UNIVERSITY OF AMSTERDAM
Faculty of Economics and Business
Master Accountancy & Control
Specialization Accountancy
Master Thesis, 17th June 2015
The rationality of stock market
reactions after a major audit failure
An event study on the reputation effect of the Tyco audit failure
By:
Ivo de Lange
10668039
Supervisor :
Dr. A. Sikalidis
2
Statement of originality
This document is written by student Ivo de Lange who declares to take full responsibility for the contents 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 contents.
3
Abstract
Background – Prior studies have shown that an auditor’s reputation is directly related to the perceived and actual levels of quality reflected by the auditor’s report. Most of these studies focus on audit quality at the moment of an IPO, or when an qualified opinion is issued. In this thesis, I study the reputation effect at the moment of an audit failure, namely the accounting fraud by Tyco in 2002 and the stock effect it had on other clients of the PwC. I focused on this aspect as this field of research is not widely explored in prior literature.
Purpose – The purpose of this paper is twofold. First, in order to provide additional support for prior studies in this field, this paper investigates whether S&P-1500 companies audited by PwC suffered negative economic impact due the audit failure concerning PwC’s work at Tyco. Secondly, This study aims to clarify how rational, in the sense of the efficient market hypothesis, investors behave with focus on the location and industry of the concerning company.
Design / Methodology / Approach – In order to investigate this topic, a quantitative case study is conducted on the Tyco accounting fraud events. The data for this study is based on all S&P-1500 constituents for 2002 and 2003. For this I used two separate Compustat databases, with the addition of hand-collected data for the two moderating variables. For the first stage, the cumulative average abnormal returns are calculated per selected event, these are tested with t-tests for significance. For the second stage, a combination of a multivariate and an ordinary least square regression is used.
Findings – The results of the CAAR analysis suggests that PwC- clients suffer significantly during the events closely related to the audit failure. Especially the first major press release, that concerned Tyco’s looted KELP fund shows a negative abnormal return between the 2.8% and 6.7%. The second stage of this research provides support for the hypothesis that the strength of this reputation effect decreases as the distance of a company towards Tyco’s headquarters increases. Lastly, I find no conclusive support for the hypothesis that this effect is influenced by the industry a company operates in.
Originality – This study is relevant for two reasons. Firstly, it adds knowledge to both actual as perceived audit quality, in which perceived audit quality is an underexposed part of the auditing research. But this research also gives more insight on the ‘fading-way’ of the reputation effect over distance. This concludes that, to some extent, investors are rational in reacting to bad audit quality.
Limitations – This thesis is subject to limitations. It appears that the used Market Model is a blunt instrument to measure a relative small effect as the reputation effect. Thereby the used models do not have excellent explanatory power, as the CAARs are basis for these regressions. Keywords: Reputation effect; Audit failure; Tyco; PwC.
4
Table of Contents
Abstract ... 3 Table of Contents ... 4 1 Introduction ... 5 1.1 Background ... 6 1.2 Research question ... 61.3 Motivation for this study ... 7
2 Literature review and hypotheses ... 8
2.1 Auditor reputation and audit quality ... 8
2.2 The stock market as a proxy for audit quality ... 10
2.3 Moderating factors for actual audit quality ... 11
2.3.1 Location of the audit failure ... 12
2.3.2 Industry of the audit failure ... 13
2.4 The Tyco events ... 14
2.5 Hypothesis development ... 17
3 Research methodology ... 19
3.1 Sample ... 19
3.2 Research design ... 20
3.2.1 The Cumulative Average Abnormal Return... 20
3.2.2 Multivariate analysis ... 22
3.2.3 Control variables ... 24
4 Results ... 26
4.1 Descriptive statistics ... 26
4.2 Cumulative Average Abnormal Returns ... 30
4.3 Multivariate regression analysis ... 32
4.3.1 The location hypothesis ... 35
4.3.2 The industry hypothesis ... 36
5 Conclusion ... 38
References ... 40
Appendix A: Predictive validity framework ... 43
Appendix B: Variable definitions ... 44
Appendix C: Event date, window and estimation period. ... 45
Appendix D: CAAR overview for each event. ... 47
Appendix E: MANOVA Analysis ... 52
5
1 Introduction
It was the major audit failure of Arthur Andersen at Enron that first exposed the extensive effects of poor audit quality at the stock markets. After this scandal it became apparent that not only Enron itself suffered from their fraud, but other clients of Enron’s auditor; Arthur Andersen experienced negative returns as well (Chaney and Philipich, 2002).
Since then the litigation and reputation hypotheses gained more academic support. In this research I shed more light on the investor’s reactions after a sudden change of auditor reputation. I do this by a stock price analysis during the audit failure of PricewaterhouseCoopers (from here PwC) at Tyco in 2002. Further, I focus on moderating variables for location and industry to measure to what extent investor’s act rational towards an audit failure.
Tyco International Ltd. (later Tyco) is currently a service focused manufacturer of security systems and fire protection equipment, traded as solid dividend share on the New York Stock Exchange under the ticker TYC (tyco.com, 2015). But until the year 2002 this company was the largest conglomerate in the U.S., operating within multiple industries and on markets all over the world. It had reached this status by perusing an aggressive acquisition strategy from the beginning of 1991. From this date the company acquired over a 1000 other smaller firms, and from 1999 also some remarkable large companies listed in the S&P-500. Until 2002 Tyco rode the bull markets, and it was a leader on it; investor and analysts frequently referring to it as; “Tyco the M&A machine” (Wall Street Journal, 30-01-2002).
The process of acquisition continued into 2001 and the board had issues controlling the organization. Due to the growing complexity of Tyco’s subsidiaries, it planned to split the company into four separate stock traded organizations in January 2002 (Wall Street Journal, 05-06-2002). But after a significant drop of share price and downgrading of ratings, this was considered too harmful for the companies’ value and was canceled. When news releases around the Enron scandal emerged, the Tyco stock suffered significantly, this was due the fact that the two giants shared a common strategy.
As the fundamental problems within the organization were not solved, the scandal surrounding the company’s CEO Dennis Kozwalski became public eventually. The signs of the fraud surfaced in July, when there was an indictment on tax evasion on Kozwalski. And by September that year, it became public that Kozwalski and his partners used over $250,- million of company funds for private benefit, in addition to this accounting legislation was ignored or offended. During this scandal the involvement of PwC as Tyco’s accountant was revealed on multiple occasions. This makes the Tyco fraud an excellent case for an event study on the reputation effect.
6
1.1 Background
Global stock markets have a need for high quality financial information. Assurance by external auditors over this information plays an important role. Lennox (1999) states that auditors are willing to provide high quality external audits because of litigation and reputation –incentives. First, auditors have an incentive to provide high quality audits because they can face litigation costs when they are engaged in an audit failure. According to the deep-pockets theory, this insurance role plays a larger role when an organization is audited by a large audit firm. The insurance role is especially more important when this occurs in a high litigation environment. Second, an auditor has the incentive to provide high quality audits, as it wants to protect its reputation as a high quality auditor. The deterioration of its reputation makes it harder to demand a quality (fee) premium to its clients. It can also lead to a misfit between auditor and client based on the demand of high audit quality, resulting in the client switching auditor. This research will focus mainly on the reputational effects of an audit failure.
A prior research about this topic by Weber, Willenborg and Zhang (2008) studied the stock and audit market effects associated with a widely publicized accounting scandal at the German company ComRoad AG. They found that the stock markets reacted negatively to that audit failure. Also, they found that clients that changed auditor after the scandal experienced a positive stock market return. This provided evidence that the reputation effect also is measurable in a reversed setting. A similar construct by Skinner and Srinivasan (2012) in Japan led to strong evidence in a low litigation environment. Through these studies the authors have separated the reputation effects from insurance effects.
1.2 Research question
In this research I want to assess if there is a measurable amount of rationalism in stock markets reactions that reflect reputational damage. The efficient market hypotheses describes that an audit failure should lead to less decision usefulness of the (audited) annual reports, when an organizations is audited by the same audit firm (Fama, 1970). A lower decision usefulness forces investors to demand a higher risk premium. This will lead to higher interest costs, but also to a lower stock price which is measurable in this study.
7
The first part of this research will focus on testing the direct stock market effect on the days that critical information on PwC’s audit failure at Tyco was released. Then, this research will specialize in the investor focus of the reputational theory, by assessing if certain groups within the stock market suffer more severe damage. This is done by testing if this effect is similar or different from what the literature suggests, based on the audit office location or on associated industries perspective. Based on these clarifications the research question of this study will be:
RQ: “How rational are the stock market reactions on (other) PwC clients after the audit failure at Tyco.”
1.3 Motivation for this study
This research will contribute to this prior research and the social construct in two ways. The first aspect is to provide confirmation on the reputational effect of an audit failure by testing the effect on the stock market. Secondly, this research will focus on the rationality of investor behavior, this is contrary to prior studies that are devoted to the audit firm perspective. Analysis of prior studies concludes that stock markets do react on audit failures, mostly through a direct stock price decline of other clients served by the same auditor (Weber, Willenborg and Zhang [2008], Skinner and Srinivasan [2012], Chaney and Philipich [2002]). But these do not provide an explanation to what extent investors react rational towards the efficient market hypothesis.
But since the beginning of this millennium there has been a lot of criticism on the efficient market hypotheses, because it doesn’t represent human irrationalities (Bradshaw, 2011). This article also provides evidence that not only investors, but also analyst tend to suffer from overconfidence, overreaction, representative bias and information bias. Earlier researches by Sloan (1996) into the use of extreme accruals by companies, found evidence that investors also have difficulties to incorporate all available information into the valuation process. They stated that investors only tend to focus on primary aspects, for example earnings per share or profit increase.
This research will continue on both of the above aspects and will try to answer the uncertainty about investor behavior during audit failures. By this I can create some understanding if investors have access to all available information and implications that (un-) official press releases reflect, and if they incorporate them into the stock price.
8
2 Literature review and hypotheses
In the previous section the scope of this research has been confined to the reputational effects of an audit failure. This section will discuss fundamental knowledge on audit firms and investors incentives and explain that they have a common interest in high quality audits. The second paragraph will present the theoretical underpinning of the research; the reputation and litigation theory. Then, I will clarify that there is a link between stock price and audit quality. This is based on prior studies in this field, but also evidence from initial public offerings. The third paragraph focusses on literature regarding the two mediating factors that are central in this research; audit office location and audit industry segment. This section will then elaborate the events used in this study. Finally, I will conclude with clearly formulated hypothesis for this research.
2.1 Auditor reputation and audit quality
The most widely used definition of audit quality is by De Angelo (1981, p.186): "The market-assessed joint probability that a given auditor will both (a) discover a breach in the client's accounting system and (b) report the breach.” This definition consists out of two factors that do frequently return in other definitions. First, the likeliness that the competence of the auditor leads to discovering the misstatement. Secondly, the lack of objectivity and independence an auditor has that decreases the likeliness that he will successfully report the misstatement. De Angelo’s explanation has been criticized as it does not longer cover the full scope of audit quality (Knechel, 2009). Although, it is still used in research often, it has become a less useful definition for standard setters and legislators.
This definition is still suitable for this research, but with additions of a second aspect from prior research. Based on the investor perspective there is a difference between actual and perceived audit quality. Jackson, Moldrich, and Roebuck (2008, p.422) described actual audit quality as “the degree to which the risk of reporting a material error in the financial accounts is reduced”. The counterpart, perceived audit quality is defined as “how effective users of financial statements believe the auditor is at reducing material misstatements” (p. 422). This second aspect is not measurable by direct audit quality research, as it contains an immeasurable investor ‘belief’ or ‘rationality’ it is only studied through analysis of stock price effects.
The relevance of audit quality has been proven in prior studies, these often focus on real audit quality. These researches provide evidence for the assumption that large auditors provide higher quality audits and through this, offer more added value to a clients’ financial statement than small auditors do. For instance, audit opinions of large auditors contain more accurate signals of financial distress in companies (Lennox, 1999). Also, companies that switch from a small to a large
9
auditor experience a positive stock market reaction (Moizer, 1997). These effects explain that perceived audit quality creates an indirect incentive for auditors to have a good track record and to be regarded as a high reputation auditor.
Francis, Reichelt and Wang (2005) find that auditors also have a direct incentive to be regarded as high reputation auditor. They studied auditor fees from U.S. based auditors from the period 2000 to 2002 in combination with auditor reputation. They found that auditors from the big-five firms had a significant fee premium of 19 percent on national basis. In addition they found significant results while testing the top-ten ranked auditors on a city basis, in combination with auditors’ industry-leadership. The authors conclude that auditors attempt to maintain a flawless reputation as it is associated with high audit quality that gives increased possibility for an audit fee premium.
Lennox (1999) describes this phenomenon in two ways; first there is the reputation hypothesis, which is the focus of this study. He explains that auditors fear the loss of future quasi rents, these are future rewards for currently performed work but without any guarantee that they will receive them. An auditor tries to maintain a flawless reputation, so that their client has no quality based incentive to switch auditor and the previous investments can be retrieved. Second, he explains the insurance hypothesis; according to the deep-pockets theory investors pressure their investees to be audited by a large audit firms, not only because they have higher audit quality, but moreover because they can meet the investor’s litigation requirements (Dye, 1993). In return the large audit firms asks a premium fee to cover this risk. The paper by Lennox (1999) provides evidence large auditors receive significantly more litigation costs after an audit failure than small auditors or auditors in a low litigation environment do, due to audit failures.
This research is performed on a sample based in the U.S., this is considered as a high litigation environment. In the previous section it was elaborated that the direct litigation effect was considered low because litigation at Tyco began after several years. The next paragraph will discuss that prior studies separated the two type of effects, through a study in a low litigation location. Although, it is considered that the insurance effects are not directly measurable to the same extent as the reputation effects, I cannot exclude all elements of it out of my sample. Therefore, I will also partly reflect on this part.
To conclude, it seems that both actual as perceived audit quality is intertwined with the auditor’s reputation. Prior research and practice shows that auditors can encounter significant losses due to decreasing audit quality, both due litigation as through reputation. It is also evident that investors have interest in high reputation auditors as this fosters the audit quality, leading to lower uncertainty in the annual accounts.
10
2.2 The stock market as a proxy for audit quality
The foundation of the used proxy will rely on the efficient market hypothesis (EHM). According to Fama’s theory (1970) a market is efficient when the market price always fully reflects all available information. He also states that when new information is released the market quickly adjusts to this new information. Although the EMH is not functional and flawless to full extent, there is no major event registered that did not had a direct effect on the stock market. In his overview paper, Bradshaw (2011) states that criticism on the EHM is mostly focused on the long-term. This is caused as analysts and investors tend to experience far more extensive uncertainties at interpreting future revenue, then they do at interpreting future direct costs.
Fama (1970) distinguishes efficient markets in three ways: the weak, the semi-strong and the strong form. The global stock markets have a status somewhere between the semi-strong and the strong form: All publicly available information is reflected in the stock price (semi-strong) and some private information, that is less easily accessible, is also incorporated in the stock price (strong). According to Healy and Palepu (2003, p. 1): “A well-functioning capital market creates appropriate linkages of information, incentives and governance between managers and investors. This process is supposed to be carried out through a network of intermediaries”. These intermediaries are professional investors, information analysts and assurance professionals. Based on this definition of the EMH a failing intermediary, for example the assurance professional (i.e. auditor) can have direct influence on the market through the EMH.
The use of the stock market as a proxy for audit quality is indirect and thereby contains more noise than a direct model such as the Modified Jones model, despite of the stock market is still more suitable for this research. As the reputation hypothesis is based on the definition of audit quality by Jackson, Moldrich, and Roebuck (2008) it contains both actual and perceived audit quality. By using a direct model it is not possible to capture the perceived audit quality in the study, which constrains us to the use of an indirect measure. Prior research used the stock markets in order to interpret audit quality in multiple times, the following studies below give guidance to this research.
A paper by Schaub (2006) studied the stock market reactions after an audit firms releases a going concern opinion on a client. He finds evidence that investors over-reaction is a stock market anomaly, because it is against the efficient market hypothesis. But he also finds that the stock market corrects itself in the coming days, as nearly 70% of the losses on the announcement date are recovered in the next five days. The author describes the over-reactions as the tendency of investors to directly drive the stock prices too low when bad news is made public and too high for good news. For my research it would mean that a direct stock market reaction would be higher than a reaction over longer term, but with less explanatory power.
11
In prior research there are different settings where investor’s valuation of (perceived) audit quality is tested on the stock market. A study by Beatty (1989) shows that there is an inverse relation between auditors reputation and the direct return, when investing in the initial public offering (IPO). She concludes that involving a nationally-known audit firm will increase the price received by the initial public offering (IPO) client. A more recent study by Albring, Elder and Zhou (2007) shows that the services of a non- big five quality differentiated auditors results in higher audit fees, but also in less underpricing during an IPO. They also find a significant effect on underpricing when the audit firm has SEC reporting experience, suggesting that investors value this as positive on audit quality. These two researches provide evidence that there is a direct link between stock price and audit quality.
But prior studies with focus in the same field as this research; Audit failures and the stock market, are also common. One very influential study by Chaney and Philipich (2002) was based on the Enron and Arthur Anderson collapse. This study focused more heavily on the Arthur Andersen events and proved that investors reacted negatively towards evidence of shredding documents and filing default. It was criticized as it did not fully focus on the Enron events, but moreover on events of Arthur Andersen. Later research focused more on the low litigation environment, as the study by Weber, Willenborg and Zhang (2008) and the study by Skinner and Srinivasan (2012). They provided evidence that audit failures also had significant stock market consequences in a low litigation environment and presumably these effects were even stronger. But these studies did not include location effects in their scope, like the study by Chaney and Philipich.
In summary, it seems that investors rely heavily on audited information. Also it appears that they do take audit quality into account when making economic decision, as seen during IPO’s. Earlier research with focus on audit failures offers fundament to expect a negative stock market reaction due reputational damage at certain critical events. Yet, there is little information available about a possible industry spillover effect, or the geographical extent to which this occurs. By using the stock market as a proxy, I can gain more insight in reputational effects. Based on both actual audit quality as on how it is perceived by investors.
2.3 Moderating factors for actual audit quality
In this literature review I recognize two possible moderating factors for audit quality. Prior research provides evidence that there are significant inequalities on audit quality between locations of an audit firm’s offices. It also suggests that audit quality differs between industry segments, due to an industry specialization effect. Prior stock market based research provide evidence that investors incorporate actual and perceived audit quality, but they give little insight on these moderating factors.
12
2.3.1 Location of the audit failure
There are two aspects of the reputation effect on the stock market that are separately recognizable. First the correction of the lower actual audit quality that was overstated by the market, this should be the most significant at the failing audit team. And second the reduced value of the brand as high quality auditor in the total market, which I expect to be likewise for the whole audit market. Although these effects cannot be measured separately, the effects can be monitored as a whole. Thereby, I expect the reputation effect to ‘fade away’ over the increasing distance between the audit failure and the auditor’s client.
Prior research provides us with information about audit quality differences between audit firm offices. Francis, Michas and Yu (2013) produced a key study through assessing the amount of client restatements at big-four offices. They provide evidence that larger offices of Big 4 accounting firms produce higher quality audits relative to their smaller offices. The reason for this is that large offices have fewer obstructions at providing client restatements. An earlier research from the same authors provided consistent evidence with earnings quality (Francis and Yu, 2009). It appears that especially large auditing firms have major inconsistencies of audit quality between offices. This suggests that audit failure provides evidence of low audit quality, but this cannot directly be assumed for other locations.
Prior studies in the same setting as this research do only cover the travel effect of (perceived) audit quality to some extent. Chaney and Philipich’ (2002) research after the Enron scandal, tested stock market returns of other Arthur Andersen clients. They also provided evidence that the Houston office suffered more severe than other U.S. located Arthur Andersen offices. Although they found significant differences they could not support any theory, due to the fact that possible explanations were inconsequential. Their case also had issues with the major punishment effect on Arthur Andersen “shredding date”. Nonetheless, the authors did describe the phenomenon of investor rationality within the reputational effects.
Another study about the Enron and Arthur Andersen events provides the reputation effect with boundaries towards the ability to travel. Barbara and Martinez (2006) reproduced the earlier mentioned study of Chaney and Philipich in Spain, but they found no significant results. They conclude that investors viewed the Enron events as a U.S. - based affair, and that investors were aware of the minor influence on the Spanish division of Arthur Anderson. A few years later Cahan, Emanuel and Sun (2009) did a replication of this study on global basis on two event dates of the Enron affair. To the contrary of Barbara and Martinez, they find negative and significant cumulative abnormal returns for both their sample in the U.S. as on global based sample. This does suggest that reputation and audit quality does spill over to other countries outside the U.S.
13
Although multiple prior studies have been performed on this subject, there is a lack of consistent evidence. It appears that investors tend to handle an audit failure with a combination of rationality, punishment and moral behavior. In this research, I rely on assumptions based on conclusions of Francis and Yu’s (2009; pp.1628) research; ‘Most of the critical decisions affecting audit quality are made at the office level, the judgments concerning these are affected by the partner’s incentives with respect to their office’. I also assume that, to some extent, investors are aware of this and take it into account when reacting to auditors damaged reputation.
2.3.2 Industry of the audit failure
A company may switch to a higher quality auditor in order to provide more credible information for investors and creditors. According to Rose- Green et al. (2011) auditors try to achieve enhanced audit quality by specializing in a specific industry. The research that the authors conducted focused on the disclosures of internal control weaknesses by audited firms, they concluded that industry specialist auditors are more likely to report these weaknesses. The authors state that if an auditor has a successful track record of multiple clients in the same industry, this can positively enhance the (perceived) audit quality to the outside world.
A research by Knechel, Naiker and Pacheco (2007) focusses on how the stock market reacts when companies switch to auditors that are considered to be industry specialists. Their findings suggest that the market does perceive audit quality differences based on industry specialization to be relevant to the valuation of a company’s market value. The other way around, this should suggest that a significant drop in reputation by an audit failure could also have effect on the markets perspective of an auditor’s industry specialism.
Another industry related factor is sector of operations. If an organization is operating in the financial service industry it influences the stock market reactions towards lower audit quality. This industry is frequently excluded from a sample during research, as the annual accounts are constructed differently and this reduces comparability. Next to this, additional regulatory obligations in the financial service industries play a role. According to Brugstahler and Eames (2003) regulated firms have greater incentives to manage earnings to maintain within the tier-one and tier- two capital restrictions. Although SEC oversight is strict the value of an external audit is significantly higher than other industries. As investors rely more heavily on the audited reports at financial institutions.
Prior literature gives little insight into the different effects of an audit failure within industries. Although earlier studies provided evidence for this subject from a different point of view, there is also no earlier research that interprets this phenomenon in the setting of this research. Nevertheless, it is highly plausible that if an auditor is involved in an audit failure, the stock market
14
reactions on organizations that operate in the same industry as the audit failure client will be stronger. This would be caused by the spill-over effect of the deteriorating auditor’s reputation in that industry.
2.4 The Tyco events
Tyco is an international security system provider that is incorporated in Switzerland, but with most of its operations in the United States. The business is known for multiple stock splits that lead to a focus on only security solutions and fire protection. Tyco’s headquarters is located in Princeton, New Jersey. To the contrary of the two other major frauds at Enron and WorldCom, Tyco is currently still in operations. The fact that this fraud was addressed internally before involving the public prosecutors, had positive influence on the company’s continuity.
I consider the Tyco scandal to be a suitable case for an event study for multiple reasons. First, this scandal is underexposed due the Enron and WorldCom scandals in the same time period. These scandals had a larger impact on the daily life of their stakeholders, as the companies ceased to exist. Moreover, the news value of these scandals was higher as the scandals were discovered externally. Still, the factual fraud committed was approximately the same size. Second, the time frame of this event is relatively short as all relevant events happen in about two years. This is beneficial for an event study because information will leak (trough informal channels) over time, this causes news releases to lose relevance with an increasing timeframe. Thirdly, Tyco managed to divert blame towards the former CEO and their auditor to some extent. This contributes towards prior literature by casting more light on the two forms of audit quality. Although there was no convictions or significant evidence on PwC’s misconduct in the first period after the scandal, according to the reputation hypothesis investors do also react on basis allegations. This provides more evidence for the perceived aspect of audit quality.
This event study on the Tyco scandal in late 2002, has a total window of events stretches from 2002 until 2003. The focus of the events is around two persons who had major part in the Tyco fraud; first Dennis Kozlowski who became CEO of Tyco in 1992, and was convicted in 2005 for illegally using over $100 million Tyco funds for private expenses (NY- Times, 01-03-2015). Second, Richard P. Scalzo who was Tyco’s engagement partner at PwC in the period 1997-2001 was barred from public accounting in 2003 for “recklessly violated the antifraud provisions of the federal securities laws and engaged in improper professional conduct” (SEC, 13-08-2003). An overview of the selected event dates has been included in Table 1 on the next page.
15 TABLE 1
# Date of event Event description Tyco stock price
1 30-Jan-2002 Kozlowski and Swarts' extensive
selling of own stock
$34,85
2 3-Jun-2002 Kozlowski indictment on tax fraud $16,05
3 12-Sep-2002 Kozlowski, Swarts and Benick's
"looting" of Tyco's KELP fund uncovered
$18,45
4 17-Jun-2003 Tyco issues restatements under
pressure of SEC
$19,92
5 13-Aug-2003 PwC auditor Scalzo permanently
barred for recklessly issuing fraudulent audit reports
$19,70
Tyco audit failure event dates
The first event date of this study is on January 30, 2002 [Event 1], the New York Times was first to report that Tyco CEO Kozlowski and the CFO Mark Swartz sold over a $100,- millions worth of their Tyco stock in the previous fiscal year (Wall Street Journal, 30-01-2002). This was to the contrary of their personal public statements that they rarely sold their own stock. After analyzing the Tyco stock I conclude that investors reacted heavily on this announcement, and it was the first time that investors were aware of misconduct at Tyco. To ease tempers of the investors Kozlowski and Swartz said that they would buy 1 million shares with their own money, with market value of 35,- per share this buyback amounted to $35,- million. Although this events doesn’t contain any direct evidence of a failing auditor, a significant result here should provide hard evidence for my hypothesis. I also include this event to control for extraordinary results.
The second event date is June 3, 2002 [Event 2]. On June 3, 2002 Kozlowski resigned from his post as CEO unexpectedly under the cover of family matters. The Wall Street Journal was quick to report that he was the subject of a sales tax evasion investigation by Manhattan District Attorney Robert Morgenthau's office (Wall Street Journal, 05-06-2002). Tyco and Kozlowski were not available for comment and Tyco stocks reacted quickly on this news, although it were mere accusations. The next day on June 4, the Morgenthau’s office released an official announcement on the rumors. They announced a criminal indictment accusing Kozlowski of (attempt to) evade state and city sales tax by a construction involving the purchase and sales of fine art. The amount of evaded tax amounted up to $2,- million. This is the first event that directly includes the auditor PwC in the scandal. This was not only because PwC auditors failed to detect these transactions, but also because PwC was involved in tax advisory services at Tyco. Still this event was relatively small in comparison with Tyco’s profits and assets, so this effect should be qualified as a perceived audit quality effect, rather than the deterioration of actual audit quality.
16
The third event date in this study is September 12, 2002 [Event 3]. This day the SEC filed a civil enforcement action against three former top executives Kozlowski, Swartz and Belnick (SEC, 12-09-2002). They were charged with granting unauthorized secret loans, not disclosing of board compensation, not disclosing of related-party transactions and fraudulent stock trading. The main part of the report concluded that from 1999 to 2001 Kozlowski retracted up to $270,- million of unauthorized funds from Tyco’s Key Employee Corporate Loan program, a sizeable amount that certainly was included in the auditors scope. This is the first official announcement that contained a direct indication that the auditors at Tyco failed.
A small year later the next relevant event occurred on June 17, 2003 [Event 4]. At this date Tyco filed their 10K report which included restatements of prior years (Wall Street Journal, 17-06-2003). Under pressure of the SEC, Tyco made restatements dating back to 1998 with a total amount of $696.1 million (pre-tax). The Wall Street Journal states that the restatement also provides legal ground for a claim, as: “Once companies restate, plaintiffs no longer have to prove past financial statements were wrong”. This restatement also draws conclusion over the quality of the audited 10K reports over these years, as the PwC auditor signed-off on these reports that contained material misstatements.
At this moment the events already uncovered the fact that the Tyco reports audited by PwC contained material misstatements, but information about the quality of the auditor was still not available. In the study by Chaney and Philipich (2006) the involvement of the auditor in the fraud became clear at the moment the fraud was uncovered, through the shredding of documents. This lead to a very strong direct stock market effect on other clients of Arthur Andersen. The first form of litigation that involved the auditor of Tyco was on august 13, 2003 [Event 5] by the SEC. The SEC stated in their press release that: “SEC Finds PricewaterhouseCoopers Engagement Partner Recklessly Issued Fraudulent Audit Report and Engaged in Improper Professional Conduct” (SEC, 13-08-2003, p1). The signing auditor Richard P. Scalzo was permanently barred from the auditing practice. I assume that this press release has a negative effect on the perceived audit quality of PwC. But in comparison with the earlier study Chaney and Philipich (2006) this information came latest in timeframe, so the possibility remains that this information’s value already was leaked in an earlier stage.
Although further litigation of the auditor would be valuable information, I will not include this in my research. The class suit case against PwC started in 2005 and ended by settlement of $225 Million in 2007 (NY- Times, 07-07-2007). As this timeframe is far past the date of the fraud it loses most relevance for an event for this study.
17
2.5 Hypothesis development
This study tests if the announcements regarding the Tyco fraud and the accompanying audit failure, affected the stock price of other PwC clients that underwent their audit assignments. As in previous studies, I expect that the accounting fraud results in uncertainty about auditor’s independence and auditor’s expertise, this deterioration of reputation should have direct external effects. These effects should be measurable on the stock price of other PwC clients, because that company’s financial statements may be viewed as less reliable.
In the previous paragraph I selected five events that are expected to have a potentially great impact on the reputation of PwC. In my first hypothesis I examine if credibility impairment that were caused by these events had negative effects on other clients. This hypothesis is the following:
H1: Stock market returns of PwC clients (excluding Tyco) are lower than stock market returns of non-PwC clients, at the moment of press-releases regarding the Tyco audit failure.
My literature review contains a description of the ability this reputation effect has to travel. Previous empirical research proved that the effect is strongest at the office that provided the audit service. It also appears that the effect does not travel to clients of the same auditors in other countries. This suggests that investors take office based decisions of an auditor into account. I will also research this phenomena by replicating the hypothesis of Chaney and Philipich (2002). I will study if clients of the office serving Tyco will suffer more severe than other PwC offices. This leads to the following hypotheses:
H2a: The negative stock effect of PwC- Boston office clients is stronger than the negative stock effects of clients at other PwC offices.
As prior research that studies the effect above more deeper is limited I will further examine the effects of location on the reputation effects. I will research if PwC clients that are located within the same region of influence as Tyco, suffered a more severe reputation effect then firms that are (geographically) distant to the Tyco. This will make the second part of my location hypothesis the following:
H2b: The negative stock effect of PwC clients (other than Tyco) is negatively associated with the distance between their corporate HQ and Tyco’s HQ.
18
In addition to location, the literature review contains a paragraph that is focused on the reputation effects within industries. It contains evidence that investors do take additional audit quality (actual and perceived) by auditor industry specialism into account, and that industry specialist firms can provide high actual quality. The reversed effect of industry specialism has not been tested in a setting comparable to this one. I expect that due a spill-over effect, PwC clients in the same industry as Tyco suffers more severe consequences from the market reaction, than a client that is not in this industry. This leads to the third hypothesis of the study:
H3: PwC clients that operate in the same industry as Tyco show a larger negative stock market return, than clients that do not operate in this same industry.
19
3 Research methodology
The research question and theoretical model of section one and two lead to the following research methodology. In appendix: A the predictive validity framework has been enclosed to recapitalize the relations made in this research. This model is also enclosed in order to strengthen the understanding of the transition from theory to model. In this section I will discuss the quality of the sample, my research design and the indispensable control variables used in this research.
3.1 Sample
This research requires a stock index that is basis for the normal return of the market, in this case the S&P- 1500. To have connecting results I also based my sample on this stock market. By using the S&P- 1500 the sample size is large enough to find significant result, yet the market will be active enough to directly incorporate large scale events. Because of the long timeframe that this research overlaps, it requires two estimation periods. This also concludes in two different samples.
The two samples consists out of all S&P-1500 index constituents on the first and fourth event date, these are retrieved from Compustat North America. From these companies, all not PwC- audited are excluded, leaving a PwC- only sample. Daily closing stock prices of 126 days prior to the first and forth event are retrieved from Compustat North America Daily (estimation period), and the closing stock prices of 40 days around the five events dates are retrieved (event period).
During the process of collecting data some companies were excluded or lost. The final samples for the research amounted up to 351 (event 1, 2& 3) and 372 (event 4&5) firms with complete data. Table 2 provides an overview of the sample sizes and the lost and excluded records. The variables used in this research are discussed in the next paragraph and an overview of these has been included in appendix B.
TABLE 2
Event 1, 2 &3 Event 4&5
30-jan-02 16-jun-03
S&P1500 Constituents 1.507 1.501
Excluded firms:
- Without CIK-keys (17) (10)
- Without Compustat data (24) (25)
1.466 1.466 - With auditor other than PWC (1.061) (1.042) 28% 405 29% 424
- With incorrect database auditor (10) (8)
- With no stock price data (8) (10)
- With Incomplete stock price data (18) (28)
- Extreme outliers (18) (6)
20
I will provide some additional explanation on the exclusion of the last four aspects. During the process of manually collecting the PwC- office for the office variable, it became apparent that the Compustat database provided incorrect auditor information for some records. These firms are excluded as they would corrupt my analysis of the stock market effect. The retrieved index constituents from Compustat on the event dates contained flaws, and the companies that were no S&P-1500 constituents at the event date one or four (first day event of sample), were excluded. Then, incomplete stock data has been removed, these firms were part of the S&P 1500, but they delisted during the period. As the incomplete data causes my sample size to fluctuate, it is more convenient to exclude delisting firms in total. Also, the outliers in stock return have been removed from the sample. All extreme outliers with more than three times the interquartile range are removed. These outliers were removed as outcome of the CAAR made it apparent that it was not caused solely by the selected events. Medium outliers are still included in the sample as they exist both positive and negative, and have small effect on the mean, median and modus.
3.2 Research design
To support the hypothesis in this study, the research is split up in two stages. The first hypothesis requires a calculation of the cumulative average abnormal return. In this research the CAAR is calculated with the use of the Market Model. Hypotheses two and three use the earlier CAARs data in multivariate regressions.
3.2.1 The Cumulative Average Abnormal Return
This event study is based on the unexpected or abnormal stock returns of PwC clients at the event dates. Cable and Holland (1999) note four possible ways to calculate abnormal in their pilot study; Capital Asset Pricing Model (CAPM), the Market Model (MM), the Mean Adjusted Returns Model (MAR) and the Market Adjusted Returns or Index Model (IM). In their study they compared the different methods and found that the Market Model was the most successful. They state that: “The market model proved valid in twenty one cases (i.e. 70% of the time)” and “The market model was found to be an acceptable simplification of the general models in all cases” (p.15, 1999). Another finding of this study is that all examined models provided relative weak support for the hypotheses. Authors state that while event studies offer a potentially powerful empirical methodology, the currently available models are blunt instruments at their best.
I used the Market Model in this study, this is partly because the chance of finding significant results is high, as explained in the study above. Another reason is that prior studies in my subject (Chaney and Philipich [2002], Barbara and Martinez [2006]) used this model and found adequate evidence with it. This is an indicator that the Market Model is suitable for this research as well. The
21
market model is a primarily used in statistics, and sometimes for valuation of investments. It relates the return of a stock towards the return of a market portfolio that is equally weighted. The model uses the intercept and the beta for each company, these are calculated with an ordinary least squares (OLS) regression over the estimation period prior to the event window. With this data the model predicts the ‘normal’ return over an event window of a predetermined amount of days around the event, reproducing the stock value as if the event had not occurred. By subtracting this outcome from the actual return, the abnormal return is calculated. The model for this research is stated as following: ̂ ̂ ̂ In which: ̂ ̂
The actual return is based on the daily closing price per company in the event window, these daily closing price are retrieved from Compustat North America Daily. In this research I use an event window of -20 days and +20 days trading days around the event date, this is mainly to avoid leakage. I use the Russell-2000 index closing prices as the normal market return. The S&P-1500 would be most fitting, but as this is a capitalization-value weighted index the return of the large stocks is overrepresented. The Russell-2000 includes most S&P-1500 firms but is not value weighted as it does not include the largest firms. Multiple indices have been compared (See Appendix D) and it appears that the differences in the calculated returns are small. Also the Russell-2000 provides the steadiest results, altogether the Russell-2000 provides the most realistic reproduction of the S&P-1500 normal market return. For the calculation of the abnormal return the estimated intercept and beta (αi and βi) are required, these are calculated with an ordinary least squares regression. The estimation period used in this regression is 126 days (half a trading year) prior to the event window. Because the timeline between the event one, two and three and the last two events is relatively long, I use a second estimation interval 126 days before the forth event window. To increase the understandability, a schematic timeline of the event date, -window and estimation period has been included in appendix C.
22
After the calculation of the abnormal returns per constituent, they are accumulated and divided by the total sample. This forms an equally weighted average abnormal return of all firms involved in the event. Formally stated as:
∑ In which:
Ultimately, the series of abnormal returns per recorded event will be summed to create the cumulative average abnormal return.
∑
The calculated CAAR is significant with a P-value of <0,05. The result of this calculation provides answer to my first hypothesis; whether PwC- clients suffer a significant negative stock reaction to the announcements on the event dates.
3.2.2 Multivariate analysis
In this research I use cross- sectional analysis to identify the association between the explanatory variables and the previously calculated CAAR. By this, I can assess if certain factors influence the stock market effect on the event dates, specified for this research; the location and industry variables. The formal notation of this regression is the following:
In which:
The Tyco audit had been performed by PwC’s Boston office, thereby I expect that the negative stock market effect is stronger on PwC clients audited by the Boston office. Thereby, I expect the regression sign for [OFFICE] to be negative. Due the limited information that is available on auditors
23
in the databases, it is not possible to directly query the auditing PwC office in the sample. This is why the auditing office variable [OFFICE] is manually retrieved from the SEC database: “Edgar”. For all companies in a radius of 500km within Princeton (NJ) the 2001 (sample 1) and 2002 (sample 2) SEC 10-K filings have been analyzed, especially exhibit 23: ‘consent of independent auditors’. As it is no obligation in the U.S. to disclose the auditing office, I considered the office where the audit opinion is signed, to also be the office that performed the audit. For the companies in my sample that are located beyond 500km from Tyco, I have assumed that they were not audited by the PwC Boston office. I first made this assumption based on the practicality, as it not possible to be competitive to local offices on a long distance assignment. And secondly I based this assumption on table 3 below, which shows that frequency of Boston -office audits decrease with longer distance.
TABLE 3
Event 1, 2 &3 Event 4&5
Distance in KM 30-jan-02 16-jun-03 from to PwC- clients in range Audited by Boston- office % PwC- clients in range Audited by Boston- office % 0 49 12 5 42% 11 5 45% 50 99 34 15 44% 39 19 49% 100 199 21 10 48% 23 9 39% 200 299 15 3 20% 16 3 19% 300 399 34 3 9% 32 3 9% 400 499 11 0 0% 10 0 0% >500 224 0 0% 241 0 0%
Total by Boston- office 36 39
Total sample 351 372
The other part of the location hypothesis is based on the distance variable [DISTANCE]. This variable is calculated through a series of combinations of sources. First, for each company in the sample the zip code of head office is retrieved from Compustat. This zip code is reformed to only contain the numerical data, although reducing the zip codes to numerical data gives a uncertainty of an approximately 20km radius, this is not relevant based on the full range of the U.S. sample. The acquired zip code is matched with geographic coordinates of the location through a zipcode-database (U.S. Zipcode Database, 2014). The distance between the coordinates of the S&P- 1500 firm and the HQ of Tyco in Princeton, New Jersey is calculated with a quantified distance formula. This study uses the measuring scale of kilometers (km) as this is most recognizable to the readers of this thesis. The distance has been rescaled from miles to km according to the following rate 1mi =
24
1.6093km. The hypothesis states that the reputation effect is to fade away over distance, thereby I expect a negative sign in the regressions.
The third independent variable in my research is the industry group [INDUSTRY], as I expect that the negative stock effect is stronger at companies that operate in the same industry as Tyco. Due this I will also expect a negative sign in the regression analysis. The industry group in which Tyco operates is 3669: Communications Equipment NEC. In the research I include the total section 3600: Electronic and other electrical equipment (excluding computer) as Tyco’s industry group. For this a dummy variable was created that gives value 1 if the company operates in the same industry as Tyco, and 0 otherwise. This variable is based on the directly retrieved 2- digit SIC industry code from Compustat.
3.2.3 Control variables
Prior studies show several factors that may influence stock price change. Control variables are used to control for the major characteristics identified. The following control variables will be included in the regression model, a summary of these is enclosed in appendix B.:
Company size [SIZE] is a control variable for multiple omitted factors that could affect the relation between earnings and return. Scott (2011) describes that firm size can function as a proxy for other characteristics such as growth and earnings persistence. He also suggests that size is positively correlated with the reliance of investors on public financial information, making company size a factor that influences stock price at an audit failure. Also, Francis, Michas and Yu (2013) found that organizations issuing a restatement are, on average, more than twice as large as non-restatement firms in assets. In the literature review I described that client non-restatements are often used in studies as a proxy for real audit quality, so controlling for size is essential. [SIZE] is measured as measured as the natural logarithm-10 of the total assets of the company.
Leverage [LEV] is to control for the risk that the capital structure of an organization bears. Dhaliwal, Lee and Fargher (1991) provide evidence for two possible ways leverage can influence this risk. First, high leverage indicates low borrowing capacity of the organization, this indicates less financial flexibility. Second, the negative effect of bad news release (in this case the audit failure) will only have effect on the return of the investors, as debtholders do not experience direct variable losses. In a highly leveraged organization, investors will have to bear more of the losses and thereby I expect that the stock market reaction is stronger. The control variable [LEV] is calculated through dividing the total debt by total assets.
Market-to-book ratio [MARKETBOOK] is a control variable for the risk and uncertainty of future performance that is embedded in the stock. Earlier studies state that auditors are more important when information asymmetry between management and the investors is greater (Liew
25
and Vasselou, 2000). I expect that the reputation effect of erroneous audits should increase with the market-to-book ratio. The proxy [MTB] is directly retrieved from Compustat North America for each company.
The standard deviation of returns [STDEV] is a control variable that mitigates the unequal stock reactions between different stocks within the market. Scott (2011) describes that some stocks are more volatile than others. As it is not possible to determine a population that is equal in this aspect a control variable is necessary. I expect that firms with a higher standard deviation will show stronger negative stock market reactions after an audit failure as well. The [STDEV] is the standard deviation calculated on the daily returns in the year the Tyco fraud took place.
The final control variable is the growth of revenue [GROWTH]. This variable is the sales growth relative to prior year. It is calculated by the sum of total revenue current year minus total revenue prior year dividing by the total revenue prior year. I expect this variable to be negative to indicate the investors’ concerns about the revenue recognition policy of their client. I assume that the market downgraded firms with high sales growth as they are expected to use the same aggressive revenue recognition method as Tyco, and that PwC would not allow clients to sustain these in the future
26
4 Results
The previous section provided the methodology and research design, which enables the statistical testing of the hypothesis. First, to provide insight in the data, the descriptive statistics of the CAARs and variables of the multivariate regression samples are provided. The next paragraph contains the results of the stock return analysis and their by an answer to the first hypothesis. After this the multivariate regression results are presented and analyzed for the purpose of answering hypothesis two and three.
4.1 Descriptive statistics
As my research is two staged this paragraph contains two different aspects as well. The first part of this paragraph contains an analysis of the descriptive statistics belonging to CAARs of the events. These descriptive statistics are plotted in table 4 and an analysis of its contents is summarized below.
This event study is based on five different event dates. CAAR calculations were performed on a 40-day window around each event date. From this period an event window is selected on basis of professional judgement. The study by Cable and Holland (1999) states that most of the abnormal returns are measurable in a window of 3 to 11 days around the event date. Prior research in my field by Chaney and Philipich (2002) and Weber, Willenborg and Zhang (2008) used relative windows of 3 to 5 days and 3 to 7 days. In this research the most convenient window is selected, with a maximum of then days around the event date. In appendix D an overview is included that provides more insight on the window selection.
As the research contains two different samples the total sample size differs between 351 and 372. The mean of the selected windows returns amounts up to the same total as the CAAR, in these samples both all of the CAARs have negative returns. Next to this the standard deviation and standard error are reported, as these are based on the event windows they provide no information on the stock market volatility (samples are too short). The minimum and maximum are relative high,
Event Window mean N sd se(mean) min p5 Median p95 max
#1 [-2d / +4d] -0.18% 351 0.0684 0.0036 -35.84% -10.99% 0.24% 9.01% 36.37% #2 [-4d / +6d] -1.85% 351 0.0712 0.0038 -36.41% -13.15% -1.39% 9.01% 19.04% #3 [-2d / +5d] -6.70% 351 0.0664 0.0035 -32.20% -18.54% -6.30% 3.88% 15.33% #4 [-1d / +7d] -0.36% 372 0.0558 0.0029 -21.66% -9.64% -0.35% 9.18% 15.64% #5 [-2d / +4d] -2.32% 372 0.0509 0.0026 -21.81% -10.73% -2.55% 6.08% 14.51% Descriptive statistics of the average abnormal return per event date (mean = CAAR)
27
as only the extreme outliers (3 * IQR) are excluded in these samples. The reason for this is that the stock market provides exact data and the medium outliers are present on both positive as negative side in the samples. The 5th and 95th percentile show that the major part of the data is within 10% difference of the mean.
The following pages include table 5 and 6. In Panel A of these tables the descriptive statistics with respect to the regression analysis are presented for the two samples. The samples contain 351 and 369 observations, resulting as further outliers in control variables have been replaced with the value of the outer fence instead of excluded. This causes the samples to show relative low standard deviations, except for the control variables Leverage and Market to Book. The variable Leverage shows a relative large difference in standard deviation, this is because the Leverage in sample one is tested positive for normality without any replacements.
In Panel B of table 5 and 6 the correlation matrices for the two samples are presented. The matrices show that there is little to no correlation between the events CAARs, this concludes that all of the events are unique. In the dependent variables there is medium correlation (max 0,25) within two variables. This is caused by the following: Office and Distance correlate negatively as they investigate the same location hypothesis. It is self-evident that the larger the distance between Tyco and a company HQ, the smaller the proportion of the companies audited by the PwC office in Boston. Next to this the variable Distance and Industry correlate positively, because of a large concentration of companies with the same industry as Tyco (industry Electronic & other electrical equipment) that is located in the west of the U.S., namely Silicon Valley (CA). Within the control variables multiple larger correlations (max 0,45) are observable between Size and Leverage and Size and Market to Book, this defines that larger firms are more leveraged and have a higher market value. I do not expect that this medium correlation will influence the outcome of my regression. Firstly, as it are control variables. Secondly, because prior studies (Francis, Michas and Yu, 2013) measure this correlation as well, without consequences in their regression. In both correlation matrices there are no observations with a correlation exceeding 0,70 or -0,70, on basis of this information I can assume that the samples are suitable for a multivariate regression.
28 TABLE 5
Panel A obs mean st. dev. min median max
OFFICE 351 0,0826 0,2757 0 0 1 DISTANCE 351 1486 1374 23 1103 4085 INDUSTRY 351 0,0712 0,2576 0 0 1 SIZE 351 3,3033 0,7072 1,8623 3,2441 5,6928 LEV 351 3,9552 9,3703 -15,4779 2,3032 27,0983 MTB 351 12,7689 8,6171 -0,2471 10,7701 44,2875 STDV 351 3,3779 2,5339 0,4747 2,6726 15,5903 GROWTH 351 0,1048 0,1998 -0,5946 0,0813 1,6149
Descriptive statistics sample #1 (for events see table 4) Included dummy variables Office and Industry
Panel B EVENT 2 EVENT 3 OFFICE DIST- ANCE INDUS
-TRY SIZE LEV MTB STDV GROWTH EVENT2 1,0000 EVENT3 -0,0862 1,0000 (0,107) OFFICE -0,0502 -0,0893 1,0000 (0,348) (0,095) DISTANCE 0,0631 -0,1322 -0,2473 1,0000 (0,238) (0,013) (0,000) INDUSTRY -0,0719 -0,1224 -0,0429 0,1868 1,0000 (0,179) (0,022) (0,423) (0,000) SIZE 0,1353 0,0796 -0,1315 -0,1716 -0,1551 1,0000 (0,011) (0,137) (0,014) (0,001) (0,004) LEV 0,0012 0,0650 -0,0971 -0,0363 -0,0631 0,0878 1,0000 (0,985) (0,224) (0,069) (0,497) (0,238) (0,100) MTB 0,0496 0,1660 -0,0321 -0,1548 -0,0949 0,3717 0,0677 1,0000 (0,354) (0,002) (0,549) (0,004) (0,076) (0,000) (0,206) STDV 0,1300 0,0484 0,0948 -0,0116 0,1437 0,1174 -0,0732 0,2110 1,0000 (0,015) (0,366) (0,076) (0,829) (0,007) (0,028) (0,171) (0,000) GROWTH -0,0951 0,1832 0,0105 -0,0857 -0,0938 0,0585 0,0392 0,0846 -0,0079 1,0000 (0,076) (0,001) (0,844) (0,109) (0,080) (0,028) (0,464) (0,114) (0,883) Pearson correlation matrix of sample #1. value within parentheses is the statistical significance
29 TABLE 6
Panel A obs mean st. dev. min median max
OFFICE 369 0,0759 0,2652 0 0 1 DISTANCE 369 1543 1402 6 1130 4085 INDUSTRY 369 0,0894 0,2858 0 0 1 SIZE 369 3,3188 0,7334 1,8117 3,2441 5,8863 LEV 369 0,7222 0,8804 -1,6719 0,4636 3,2841 STDV 369 2,0115 1,0808 0,2381 1,7994 4,5817 MTB 369 11,9536 7,6502 -1,3563 10,6994 29,5880 GROWTH 369 0,1027 0,1949 -0,5707 0,0813 1,6149
Descriptive statistics sample #2 (for events see table 4) Included dummy variables Office and Industry
Panel B EVENT
5 OFFICE
DIST- ANCE
INDUS-
TRY SIZE LEV MTB STDV GROWTH EVENT5 1,0000 OFFICE -0,0032 1,0000 (0,951) DISTANCE 0,0409 -0,2429 1,0000 (0,434) (0,000) INDUSTRY 0,1397 -0,0539 0,2274 1,0000 (0,007) (0,301) (0,000) SIZE -0,2891 -0,1145 -0,1534 -0,1590 1,0000 (0,000) (0,028) (0,003) (0,002) LEV -0,1073 -0,1142 -0,0821 -0,1622 0,4111 1,0000 (0,039) (0,028) (0,116) (0,002) (0,000) MTB 0,0129 -0,0540 -0,1529 -0,1395 0,3834 0,0937 1,0000 (0,805) (0,301) (0,003) (0,007) (0,000) (0,072) STDV 0,0846 0,0276 -0,1689 -0,0903 0,1309 -0,0008 0,2638 1,0000 (0,105) (0,597) (0,001) (0,083) (0,012) (0,988) (0,000) GROWTH 0,0509 0,0160 -0,0785 -0,1189 0,0126 -0,0236 0,1051 0,2649 1,0000 (0,330) (0,760) (0,132) (0,022) (0,809) (0,652) (0,044) (0,000) Pearson correlation matrix of sample #2. value within parentheses is the statistical significance
