Amsterdam Business School
The effect of auditors’ reputation loss on their listed clients
The case of KPMG Netherlands during a timeline of bad exposure
Name: S. Taylor
Student number: 10447431 Date: 22 June 2015
Word count: 12599
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam First Supervisor: Ir. Drs. A.C.M. de Bakker
Statement of Originality
This document is written by student Stivin Taylor who declares to take full responsibility for the con-tents 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.
Abstract
This paper researches the effect of the loss of reputation; due to bad publicity and audit failures surrounding a big four firm KPMG Netherlands on their clients stocks.
Previous studies performed by (Weber et al. 2008) involved a single specific event sur-rounding KPMG Germany and their client COMROAD AG that committed fraud in revenue recognition; the research shows that such event influences stock prices, as reputation in the German setting does matter.
This specific case involves KPMG Netherlands, which was connected to several fraud and irregularity cases that were reported in the Dutch media, these bad exposures in the period between 2012 and 2014 provide us with a broad timeline that can function as good basis to re-search if audit failure and reputation loss effects public listed clients in a Dutch setting, confirm-ing if reputation matters and how it is perceived by the market as they value companies based on the quality of audit and reliability of the reporting that is provided by these big four firms.
Relevance of this study is to research if in the Netherlands the reputation of an audit firm KPMG matters to investors and therefore will re-evaluate their investees once bad exposure surrounding their auditor comes to light. The audit market especially in the Netherlands is under massive attention from the general public and the AFM due to many worldwide cases of fraud, the general perception of an auditor and trust in the vocation is very low.
The unfortunate events surrounding KPMG Netherlands provide a good spread of events over a period of two years, the results however show, with one exception, no significant negative cumulative abnormal returns on the KPMG listed clients during the events that oc-curred in 2012 to 2014.
Contents
1 Introduction ... 6
2 Literature review ... 8
2.1 Theoretical literature on audit quality ... 8
2.1.1 Agency theory ... 8
2.1.2 Reputation models in repeat-purchase settings ... 9
2.1.3 Insurance models in repeat-purchase settings ... 10
2.2 Empirical literature on audit quality ... 11
2.2.1 Effect of bad exposure ... 11
2.2.2 Other studies ... 12
2.3 Event studies ... 13
3 The Dutch setting and KPMG Netherlands ... 16
3.1.1 The Dutch setting ... 16
3.1.2 KPMG Netherlands/Timeline of events ... 17
4 Hypotheses development ... 20
5 Methodology and Data ... 22
5.1 Research Methodology ... 22
5.2 Data Collection ... 26
5.3 Data filtering ... 27
6 Results ... 29
6.1 Results for Hypothesis 1 ... 29
6.2 Results for Hypothesis 2 ... 30
6.3 Results for Hypothesis 3 ... 32
6.4 Reliability Check ... 33
7 Conclusion... 38
7.1 Conclusion of the Hypotheses ... 38
7.2 Limitation of the research ... 39
7.3 Future research ... 39
8 Literature ... 40
Appendix A: Composition of the indexes ... 42
1 Introduction
In 2008 a study by Weber revealed that German investors and supervisory boards react to revela-tions of substandard audits by a firm with a reputation for high quality. From a stock market standpoint, investors appear to respond negatively to events relating to COMROAD AG, the firm they used in their research paper (Weber et al. 2008). The findings showed that unlike pre-vious researches performed in the UK and the US by Khurana and Raman (2004); Lennox (1999) and Willenborg (1999), not the insurance rationale, which states that companies and in-vestors contract a big-four auditor for the possibility of litigation and a higher probability of re-ceiving compensation when bad quality is provided, but that reputation of the firm was of sig-nificance in affecting the market and behaviour of investors.
The theory used is the reputation rationale, which states that a reputable firm will provide high quality because they earn quasi-rents that they fear losing should they “cheat” by opportun-istically providing low quality (Klein and Leffler 1981). High quality is perceived to exist when an audit is performed by a big four auditor. DeAngelo (1981) states that the size of a firm is a proxy for quality, therefore investors value the information higher when they are obtained through a big four firm, where according to Dye (1993) the costs of failure would be compensated by the big four auditor because they tend to be wealthier and have deeper pockets, thus referring to the insurance rationale.
This research will take a similar approach to see if these findings apply for a Dutch setting where a litigation risk is comparable to Germany, which is capped, by investigating if KPMG Netherlands who enjoyed continuous bad exposure over a period of 2 years (2012 – 2014) in connection to audit failures, fraud and corruption will affect their clients’ value.
The research is set out as follows; In the first section of this research literature review is conducted and relevant theories are discussed before introducing the previous studies that were performed. Also defining what an Event Study entails, which is necessary to comprehend how this research will be designed. In the third section the Dutch setting and the identified events that are the bases of this research will be introduced.
In section four we will introduce our hypotheses, followed by the research methodology for this event study and collected data in section 5. The last two sections contain the data analysis and conclusions.
The research question that will be answered in this research is:
The Netherlands provides a good and controlled base to perform this research; similar to Ger-many the Netherlands is also a low litigation country. In the Netherlands it’s difficult for inves-tors to sue audiinves-tors, the auditor has a protected profession and is as individual responsible for his actions, however these penalties are limited to a fine of EUR 7.500 and a permanent deletion from the register of public accountants, thereby losing their certificate. The accounting firms usually settle certain litigations to minimize reputation damage.
2 Literature review
In this chapter prior research will be discussed in order to clarify the definition of audit quality in the relation between the firms and their auditors. The theories that underlie this paper, which are the agency theory, insurance rationale and reputation rationale will be explained. Firstly we will be discussing the Agency theory, which explains the relation between the management, owner and auditor and the need for high quality information. This will be followed by the insurance and reputation rationale will be explained in relation to the audit quality and prior studies that show differential quality arises either because reputable firms have greater incentives not to perform a low-quality audit (reputation rationale) or because wealthier firms have a stronger bond to ensure a high-quality audit (insurance rationale). Furthermore a summary of prior studies will be made in order to create a clear bases to argue if the events in which KPMG Netherlands is presented provide a good setting to examine the reputation rationale and its effect on their clients, which we test by studying whether an auditor’s reputation has financial impact to the clients that are contracted by KPMG Netherlands.
2.1 Theoretical literature on audit quality
2.1.1 Agency theory
Management can increase the credibility of their financial reporting by having it audited by an independent auditor who will give his opinion. Shareholders rely on this independent opinion in order to react to risks, set the level of compensation and to decide whether further investments in the company are viable. The higher the quality provided by the auditor, the more assurance there is for the shareholders that the financial reporting does not contain any material misstate-ments when the auditor issues an unqualified opinion. The auditor provides the monitoring for the performance and activities of management to shareholders; this monitoring, also called agen-cy costs, is in the form of an audit fee. When looking at the agenagen-cy theory we identify three par-ties that are involved, the main theory states that the relation between principal (owner) and agent (management) exists in which the agent performs work in order to manage and create a higher value for the principal, because the principal does not have the means and ability to moni-tor the agent himself there is a third party identified. This third party is the independent audimoni-tor who is assigned by the principal to monitor the agent and audit the performance since there is an economical conflict of interest between the agent and principal. The choice for a certain auditor is based on their quality of work and reputation; the principal will pay a premium (agency costs) to assign an auditor with a great reputation to minimize the risk of mismanagement by the agent,
the auditor will do so by performing an audit on the financial statements of such entity and providing an unqualified opinion. The users of the financial statements are then guaranteed by the auditor that the information they use for their decisions are free of noise (unintentional mis-statements) and bias (intentional misstatements, also known as fraud) (Moizer 1997). Barton (2005) states that for shareholders it’s not possible to measure audit quality directly, instead they will measure reputation and brand name of the auditor as an indicator for reliability of the finan-cial reporting. When the relation between agent and principal is disturbed, uncertainties arise about the provided audit and shareholders will doubt the quality of the audit performed.
2.1.2 Reputation models in repeat-purchase settings
The relation between reputation and quality is made clear by Klein and Leffler (1981) as they present a model in which higher audit fees, referred as price premiums exist “to motivate com-petitive firms to honour high quality promises because the value of satisfied customers exceed the cost of savings of cheating them” (p.623). An audit firm earns fees, known as quasi-rent for providing assurance on the financial statements. Customers are willing to pay these quasi-rents as they see it as an investment in order to obtain high quality information from the investee that is audited. This willingness is translated into a “premium” that the customers (shareholders) want to pay in order to receive a higher return on their investments (company) creating a market in which audit firms provide high quality because they earn a fast stream of profits that discourages them to cheat by providing low quality. Once a company is caught “cheating” by the customer, the punishment will be curtailing purchases. But what is the relation between a big-four firm and higher quality and why are quasi-rents higher at a big-four firm compared to non-big-four firm?
In order to explain what the relation is between firm size and audit quality, we need to know the definition of audit quality. DeAngelo (1981) defines audit quality as a calculated joined prob-ability that an auditor will (a) trace any omission in the clients financial administration and (b) will report these findings. The probability that an auditor will be able to find all these omissions depends on several factors such as (i) whether the magnitude and nature of the audit work that needs to be performed are suitable for this specific client; (ii) how capable the auditors are in order to do their work well, and (iii) the independence of the audit firm and how certain it is that they will report any misstatements and omissions when found (DeAngelo 1981a). DeAngelo concludes in her other research (DeAngelo 1981b) that auditor size is a proxy for quality, she argues that for larger audit firms the disincentives to “cheat” outweigh the incentives. She states that for an incumbent audit firm, for whom the quasi-rents are client specific (i.e., the repeat purchase is by the same customer), the incentive to opportunistically provide a low-quality audit
is to retain the present value of the client’s quasi-rents. In this sense, cheating results in a benefit as the auditor continues to receive rents without incurring the additional costs of exerting high-quality effort. The disincentive to cheating relates to quasi-rents from all other clients, therefore, as audit firms grow their clientele the costs of cheating exceed the benefits, as there is a greater fear of losing clientele once they learn their auditor is providing low-quality audits. Resulting in possibly ending their purchases (audits) from this firm.
Shapiro (1983) and Klein and Leffler (1981) refine these insight further by adding a multi-period setting with free entry in which reputation is considered a cost of (not a barrier) entry to charge premiums. In this model, companies earn a price premium either as an incentive to pro-duce high quality or as a return on their investment in reputation. Either way we see the relation that exists between firm size, audit quality and reputation in order to obtain and maintain quasi-rents, thus explaining the importance of reputation for the profitability and existence of the audit firm.
2.1.3 Insurance models in repeat-purchase settings
Another model that is referred to in this research is the insurance model; the relation between this model and audit quality is based on litigation possibilities for the customer. The insurance rationale for audit quality arises to the extent that it enables financial statement users to recover damages from auditors in the case of audit failures (Weber et al. 2008). Simunic (1980) models audit fees as a linear combination of marginal cost and expected losses from shareholder litiga-tion in a manner that interrelates these components, i.e. addilitiga-tional effort increases the resource cost of the audit and decreases expected litigation losses. By expending more effort, auditors more likely detect material misstatements and satisfy professional standards, thereby mitigating each of the conditions necessary to bring successful litigation. Dye (1993) models the audit as enhancing allocation of resources and providing investors with a claim on the auditor in the event of an audit failure, and demonstrates conditions under which an auditor’s wealth serves as a bond for audit quality. Because larger firms tend to be wealthier (i.e. they have “deeper pock-ets”), depending on the institutional setting, larger/wealthier audit firms have greater incentives to provide high-quality audits. This model is more common in the United States where its easier to sue a company and/or civilian as their ruling outcomes provide a higher compensation to cover the effort of their litigation. Therefore confirming the relation between audit quality and firm size, due to the fact that the insurance rationale for those regions where litigation is a higher risk and providing high quality assurance is essential in order to avoid such litigations.
2.2 Empirical literature on audit quality
2.2.1 Effect of bad exposure
In prior studies the researchers have looked into the relation between a certain event and the reaction on the market. The focus is on events in which companies are coming out with bad news regarding bankruptcy or fraud to identify if investors will change their view of an auditor when they receive such news, in a next section event studies will be clarified further. Baber, Ku-mar and Verghese (1995) studied the impact of a United States based accounting firm Laventhol & Horwath (L&H) who filed for bankruptcy in November 1990. At that time L&H was the sev-enth-largest accounting firm in the country, the bankruptcy filing came unexpectedly. Baber et al. (1995) studied the event that L&Hs’ bankruptcy filing was made public and the reaction on their clients’ shares, the results show a negative abnormal return of about 2 percent in L&H’s client portfolio. These results were consistent with the theory that investors rely on their auditors as a source of high quality information, however this could support both a reputation rationale as well an insurance rationale for audit quality.
The fall of Arthur Anderson, another United States accounting firm that was one of the big-five at that time shook up the audit market as well. Chaney and Philipich (2002) studied the stock market reaction to four Enron/Andersen events, focusing on Andersen’s admission on January 10, 2002 that they destroyed a large number of Enron-related documents. They docu-ment significant negative cumulative abnormal returns (CAR) to Andersen’s clients of 2% at the time of the shredding disclosure, and 1% at the release of a report from Enron’s Board and An-dersen’s hiring of Paul Volcker to head an independent oversight board (Weber et al. 2008). Chaney and Philipich (2002, p. 1243) interpret these negative returns as damage to Andersen’s reputation for high quality, which “ [paid] the ultimate market price for loss of reputation.” However, because Enron so closely follows other big Andersen audit failures at Sunbeam and Waste Management, it is difficult to disentangle the reputation from insurance rationales for au-dit quality in the case of Enron/Andersen. In reacting to the shredding disclosure, within just months of Andersen’ s Sunbeam and Waste Management settlements, Assistant U.S. Attorney General Chertoff said Andersen was a “ recidivist . . . [and that] the shredding at Enron was the worst case of corporate obstruction . . . he had ever seen” (Alexander et al. 2002). Thus, it seems reasonable that a portion of the returns Chaney and Philipich (2002) document may reflect the stock market’ s reassessment of Andersen’ s prospects for survival, and the corresponding possi-bility that Andersen’ s insurance coverage to investors could be withdrawn.
Whereas the L&H and Andersen incidents involve concerns regarding whether auditor resources would continue to be available to indemnify investor losses, the situation differs for
KPMG and ComROAD research by Weber (2008). Because auditor liability to investors is lim-ited under German law, despite the losses to ComROAD investors, the viability of KPMG Germany was arguably never in serious question. Moreover, even if KPMG was to suffer a simi-lar fate as Andersen and L&H, investors in their German clients would likely retain the insurance coverage specified by German law (Weber et al. 2008). Weber (2008) shows that the market reac-tion to the performance of low quality audit resulting in a 3 percent negative return for KPMGs’ client portfolio. The KPMG Netherlands/incidents that are combined in a short period of time enables us to abstract from the insurance rationale for audit quality and focus on the market re-action to events that are, more than for the L&H and Andersen incidents, damaging an auditor’s reputation.
2.2.2 Other studies
Other studies conclude that the insurance rationale for audit quality appears to dominate the reputation rationale. For instance the research performed by Lennox (1999) focused on audit failures in the United Kingdom in the period 1987 to 1994. He finds that while larger U.K. audi-tors are more likely to be sued and criticized in the business press (e.g., for rendering a clean opinion to an ex post bankrupt company), such firms do not suffer a decrease in market share or an increase in client defections. Overall, he interprets this as consistent (inconsistent) with an insurance rationale (reputation rationale) for audit quality. Weber (2008) refers to a study of small U.S. initial public offerings (IPOs), a context with substantial diversity in auditor type, which is conducted by Willenborg (1999), the findings of the paper are that the inverse relation between auditor size and underprizing is also evident for start-up IPOs, a setting wherein Willenborg (1999) argues the reputation rationale for audit quality is less important. He also finds audit fees are positively associated with IPO proceeds (the upper limit of the auditor’s coverage to inves-tors). He concludes that “ . . . results suggest the importance of an insurance-based demand for IPO audits” (Willenborg 1999, p.237; Khurana and Raman 2004, p. 475) find that the presence of a large auditor is inversely related to a measure of the ex ante cost of equity capital in the United States but not Australia, Canada, or the United Kingdom (i.e., lower investor litigation countries than the United States), and conclude “. . . litigation concerns, rather than reputation protection, drive the perceived higher quality and financial reporting credibility of Big four au-dits.”
2.3 Event studies
Many of the aforementioned studies are known as “event studies”. This is a method used to as-sess if the value of a company changes when certain events occur. An event can be defined as a moment in which information is made publicly known to investors and other stakeholders that can use this information for their future decisions regarding the company. In order to measure the value of a company and the impact of a certain event, the studies make use of the stock pric-es. This method of research is used to measure the impact at events that inform about mergers and takeovers, press releases regarding results etcetera for many years now. According to MacKinlay (1997) the first published event study dates from 1933.
According to Brown and Warner (1980) an event study is a test of market efficiency that is formed by the occurrence of an event (news publication) that can impact the market returns for a certain companies’ share. The increase of the share price indicates that the shareholder as-sumes an increase in the cashflow. Because information is widely accessible for everybody, shareholders will react instantly which will result in the increase of price for the currently under-rated shares to a fair level, here the market adjusts by itself. This is called market efficiency. Brown and Warner (1980) also state that when no abnormal returns are measured during an event, then the outcome is not consistent with the efficient market hypotheses because the event obviously did not contain new information for the market to react.
An event study has three important measure periods (see figure 1). The first measure period is called the estimation window, this is the period prior to the event. The second measure period is called the event window, this period is in which the actual event occurs. The last meas-ure period is called post event window, this is the period after the event has occurred.
Figure 1. The timeline for an event study, MacKinlay (1997)
MacKinlay (1997) states that it is of importance to identify an event (also called event window) in which the share price is measurable. A good example of such event window is an earnings an-nouncement. The event date is the day the earnings announcement is made, it is customary to
define the event window to be larger than the specific period of interest. This permits examina-tion of periods surrounding the event (MacKinlay 1997). When there is a study involving a large sample (amount of companies and share prices), it’s common to classify events based on charac-teristics such as industries. The estimation window is the period prior to the event, based on this window the normal expected quotation is calculated. The most accepted way to measure this is based on the period prior to the event; MacKinley (1997) illustrates an example in his paper where daily data over a period of 120 days is being used with exemption of the event date in or-der to calculate abnormal returns.
There are different ways to calculate normal returns for a share price. These methods can be grouped into two categories - statistical models and economical models (MacKinlay 1997). A statistical model is based on statistical assumptions without being influenced by economical con-siderations. The economical model on the other hand is not based on statistical assumptions but solely on the decision made by investors. However MacKinley (1997) does point out that the economical model cannot be used without statistical consideration, because this gives it a more precise way to measure normal returns based on economical limitations.
The Constant Mean Return Model is a statistical model that measures the mean return of an asset over the estimation window. This mean is assumed also to be valid for the event win-dow. The calculation of the abnormal returns is based on this assumption . MacKinley (1997) states that although the Constant Mean Return Model is one of the simplest models, it often yields results similar to those of more sophisticated model as found by Brown and Warner (1980, 1985), however they only analyzed short-horizon event studies, which are shorter than one year.
The market model is a statistical model that relates the return of any given security to the return of the market portfolio. For the market portfolio a portfolio like the S&P 500 can be used. The relationship between the returns of a security and the market returns is determined over the estimation window. This relationship is assumed also to be valid for the event window. The calculation of the abnormal return is based on this assumption. The market model repre-sents a potential improvement over the constant mean return model. By removing the portion of the return that is related to variation in the market’s return the variance of abnormal return is reduced (MacKinley 1997).
Economic models can be used to provide more constrained normal return models. The two most common economic models are the Capital Asset Pricing Model (CAPM) and the Arbi-trage Pricing Theory (APT). The CAPM is an equilibrium theory where the expected return of a given asset is determined by its covariance with the market portfolio. The APT is an asset pricing
theory were the expected return of a given asset is a linear combination of multiple risk factors. In the seventies the CAPM was the most popular model, however due to limitations and re-strictions presented with the use of this model, its implementation has almost ceased. The APT is an addition to the CAPM however the gains of using APT versus the market model are small.
Based on prior studies such as Chaney and Philipich (2002), this study will use the market model as described earlier. In the next chapters the identified events will be shared and an hy-potheses will be described.
3 The Dutch setting and KPMG Netherlands
3.1.1 The Dutch setting
Several characteristics of the Netherlands are similar to Germany and the paper conducted there by Weber (2008). The characteristics are important to our study. To begin, just like in Germany it is difficult for clients and investors to sue auditors for damages associated with misstated fi-nancial statements. Clients may sue for breach of contract, but Dutch law imposes a high bar by requiring they demonstrate the auditor acted intentionally or wreck less disregard for the truth. Investors may bring action under civil liability, but Dutch law requires they show an implied con-tract or that the concon-tract has “ protective effects to third parties,” either of which are arguably quite difficult to demonstrate (European Commission 2001). In addition to the difficulty in bringing investor suits against auditors, in most instances there is a limit on auditor civil liability. Dutch accounting firms usually settle damages to their client for negligence, and, since 1998, auditors are liable to third parties for negligence, though damages are also subject to cap.
The Autoriteit Financiële Markten (AFM, similar to SEC) can issue penalties to the ac-counting firms for non-compliance with the Wta (wet toezicht op de accountants organisaties), however these penalties are also capped. At the time of the first event of audit failure for KPMG Netherlands discussed in this paper, the maximum amount for negligence was € 4 million per audit as mentioned in the Wta. While third parties can sue under tort law, where damages are not capped, to successfully do so they must demonstrate immoral violation of professional duties by the auditor with the intent to damage. In addition to these characteristics of their legal system (and in contrast to the American Institute of Certified Public Accountants Code of Ethics), The Netherlands does not permit auditors to advertise or to engage in direct, uninvited solicitation of competitor’s clients. As discussed earlier a critical assumption of reputation models is that if a high-quality provider is caught cheating, consumers punish them by curtailing purchases (section 2.1.2)
The Netherland’s restrictions on auditor competition should bias against finding market-share evidence of a reputation rationale for auditing. Lastly, because of the institutional details of board structure and stakeholder monitoring, independent audits of public companies are argua-bly less important in the Netherlands. To the extent that auditing serves as a monitoring device to mitigate conflicts of interest (Watts and Zimmerman 1983) in countries where the monitoring of management is already “high” the demand for audit quality should be lower. Following this, Dutch investors may be less likely to place as much reliance on audit quality and Dutch auditors are less likely to be subject to dismissal when their reputation becomes tarnished. This should
bias against finding either stock market or market share evidence of a reputation rationale for audit quality. To summarize, in the Netherlands, the difficulty of successfully bringing suit against auditors coupled with a cap on client and investor damages for negligence reduces the insurance rationale for audit quality. As such, the Dutch setting provides an opportunity to ex-amine the reputation rationale for audit quality. However, because of certain institutional details, the Dutch setting also imposes obstacles to finding evidence consistent with a reputation ra-tionale for quality.
3.1.2 KPMG Netherlands/Timeline of events
Unlike the research performed by Weber (2008) in Germany, in this chosen Dutch setting we will not focus on a single event, but several events in period 2012 – 2014 in which KPMG Neth-erlands was faced with bad exposure. In all cases KPMG was mentioned in fraud, corruption and audit failure cases at big companies, which are or were served by KPMG Netherlands during a that two year period. These events give us good bases to verify if the results found in Germany also apply in the Netherlands, because there is a timeline of events impacting KPMG Nether-lands we believe to achieve a better result to the question if the clients pay the cost of auditors’ reputation loss.
In the Netherlands the following events took place and were published in the main-stream financial media in the Netherlands, sources such as DeAccountant, het Financieel Dagblad,
Elsevier and de Financiële Telegraaf:
I. Vestia is a real estate corporation that serves a large civil purpose in the renting of houses to low income individuals – on April 26st of 2012 KPMG announced that they cannot longer uphold their unqualified opinion given with the annual report of 2010 and there-for revoked their independent auditor’s report. On May 3rd 2012 KPMG also revoked the signing rights of one of the their accountants, shortly following a litigation from the AFM, which is the Dutch Financial market supervision authority.
II. Criticism and penalty from the AFM – on June 6th 2013 the AFM issued a penalty to
KPMG after conducting reviews of their audit work and quality, they concluded that the internal quality standards at KPMG Netherlands were under par and therefore fined KPMG Netherlands for an amount of EUR 881.250.
III. Criticism from the AFM – on September 25th of 2014 the AFM issued their findings of
the research they conducted at the Big Four accounting firms regarding quality of audit files and work performed. This research was a re-performance of the audit work con-ducted by the Big Four firms, the AFM selected 10 audit files from each firm and the re-sults show that KPMG Netherlands scored poorly as 7 of the 10 files were inadequate where as the other firms had 2 or 3 files with remarks each.
IV. Imtech is a large IT company situated in several countries – On December 10th 2012
KPMG Netherlands accountant was in the news for possibly signing off on a falsified annual statement of Ventilex, a daughter company of Imtech. In august of 2013 the VEB also started prosecuting KPMG Netherlands for the fraud committed by the Polish and German subsidiaries of Imtech during the KPMG audit contract.
V. Ballast Nedam is a large Dutch building contractor, Ballast Nedam also has building pro-jects abroad – on November 1st 2013 Ballast Nedam was mentioned in a bribery case in
Saudi-Arabia where money was paid to obtain contracts, KPMG as their auditor was blamed because they were aware of these projects and three old KPMG Partners were found to be involved in these bribes.
VI. SBM Offshore is a provider of floating storage, production and offloading – As a result of the Ballast Nedam bribery affair, the Dutch public prosecutor also conducted a re-search on other companies that had projects abroad and found that SBM Offshore paid EUR 250 million in bribes to assure a contract, their annual report was also audited by KPMG Netherlands and provided with an unqualified opinion. This information was first published on February 7th 2014.
VII. Building of the KPMG Head office in the Netherlands – on April 17th of 2014 a
publication was that the Dutch Tax authority and the Public Prosecutor is currently in-vestigation a joint venture that was set up by KPMG Netherlands in connection to tax fraud. It was a initiative of several KPMG Netherlands partners to build a new head of-fice, there was EUR 16 million involved and the high rent they charge the other KPMG partners that will be situated in that building.
VIII. Van Weyl is a large meat processor – on June 16th of 2014 the CEO and CFO
were sentenced to two and one year prison by the Almelo Court for the large scale fraud that committed in the years preceding their conviction, During the years in which the fraud was committed KPMG Netherlands was the auditor. The fraud was made public in 2012, on February 1st 2012 KPMG lost a first trial regarding the publication of the audit
4 Hypotheses development
As explained in previous chapters the Dutch setting provides a good base to further examine if reputation matters and if this can translate to market reaction by measuring the revaluation by stockholders after the aforementioned events occur. The events that occurred in the period 2012 – 2014 as described in chapter 3.1.2 are the events, which will be used to answer the re-search question “Do clients pay the cost of auditors’ reputation loss?”; this is done through the following hypotheses that are formed:
H1: KPMG reputation loss event (I to VIII) show a negative abnormal return for their listed clients.
This hypothesis is to research if the events that occurred containing information that cause repu-tation damage to the accounting firm KPMG Netherlands will impact the clients that contracted KPMG as their auditor during the year that the event occurred. This hypothesis is set up to test the reputation rationale as explained in chapter 2.1.2. This hypothesis will be conducted 8 times, corresponding the eight events that occurred in the period 2012 – 2014. For each event a sepa-rate assessment will be made to reject or not reject this hypothesis.
H1a: KPMG reputation loss event (I to VIII) show a worse abnormal returns for their listed clients compared to non-KPMG listed clients during the same event
This hypothesis is only conducted when for one or more events H1 is found true, hypothesis H1a is designed to correct for the market itself by looking at the returns for clients that didn’t contract KPMG as their auditor during the same events. Hereby correcting for any global market reaction or the economic climate at that time that possibly also influenced the market. Since we focus on the Dutch Market EURONEXT, which consist of three indices with each 25 compa-nies, we can easily collect data from the whole market for this hypothesis and indicate who their auditor was at that time in order to answer this hypothesis correctly.
H2: KPMG reputation loss event (I - VIII) shows a greater negative impact on AEX listed clients compared to AMX and AScX listed clients
After looking at the market as a whole, it is also relevant to calculate the difference between the three different indices that form the Dutch EURONEXT. The indices consist of 25 companies each that based on their size and ratios are divided between the three classes AEX (the main Dutch index), AMX (midcap, middle sized companies) and AScX (Small cap index, smaller listed companies). The hypothesis states that the events that occurred have a greater negative impact on the AEX, containing larger companies, since larger companies have more aggressive and ac-tive shareholders that react to new information compared to the other indices AMX and AScX.
H3: KPMG reputation loss events that contain audit failure (I, IV, VI and VIII) show a lower abnormal return for KPMG clients compared to events that contain other type of failures (II, III, V, and VII)
This hypothesis is set up to measure if audit related failures, directly related to the auditors’ per-formed work and audited financial statements, have a greater influence on the market as a whole looking at KPMG and non-KPMG clients. As described in earlier chapters the audit market was under pressure and quality of audits where generally criticized. The question that needs to be answered is if the market reaction is focussed on KPMG Netherlands when we specify the theme of the event as an “audit failure” versus “other events”.
Reliability checks
Prior studies such as that of Brown and Warner (1980) indicate that the use of a different statisti-cal model does not lead to significant different findings; To check this assumption, hypotheses 1 to 3 will be answered again using the Constant Return Model. This is done to confirm that the chosen method of measurement is reliable. In addition hypothesis 1 to 3 will also be measured again using different lengths of estimation- and event window, in order to check if the length affects the outcome for the events.
5 Methodology and Data
In this chapter the research methodology will be further elaborated upon applying the theories of MacKinlay (1997) and Kothari and Warner (2004). The data collection and preparation proce-dures will also be clarified in this chapter.
5.1 Research Methodology
In order to research the impact of reputation loss of KPMG Netherlands on their listed clients, an event study is conducted where the identified eight events (section 3.1.2) will be measured against the stock prices of the listed companies during the events, both of KPMG and non-KPMG clients. ‘Event studies examine the behaviour of firms’ stock prices around corporate events’ (Kothari and Warner 2004). Event studies have evolved in the past several decades to become an important part of financial economics. The purpose of this event study is to find evi-dence that share prices show an abnormal return during a selected event compared to the normal expected return in the period preceding the event.
This event study qualifies as a “short-horizon” study because we have an estimation window, which is shorter than one year. According to Kothari (2004) a shorter estimation window is more reliable than a “long-horizon” study, which has an estimation window of over a year. The relia-bility of a short-horizon event study is due to its short nature, allowing it to be trouble-free and straightforward, containing less inferences (Kothari and Warner, 1997, p.301) as new studies on the statistical properties of long-horizon estimation windows in the late 1990s have shown.
In this chapter the approach as described by Kothari (2004) and MacKinley (1997) will be followed to set out the research methodology for this event study. The steps taken to perform this event study are set out below.
Step 1. Defining the estimation-and event window (𝐿1 𝑎𝑛𝑑 𝐿2)
The initial task of conducting an event study is to define the event of interest and identify the period over which the stock prices of the firms involved in this event will be examined; this is called the event window. In this study there are eight events as described in chapter 3.1.2 which occurred, announcing information about rulings against KPMG due to an audit failure, fraud, corruption and lack of audit quality. The event will be the date this announcement was made and the event window will include the day of the announcement. The exact date of the announce-ment is described, that is the moannounce-ment information is made public and is available to investors.
According to MacKinley (1997) it is customary to define the event window to be larger than the specific period of interest, which is the date of the announcement of the event. This permits examination of periods surrounding the event. Most commonly used event window con-sists of at least the day of the announcement and one day after the announcement to capture the price effects of announcements which occur after the market closes on the announcement day.
An estimation window needs to be set up to calculate the normal market return during a specific period prior to the event, this in order to measure any fluctuations during the event win-dow as said above against the measured normal course of the stock price in the market during the estimation window. For the estimation window the duration as described by MacKinley will be used in this research, which is an estimation window of 120 days prior to the event window. The event period itself is not included in the estimation window to prevent the event from influ-encing the normal performance parameter. Refer to figure 1 for further information on the posi-tioning of the estimation window and event window.
Step 2. Calculation the daily returns
The daily change in stock price is necessary during the entire event (estimation and event win-dow), the formula used to calculate these daily returns is:
𝑅𝑖𝑡 =𝑃𝑖𝑡− 𝑃𝑖𝑡−1 𝑃𝑖𝑡−1
Here by calculating 𝑃𝑖𝑡, which is the current stock price (today) and 𝑃𝑖𝑡−1, which is the old stock
price (day -1), the variable 𝑅𝑖𝑡 shows the return for a certain stock (𝑖) on a certain day (𝑡). This formula will be used for each day during the windows (estimation and event) for each firm that is part of the event.
Step 3. Calculation of the normal returns
The market model is a statistical model, which relates to the return of any given security to the return of the market portfolio (index). The normal returns are calculated using a linear regres-sion, as previous literature described there is a connection between the market returns and a spe-cific firms’ stock price. The formula below is used to estimate the market returns for all the days in the estimation window based on the ordinary least squares (OLS) linear regression method.
𝑅𝑖𝑡 = 𝛼𝑖 + 𝛽𝑖𝑅𝑚𝑡+ 𝜀𝑖𝑡
Where 𝑅𝑖𝑡 stands for the daily return of the market for a given day (𝑡) during the estimation window. 𝜀𝑖𝑡 is the zero mean disturbance term, which indicates the portion of the equation that
the market model. 𝛽𝑖 Indicates the relation between the share price and the market the firm is in, this being AEX, AMX or the AScX.
Step 4. Calculating the expected market return
With the estimator’s 𝛼̂𝑖 𝑎𝑛𝑑 𝛽̂𝑖 the expected return is calculated without the need for the actual
event to occur, this expected return is calculated based on the market development that is meas-ured during the estimation window. The normal returns can be calculated during the event win-dow using the following formula:
𝐾𝑖𝑡 = 𝛼̂𝑖+ 𝛽̂𝑖𝑅𝑚𝑡
In this formula 𝐾𝑖𝑡 stands for the normal return, also called the expected return or estimated
return, for a certain firm (𝑖) on a certain day (𝑡). 𝑅𝑚𝑡 stands for the return of the market on a given day (𝑡). (This formula is relevant for all day in all the windows (estimation and event win-dow).
Step 5. Calculating the abnormal return
The abnormal return is the measurement of the actual return compared to the normal return as calculated earlier. To calculate the abnormal return the following formula is applicable:
𝐴𝑅𝑖𝑡 = 𝜀̂𝑖𝑡 = 𝑅𝑖𝑡− 𝐾𝑖𝑡
Showing that the abnormal return is equal to the difference between the actual measurements and estimated normal return, this is also indicated by the value 𝜀̂𝑖𝑡 showing that the part that is
not explained by the model (the market) is considered abnormal for any firms’ share price (𝑖) on any given day (𝑡).
Step 6. Calculating the disturbance variance
The next step is to estimate the disturbance variance for the estimation window. This is done for all the firms in the sample per event as will be clarified in chapter 4.2. For all the firms the dates are numbered, where as the event day is identified as day 𝑡 = 0.
𝜎̂𝜀2𝑖 = 1
𝐿1− 2 ∑ (𝑅𝑖𝑡 − 𝛼̂𝑖 − 𝛽̂𝑖𝑅𝑚𝑡)
2 −1
𝑡=−𝐿1
The returns for a specific share (𝑖) during a certain period (𝑡) in a specific market are indicated by 𝑅𝑖𝑡 and 𝑅𝑚𝑡. 𝐿1 refers to the length of the estimation window, which in this research is set to
Step 7. Calculating the average abnormal returns
For a population in an event consisting of 𝑁 firms, the cross-sectional average abnormal return for day (𝑡) is measured using the following formula:
𝐴𝑅 ̅̅̅̅𝑡 = 1
𝑁 ∑ 𝐴𝑅𝑖𝑡
𝑁
𝑖=1
The average abnormal return 𝐴𝑅̅̅̅̅ will be calculated for all the days during all the windows (esti-mation and event window) with all relevant shares.
Step 8. Calculating the aggregation of abnormal returns for an event window
The abnormal returns must be aggregated in order to draw overall inferences for the event of interest. The aggregation is along two dimensions-through time and across securities. The aggre-gated abnormal return is measured by adding all underlying abnormal returns for that certain event together. To compute the different cumulative abnormal returns (𝐶𝐴𝑅) between day 𝑡1
and day 𝑡2 the following formula can be used:
𝐶𝐴𝑅
̅̅̅̅̅̅(0, 𝑡1, 𝑡2) = ∑ 𝐴𝑅̅̅̅̅̅𝑡 𝑡2
𝑡=𝑡1
This formula shows that the CAR of a certain firm (𝑖) during the event window (in which for the event window is 𝑡1 = 0 for the first day, 𝑡2 = 𝐿2− 1 is used to indicate the last day of the
event window) is equal to the sum of the average abnormal returns of a firm during the event window. The formula is used specifically for the calculation of the CAR for all the days in all the windows (estimation and event window).
Step 9. Calculating the variance of cumulative abnormal returns
In this step the variance of the cumulative abnormal return of an event window is calculated using the following formula:
𝑣𝑎𝑟(𝐶𝐴𝑅̅̅̅̅̅̅(0, 𝐿2− 1)) = 𝐿2 𝑁2∑ 𝜎̂𝜀𝑖 2 𝑁 𝑖=1
See step 6 for the calculation of the disturbance variance.
Step 10. Testing the results using the T-value
The type of sample identifiable with a one-sample T-test for hypotheses 1, 1a and 2. All the data is used once for each event. The following test statistic is used:
𝑇1 = 𝐶𝐴𝑅̅̅̅̅̅̅ (0, 𝐿2− 1) √𝑣𝑎𝑟(𝐶𝐴𝑅̅̅̅̅̅̅(0, 𝐿2− 1))
Step 11. Measuring significance of the abnormal return using the P-value
The last step in analysing the data and coming to a conclusion on the hypotheses is the calcula-tion of the significance level, expressed as the P-value. The commonly used significance levels are 𝛼 = 10%, 𝛼 = 5% and 𝛼 = 1%. We apply a T-test with 𝑑𝑓 = 𝑛 − 1. We conduct a one-tailed test.
5.2 Data Collection
For this research Datastream, which is a database, was used to extract the relevant data needed in order to answer the hypotheses. As mentioned in previous sections the focus is on the Dutch setting and stock market, therefore the most important data is the daily share price of the firms that were in the Euronext Netherlands (AEX, AMX and AScX) in the years 2012, 2013 and 2014.
The first step was to identify the composition of the indices (AEX, AMX and AScX) for each event. The eight events occur during a time span of three years, from 2012 to 2014. Since the composition of the Dutch indices (measuring method) changes twice a year (February and August), the choice was made to select the composition of indices based on 31st of December.
This data was extracted from Datastream using the search tools to select the index and defining the indexes composition per 31-12-2012, 31-12-2013 and 31-12-2014, this resulted in a list per index of firms that were part of Euronext Netherlands during the mentioned periods. Now knowing which firms were part of the indices a new search in Datastream was conducted to find the daily share stock prices for these firms, choice was made to select a broader period of data than just 2012-2014 in order to be flexible when conducting analyses and setting estimation windows that might extend to other years.
Data for these firms were extracted over a period of about 4 years, starting from 01-01-2011 to 28-05-2015 (being the day the extraction was made). Once this data was extracted a check on reliability was made by looking up stock prices randomly over several periods of time for several firms on third party websites such as www.bloomberg.com and www.fd.nl, no relia-bility issues arose. The stock prices were found to be accurate and suitable for further use.
To be able to conduct the hypotheses and chosen methodology the total index price for each day was also extracted using Datastream, this was also done for the same period of 1 years and a
check was made to verify reliability and accuracy in the same manner as mentioned above, no issues were found.
The data now consisted of 99 firms that were part of an index in the period 2012-2014, showing a combined 118.450 points of data (daily stock prices) over a period of about 4 years (1150 days). Each index in the Netherlands consists of 25 firms which would result in 225 firm years over a period of three years, however some firms never change or leave the stock market so their data was only extracted once to prevent duplicate data and corruption of the results. After obtaining this data the next step is to identify the auditor, since we are testing the impact of reputation loss for KPMG Netherlands and the impact on their clients, it’s necessary to identify who their clients were during 2012, 2013 and 2014. In order to do this we have used database of the Dutch Chamber of Commerce, which we accessed via Company.info. This is a professional company that provides companies that are licensed to access their web based database in which they collect data about the firms such as; Financial Statements, Benchmarking figures, Sector, Adress and contact information etcetera.
Company.info was used to find information about the auditor of the company during a specific year and their statutory seat (for example being the Netherlands or Luxembourg). This data collection was performed manually for the 99 firms, resulting in a table where per firm it was clear who their auditor was.
5.3 Data filtering
Now that the data was obtained and the information about the auditor was clear, the data needed to be filtered, by setting clear standards during the introduction, hypotheses and methodology we eliminated companies that do not fir in this research.
First looking at the firms statutory seat using Company.info we noticed that some firms had their main seat outside the Netherlands, meaning they also had their stocks in a different index such as Euronext Belgium. This research focusses on the Dutch setting and believes that Belgium stakeholders will be less likely to respond to the Dutch events as they have a different market, information and also in many cases have a local auditor. This resulted in an elimination of three firms from the sample that listed elsewhere as well.
Looking further at the data, one firm made use of a join-audit, meaning that two audit firms work together to perform and audit. As we focus on the impact of a given event on the clients of a one specific audit firm (being KPMG or Non-KPMG) we choose to eliminate this firms from the sample.
After extracting the data from Datastream we noticed that some firms were marked “dead”, meaning they left the market either voluntary or by take-over or bankruptcy. The data contained the specific date that a firm was dead, the choice was made to exclude the firms that were dead during the year (not completing a full year being in 2012, 2013 or 2014). This resulted in an elimination of ten firms depending on the event and the year it was related to, as some of these firms for example Dockwise (dead 12/04/13 only eliminated for events that occurred in 2013 and 2014) was part of the index during 2012 for a full year. To make clear what the compo-sition of the index and their firms were per event after filtering the data, a table was composed, see appendix A.
To clarify, the elimination of the firms from the sample as mentioned above was done by deselecting their data during the analysis, in no stage of the research any data was deleted. The integrity of the data was under no circumstances compromised.
6 Results
In this section the results will be presented, the results will be shared for each event (I to VIII) separately and a brief walkthrough will be given for each hypothesis. For the analyses a tool in Microsoft Excel was developed. Considering the amount of data gathered and the relatively low number of used formulas Excel is a suitable tool to be used
6.1 Results for Hypothesis 1
H1: KPMG reputation loss event (I to VIII) show a negative abnormal return for their listed clients.
This hypothesis is set to measure the effect of reputation loss events on KPMG listed clients. Looking at the abnormal returns for the portfolio for each event we notice that some events don’t have a negative impact on the market, specifically on KPMG listed clients as they show a positive abnormal return during the event window. Indicating that the events did not affect the their stock prices negatively at all, this is measured the strongest during event VII in which KPMG was accused of committing tax fraud during the built of their new head-office.
For all the events we measured the significance of the results, as shown in the table be-low (Table 1) none of the eight events show a significant cumulative negative abnormal return (CAR) under the Market Model, therefore no statistical support for hypothesis 1 was found. The most negative abnormal return (-1,1%) was measured during event III, this was when the market had access to the new information, being the publications of the AFM on the audit quality of all the Big-Four accounting firms, in which KPMG Netherlands came out as the worst firm in terms of quality.
Table 1. Table of the statistical tests conducted, summarizing the calculated average Cumulative Abnormal Returns for the events, the corresponding T-value and the P-value. None of the results are found to be significant under 𝛼= 1%, 5% or 10%
To illustrate this further, the results are shown in the graph below (Fig.2):
Event I Event II Event III Event IV Event V Event VI Event VII Event VIII
CARavg 0,0050 -0,0015 -0,0074 0,0036 0,0054 -0,0039 0,0052 -0,0008
T-value 0,859 -0,193 -0,863 0,583 0,965 -0,691 0,925 -0,134
P-value 0,800 0,423 0,194 0,717 0,827 0,245 0,816 0,447
Figure 2. Graph of cumulative abnormal return for reputation loss events from event day 0 to event day 2. The abnormal return is calculated using the Market Model as the normal return measure
For hypothesis 1 there was no significant statistical evidence found for the claim that KPMG clients will show a negative CAR for all the events I to VIII, therefore H1 was rejected for all the events.
6.2 Results for Hypothesis 2
H2: KPMG reputation loss event (I - VIII) shows a greater negative impact on AEX listed clients compared to AMX and AScX listed clients
Under Hypothesis 2 we will test if the size of the company that contracted KPMG as their audi-tor during an event have effect on their share prices when new information is made publicly (events I to VIII), assuming that larger companies will show a worse abnormal return compared to smaller KPMG clients as their shareholders react to information faster and are more transient. The indexes in the Euronext Netherlands functioned as a proxy for size in which the AEX was the classified as large (main index of the Netherlands) opposed to clients that were listed in the AMX and AScX (Midcap and Smallcap).
The data analysis consisted of a selection of KPMG clients which were listed in the AEX for which the cumulative abnormal return was measured. This was then set off against the data of KPMG clients that were listed in the indexes other than AEX. The results show that under most events the AEX clients of KPMG have a worse cumulative abnormal return compared to Non-AEX listed clients of KPMG, which is in line with the hypothesis. However looking at the
-0,012 -0,01 -0,008 -0,006 -0,004 -0,002 0 0,002 0,004 0,006 0,008 0 1 2 C A R Event day
CAR under Market Model [H1]
level of significance, only under event I, where KPMG was no longer able to uphold their un-qualified opinion that they issued on the annual statements of 2010 of Vestia, we find a signifi-cantly worse CAR (-2%) for KPMG clients that are listed under the AEX compared to KPMG clients listed at Non-AEX indexes. See the table (Table 2) below for further information on the results.
Table 2. Table of the statistical tests conducted, summarizing the calculated average Cumulative Abnormal Returns for the events I to VIII, the corresponding T-value and the P-value. Event I showing a probability of 4.1% which is significant under an 𝛼 =5% (**)
When we plot event I (Fig. 3) using the separate data we gathered for this hypothesis, being the abnormal return and the cumulative abnormal return, it is easier to understand how the firms in the different groups (AEX vs Non-AEX) reacted during the event window.
Figure 3. Graph of cumulative abnormal return for reputation loss events from event day 0 to event day 2. The abnormal return is calculated using the Market Model as the normal return measure for Event I
For hypothesis 2 there was no significant statistical evidence found for the claim that KPMG clients listed under the AEX will show a worse CAR during the events II to VIII, therefore H1 was rejected for those events. However we found a significantly worse CAR (-2%) for the listed KPMG clients in the AEX under event I, compared to the KPMG clients in the other indices.
Event I Event II Event III Event IV Event V Event VI Event VII Event VIII
CARavg -0,0202 0,0032 -0,0111 -0,0085 -0,0114 0,0125 -0,0054 0,0099 T-value -1,739 0,220 -0,579 -0,674 -1,090 1,092 -0,474 0,828 P-value 0,041** 0,586 0,281 -0,474 0,138 0,854 0,318 0,792
Statistical tests [H2]
-0,0150 -0,0100 -0,0050 0,0000 0,0050 0,0100 0,0150 0 1 2 C A R Event dayCAR under Market Model [H2]
Concluding that the size of the company does affect the CAR of the firms for event I, with a significance level of 𝛼 = 5% H1 was not rejected for event I.
6.3 Results for Hypothesis 3
H3: KPMG reputation loss events that contain audit failure (I, IV, VI and VIII) show a lower abnormal return for KPMG clients compared to events that contain other type of failures (II, III, V, and VII)
Under Hypothesis 3 we made a difference between two kinds of reputation loss events, being events which can be classified as “audit failure” events and “non-audit failure” events.
This classification is made to measure if audit failures cause a worse abnormal return for KPMG listed clients compared to the events containing non-audit related information, such as corruption, tax evasion, quality reports and other information. As described in the theory share-holders react to new information, since they rely on the audited financial statements, we believe that their reaction on the KPMG audit failure events is greater as they will reconsider the possi-bility that the financial statements on which they rely might also contain misstatements, resulting in a bigger CAR for the KPMG listed clients as the shareholders re-evaluate their position in the firm.
Looking at the nature of the events as described in section 3.1.2 the following classifica-tion of the events was done manually, see the following table (Table 3):
Table 3. Table showing the classification of events based on their nature
Using the developed tool in Microsoft Excel we measured the effect between these two types of events, the result are shown in the table below (Table 4). Looking at the results shown below the CAR is positive, which is in line with the outcome of Hypothesis 1 as there was no significant outcome found there either.
AF NAF I II IV III VI V VIII VII AF = Audit Failure NAF = Non-Audit Failure
Table 4. Table showing the test statistics for hypothesis 3, the results show that there is no statistical evidence that audit failure events show a worse abnormal return compared to non-audit failure events. None of the P-values are found to be significant under 𝛼= 1%, 5% or 10%
Concluding for hypothesis 3 that the outcome does not support the hypothesis as there is no significant evidence that the nature of the reputation loss event causes a bigger CAR for KPMG listed clients, hereby rejecting H3 for all the events I to VIII.
6.4 Reliability Check
6.4.1 Change the length of the estimation window and event window
In order to check the reliability of the chosen model and length of the estimation- and event window, the data is analyzed once again under different combinations of the length of estima-tion- and event window as shown in Appendix B.
The main estimation window and event window as chosen in the methodology is based on the research of MacKinley (1997), however there is a possibility that the market needs more time to react to new information ( increasing the event window). Also the possibility that the estimation window is more effective when its shorter than 120 days to be more representative and to prevent other events overlapping the days is something that needs to be checked.
A combined check was performed in which for Hypothesis 1 (main hypothesis) the var-ious lengths of estimation- and event window were calculated for all the events (I to VIII), in the matrix as shown in Appendix B the combined results are shown by significance, the levels shown are indicated as follows (Table 5):
Table 5. Table indicating the legend for Appendix B, which is the matrix set up to measure significance of the impact of the different lengths of estimation- and event windows for Hypothesis 1
CARavg (AF-NAF) 0,0006
T-value 0,139
P-value 0,555
Statistical tests [H3]
Legend for Appendix B 0 = No significance
* = Significant under alpha 10% ** = Significant under alpha 5% *** = Significant under alpha 1%
The matrix includes the measurement under the Market Model and the Constant Mean Return Model (CMRM), as described in the next section this is also a reliability check to measure if a different statistical model will have a significant impact on the results.
The results indicate that there is a significant outcome for some of the events under H1 when using a shorter estimation window as well as extending the event window. These findings are found mostly for event I, II and III under H1, see the table (Table 6) below for the summary of the findings.
Table 6. Table showing the summary of the results shown in Appendix B, which is the matrix set up to measure significance of the impact of the different lengths of estimation- and event windows for Hypothesis 1
Looking at the results the following analysis can be made for each event (I to III);
Event I occurred on April 26th 2012, which consists of the publication that KPMG
Netherlands can no longer uphold their unqualified opinion for Vestia for the audit of 2010 shows that the market needs more time to react to new information, as the results become more significant when the event window increases to (0,+10) and (0,+20) under all the estimation windows calculated for CMRM. This indicates that the method used being the CMRM would have a significant impact on the results when the event window increases.
Event II occurred on June 6th 2013, the AFM fined KPMG Netherlands after reviewing
their work, coming to the conclusion that KPMG Netherlands has a low level of quality in per-forming their audits and documentation, this event shows significance (negative CAR) when the event window is increased to (0, +20) under both the MM and CMRM. There is no direct rela-tion between this finding and our previous finding in secrela-tion 6.1 for H1, the possibility is that the market needed more time to react to this new information and considering that the results become more significant when the estimation window is shorter (60 and 30 days).
Event III occurred on September 25th of 2014, where the AFM published their report
af-ter reviewing several audit files of the Big Four firms, their conclusion was formed about the quality of the Big Four firms in the Netherlands, showing that KPMG Netherlands was the worst by far. This Event also showed the highest negative CAR under H1 in section 6.1, howev-er the results whowev-ere not found significant. Looking at the results shown in table 6 we can conclude that under the CMRM the results are found to be significant when the event window increases, this could indicate that the market needs time to process this new information and re-evaluate MM CMRM MM CMRM MM CMRM MM CMRM MM CMRM MM CMRM MM CMRM MM CMRM MM CMRM MM CMRM MM CMRM MM CMRM
Event I ** *** ** *** *** *** **
Event II ** ** * * ** * *** **
Event III * *** * * * *** *** *** *** * ***
B.4 C.4 D.4
Significance Level (P) measured in % (summary of Appendix B) Summary
their position in the firms that are audited by KPMG Netherlands. Under the MM we notice a significant result when the estimation window is shortened to 30 days with an event window of (0,+2) and (0,+5). The short estimation window could be showing this result because the public was aware that such research was being conducted by the AFM and the market could have cor-rected itself for this upcoming event.
As concluded above under certain combination of estimation- and event window length for event I, II and III of H1 a significant result was found, that differs from our previous find-ings under the Market Model with an estimation window of 120 days and event window of (0, +2).
6.4.2 Constant Mean Return Model
As described in section 2.3 MacKinley (1997) states that there are several statistical models such as the Constant Mean Return Model (CMRM), Market Model and the CAPM, we chose to use the Market Model for our research. Brown and Warner (1980, 1985) state that the use of the CMRM does not lead to significantly different results.
The reliability check is performed to determine if our data analysis for the Hypotheses 1 to 3 will show significantly different results when measured under the CMRM opposed to the chosen Market Model (MM).
Hypothesis 1: KPMG reputation loss event (I to VIII) show a negative abnormal return for their listed
cli-ents.
Under the CMRM no significant results were found, which is also presented in Appendix B (sub table A.1) with the estimation window equal to 120 days and event window equal (0,+2). This result is also illustrated in the graph below (Fig.4), the results corresponds with the results shown under the MM, concluding that the reliability check did not lead to a significantly different out-come for events I to VIII.
