4 independent investment advisors and 5 to all others.

(33)

4 Methodology and Results

The first level of interest is the information revealed through the passage of time; this is examined by the hazard rate of completion. The second level is examining deal and company information by non-parametric and by semi-parametric measures to detect relevant characteristics concerning the hazard rate of completion. The third and most important level of interest of this thesis is to ultimately analyze the effect of ownership structure on the under-/overreaction to the passage of time represented by hazard rates graphically, then parametrically.

4.1 Actual Hazard Rate of Completion of Merger Deals

The time to occurrence of a specific event is the broad definition of survival time (Lee & Wang, 2003). Survival time analysis is used due to certain aspects of the analysis data, which will be hinted at throughout this section. For this thesis, survival time is defined as the time after the announcement of a merger until the completion of the merger. In survival analysis, time is not seen as calendar time, but as time after a specific date. Time, t, is set to 0 at the announcement date for all merger deals and counts the number of weeks after the announcement date. The survival function, S(t), is set as the probability that the deal survives until time t, meaning it neither completed nor was withdrawn. The probability density function of the survival time, f(t), is defined as the limit of the probability that the deal completes. The hazard function, h(t), is defined as the fraction of deals that complete during each period t conditional on that it survived until time t, taking into account that once the deal completed it cannot withdraw, and once the deal withdrew it cannot be completed. The hazard rate thus is simply the unobserved rate at which the deal completes, controlling for the occurrence and the actual timing of the event at interest. The interaction between the functions is better shown mathematically:

(34)

(2)

For the calculation of the hazard rates, the standard Nelson-Aalen estimator is used. The Nelson-Aalen estimator is a nonparametric statistic; it does not make assumptions about the probability distribution. The shape of the hazard rate reveals important information of the passage of time. For example, a twice as high hazard rate would mean that the probability that the deal will complete conditional on it not having completed or withdrawn is twice as high compared to the other time period. With time passing, in this case weeks after announcement, obvious and less obvious factors have an effect on the hazard rate of completion of a merger deal. Thus, the hazard rate is calculated based on mergers with an event between 5 to 250 business days. This boundary has to be set because a longer than one year duration hints at other difficulties, external or internal, influencing the probability that the deal will complete which are more and more difficult to characterize. As the hazard rates for each of the deals differ, it makes sense to analyze smoothed versions of the hazard curve. This is further reasonable because the variables included in the hazard rate have not been adjusted but are taken as given or as described in section 3.2, but not manipulated or adjusted. As many variables, hazard rates also have outliers. These are especially situated in the beginning and end of the time frame. An explanation for this is that there is high discrepancy of the probability of an event occurring right after the announcement, and very late after the announcement.

A very and an only slightly smoothed version of the hazard rate of completion are displayed in figure 3 and 4, respectively. Missing values at the very beginning and end of the time range are due to the explanations given above. Starting at week 8 after the announcement to approximately week 35 after the announcement, the very smoothed line steadily increases from a hazard rate of approximately 2.5% to 8%. From approximately week 35 to 44, there is a steeper increase from 8% to 17%, whereas the

(35)

curve hints at a stabilization of the hazard rate. The less-smoothed curve is quite similar, however, in general the hazard rates are higher (outliers not omitted or rates

smoothed) and there is a small hump around week 20. Further, the increase after week 40 is higher.

In real terms this means that the probability that a deal will complete, conditional on it not having been completed or withdrawn increases with time. The probability that the deal will complete in the next week is higher than the probability that it completed in the previous week. In comparison, a horizontal line would mean that over time the probability of completion does not change. As this is not the case, we can easily

Figure 2 Hazard Rate of Completion of All Deals (Smoothed)

(36)

conclude that the passage of time contains information about the completion of the deal.

Giglio and Shue (2014) derive a different hazard rate shape. The conducted hazard rate has a hump shape with hazard rates starting close to zero and increasing approximately until week 25 up to approximately 7% and then decreasing until week 50 down to 2%. Their explanation is that at the very beginning the deals are less likely to complete, which perhaps may be due to the time needed to ultimately complete the deal, and in the following weeks the completion of a deal in the next week is more and more likely. This reasoning is applicable to the findings of this data analysis as well. However, after a steady increase in the findings of Giglio and Shue (2014) the reasoning for the decrease might be that if a deal has not been completed after a long time, the probability that it will complete in the following week will be less likely. Whereas this dataset suggests that the announced deals will complete to a high probability: if it did not complete in the previous week, the probability that it will increase in the next is more likely. After having examined the descriptive statistics, this seems more reasonable for the data, as significantly more deals complete than withdraw.

4.2 Heterogeneity Issues in Hazard Rate of Merger Deals

The hazard rate is calculated by including multiple deals with different characteristics. As seen in comparison with the hazard rate curve constructed by Giglio and Shue (2014), differences in the dataset lead to different shapes of the curve. Whereas the major differences between the datasets are not obvious, it is important to be aware of the differences in the dataset used for this analysis. This is because to ultimately analyze the effect of ownership structure on hazard rates, these influential factors need to be accounted for. Since Lancaster (1979) it is widely recognized that it is important to account for unobserved heterogeneity in hazard models. Possible factors, categorical predictors, can be analyzed by comparing hazard rate curves. This graphical analysis is

(37)

the thought behind splitting the characteristics into approximately equally sized groups as described in section 3.2 “Variable Preparation”. In survival time analysis, it is common to specifically examine the parallelism of the hazard rates. If the hazard rates are approximately parallel, then the groups which were derived according to different characteristics are proportional, meaning the deals do not differ according to that characteristic. The rate (height of the curve) is not the factor of interest, but whether the timing and the structure of the curve are comparable. Differing timings of increases and decreases, i.e. non-parallel hazard curves, imply relevancy of the characteristic - the characteristic might influence hazard rates – the deals are heterogeneous.

To conclude which characteristics are relevant, the analysis will further be extended by conducting a non-parametric test: the log-rank test of equality of survival curves. The test compares the two or more than two curves of the Nelson-Aalen survival functions. Equation 1 and 2 show the relation between hazard and survival function. It shows that the log-rank test of equality also makes sense when the relevant variable actually is the hazard rate. The H0 hypothesis states that there is no difference

between (true) survival curves.

Observable characteristics before a deal announcement will be considered as these are incorporated in the pricing of the stock, which builds the basis of calculating the hazard rate observed in the market, in this paper referred to as “market hazard rate”. The market hazard rate will be explained more thoroughly in the next subsection. Giglio and Shue (2014) include the following characteristics for which data is available for this paper as well: attitude, calendar time, size of target, financing structure, whether the deal is industry diversifying, merger arbitrage spread, whether the deal took place during a merger wave, premium at announcement, target share price, and the amount of the termination fee. For this analysis we further will consider the acquirer’s leverage, size of acquirer, and relative deal size. Moreover, it is in the nature of the research question to take a first look into differences in the dataset regarding hazard rates and

(38)

institutional ownership. Figure 4 displays the graphs used to examine the different heterogeneity characteristics.

Deal attitude is split into three categories: hostile, friendly, and other. Different analysis have been conducted to examine whether deal attitude has an effect on the returns of a deal (e.g., Servaes, 1991), whether hostile bids were damaging (e.g., Schwert, 2000), or whether it plays a role in bidding contests (Malmendier, Moretti & Peters, 2012), all finding different levels of relevance of the characteristic. The graphical analysis displayed in figure 4A of the characteristics and hazard rates reveals that the hazard rates of hostile, friendly and other (unsolicited and deals where the deal attitude information was not available) are not very proportional. On another note, it is interesting to see that the hazard rate of a hostile deal is very close to zero, in contrast to other deals having a hazard rate around 5%. This big difference may be due to the low number of observations of hostile deals.

Changes in law, the industry environment, financial environment (such as inflation rate, currency, interest rates), political environment and similar might have an effect on the completion of the merger deal. The characteristic calendar time captures these events. The calendar time is split into 80s, 90s, 2000s and 2010s. Again, the graph (figure 4B) in majority displays a continuous growth in hazard rate and looks very proportional, whereas the hazard rates for deals from the 2010s are constant for a few weeks. The 80s in the context of heterogeneity issues can be disregarded because only 3.1% of the observations represent those calendar years. The disregard of this line is further supported by the explanation that the analysis of this paper considers the years 1985 – 2014, thus only 5 instead of 10 years of the 80s are included. Nevertheless, the existing line appears quite parallel.

Whether the deal took place during a merger wave is related to the characteristic calendar time in the sense that merger waves evolve because of different factors. Martynova and Renneboog (2008) define takeover wave by the number and

(39)

(40)

cont

(41)

the total value of takeover deals over time, whereas this again is based on a number of political, economic and regulatory changes. There is evidence that companies experience negative abnormal returns when the deal took place during a merger wave (e.g., Agrawal et al., 1992 and Moeller et al., 2004), which hints at possibly also affecting the probability that the merger completes. After all, competition in the market might drive deals to complete faster, or a “merger” trend would influence the number of deals announced in a specific time window. The definition of Martynova and Renneboog (2008) is applied to the data collected for this analysis. The curves in figure 4C representing deals which occurred during a merger wave and curves which did not are parallel, hinting at no heterogeneity issues concerning this characteristic.

Giglio & Shue (2014) find that with respect to the size of the target the location of the peak of the hump varies. Further, it seems reasonable that a larger company is more challenging to acquire and the positive effects of acquiring a large versus a small company might differ. Figure 4D, however, displays that in this dataset there is no major visible difference between the hazard rates of deals where the target company was smaller compared to deals with larger target companies. This is not as surprising as it seems because factors relating to the size of a target company are, as other “obvious” characteristics, probably considered before announcing a merger leading to an “average” hazard rate of completion.

A great amount of literature has tried to explain differences in cash versus stock acquisitions, for example Huang and Walkling (1987) found that announcement returns are higher for cash deals than equity deals. Malmendier, Opp, and Saidi (2016) analyzed failed takeover attempts and find that there is a link between failed takeovers and the financing structure of the deal. Giglio and Shue (2014) also find differences in the hazard rates for cash and equity deals. In line with the other theoretical findings, figure 4E shows two increasing but non-parallel hazard curves. The hazard rate of completion for cash deals is higher approximately until week 28. Then, the hazard rate

(42)

of completion for equity deals increases while the hazard rate of completion for cash deals nearly stays constant. Both curves experience a strong increase starting week 35. Diversifying mergers are between two companies from different industries, and non-diversifying deals are between companies from the same industry. The literature does not agree on the different effects of industry diversifying deals: Whereas Haugen and Udell (1972) find that over the longer-term unrelated deals performed better than related deals, Bruner (2002) finds that the short-term returns are positively related to similarities of the industry type. Concerning hazard rates we see high similarity of the curves in figure 4F, i.e. no heterogeneity issues visible.

The category of lower arbitrage spread includes the deals with relatively higher target prices after the merger announcement compared to the offer price, whereas the category of higher arbitrage spread includes the deals with relatively higher offer prices compared to the target price after the merger announcement. According to the graphical analysis in figure 4G, there is not much difference in the timing of the increases and decreases of the hazard rates. This was also concluded by Giglio and Shue (2014).

It seems plausible that deals with a higher premium have a higher premium due to their high attractiveness. The target company might face competing opportunities and switch strategies after announcement. On the other hand, the attractiveness might imply a high probability of completion due to the opportunities which the acquirer desires to tackle. Figure 4H hints that the arguments are well balanced, the curves of deals with low and high premium are very parallel.

According to Giglio and Shue (2014) low-priced stocks are probably smaller and more illiquid than high-priced stocks. In contrast to the findings of Giglio and Shue (2014), figure 4I reveals, however, that the characteristic share price of targets does not seem very relevant: hazard rates of completion of deals with target companies with a lower share price almost stagnate in weeks 25 to 30 instead of a constant increase, but