Loading...

Measuring M&A-Success in Cross-border Banking M&A

An Assessment of Methodologies

Diploma Thesis 2007 91 Pages

Business economics - General

Excerpt

List of contents

List of figures

List of tables

List of abbrev1at1ons

List of symbols

1 Introduction

2 Assessment of methodologies
2.1 Overview of methods to measure M&A-success
2.2 Event study methodology
2.2.1 Basic concepts
2.2.2 Critical issues
2.2.3 Long-horizon event study methodology
2.3 Operating performance analysis
2.3.1 Basic concepts
2.3.2 Critical issues
2.4 Frontier efficiency methodology
2.4.1 Basic concepts
2.4.2 Critical issues
2.5 Case study research
2.5.1 Basic concepts
2.5.2 Critical issues
2.6 A claim for a multi-methodology approach

3 M&A-success in European banking - the case UniCredit-HVB
3.1 Introductory remarks
3.1.1 The selection of the merger case
3.1.2 Presentation of transaction partners
3.1.3 Transaction overview
3.2 Qualitative analysis
3.2.1 Evaluation framework - determinants of M&A-success
3.2.2 Ex ante determinants
3.2.2.1 How to evaluate M&A-strategy?
3.2.2.2 Market-based-view
3.2.2.3 Resource-based view
3.2.3 Ex interim determinants
3.2.4 Ex post determinants
3.3 Quantitative analysis
3.3.1 Short-horizon event study
3.3.2 Long-horizon event study
3.3.2 Operating performance analysis

4 Conclusion

References

Appendix I
Appendix 1: The Dodd and Warner (1983) test-statistic
Appendix 2: UniCredit’s long-term share price performance
Appendix 3: Results of the OLS-regression analysis
Appendix 4: CARs around the leakage date (MIB 30/ DAX 30)
Appendix 5: List of eligible long-horizon event study peers
Appendix 6: List of eligible operating performance peers

Abstract:

The recent surge in European cross-border bank mergers and the limited amount of studies investigating the value-consequences of these transactions call for further research. This paper assesses the methodologies that are in use to measure M&A- success in order to support a rigorous use of these approaches in future analyses. Moreover, the assessment of the bank merger between Italy’s UniCredit and Germany’s Hypovereinsbank exemplifies the accurate application of these tools as well as the usefulness of appraising individual deals within a multi-methodology framework.

List of figures

Figure 1: Overview of methodologies to evaluate M&A-success

Figure 2: Overview of X-efficiency frontier methodologies

Figure 3: Steps to design a case study

Figure 4: UniCredit’s pre-merger structure

Figure 5: Overview of UniCredit’s tender offer

Figure 6: Case study evaluation framework

Figure 7: UniCredit’s strategic move within the CEE-region

Figure 8: Diversification effect of the loan portfolio

Figure 9: Main integration steps

Figure 10: UniCredit’s post-merger structure

Figure 11: CARs around the leakage date

List of tables

Table 1: Potential blases of short-horizon event studies

Table 2: Potential biases of long-horizon event studies

Table 3: Overview of selected bank performance ratios

Table 4: Potential biases of operating performance studies

Table 5: Key forces of change within the European banking sector

Table 6: Selected integration activities within UniCredit’s divisions

Table 7: CARs of the short-horizon event study

Table 8: UniCredit’s long-horizon BHARs

Table 9: UniCredit’s pre- and post-merger operating performance

Table 10: HVB’s pre- and post-merger operating performance

List of abbreviations

illustration not visible in this excerpt

List of symbols

illustration not visible in this excerpt

1 Introduction

The merger and acquisition (M&A)1 activity in the global banking industry over the last decades can be characterised by two things: firstly, M&As constitute an insecure way to increase corporate value and secondly, they usually have taken place within domestic domains. One of these issues has surely changed. Whereas during the period 2000-2004 cross-border bank mergers accounted for already 14% of the total value of euro area bank M&As, this figure increased to 38% for the years 2005-2006, projected to raise further in the future.2 Notably, overall deal values are increasing for both domestic as well as cross-border transactions. In 2005 the Italian UniCredit and the German Hypovereinsbank decided to create the “First Truly European Bank” in a €19,2 billion merger, topping Spanish Santander’s successful bid for British Abbey National of about €12,9 billions one year prior. Dwarfing all previous bank mergers, offers for Dutch ABN AMRO totalled €71 billions in May 2007.

In view of the paramount values being transferred, it is essential to understand the consequences resulting from these transactions. Several methodologies have been developed to appraise whether mergers are successful. In this paper the quest for M&A-success is ultimately linked a deal’s potential to increase shareholder value. Undeniably, due to the surge in bank mergers and the rather limited amount of literature in comparison to U.S. evidence, further research investigating the value- consequences of European cross-border banking M&As is at hand.

This paper goes back to the basics, assessing the methodologies that have been used to measure M&A-success in order to provide an overview of the merits and potential pitfalls of the respective approaches (section 2). In addition, the accurate application of these methodologies is exemplified, taking Europe’s present largest cross-border bank deal between Italy’s UniCredit and Germany’s Hypovereins­bank (section 3). Capturing the idiosyncrasies of this merger, the multi­methodology analysis breaks down into a qualitative (section 3.2) as well as quan- titative component (section 3.3). The former sheds light on the factors that moti­vated UniCredit’s cross-border acquisition and qualitatively appraises the deal by a set of prespecified determinants of M&A-success. The quantitative evaluation makes use of a short- and long-horizon event study as well as an operating per­formance analysis in order to quantify the value-consequences of this merger. Scrutinising this transaction might also help to judge present and future cross­border M&As that likewise intend to create European banking champions. The paper concludes by summarizing the main findings and drawing implications for future research (section 4).

2 Assessment of methodologies

2.1 Overview of methods to measure M&A-success

Before taking a close look at the different methodologies, this section intends to give an overview of and categorize methods that have been used to measure the success of M&A-transactions. Following Pautler (2003), evidence on M&As falls into two methodological streams:3 direct and indirect approaches (see figure 1):

illustration not visible in this excerpt

Figure 1: Overview of methodologies to evaluate M&A-success

Source: Own illustration.

Indirect (static) analyses relate the degree of local market concentration or certain firm characteristics such as size to variables of corporate performance like profits, prices, efficiency or quality of service. Although these studies do not focus on M&As directly, they allow inferences on potential consolidation consequences.4 If, for instance, enlarged market concentration coincides with enhanced profitabil­ity and efficiency, mergers that increase market share are likely to enjoy these benefits. In reference to the banking industry, higher concentration proved to de­crease cost efficiency levels but on the other hand enhanced market power enables banks to increase prices and thus maintain profit levels.5 Moreover, concentration has revealed to reduce quality of service as proxied by bank opening hours.6 Con­versely, cross-border M&As might help to diminish local market power of incum­bent banks, resulting in enhanced efficiency levels and favourable prices.7

This paper, however, presents and assesses the four direct (dynamic) methodolo­gies which can be used for the evaluation of the European cross-border bank merger case. Event studies distinguish themselves as they rely on stock price data to directly measure changes in shareholder value.8 Operating performance and frontier efficiency analyses, by contrast, make use of financial statement data to determine changes in profitability and efficiency levels following M&As. Hence, any inferences on changes in corporate value presuppose that improved financial performance and higher efficiency levels translate into increases in shareholder value. Case studies examine the details of individual transactions and particularly provide insights into M&A-motives and implementation of integration strategies.9

Event studies find that for the U.S. banking sector M&A-deals deliver premium returns to targets while returns to bidders are negative or not statistically signifi­cant.10 This common picture is also found for European cross-border bank merg­ers. Results on aggregate value creation, however, seem to deviate.11 Within Europe, early studies report negative or zero wealth gains (Cybo-Ottone and Mur- gia (2000); Beitel and Schiereck (2001)) whereas more recent papers find evi­dence that European cross-border bank M&As increase shareholder value on a net basis (Beitel et al. (2004); Ismail and Davidson (2005)). Results from operating performance and frontier efficiency analyses are also not consistent.12 Conversely, case studies report more optimistic prospects of success for bank mergers.13

2.2 Event study methodology

2.2.1 Basic concepts

The fundamental objective of the event study methodology is to determine whether there are any abnormal stock returns following a specific event such as a merger. In fact, this methodology has evolved into the standard method to evalu­ate security price responses following corporate events in nowadays research.14

Although there is no exclusive pattern of how to conduct an event study, there is a general track to follow.15 The first job is to define the event of interest which is in our case the announcement of an M&A-transaction. In addition, the event window needs to be defined, that is, the time-frame surrounding the event in which we expect the induced abnormal returns to be realized. At this stage studies depart into short- and long-horizon studies with long-horizons generally embracing event windows of one year or more.16 But since the choice of the event window range is a particular sensitive issue within this methodology, it requires for now to stay with the short-horizon notion, i.e. event windows of less than one year, typically containing just a couple of days around the event under investigation.

The estimated abnormal return [Abbildung in dieser Leseprobe nicht enthalten] of security i on day τ of the event window is given by the difference between the actual [Abbildung in dieser Leseprobe nicht enthalten] and expected normal return [Abbildung in dieser Leseprobe nicht enthalten].

illustration not visible in this excerpt

Several techniques are available to estimate RiT : mean-adjusted return models,market models, market-adjusted return models and factor models.17 Apart from the market-adjusted return model, all others make use of a period prior to the event window, referred to as the estimation or observation window. The observed return-profile which is believed to be free of any event influence, is projected as an estimate for the expected return RiT onto the event window.18 This return cor­responds to the yield one would expect in the absence of the M&A-deal.

The mean-adjusted-model, also know as the constant-mean-return model, assumes that expected returns of a security remain constant.19 Thus, [Abbildung in dieser Leseprobe nicht enthalten] equals the arith­metic average return inside the observation window T, ranging from t1 to t2.

illustration not visible in this excerpt

The market model relates the return of a security i to the return of a market index Rm and therefore follows in concept the Capital Asset Pricing Model (CAPM). The following linear regression specification is used with cci and β being OLS- regression coefficients to be estimated within the observation window and sit representing the model’s prediction error assuming E(eit = 0) :20

illustration not visible in this excerpt

where coefficient β quantifies a security’s systematic risk by measuring the rela­tive impact of changes in Rmt on Rit .21 After the coefficients have been specified for each firm, daily normal returns in the event window can be estimated by:

illustration not visible in this excerpt

This approach prevents abnormal returns from being affected by market-related returns as these are captured in the estimated normal returns via equation (4).

In a similar fashion CAPM estimates normal returns by means of:22

illustration not visible in this excerpt

At first glance this specification might potentially supersede the market model. Only ß is estimated during the observation window. Any fluctuations of the risk­less interest rate Rf are accounted for in the CAPM whereas in the market model the regression constant remains unchanged throughout the event window to the firm-specific estimate ai .23 However, the derivation of equation (5) relies on a number of assumptions24 that translate into restrictions for the applicability of the model. For instance, Banz (1981) finds that in the CAPM small firms realise on average higher risk-adjusted returns than large entities.25 This size-anomaly26 points to a misspecification in that ß of equation (5) does not explain the cross­section of average returns.27 In fact, firm-specific approaches like the market model bypass such bad model problems as they do not rely on an incomplete de­scription of average returns but employ firm-specific return estimates.28

Somewhat different, the market-adjusted-return model assumes that stocks’ nor­mal returns mirror the market return and therefore no estimation of firm-specific parameters is required. Being a special case of the market model with a = 0 and ß = 1, equation (4) can be restated as:

illustration not visible in this excerpt

Multiple factor models supplement regression specifications in style of equation (4) with additional factors such as industry indexes or portfolio returns of traded securities.29 Fama and French (1993), for example, propose a three-factor model that, in addition to an overall market return, caters for differences in firm size and book-to-equity values.30 For the banking industry, a two-factor-model has been advocated that, next to market returns, captures changes in interest rates since these have shown to significantly influence banks’ stock price performance.31 However, researchers found limited explanatory power of additional factors which translates into support for the standard market model.32

Moreover, the net gain in shareholder wealth, i.e. the overall wealth effect of a deal, irrespective of the sharing between the transaction partners, is of interest. The combined abnormal return of acquirer i and target j is estimated by:

illustration not visible in this excerpt

where Vi and Vj are designated weights like the market value of a firm’s equity prior to the merger announcement.33

After the computation of abnormal returns on day τ of the event window, they are aggregated across time and securities.34 The cumulative abnormal return (CAR) of security i across the event window is given by:35

illustration not visible in this excerpt

In case of more than one M&A-transaction under investigation, abnormal returns are averaged by the number of deals ( n ) for the sample of both acquiring and tar­get companies, to obtain averaged cumulative abnormal returns:36

illustration not visible in this excerpt

The last step of conducting an event study is to test the results for statistical sig­nificance. Typically, the null hypothesis to be tested is whether the measured ab­normal returns equal zero.37 For example, following MacKinlay (1997) the test- statistic is given by:

illustration not visible in this excerpt

where the denominator constitutes the estimated standard deviation of the CAR over the event window.38 Having explained the basic event study method, the next section intends to shed light on those issues that have been subject to debate.

2.2.2 Critical issues

In order to assess the event study methodology in its form explained so far, focus is directed towards underlying assumptions affecting the abnormal return metrol­ogy as well as statistical properties of event study models that concern the specifi­cation of test-statistics.

To begin with, the short-horizon event study method assumes markets to be effi­cient. If an M&A-deal is announced, market participants instantly price in the new information so that the resulting change in stock price, i.e. the abnormal return, reflects the change in expected future earning streams due to the merger. Thus, at least the semi-strong efficient market hypothesis is required to hold.39 But if capi­tal markets take a long time to react to new information or worse, prices (tempo­rarily) do not reflect fundamental values, defined as the discounted sum of ex­pected future cash-flows, abnormal returns may not accurately capture long-term value-consequences resulting from M&As. De Bondt and Thaler (1985), for ex­ample, find evidence that markets overreact to the arrival of new information and due to limits of arbitrage prices converge to fundamental values only gradually.36 Accordingly, stock price reactions following a merger announcement may be dis­torted as traders overweight current and recent information surrounding the deal and underweight prior data.

Moreover, high inflation potentially penetrates market efficiency since investors might capitalise future earnings at a nominal instead of the lower real rate and neglect depreciation in the real values of corporate liabilities.40 Both errors result in an undervaluation of equity. This is particularly relevant in the context of M&A-transactions that might substantially impact on future earning streams as well as increase financial leverage41.

Secondly, the event study methodology assumes that events are unanticipated42 Most importantly, the event has to deliver new information to traders. Whereas some corporate events such as the sudden death of a CEO are by nature a surprise, M&A-announcements carry the risk that information leaked to the market in ad­vance. In this case, abnormal returns measured around the merger’s announce­ment do not picture the total change in shareholder value.

However, there is an additional quandary to anticipation. Especially during a merger wave investors are likely to speculate on deals taking place.43 Problemati­cally, negative abnormal returns may also reflect markets’ disappointment that a better potential merger is less likely to occur as a result of the current transaction. For instance, the market anticipated the acquirer to be the target.44 In this situa­tion, abnormal returns do not allow inferences on a deal’s value-consequences.

Thirdly, it is necessary to assume that no confounding effects of other corporate events take place that trigger share price reactions during the event window. This assumption is particularly restrictive when examining large multinational compa­nies, for which significant events occur frequently. A possible solution is to with­draw firms from the sample that experienced confounding events.45

Regarding statistical properties of event study models, non-normality of daily stock returns, non-synchronous trading and clustering effects might bias standard parametric tests for significance. Brown and Warner (1980, 1985) investigate these issues and show that under certain restrictions none of them dilute the (short-horizon) event study methodology.47 Even if the distributions of individual stock returns are fat-tailed, a sufficiently large sample size can be presumed to convert distributions of cross-section average abnormal returns back to normal.48 Autocorrelation, induced by non-synchronous trading, only has a limited impact.49 Clustering effects occur when events of firms included in the sample fall in the same calendar time whereas industry clustering concerns events concentrated in the same industry.50 Both might induce cross-sectional correlation of abnormal returns which renders the assumption of zero covariance between them incor- rect.51 Problematically, ignoring such dependencies might bias the estimated cross-sectional standard deviation of abnormal returns downwards so that test- statistics are biased upwards.52 However, regarding event-date clustering, Brown and Warner (1980, 1985) find that tests which assume non-zero cross-sectional dependence perform worse than those assuming independence.53 Test-statistics are also not misspecified for the case of risk clustering.54 In fact, more compli­cated approaches have shown to underperform the standard market model ab­stracting from these issues.55

Another sensible issue is that of event-induced variance. This is particularly rele­vant in the context of M&A-transactions since the mere announcement is not a guarantee for final accomplishment. White knights, antitrust-authorities or de­fence mechanisms launched by targets might threaten the deal. This uncertainty potentially increases the variance around announcements, i.e. within the event window. Again test-statistics might overreject. However, approaches to tackle this issue and verify the robustness of results are available.56 To sum up, within the short-horizon framework, problems related to statistical properties can be dealt with and the standard market model has shown to be the preferred approach.57

Nevertheless, inefficient markets and the anticipation of deals might bias short- horizon event studies that focus on announcement date returns. Moreover, it is conceivable that complex M&As are surrounded by uncertainty on the day of their announcement. When this uncertainty vanishes over time as additional terms and conditions or prospects of success become apparent, stock price adjusts ac- cordingly.58 Furthermore, apart from information leakages, insider trading can cause a run-up of abnormal returns as well. Consequently, even if markets are efficient and information did not leak to the public in advance, pre- and post­announcement abnormal returns can occur. Table 1 summarizes potential biases, inherent in the short-horizon event study methodology:59

illustration not visible in this excerpt

Table 1: Potential biases of short-horizon event studies

Source: Own illustration.

The third bias reported in table 1 cannot be addressed. But in order to deal with the leakage and uncertainty bias, events windows can be shifted to other than the announcement date.60 For example, Houston and Ryngaert (1994) use an informa­tion leakage date to adjust their event windows.61 Recently, Duso et al. (2006a) employ the date on which the European Commission decides upon the imposition of merger remedies. This allows for insights into the outcome of the bargaining process between authority and merging parties.62 But if the precise moment of the arrival of new information cannot be pinpointed, the event window can be ex­tended, for instance, by a couple of days prior to the announcement to capture potential insider-trading effects. In fact, short-horizon event studies generally in­clude a couple of days or weeks around to the event date to cater for such effects.

However, if post-event abnormal returns are dispersed over long time periods ei­ther because markets are inefficient or uncertainty regarding the merger’s pros­pects of success and its terms and conditions vanishes only gradually, this war­rants the extension of event windows to cover a range of one year or more.63 Such long-horizon studies are presented and examined next.

2.2.3 Long-horizon event study methodology

Long-horizon models embrace basic concepts outlined above as well as new mod­els to measure abnormal returns and calibrate test-statistics. Recall that for short- horizon models abnormal returns are cumulated if the event window contains more than one period of time. Likewise it is possible to compute long-horizon CARs of event windows that span over one year or more. In this case periodic abnormal returns usually do not refer to days but rather to months. However, the procedure of cumulating short-term returns over long intervals is problematic if measurement errors in securities’ returns occur. Especially the bid-ask spread in closing prices has been shown to upward bias single-period stock returns, in par­ticular those of low-priced firms.64 When cumulating abnormal returns over long intervals, the measurement error adds up. A way to mitigate this cumulation bias is to calculate buy-and-hold abnormal returns (BHAR) 65

There are two ways to obtain long-horizon BHARs. One is to compound (monthly) abnormal returns Ara instead of cumulating them.66 Hence, in contrast to equation (8) the BHAR of security i is calculated by:67

illustration not visible in this excerpt

Alternatively, BHARs can be defined as the prespecified multi-period com­pounded return difference between two investment strategies: buying and holding an event firm i compared to a control firm b or a benchmark portfolio bp that did not experience the event.68

illustration not visible in this excerpt

Control firms are identified and reference portfolios are constructed by matching companies on certain characteristics such as size or market-to-book ratios that are presumed to proxy for the expected return on a security.69 Thus, this measurement methodology is also referred to as the characteristic-based matching approach70

Following equation (9), once security-specific BHARs have been computed, they are averaged using equal- or value-weights gt :71

illustration not visible in this excerpt

In fact, cross-border M&As allow to hedge macroeconomic risk and thus to re­duce the merged company’s systematic risk. This entails that the estimated pa­rameter for risk-adjustment (in the market model captured via β) changes due to the event. Kiymaz and Mukherjee (2001) illustrate that within the market model, cross-border mergers might significantly reduce the post-event β compared to pre-event-estimates.72 Whereas short-horizon studies are not biased by a crude estimate of β, long-horizon models are likely to be misspecified73 To address this bias, post-event or pooled risk parameters can be estimated for the calculation of ARiT in equation (11). Nevertheless, the hazard of risk/parameter-shifts sup­ports the characteristic-based matching approach since no parameter estimation is required to calculate BHARs according to equation (12).

Another procedure to circumvent the risk shifting bias is to calculate calendar­time abnormal returns (CTAR)74 In each calendar month an event portfolio is created that contains all firms that experienced M&A-transactions within the pre­vious E months (length of the event window). The portfolio is rebalanced each month to include newly-merged entities and to remove firms that reached the end of the E month period.75 The CTARs are given by:76

illustration not visible in this excerpt

If we substitute CTAR, by the regression constant ap and use the CAPM77 to approximate E (r) , equation (14) can be rearranged to its usual form:

illustration not visible in this excerpt

This expression reveals that the time-series of monthly equal or value-weighted excess returns of portfolio RpT over the risk-free rates RfT is regressed on the CAPM market factor. The intercept of the regression line ap (also know as Jen-sen-alpha) denotes the measure of average monthly calendar-time abnormal re­turns of the portfolio of event firms over the E month post-event period.78 If M&A-deals do not increase corporate value ap would equal zero.

Having explained the basic concepts of the long-horizon return metrology, critical issues are discussed next. Generally speaking, within the long-horizon framework, it is more difficult to obtain unbiased estimates for abnormal returns and their estimated standard deviation, i.e. for both, the numerator as well as denominator of test-statistics. Along these lines Kothari and Warner (1997) state that ”long-horizon event studies involve many related considerations that do not arise or are less important with short-horizons.” 79

Although long-horizon event study models do not require efficient markets or events to be unanticipated, we might view the assumption of no confounding ef­fects as most critical. Especially large companies might experience a variety of events that influence stock prices. Consequently, longer event windows increase the likelihood that abnormal returns are confounded by other corporate incidents.

Moreover, CARs, BHARs and CTARs might be subject to a new listing bias. Whereas sample firms have long post-event histories of returns, benchmark indi­ces (proxying for the return of the benchmark portfolio) might be altered by new stock listings occurring during the long-horizon event window. If newly-listed firms underperform the market, then the return on the benchmark index is biased downwards and hence long-run abnormal returns biased upwards.80 A way to cir­cumvent this problem is to model expected returns via, for example, size-decile or book-to-market portfolios that by definition require prior data to be constructed.81

In addition, the utilisation of an equally-weighted market index might introduce a rebalancing bias to the calculation of BHARs since the compounded returns on an index are computed assuming periodic rebalancing whereas returns on sample firms are compounded without rebalancing.82 The preservation of equal security- weights in an index demands that winner stocks are sold and losers bought at the end of each period. As returns reverse, these rebalances result in an inflated return on the benchmark index.83 Consequently, BHARs are deflated as the return on benchmark portfolio is inflated. To circumvent this bias, non-rebalanced, i.e. value-weighted benchmark portfolios can be employed.84 However, such a proce­dure might be inconsistent with the general event study conception which attrib­utes equal- rather than value-weights to sample observations.85

Another drawback of BHARs concerns the selection of the non-event benchmark portfolio and control firm. Banks undertaking M&As can be systematically differ­ent from those choosing other modes of expansion like de novo branching. A se­lectivity bias arises because benchmark returns do not only capture returns in the absence of an M&A-transaction but also entail value-enhancing non-M&A- strategies. Thus, improvements from consolidation might be understated.86

Next to the issue of modelling long-horizon abnormal returns, statistical proper­ties are analysed which affect the denominator of test-statistics. Long-horizon event windows expose themselves to both, calendar-time and industry clustering that potentially introduce cross-correlation of CARs and BHARs. Already minor cross-dependences lead to overreactions of the null hypothesis due to the down­wardly-biased estimate of the standard deviation of cross-sectional abnormal re- turns.87 It is impossible to eradicate this bias since the number of return covari­ances to be estimated exceeds the number of time-series observations.88 CTARs are unbiased in this regard since the cross-correlation is reflected in the variation of the time-series of monthly excess returns of the rebalanced portfolios.89

Another bias might arise due to the delisting of sample firms.90 If the sample size decreases, the portfolio’s return-volatility might be increased.91 Hence, the esti­mated variance of the observation window might understate the true variance of the post-event sample which in turn biases test-statistics to overreject the null­hypothesis. A way to mitigate this drop-out bias is to estimate the (higher) vari­ance from the time-series of event window abnormal returns. However, as the true event window variance is not constant but changes as drop-outs occur, this esti­mate might be too high for the event month when the sample is still complete.92

In particular BHARs exhibit cross-sectional distribution properties that are sig­nificantly skewed 93 This problem arises as the lower bound for returns is -100% whereas the upside is unbounded. Moreover, individual sample firms are more volatile than the benchmark portfolio, i.e. it is more common for a sample firms to have a return of more than 100% whereas this is rather exceptional for an index return. Thus, abnormal returns, measured as the difference between the two, are positively skewed which results in downwardly biased test-statistics.94 Lyon et al. (1999) propose two methods that remove skewness biases.95 One is to employ a bootstrapped skewness-adjusted t-statistic. Secondly it is possible to substitute each sample firm by a control firm that matches the sample firm on certain char­acteristics to construct a pseudo-portfolio.96 Repeating this process yields a large number of simulated matched-sample portfolios of which the BHARs are calcu­lated in order to estimate an empirical distribution of long-horizon abnormal re­turns. Alternatively, Cowan and Sergeant (2001) recommend to winsorise abnor­mal returns to rule out distortions arising from extreme observations and compute two groups test-statistics.97 However, bootstrapping and pseudo-portfolio-based test-statistics assume independence of event-firm abnormal returns which is vio­lated in presence of clustering effects.98 Likewise, Cowan and Sergeant (2001) find that clustering renders all advocated test-statistics misspecified.99 Conse­quently, even if more sophisticated procedures are used, the null hypothesis of zero abnormal returns might still be overrejected.100

illustration not visible in this excerpt

Table 2: Potential biases of long-horizon event studies

Source: Own illustration.

In view of the multiple sources of biases (see table 2) the question arises whether the long-horizon approach constitutes a valuable alternative to short-horizon mod­els. Whereas the assessment of the short-horizon methodology points to the effi­cacy of the market model, there is no such recommendation within the long- horizon framework. BHARs in conjunction with non-parametric procedures such as bootstrapping have initially been recommended.101 Subsequent papers, how­ever, discard this notion and recommend the usage of CTAR.102 But as seen above this approach also incorporates potential biases. Barber and Lyon (1997) advocate the matched-characteristic control firm approach because of its potential to elimi-

[...]


1 Throughout this paper the terms merger and acquisition are used synonymously.

2 See European Central Bank (2007), p. 35.

3 Berger et al. (1999) refer to static and dynamic approaches synonymously, see pp. 152 et seqq.

4 See Berger et al. (1999), pp. 136-137.

5 See Berger and Hannan (1998), pp. 455-456.

6 See Berger and Hannan (1998), p. 464.

7 See Berger and Humphrey (1994), p. 22.

8 See Pilloff and Santomero (1998), p. 64.

9 See Pautler (2003), pp. 166-167.

10 See Tourani Rad and Van Beek (1999), p. 533 for an overview.

11 See Pilloff and Santomero (1998) for an overview of U.S. evidence.

12 See Vander Vennet (2002) or Pilloff and Santomero (1998) for reviews of the related literature.

13 See Calomiris and Karceski (1998); Rhoades (1998); Mußhoff and Schiereck (2005).

14 See Binder (1998), p. 112.

15 See Campbell et al. (1997), pp. 151-152.

16 See Kothari and Warner (2006), p. 7.

17 See Peterson (1989), pp. 39 et seqq. and MacKinlay (1997), pp. 17 et seqq.

18 As for event windows, the time span of observation windows varies substantially across studies,see Rhoades (1994), p. 5. However, this variation is less subject to debate.

19 See Peterson (1989), p. 41.

20 See MacKinlay (1997), p.18.

21 For that reason this model is also known as the market-and-risk-adjusted model, see Brown and Warner (1980), p. 206.

22 Analogous to equation (3), the regression-specification used to estimate ß in the observation

23 See Binder (1998), p. 118.

24 See Black (1972), p. 444 for an overview.

25 See Banz (1981), p. 16.

26 See Fama and French (1996), pp. 75 et seqq. for further anomalies troubling the CAPM.

27 See Fama and French (1992), p. 428.

28 See Fama (1998), pp. 292-293.

29 Note that the market model exemplifies a one factor model.

30 See Fama and French (1993), p. 51.

31 See Flannery and James (1984), p. 1152.

32 See MacKinlay (1997), p. 18 and Bessler and Murtagh (2002), p. 462 in reference to the banking

industry.

33 See Houston and Ryngaert (1994), p. 1161.

34 Note that the order is irrelevant.

35 The notation follows the single-entity perspective.

36 See MacKinlay (1997), pp. 21; 24.

37 See Kothari and Warner (2006), p. 11.

38 A common modification is to standardise each abnormal return, using an estimator of its stan­

dard deviation, see MacKinlay (1997), p. 24. Frequently the Dodd and Warner (1983) z- statistic has been used, see appendix 1.

39 See Bromiley et al. (1988), p. 29. The semi-strong efficient market hypothesis states that prices

reflect all publicly available information regarding the security.

40 See De Bondt and Thaler (1985), p. 793.

41 See Modigliani and Cohn (1972), p. 24.

42 Section 2.3.2 explains why increased financial leverage is often found after bank mergers.

43 See McWilliams and Siegel (1997), p. 630.

44 See Calomiris (1999), p. 616.

45 See Calomiris and Karceski (1998), p. 26.

46 See McWilliams and Siegel (1997), p. 637 for further proposals.

47 See Brown and Warner (1985), pp. 25 et seqq.

48 Brown and Warner (1985) show that for a sample size of 50 securities mean excess returns are

close to normal, see p. 10. Alternatively, nonparametric test can be employed that do not rely on assumptions regarding the distribution of abnormal returns, see MacKinlay (1997), p. 32.

49 See Brown and Warner (1985), p. 19. Non-synchronous trading activity also does not seem to notably bias OLS-market-model-parameters, see p. 16. Therefore, any adjustments of OLS- parameters to account for non-synchronous trading according to Scholes and Williams (1977) can be deemed unnecessary, see Beitel and Schiereck (2001), p. 10.

50 See Brockett et al. (1999), p. 210.

51 See MacKinlay (1997), p. 24.

52 See Kothari and Warner (2006), p. 13.

53 See also Boehmer et al. (1991) p. 266 for additional evidence.

54 Brown and Warner (1980) use the term risk clustering which can be viewed as a proxy for in­ dustry clustering, see Dyckman et al. (1984), p. 23.

55 See Brown and Warner (1980), p. 249.

56 See Boehmer et al. (1991), pp. 255-256 for an overview.

57 See also Cable and Holland (1999) who confirm the supremacy of the market model, see p. 339.

58 See Caves (1989), p. 154.

59 Note that confounding events also constitute a bias. However, short-horizon studies’ advantage is that the likelihood of other non-transaction-specific abnormal return influences is mini­mized.

60 The key-criterion is to capture the arrival of new information.

61 See Houston and Ryngaert (1994), p. 1160.

62 See Duso et al. (2006a), p. 2.

63 Note that market inefficiency refers to the valuation of the M&A-transaction under investiga­ tion. The assumption of efficient markets is still required to assume that previous corporate events are fully reflected in stock prices. Otherwise abnormal returns are persistently distorted by the delayed valuation of past events. An extension of the event window would be useless.

64 See Conrad and Kaul (1993), p. 40.

65 See Roll (1983), p. 374 and Conrad and Kaul (1993), pp. 46-48.

66 See Kothari and Warner (1997), p. 313. The authors also note that both BHAR measures yield similar results, see p. 313.

67 Recall that the index i refers to securities of acquiring firms. Separate calculations are necessary to gauge the value effects of target firms j or the combined entity i u j, see section 2.2.1.

68 See Mitchell and Stafford (2000), pp. 296-297.

69 See Barber and Lyon (1997), pp. 353-355.

70 See Kothari and Warner (2006), p. 27.

71 See Mitchell and Stafford (2000), p. 297.

72 See Kiymaz and Mukherjee (2001), p. 263.

73 See Kothari and Warner (2006), pp. 25-26 for a numerical example.

74 See Abhyankar and Ho (2007), p. 70.

75 The varying number of portfolio firms introduces heteroskedasticity that needs to be controlled for, see Franks et al. (1991), pp. 88-89 and Mitchell and Stafford (2000), pp. 316-317.

76 See Mitchell and Stafford (2000), p. 318.

77 Alternatively, the Fama and French three-factor asset pricing model can be used, see Kothari and Warner (2006), p. 30 or Abhyankar and Ho (2007), p. 70.

78 See Kothari and Warner (2006), pp. 30-31.

79 Kothari and Warner (1997), p. 303.

80 See Ritter (1991), pp. 4 et seqq. for evidence of underperforming initial public offerings.

81 See Barber and Lyon (1997), p. 361.

82 CARs are not subject to this bias since cumulating abnormal returns of sample firms implicitly rebalances winner and loser stocks periodically, see Conrad and Kaul (1993), p. 40.

83 See Barber and Lyon (1997), pp. 348-349 for explanations of the negative autocorrelation im­ plied by return reversals.

84 See Mitchell and Stafford (2000), p. 298.

85 See Barber and Lyon (1997), p. 354.

86 See Calomiris (1999), pp. 616-617.

87 See Kothari and Warner (2006), p. 35.

88 See Fama (1998), p. 295.

89 See Fama (1998), p. 295.

90 This might be particularly likely for a sample focusing on M&A-targets.

91 See Kothari and Warner (1997), p. 324.

92 See Kothari and Warner (1997), p. 326.

93 These are less pronounced for CARs, see Kothari and Warner (1997), p. 312.

94 See Barber and Lyon (1997), p. 347.

95 See Lyon et al. (1999), p. 166.

96 See Lyon et al. (1999), p. 175.

97 See Cowan and Sergeant (2001), p. 762.

98 See Mitchell and Stafford (2000), pp. 290-291.

99 See Cowan and Sergeant (2001), p. 758.

100 See Kothari and Warner (2006), pp. 38-39 and Mitchell and Stafford (2000), pp. 307; 324.

101 See Kothari and Warner (1997), p. 337.

102 See Fama (1998), p. 295 and Mitchell and Stafford (2000), p. 326.

Details

Pages
91
Year
2007
ISBN (eBook)
9783640817092
ISBN (Book)
9783640820887
File size
858 KB
Language
English
Catalog Number
v165733
Institution / College
University of Münster – Finance Center Münster
Grade
sehr gut
Tags
M&A value effects event study case study UniCredit-HVB merger

Author

Previous

Title: Measuring M&A-Success in Cross-border Banking M&A