The value relevance of goodwill impairment losses: The roles of the institutional environment and industry structure.

The value relevance of goodwill impairment losses: The roles of the

institutional environment and industry structure.

Meinard Asjes (S2173468)

Master Thesis for Msc Finance and Msc Accountancy

University of Groningen

Faculty of Economics and Business Supervisors: dr. Mukherjee & dr. Forbes 13 June 2016

2

The value relevance of goodwill impairment losses: The roles of the

institutional environment and industry structure.

Meinard Asjes

Abstract

The objective of this thesis is to expand current knowledge on the value relevance of goodwill impairments by evaluating the effects of the institutional environment and industry structure. I find that goodwill impairments are value relevant to investors in Europe and are taken into account negatively when evaluating the market value of equity of a firm. Also, results show that goodwill impairments are evaluated significantly more negative in common law countries as opposed to civil law counties. Lastly, investors do not consider goodwill impairments for financial companies or in highly-regulated industries to be value relevant.

JEL Classification: M40, M41

Keywords: Value relevance; Goodwill impairment; Institutional environment; Industry structure; Civil law; Common law; Regulatory influence

Introduction

In 2005, IFRS was introduced in the European Union to enhance the comparability of the financial statements across countries and across industries for all listed firms in Europe. Besides, IFRS was designed by the regulatory accounting bodies to increase the quality level of the financial reporting. Prior literature shows mixed results about the envisaged quality improvement in single-country or single-industry settings from a value relevance perspective, which tests for the association between the financial statements and the market value of the company. For example, the value relevance has increased after the adoption of IFRS in the European banking sector (Agostino, Drago & Silipo, 2011), whereas the accounting quality in the United States deteriorated under IFRS compared to U.S. GAAP for cross-listed firms (Barth, Landsman, Lang & William, 2006).

More importantly, there are concerns that the comparability of the financial statements across different countries is limited due to the variation in the levels of IFRS adoption (Nobes, 2015). Nobes (2015) also notes that these country differences influence other important areas of the

3 financial statement that are less visible or less known to investors leading to different interpretations, such as the choices regarding the recognition and impairment testing of goodwill. The recognition of goodwill on the balance sheet and the amortization through the profit and loss statement under Local GAAPs was changed to recognizing only the goodwill arising from Business Combinations as set out in IFRS 3. The amortization-regime is replaced by an impairment testing regime, which requires companies to test for an impairment on an at least annual basis.

There has been research on the value relevance of impairments well after the introduction, but the transitional phase and comparison to Local GAAPs have been of particular interest to scholars. This thesis aims to extent the work by Laghi, Mattei & Di Marcantonio (2013), who investigate the value relevance of goodwill impairments in a multi-country setting after the transitional phase. However, their results do not provide evidence that goodwill impairments are value relevant to investors nor do they touch upon the possibility of country influences.

This thesis contributes to current literature by focusing on the long-lasting impact of IFRS on the value relevance in Europe after the transition. I examine whether goodwill impairments are value relevant to investors in an 8-year period from 2007 onwards, two years after the IFRS introduction. It is of interest to the regulatory accounting bodies to evaluate the effects of regulation for investors and to reflect on the intended goals of IFRS. I extent the work of Laghi et al. (2013) by applying a multi-country setting of 8 European countries, which could reveal national differences that hinder the uniform interpretation and application of the guidelines as set out in IFRS. The countries are grouped according to their legal system and institutional background to test for value relevance differences.

IFRS has focused on introducing a broad set of rules that are applicable to almost all firms. However, the influence of the business model or industry structure could hinder the value relevance of inter alia goodwill impairments, because the information is not properly aligned to investors’ needs. Most literature on the value relevance of accounting metrics have excluded, for example, the financial sector due to their different business model and financial statements. However, it is interesting to include all industries in the thesis to obtain a complete view of the possible effects. To formally test for value relevance differences between the financial sector and other sectors, the sample is separated in financial companies and non-financial companies and separated in highly regulated industries versus non-regulated industries.

4 Lastly, through an extension of the accounting-based valuation model of Ohlson (1995) the value relevance of other potentially important metrics for investors is measured. The model is extended by splitting up the earnings component in an expected part and an unexpected part, being the latest analysts’ earnings consensus before the annual report and the earnings surprise respectively. Having a Big 4 auditor is added to the model to proxy for high quality financial statements as a control variable. Also, the Ohlson model is adjusted to include leverage to proxy for the degree of debt financing in the company, which to my knowledge has not been done before.

The results provide evidence that the goodwill impairments under IFRS are value relevant to investors in Europe and are negatively associated to the market value of equity. Contrary to the findings of Laghi et al. (2013), it implicates that the new impairment testing regime embedded in IFRS 3 is considered sufficiently reliable by investors as the outcomes of the test enter into the valuation assessment of a firm’s share price. The results also show that the institutional environment affects the value relevance of goodwill impairment and other accounting metrics, when split up in civil law and common law countries. The impairment is taken into account significantly more negative in common law countries than in civil law countries. Besides, the former takes into account an earnings surprise negatively, whereas an earnings surprise is positively associated with market value in civil law countries. Lastly, I find that industry differences affect the value relevance of goodwill impairment and other accounting metrics as well. These results are similar to the results of Laghi et al. (2013). Extending their work I find that non-financial companies as well as non-regulated industries take into account a goodwill impairment significantly more negative than their opposites.

The control variables applied show significance in most models. Leverage is significant in the market value of equity, indicating that the capital structure is taken into consideration in the valuation by investors. Although it has been under much debate already, it does again provide evidence to reject the capital structure irrelevance theorem by Modigliani & Miller (1958).

Concluding, the value relevance of accounting metrics is thus influenced by institutional and industrial factors, which might hinder the uniform application and interpretation across countries and industries. A suggestion for future research would be to test for these differences in isolation, whereas in this thesis it is likely that both forces are able to influence and bias each other in the results.

5

Literature review

Background on goodwill impairment testing

There are two types of goodwill, one being internally-generated goodwill and the other being goodwill that is recognized as a result of business combinations. Internally-generated goodwill can be, for example, the research costs incurred in an early phase of an internal project. The International Financial Reporting Standards (“IFRS”), however, do not consider these type of costs as intangible assets nor as goodwill and hence require these costs to be treated as an expense in the profit and loss statement, contrary to Local Generally Accepted Accounting Principles (“GAAP”) in certain countries. As a result, it cannot be recognized on the balance sheet (IAS 38: Intangible Assets) preventing its “articulation” in the accounts. The second type of goodwill arises as the result of a business combination and is the only type of goodwill to be recorded on the balance sheet for IFRS compliance purposes. The goodwill in a business combination reflects the difference between the fair value of the assets of the acquired company and the acquisition price paid by the acquirer (Pounder, 2013). It is determined through the allocation of the purchase price to all tangible and intangible assets and these identified assets are put at fair value at the valuation date. The unallocated part of the price paid, being the difference between the purchase price and the fair value of all assets, is recognized by the acquirer as goodwill (Shalev, Zhang & Zhang, 2013). It is described by Pounder (2013) as being the amount by which “the whole is greater than the sum of its parts”.

With IFRS 3, Business Combinations, a new accounting treatment of goodwill has been introduced. All European listed companies were required to comply with IFRS starting from 1 January 2005 and as at that date the Local GAAP reporting regimes are replaced. Most of the preceding Local GAAPs are based on a capitalize-and-amortize regime (Carlin & Finch, 2010), which is substantially different from the new IFRS regime of impairment testing. The previous method of treating goodwill was based on the premise that there is a maximum period that a company can benefit from goodwill. Within this period, the goodwill is reduced each year through systematic amortization until it is completely written down. IFRS, however, does not carry these features as the goodwill could theoretically remain on the balance sheet for an indefinite period of time (Carlin & Finch, 2010). Instead, IFRS 3 prohibits the systematic amortization of goodwill and requires companies to test for impairments on an at least annual basis, or even more frequently,

6 if there are any special events that indicate that there could be an impairment of the asset (AbuGhazaleh, Al-Hares & Haddad, 2012).

The methodological framework of applying IFRS in relation to impairment testing is laid out in IAS 36 Impairment of Assets. The impairment test is performed at the level of the smallest group of identifiable assets that generate cash inflows that are largely independent of the cash flows from other groups of assets. This is known as a Cash Generating Unit (“CGU”). There is an indication of an impairment, when the book value of the assets within the CGU is lower than the recoverable amount of these assets. IFRS has prescribed two approaches to determine the recoverable amount, which are the “Fair Value Less Cost To Sell” approach and the “Value-in-Use” approach. The former is the price that would result in a transaction between market participants for a similar asset (i.e. the fair value) minus the theoretical costs incurred for the sale of the asset. Finding an asset with exactly the same underlying characteristics (i.e. pay-off structure, risk characteristics, maturity) is not easy, which is one of the reasons that the Value-in-Use approach is applied by companies in the vast majority of impairment tests (Carlin & Finch, 2011). The Value-in-Use can be determined through a discounted cash flow model, which requires a view on the CGU’s future cash flows, their timings, risk profile and future growth. The present value of these cash flows is called the Value-in-Use and is compared to the book value of the CGU’s assets.

The IFRS treatment of goodwill is designed to increase the information content for the users of the financial reports and to provide more relevant information by reflecting the economic value of goodwill more closely (AbuGhazaleh et al., 2012). Relevance in accounting is usually referred to as “the pertinence of an economic construct to a user’s decision” (Kadous, Koonce & Thayer, 2012), which is always a challenge to establish when the nature of the assets is intangible such as goodwill (Bens, Heltzer & Segal, 2011). However, regulatory accounting bodies always try to strike a balance between relevance and reliability, of which the latter addresses the quality of measurement and predictability of that economic construct (Kadous et al., 2012). The new approach makes the impairment charges in fact less predictable in timing, in magnitude (Carlin & Finch, 2010) and more surprising to investors. Also, eliminating the systematic mechanism, i.e. the amortization, makes the need for users of the financial statements to evaluate the impairment more compelling (Hayn & Hughes, 2006). Critics argue that both the reliability and the relevance are questionable in the new approach and, additionally, have raised their concerns on the

7 managerial discretion that is conveyed in the impairment testing calculations (Schultze, 2005; Sahut & Boulerne, 2010).

Management discretion

A significant stream of research has tried to discover factors that can explain or predict whether an impairment should be taken in relation to the actual impairment taken. Especially the role of management has been under fierce debate in the impairment testing process. Before actual disclosure, management has to determine the magnitude of the impairment, if any, based on Value-in-Use approach as described above. Management is able to influence the outcomes of the impairment results by for example choosing a discount rate lower than the real weighted average cost of capital (“WACC”) of the CGU. The expected future cash flows are then discounted with a lower rate and thus a lower risk factor, which overstates the present fair value of the asset. As a result, the chance of an impairment is reduced. Carlin & Finch (2010) found that the discount rate used by Australian companies for impairment testing purposes was lower for more than half of the sample than independently estimated discount rates based on market data, suggesting that there is significant opportunism in choosing the appropriate discount rate.

The same holds for management’s expectations (i.e. estimates) about the future cash flows that the asset can generate, which is also subject to possible managerial optimism (Sahut & Boulerne, 2010). Both these forms of discretion are part of the private information that management possesses, but which is hardly observable for external auditors (Ball, Robin & Wu, 2003). Private information and estimates open the door for management to engage in earnings management (Sevin & Schroeder, 2005), which can significantly affect the stock price of the company. What increases the managerial discretion even more is that the impairment test is usually performed on the reporting-unit level, which is similar to a CGU. As a result, financial information at this level is not directly available for investors and therefore nearly impossible to collect and analyze (Lapointe-Antunes, Cormier & Magnan, 2009).

Besides the magnitude of the goodwill impairment, management can also exert significant discretion in the timing of reporting the impairment loss (Hayn & Hughes, 2006). Under SFAS 142, there is significant evidence that management avoids timely recognition of goodwill impairments due to, among other factors, conflicting agency-based incentives while there are sufficient market indications suggesting a write-off was justified (Ramanna & Watts, 2012).

8 Another important factor in the timing of the impairment charges is a possible violation of contracts with third parties due to special covenants. When debt covenants require a certain level of earnings and the company is on the verge of a breach, evidence suggests that an impairment is deliberately delayed by management to prevent missing the earnings benchmark (Beatty & Weber, 2006). The main driver for the untimely recognition is that a breach of the covenant usually results in a higher interest rate on corporate debt, which in turn reduces future profitability.

All these forms of managerial discretion described above have the same effect on the goodwill impairment charge in the profit and loss statement: the reliability of the impairment loss is reduced, which in turn could diminish its potential value relevance to investors.

Value relevance of accounting entries

In light of this thesis to test whether goodwill impairments have value relevance for investors, it is necessary to, firstly, define value relevance. In this research, the definition of Barth (2000) will be followed, being: “There is value relevance of an accounting entry when the entry is associated with some measure of market value of the company”. Thus, it can be regarded the capability of the information in the company’s accounting statements to sum up the market value of the company (Laghi et al., 2013). This implies that high value relevance is basically a strong association between a firm’s stock price, its earnings and equity book value, which together reflect the underlying economics of the firm. The measurement of value relevance will be elaborated on in the Methodology section.

Research has already focused on changes in value relevance of the financial statements between the locally accepted accounting standards and the newly introduced IFRS compliance. Accounting bodies designed IFRS to increase the comparability of the financial statements between firms and to enhance the quality of financial reporting for listed firms (Daske, Hail, Leuz & Verdi, 2008). Prior literature shows mixed results on the envisaged quality improvement of financial reporting through IFRS introduction. Paananen & Lin (2009) show that the financial reporting quality in Germany deteriorated after the introduction of IFRS, which seems counterintuitive given the goals set out by the accounting regulatory bodies. Ball, Kothari & Robin (2000a) argue that this might be due to a shift in the intended users of the financial statements. This (slow) shift in users is driven by signs of increasing economic importance of debt markets at the cost of the economic importance of equity markets. As a result of measuring the value relevance through an equity holder’s

9 perspective it could be decreasing for equity holders and increasing when measured from a perspective of debt holders.

Also, Barth et al. (2006) find in their sample that companies outside the United States that report in accordance with IFRS provide accounting information of lower quality than firms in the United States that report in accordance with U.S. GAAP. This is challenged by Leuz (2003) who does not find an increased information asymmetry between the regimes and Bartov (2005) who does not find significantly different quality levels in IFRS reporting compared to U.S. GAAP reporting. Contrary to the literature concluding a deterioration of accounting quality, there are several scholars who find that the value relevance of the accounting information has actually improved after the adoption of IFRS. For example, the information content of earnings and book value of assets has increased in the European banking industry (Agostino et al., 2011). Besides, the initial IFRS adoption by French companies is perceived by investors as an increase in the quality of the financial statements provided (Cormier, Demaria, Lapointe-Antunes & Teller, 2009). Lastly, Jermakowicz, Prather-Kinsey and Wulf (2007) find a significant value relevance increase of earnings for companies in Germany after the compliance to IFRS relative to market prices.

Beyond the more general accounting entries in the profit and loss statement, there has been research on the value relevance of more specific entries such as the value relevance of goodwill impairments. Before the implementation of IFRS 3, results evidence that amortization of goodwill does not have informational value to investors (Moehrle, Reynolds-Moehrle & Wallace, 2001). There are even indications that the previous straight-line amortization treatment of goodwill adds noise to the information content, which makes it harder for the investors that use earnings as a predictor of future profitability (Jennings, LeClere & Thompson II, 2001). In 1996, the research of Jennings, Robinson, Thompson II & Duvall (1996) already advised the regulatory accounting bodies to record goodwill as an asset that is reduced only in value when there is a clear indication of an impairment.

In the transitional phase to IFRS 3, Lapointe-Antunes et al. (2009) find a significant negative relationship between impairment losses and the share price of companies in a Canadian setting. Chalmers, Godfrey & Webster (2011) find that a goodwill write-off has value relevance as it aligns the financial statements with the underlying economic circumstances of the company in an Australian context. Both papers conclude that the impairment of goodwill enters into the valuation

10 assessment of investors, indicating that the results of the IFRS approach are considered sufficiently reliable by these investors and thus provide value relevant information. However, there have also been studies that do not find a significant relation between a goodwill write-off and the value or share price of the company. Following the change in regulation in Sweden, Hamberg & Beisland (2014) note that the value relevance of goodwill impairments is significant and negative under Swedish GAAP in addition to the general systematic amortization in place. However, in the IFRS period of their sample the coefficient lacks significance, indicating that the goodwill impairment has lost its value relevance to investors (Hamberg & Beisland, 2014). A possible explanation of a decrease in or even complete loss of value relevance is that investors do not take into account unverifiable estimates resulting from the goodwill impairment test (Watts, 2003), which supports the view of the managerial discretion inherent in testing process.

H1: Goodwill impairments are not value relevant to investors.

Research has focused on the value relevance of these entries mostly in a single country study. However, critics have raised their concerns about differences in institutional context that could possibly hinder the comparability of IFRS across countries due to significant scope in variation in applying IFRS (Nobes, 2015). Nobes (2006) suggests that these variations are not randomly applied, but rather stem from systematic country differences. National differences in hardly observable areas, such as an impairment, are therefore very probable to have an effect on the value relevance.

To my knowledge, the only research on differences between countries was done by Laghi et al. (2013). They perform an analysis on the value relevance of goodwill impairments across European countries. Their paper could not provide evidence that goodwill impairments are value relevant in most of the countries included in their sample except in France nor does the paper analyze possible differences between the countries. This thesis extends their work by taking into account a longer time period, additional countries and an analysis on the differences between countries. Lastly, a theoretical explanation is provided to explain the institutional and industrial forces, which was not provided in the paper of Laghi et al. (2013).

The institutional environment

Although no countries are the same, scholars have identified two broad families of law and legal institution, namely common law and civil law (the latter is also known as code law). The

11 classification of a country as either common law or a civil law reflects inter alia the historical background of the system of law, its judicial application within the system and other legal institutions (Glendon, Gordon & Osakwe, 1992). Common law was originally formed in the United Kingdom and relies on the active role of the judicial system to develop rules, where precedents and similar case rulings are the main sources to be used in court. As it is heavily dependent on prior rulings and interpretation and to a lesser extent on written laws, common law is usually characterized as being uncodified. The alternative to common law within the legal origins schema is civil law, which is regarded as being a comprehensive written statement of laws. Its main sources used in court are not precedent cases, but rulings rather depend on the writings of legal scholars and more strict application of written law. Civil law is divided into three different families, being the German, French and Scandinavian civil law. The common law and civil law were mainly spread throughout the world by means of colonialization.

Based on this classification, the seminal work of La Porta, López-de-Silanes, Shleifer & Vishny (henceforth “LLSV”, 1998) shows that the legal origin of a country is strongly related to the degree of investors protection measured through a function of voting rights, bankruptcy law and accounting quality. Common law countries offer investors the strongest protection, whereas French civil law countries have the weakest investor protection. Scandinavian civil law countries fall in the middle, which also holds for German civil law countries with the exception that these countries have the strongest (secured) creditor protection (LLSV, 1998). Similar results are found for differences in accounting quality among the legal families. Given the active role of judges in common law countries described above, it is not surprising that these countries offer better investor protection as judges are expected to rule on new situations based on similar cases. In contrast, judges in civil law countries are not supposed to rule beyond written laws and thus these laws cannot offer the same level of protection as is the case in common law countries (LLSV, 1998). Subsequently, LLSV (1997) provide evidence that a country’s degree of investor protection and its legal origin are related to a firm’s ability to raise external financing either through debt or equity. These factors are also explanatory for the relative size and extent of the financial markets. Especially, the civil law countries show much narrower stock markets than common law countries. Consistent with the notion that German civil law countries have the strongest creditor protection, results show that these countries relatively have the highest debt financing (LLSV, 1997). It indicates that investor protection does not only influences the size of the capital markets, but

12 influences corporate ownership as well. As a result of weak investor protection, civil law countries show more concentrated corporate ownership than common law countries, in which ownership is more widely dispersed among equity holders as they are offered better protection (LLS, 1999).

In conclusion, financial markets are thus strongly related to their legal origins. Characterization of a country’s financial market before the legal origin theory was along the lines of a financial market being either market-centered or bank-centered. Results are comparable to the legal origin theory as, for example, Germany is classified being a bank-centered market, meaning that the bank provides an important part of the firm’s capital. In a market-centered system, such as the United Kingdom, financing is rather provided through a dispersed and large number of unrelated shareholders in the form of equity (LLSV, 2000). As one can infer, there are great similarities between classifying a country as being bank-centered and being a civil law country. A limitation to the financial market classification, however, is that it is not a strong classification in the case where countries have both an underdeveloped equity capital market and an underdeveloped debt financing market.

Previous research argues that differences in the value relevance of accounting metrics across countries is attributable to the different accounting regimes. Ali & Hwang (2000) find that the value relevance of accounting metrics is lower for bank-centered countries relative to the market-centered countries, which support the notion that banks have direct access to a firm’s financial information and thus enjoy some private benefits of control to offset monitoring costs. Also, Ali & Hwang (2000) find greater value relevance for common law countries than for civil law countries supporting the view that the financial statements are more relevant to a dispersed group of equity shareholders. Arce & Mora (2002) suggest that significant value relevance differences across countries can be eliminated through harmonization of accounting principles. They note that, if the differences are in fact due to institutional factors, harmonization is unnecessary as it will not reduce these differences across countries. Arce & Mora (2002) find differences in value relevance similar to those reported by Ali & Hwang (2000). They conclude by recommending accounting harmonization to resolve the differences as they argue that institutional factors have decreased in last decades. Harmonization has taken place through the introduction of IFRS as per 1 January 2005 for all listed European companies, suggesting that significant cross-country differences in value relevance should have disappeared.

13 Research on the effects on the institutional environment shows that investors place greater emphasis on certain accounting metrics depending on the legal origin and the orientation of the financial markets, but focus has mainly been on the traditional items in the financial statements. For example, as common law countries and market-centered financial markets are mostly driven by equity shareholders, investors place greater value relevance on earnings than on the book value of equity (Arce & Mora, 2002). The opposite result was found for civil law countries. The results are not surprising as the shareholders in the common law countries get paid out whatever is left after all costs incurred and are the last stakeholders to receive their part in the form of dividends. Debt holders have senior claims over shareholders making them less concerned with the net income of the company. This could be an indication that a goodwill impairment charge would be more value relevant in common law countries as the charge flows directly through the profit and loss statement onwards to the residual claims of shareholders. Debt holders usually have a right to sell certain tangible assets in case of default, so any write-down on intangibles does not impact their claim on the underlying assets of the loan. Lastly, a goodwill write-down is an accounting entry without any cash changing hands, which could reduce its value relevance to investors. Based on previous research and the theories above, the following formal hypothesis will be tested:

H2: There are no significant value relevance differences in goodwill impairment between common law countries and civil law countries.

Industry differences

As Keener (2011) mentions, the European literature on value relevance predominantly applies a single country or single industry perspective without comparing the results across countries or industries, yielding little insight in the broader or international context of the results. Firms in a specific industry throughout Europe are more likely to operate in the same circumstances and, after the IFRS implementation, follow the same accounting standards. Also, the different business models across industries could, at least in theory, give rise to variation in accounting practices (Nobes, 2015) and influence the usefulness of certain accounting metrics.

Financial institutions, for example, are subject to the Basel regulation, which determines the regulatory capital required in relation to the operational risks of these institutions. These regulations directly impact the book value of equity, thereby potentially affecting its usefulness and value relevance (Laghi et al., 2013). Strong regulation in industries, such as healthcare and utilities, could also affect the value relevance to investors as the earnings in these sectors are

14 usually straightforward (Blacconiere, Johnson & Johnson, 2000). For example, for utilities it is not uncommon to be rewarded on the basis of cost-plus pricing, in which the company gets a fixed profit margin on the costs incurred. To create an incentive to become more efficient, the government can apply a decreasing profit margin scheme. The reward scheme is in place for a couple of years and is known to investors, in turn making earnings fairly predictable. Blacconiere et al. (2000) argues that the market value of equity for these companies is approximately equal to the book value of equity resulting in high value relevance. Subsequent deregulation of the utilities shows that earnings are becoming more value relevant than book value of equity, because earnings without regulation provide information to investors about the future prospects (Blacconiere et al., 2000). Other factors that increase the predictability of earnings are inter alia the stability of consumer demand in a particular industry. With a fairly stable demand over the years, it should be less difficult to predict normal earnings for investors relative to high growth and highly uncertain industries such as IT.

Ballas & Hevas (2005) find that the value relevance model for the consumer discretionary industry has the lowest explanatory power (measured through R-squared) whereas the healthcare sector has the highest power, but no conceptual explanation for the differences is provided. Besides, research has been done on the relative differences in the valuation multiples of firms across countries and across industries by looking at the explanatory power of industry-, country-, and firm-specific variables. It shows that the industry and country influences play a relatively limited role in the valuation of the firm, whereas firm-specific variables are able to explain a significant part of the variation in the observed valuation multiples (An, Bhojraj & Ng, 2010). Based on the above, the following hypotheses will be tested:

H3: There are no significant value relevance differences in goodwill impairment between common law countries and civil law countries

Analysts’ earnings consensus and earnings surprises

The accounting-based valuation model of Ohlson (1995) to measure value relevance is adjusted here to allow other variables, not part of the original work, to be included in the model to test their value relevance to investors. Ohlson’s model theorizes that the market value of the firm is a function of the present value of expected dividends and abnormal earnings. The present value of dividends proxies for the expected normal earnings and is in fact capitalized equaling the book value of equity (Ohlson, 1995). Subsequently, Collins, Pincus & Xie (1999) have amended the

15 model by replacing the abnormal earnings by current earnings and the present value of dividends by the book value of equity as they show that the book value of equity is an important omitted variable that serves as a proxy for the normal earnings in case of a loss-making firm. This thesis amends the model in two ways by using analysts’ earnings consensus as a proxy for current earnings, but also includes the earnings surprise to account for the abnormal or unexpected part of earnings that was part of Ohlson’s original model.

Analyst earnings estimates are an important source of information for the average investors as analysts act as an information intermediary between the investors and the firms they are following (Yu, 2010). The earnings consensus, which is the median of all analyst’s forecasts, should provide investors with a good proxy of the expected earnings of a specific firm. It is a more accurate estimation of earnings than the earnings estimated through historical time-series, because the forecasts allow for the incorporation of more timely information as well as current economic conditions (Healey & Papulu, 2001). Before the introduction of IFRS, it was shown that the that the degree of investors protection is positively related to the forecasting accuracy meaning that the accuracy is higher in common law countries (Hope, 2003b). A direct result of the harmonization of accounting standards is the increased number of foreign analysts covering firms. With uniform standards, it is argued that the costs of following a foreign firm are lower than when accounting standards differ from the analyst’s country standards. Also, forecasting errors as well as the earnings forecasts dispersion have decreased in 17 countries after implementation of IFRS, indicating that the forecasting accuracy of analysts has improved (Wang, Young, & Zhuang, 2008). The real earnings usually only become known to the public after the release of the annual report, which determines the unexpected part, or the surprise, of the earnings relative to earnings consensus. Subsequently to the release of the new information, investors update their beliefs on the future outlook of the company and the information contained in the annual report is impounded in the market value of equity (assuming market efficiency and rational investors).

Based on the industrial and institutional environment above there could be significant differences in the ways that investors take into consideration the analysts’ consensus as well as the earnings surprise. For example, Rountree, Weston & Allayannis (2008) find evidence that investors value smooth earnings as more volatility in earnings is associated with a lower market value. If investors indeed prefer smooth earnings, then the coefficient on the earnings surprise should be negative. A

16 priori, it is not clear what the effects of replacing current earnings by earnings consensus and the addition of earnings surprise will be in the accounting-based valuation model. Therefore no separate hypothesis are developed.

Corporate governance as control variables

The sole introduction of IFRS is not enough in itself to obtain high quality financial reporting, and therefore adequate corporate governance is of great importance (Verriest & Gaeremynck, 2009). These systems of corporate governance are put in place as ways of reducing possible managerial discretion in decision-making as outlined above (Wines, Dagwell & Windsor, 2007). The role of an effective corporate governance system is to act as a constraint on management by reducing their optimism or earnings management and to provide a signal of reliability to the investors (Lapointe-Antunes et al., 2009). On the other hand, Wines et al. (2007) note that the many aspects of the corporate governance system, such as the appointment of a director on the board, are self-regulated by the company and lack transparency to outsiders. In turn, this lack of transparency reduces the reliability of the signals sent to investors. However, firm-specific corporate governance characteristics can significantly impact the decision made by management to impair goodwill. There has been little research on these characteristics compared to research on the accounting treatment of goodwill. This is remarkable as it is widely recognized that corporate governance structures are important conditions for the quality and reliability of financial statements (Verriest & Gaeremynck, 2009). Although part of the corporate governance is already implemented through the legal environment, this thesis will include having a Big 4 auditor and analyst consensus as control variables in the corporate governance domain, for which no separate hypotheses will be developed. The auditor and the analysts are professional agents acting as a gatekeeper for both the firm and investors by providing assurance on the financial statements and company analyses respectively (Von Koch, Nilsson & Eriksson, 2014).

External auditing has an important function in ensuring an effective corporate governance system as it provides “an independent assessment of the accuracy and fairness with which the financial statements represent the results of operations” in accordance to for example IFRS (Hope, Langli & Thomas, 2012). Early work of DeAngelo (1981) focuses on the theoretical background of audit quality in relation to the size of the audit firm. She argues that audit quality is hard and costly to evaluate for investors and they therefore develop a surrogate, i.e. size, for audit quality. As the size of the audit firm increases, its total fees grow due to the increased client portfolio. This makes a

17 larger audit firm less sensitive to retaining any one client. Therefore, they will be more likely to report a breach in the client’s accounting systems and deliver higher quality audits than small firms (DeAngelo, 1981). Lastly, due to more reputational capital in place large audit firms have more to lose when investors discover misrepresentations, which results in a damaged auditor reputation. The spill-over of the damage to other clients makes it harder for the company to retain the client base (DeAngelo, 1981). Given that large audit firms have more to lose, they are more incentivized to deliver high quality audits. This seminal paper resulted in a huge amount of accounting literature using Big N, most commonly the Big 4, as a proxy for audit quality. Big 4 is defined here as the four largest audit firms measured through revenues in the world, being KPMG, EY, PwC and Deloitte. Despite some exceptions, most scholars find results in line with the theory outlined above and observe a significantly positive relation between earnings quality and Big 4 auditors (see for example Francis & Wang, 2008). Also, the value relevance to investors of the financial statements is higher for firms that have a Big N auditor (Lee & Lee, 2013). Stokes & Webster (2010) argue that Big 4 auditors enforce stricter compliance with IFRS due to their high reputational capital and find that the goodwill impairment charge better reflects the true economic value of goodwill only in the presence of the sort high quality accounting Big 4 auditors bring.

Lastly, leverage is added to the model as a control variable and measures the effect of the capital structure on the market value of equity. Flowing from the work of Modigliani & Miller’s (1958), it is argued that the financing choice of the firm does not affect the value of the company, i.e. the capital structure is irrelevant. Although there has been much debate on and proof against their proposition, it is of interest to add the variable to the accounting-based valuation model of Ohlson (1995) as there are only variables in the model that affect equity. If Modigliani & Miller’s (1958) theory holds, the variable should be insignificant in the estimations. No formal hypotheses are developed for the control variable. However, from the institutional perspective one would expect that the degree of debt financing is higher in civil law countries than in common law countries, but a priori it is unclear how this would influence the value relevance.

18

Methodology

This thesis will build on and extent the research of Laghi et al. (2013). IFRS 3 was introduced in 2004 and listed companies were to comply with the new regulation as per 1 January 2005. There has been research on the transitional phase of complying with Local GAAP to IFRS and its effects on the value relevance of goodwill impairments. Therefore, this thesis will exclude the transitional phase and will focus on measuring the value relevance of the impairment charge after possible start-up write-downs due to the transitional nature of regulation. The sample of firms will consist of the following countries: the Netherlands, Germany, France, Spain, Belgium, the United Kingdom, Italy and Portugal. Flowing from the work of LLSV (1998) the countries are divided according to their common law and civil law classification. The common law group will consist of the United Kingdom and the Netherlands. The Netherlands was originally part of the French civil law countries, but Ballas & Hevas (2005) note that the Dutch accounting profession and academics strongly influence the accounting standards and therefore the Netherlands exhibit significant similarities with common law countries. France, Spain, Belgium, Italy and Portugal are all French-origin civil law countries and Germany is classified as the only German-origin civil law country. The data period is set at a period between 2007 and 2014. The data of 2005 and 2006 is excluded to correct for the possible effects of the transitional phase as described above. The sample period is thus drawn from the onset of the financial crisis to our gradual emergence from it today.

To test for the value relevance of the accounting metrics, we will follow the approach that has been used in prior research relating to this topic (e.g. Lapointe-Antunes et al., 2009; AbuGhazaleh et al., 2012; Laghi et al., 2013), which is an adjusted accounting-based valuation model that views the market value of the company as being the book value of its equity and the current earnings (Ohlson, 1995; Collins et al., 1999). The use of a multivariate Ordinary-Least-Squares (OLS) regression could result in cross-sectional dependence, which occurs due to unobserved common factors. However, none of the current literature that applies an OLS regression in combination with an accounting-based valuation model has mentioned the possible influence of cross-sectional dependence. Contrary to prior research, this thesis will measure the value relevance through a Fixed Effects regression, but will perform several tests to check the issues of cross-sectional dependence and a possibly incorrect specification of the regression.

19 (1)

𝜇𝜇

𝜐𝜐

𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑖𝑖𝑖𝑖 = 𝛼𝛼 + 𝛽𝛽1𝐵𝐵𝑀𝑀𝑀𝑀𝑀𝑀𝑖𝑖𝑖𝑖+ 𝛽𝛽2𝐶𝐶𝑀𝑀𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖+ 𝛽𝛽3𝐸𝐸𝑀𝑀𝐸𝐸𝐸𝐸𝐶𝐶𝐸𝐸𝐸𝐸𝑖𝑖𝑖𝑖 + 𝛽𝛽4𝐸𝐸𝑀𝑀𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖𝑖𝑖+ 𝛽𝛽5𝐶𝐶𝐺𝐺𝑀𝑀𝑖𝑖𝑖𝑖 + 𝛽𝛽6𝑀𝑀𝐸𝐸𝑀𝑀𝑖𝑖𝑖𝑖+ 𝛽𝛽7𝐵𝐵𝐺𝐺𝐶𝐶4𝑖𝑖𝑖𝑖+ 𝜇𝜇𝑖𝑖+ 𝜐𝜐𝑖𝑖𝑖𝑖

MVAL= the market value of the equity at the end of the fiscal year in which the goodwill impairment is taken. The variable is deflated by the market value of equity at the

beginning of the respective fiscal year.

BVAL= the book value of the equity at the end of the fiscal year in which the goodwill impairment is taken minus the carrying value of the goodwill at the end of the fiscal year. The variable is deflated by the market value of equity at the beginning of the respective fiscal year.

BVGW= the book value of the goodwill at the end of the fiscal year plus the reported goodwill impairment charge in that fiscal year. The variable is deflated by the market value of equity at the beginning of the respective fiscal year.

EARNCON= the latest earnings forecast consensus by analysts before the earnings announcement plus the reported goodwill impairment. The variable is deflated by the market value of equity at the beginning of the respective fiscal year.

EARNSURP= the earnings surprise measured as the difference between actual earnings and the earnings forecast consensus plus the reported goodwill impairment. The variable is deflated by the market value of equity at the beginning of the respective fiscal year. GIL= the goodwill impairment charge that is reported by the company in the fiscal year. It is expressed as a positive number. The variable is deflated by the market value of equity at the beginning of the respective fiscal year.

LEV= the debt-to-equity ratio of the company based on the net debt and the market value of equity at the end of the fiscal year. The variable is deflated by the market value of equity at the beginning of the respective fiscal year.

BIG4= a dummy variable that proxies for a firm having a Big 4 auditor. When the company has appointed a Big 4 auditor, it takes the value of 1 and 0 otherwise.

In the model equation, represents the time-varying error term, whereas the idiosyncratic error or disturbance term that varies across time and cross-section is represented by . Further discussion on the interaction effect between these terms is out of scope for this thesis.

20 The original accounting-based model is altered to separate the book value of goodwill from the book value of equity. Also, the goodwill impairment is separated from book value of goodwill by adding it back. The impairment is also added back to the earnings consensus and the surprise to prevent its inclusion and thus its effect in the model twice. Additionally, leverage and Big 4 auditor are added as an extension of the original model and serve as control variables.

The deflation of variables is necessary to reduce scale problems that are associated in the value relevance model as variables are affected by differences in firm size. Large listed firms usually have large variables (i.e. high revenue or net income) compared to relatively small listed companies and results of a regression can be driven mostly by these relatively small subset of large sample firms. In turn, this scale effect can cause heteroskedasticity, and consequently inefficiently estimated coefficients and an inflated R-squared.

Prior research applying a value relevance model deflated the accounting variables by the number of outstanding shares at the end of the year in which the goodwill impairment is taken (Lapointe-Antunes et al., 2009). However, it is argued that the number of shares outstanding is an arbitrary choice that could give rise to new size differences (Gil-Alana, Iniguez-Sanchez & Lopez-Espinosa, 2011). Therefore, other variables are preferred over shares outstanding. Literature has not formed an uniform idea on the best variable to use as a deflator, but the majority of the scholars recommends to use the market value of equity at the beginning of the fiscal year. Concluding, in this research all variables will be deflated by the market value of equity at the beginning of the fiscal year. As part of the robustness checks other deflators are considered as well.

21

Results

Descriptive statistics

As described above, the companies are all listed on stock exchanges in Europe. The sample includes all companies that have been listed in the full period from 2007 to 2014 or were listed in at least two years in the sample period. The two-year period is required in order to deflate the variables by the lagged market capitalization of equity. The sample contains only companies listed on the largest stock exchanges in the Netherlands, Germany, the United Kingdom, Belgium, France, Spain, Portugal and Italy. For example, in the Netherlands there are three stock exchanges composed by trading activity. Only the companies from the AEX, containing the most actively traded companies, are included in the sample. The constituents and tickers per index in the sample period were collected from Bloomberg. To obtain the other variables described in the methodology section S&P Capital IQ was used. S&P Capital IQ is a financial database that can be utilized to analyze company fundamentals of both public and private companies. All data in S&P Capital IQ can be traced back to the original source, which usually is the annual report of a company. This function makes the data transparent to verify and enhances the quality of the sample obtained.

Based on the Bloomberg tickers and financial data of S&P Capital IQ, the total number of firm-year observations is 3,896 (Table 1). The sample size is reduced by 20% due to limited information for some companies and the listing requirement of two years. In total, there was not enough information available to obtain a complete observation in the model for 861 observations. For 512 observations there was no market value of equity available for the fiscal year and for 67 datapoints no lagged market capitalization could be derived from S&P Capital IQ. Also, there were no earnings consensus estimates for 270 datapoints and for 10 datapoint there was no reported net income. Lastly, 2 observations did not have a reported book value of equity, which brings the total number of missing observations to 861. The final sample consists of 3,035 complete firm-year observations in period from 2007 to 2014. When the control variables are added to the model another 223 observations are lost. This is mainly due to missing information on the auditor of the financial statements.

22 Table 1

Sample selection across total observations*.

Country Index Constituents Total firm-year

observations Missing information Included firm-year observations Netherlands AEX25 25 376 -128 248 Belgium BEL20 20 240 -37 203

United Kingdom FTSE100 100 1,360 -338 1,022

France CAC40 40 440 -43 397 Germany DAX30 30 328 -75 253 Spain IBEX35 35 432 -103 329 Italy FTSEMIB 40 488 -76 412 Portugal PSI20 20 232 -61 171 Total 3,896 -861 3,035

*Please note that the table above is for the basic model only exlcuding the control variables LEV and BIG4. The number of firm-year observations for the total model is 2,812.

Previous research on goodwill impairments has only included companies in their sample that reported positive book values of goodwill on their balance sheet or only included companies that recorded an impairment, which could bias the outcomes due to significant sample selection. This research has not made any selection bias based on balance sheet or income statement data. The only sample selection that is applied is to only include the companies that are traded most, i.e. are listed on the most actively traded exchanges. This should not impact the outcomes of the results as there is no relation to goodwill impairments nor to the value relevance of accounting metrics. Also, by focusing on the most actively traded companies it is likely that these companies are important in Europe from an economic and social perspective.

Subsequently, the companies included in the final sample are grouped based on their primary industry sector, as provided by S&P Capital IQ, to test for possible cross-industry effects. The primary industry is regarded as a coarse seperation between companies, but any further breakdown would reduce the sample size of some industries to only 20 observations. Based on retaining sufficiently large industry samples, the division by primary industry results in 10 different sectors.

23 Table 2

Sample selection across industries.

Index AEX25 BEL20 FTSE

100

CAC40 DAX30 IBEX 35 FTSE

MIB PSI20 Total Industry Industrials 78 1 172 72 32 68 38 28 489 Financials 42 60 236 58 47 85 165 20 713 Consumer Staples 23 24 91 32 15 19 16 16 236 Materials 28 30 126 40 59 16 16 38 353 IT 16 8 18 32 15 12 8 0 109 Energy 29 0 62 24 0 16 32 8 171 Telecommunication 9 16 32 8 8 16 8 16 113 Consumer- Discretionary 23 16 193 87 32 33 70 24 478 Healthcare 0 28 38 16 31 8 15 0 136 Utilities 0 20 54 28 14 56 44 21 237 Total 248 203 1,022 397 253 329 412 171 3,035

*Please note that the table above is for the model exlcuding the variables LEV and BIG4. The number of firm-year observations for the total model is 2,812.

As could be inferred from Table 2, companies in the financial sector account for more than 20% of the total observations whereas the Information Technology sector only accounts for 3% of the total number of observations. However, S&P Capital IQ does not allow for a more general division in primary industry sectors. Besides, the sample observations of the FTSE 100 account for one third of the total sample. These size differences are inevitable as the largest indexes of the countries are inherently of different sizes.

The following table shows the distributional statistics of the sample after deflation. The distributional characteristics show that the minimum value and the maximum value of some of the variables are widely dispersed. However, removing all outliers might introduce a new selection bias that affects the true regression line of the dependent variable (Winship & Mare, 1992).

Table 3

Descriptive statistics after deflation by lagged market value of equity

Variable Mean Median Min Max Std dev.

Market value of equity 1.116 1.049 0.055 21.074 0.684

Book value of equity 0.568 0.364 -4.884 16.847 0.984

Book value of goodwill 0.321 0.167 0.000 8.252 0.663

Earnings consensus 0.074 0.071 -1.515 4.024 0.150

Earnings surprise -0.037 -0.003 -3.915 0.636 0.208

Goodwill impairment 0.012 0.000 0.000 1.555 0.074

Leverage 2.413 0.544 0.000 25.212 1.183

24 Lastly, it is necessary to test whether the explanatory variables are highly correlated with each other. A simple method to investigate the presence of near multicollinearity is to look at the matrix of correlations between the individual variables. A rule of thumb here is that a correlation of 0.8 is considered high (Brooks, 2008). The results in Table 4 do not show near multicollinearity issues as the strongest correlation is between the book value of equity and the market value with a value of 0.45. As a first observation, the goodwill impairment is negatively correlated to the market value of equity, which is line with the prediction that it should lead to a lower market value. The other variables in the correlation matrix have the expected sign in relation to the market value of equity. Additionaly, a Variance Inflation Factor (“VIF”) test is performed as a severity measure of multicollinearity. It serves as an additional check on the correlation outcomes to rule out multicollinearity issues in the sample. In general, the values of the VIF test should not exceed 5.0 in small samples and 10.0 in large samples (Brooks, 2008). The VIF test (untabulated) for the sample shows that the highest value is 2.33 for the goodwill impairment, which is still below the treshold of 5.0 for small samples. Therefore, multicollinearity is not an issue in the sample.

Table 4

Pearson correlation coefficients including the respective p-values between brackets.

Market value of equity Book value of equity Book value of goodwill Earnings consensus Earnings surprise Goodwill impairment Leverage Big 4 Market value of equity 1 (-) Book value of equity 0.450 (0.000) 1 (-) Book value of goodwill 0.188 (0.000) -0.018 (0.000) 1 (-) Earnings consensus 0.214 (0.023) 0.205 (0.000) 0.328 (0.000) 1 (-) Earnings surprise -0.232 (0.101) 0.082 (0.000) 0.167 (0.000) -0.181 (0.000) 1 (-) Goodwill impairment -0.153 (0.000) -0.344 (0.011) -0.247 (0.000) -0.432 (0.0000 -0.431 (0.000) 1 (-) Leverage -0.088 (0.000) 0.264 (0.000) 0.117 (0.000) -0.051 (0.002) -0.251 (0.000) 0.245 (0.000) 1 (-) Big 4 Auditor 0.004 (0.817) 0.017 (0.367) -0.033 (0.077) 0.023 (0.227) -0.005 (0.739) 0.001 (0.775) -0.026 (0.169) 1 (-)

25 Fixed Effects assumptions

The Fixed Effects regression is applied in the value relevance model as outlined in the methodology section. As the data consists of a time series of firms that have at least two observations in the sample, the basic structure that is used in Eviews is a Dated Panel. This structure allows for incorporation of fixed and random effects in the model to counter intra-group clustering (Devalle, Onali & Magarini, 2010), which is not possible when using an unstructured and undated sample.

Firstly, to test whether it is appropriate to include fixed or random effects, a model without either of the effects is estimated. The model is then altered to include both Period fixed effects as well as Cross-section fixed effects. Its appropriateness is tested for by applying the Redundant Fixed Effects – Likelihood ratio with the null hypothesis that fixed effects are redundant. Eviews tests for the Cross-section effect, Period effect as well as the interaction of the fixed effects. Based on the Test-statistics and their associated p-values (p<0.0000 for all three measures), the null hypothesis of redundant fixed effects is strongly rejected. Lastly, the model is altered to only include cross-sectional random effects and a Correlated Random Effects – Hausman Test is performed. The Hausman Test is employed to compare the fixed effect and random effect models, where the random effects model is preferred under the null hypothesis. The null hypothesis is strongly rejected as the Chi-Squared Statistic and associated p-value (p<0.0000) provides evidence that a fixed effect model is the more appropriate one. Concluding, the regression model will apply Cross-section fixed effects and Period Fixed effects as the tests employed above provide evidence that it is appropriate to do so.

The last correction in the specification of the model is to account for heteroskedasticity. Heteroskedasticity issues cannot be estimated in the Panel Data structure in Eviews and are therefore estimated in an unstructured and undated setting. White’s Heteroskedasticity Test can test the null hypothesis of no heteroskedasticity against the alternative hypothesis of heteroskedasticity. For the sample the null hypothesis is strongly rejected (p-value<0.0000), indicating the presence of hetereskedasticity. However, it should be noted that the test only indicates heteroskedasticy of an unknown and general form (Eviews Userguide, 2013), whereas correcting for heteroskedasticity in panel data requires the choice of White’s cross section, White’s time or White’s diagonal-correct coefficient estimates of the principal diagonal (or variances) in the variance-covariance matrix. Unfortunately, there is no specific test that can be employed to

26 form an idea on which kind of White’s coveriance method is the most appropriate. The estimation results of the three options are depicted in Table 5.

The basic model includes the total sample with all observations, but without any industry or country separation included. Leverage and Big 4 are added only in the Total model. The results of the regressions (Table 5) are altered through the model specification with White’s cross section, White’s period and White’s diagonal covariance methods respectively.

Table 5

Fixed Effects regression of the basic model*

No White errors White Cross-section White Period White Diagonal

Model 1 Model 2 Model 3 Model 4

Variable Coefficient T-Stat. Coefficient T-Stat. Coefficient T-Stat. Coefficient T-Stat.

Intercept 0.58323 (0.0000) 42.107 0.58323 (0.0000) 9.073 0.58323 (0.0000) 5.409 0.58323 (0.0000) 8.010 Book value of equity 0.55169 (0.0000) 41.066 0.55169 (0.0000) 6.477 0.55169 (0.0000) 4.233 0.55169 (0.0000) 5.969 Book value of goodwill 0.54964 (0.0000) 23.818 0.54964 (0.0000) 10.200 0.54964 (0.0001) 3.847 0.54964 (0.0001) 5.976 Earnings consensus 0.75249 (0.0000) 10.800 0.75249 (0.0000) 4.360 0.75249 (0.0000) 4.189 0.75249 (0.0000) 4.888 Earnings surprise -0.12002 (0.0422) -2.033 -0.12002 (0.6122) -0.507 -0.12002 (0.6346) -0.475 -0.12002 (0.6023) -0.521 Goodwill impairment -2.09675 (0.0000) -11.752 -2.09675 (0.0000) -5.427 -2.09675 (0.0000) -4.668 -2.09675 (0.0000) -6.275 N 3,035 3,035 3,035 3,035 Adjusted R-Squared 59.41% 59.41% 59.41% 59.41% F-statistic 10.934 (0.000) 10.934 (0.000) 10.934 (0.000) 10.934 (0.000) DW Statistic 2.087 2.087 2.087 2.087

Fixed effects Cross sectional & period effects

Cross sectional & period effects

Cross sectional & period effects

Cross sectional & period effects *Fixed cross sectional effects and fixed period effects are added as the null hypotheses of random effects were rejected through the Hausmann test and counter possible intra-group correlation.

Overall, Model 1 in table 5 (without White’s errors) yields good results in terms of statistics. Autocorrelation is usually a problem when the Durbin-Watson Statistic (“DW Statistic”) is below 1.0 or above 3.0, indicating the presence of positive autocorrelation and negative autocorrelation respectively. The DW-statistic has a value of 2.087 in all models meaning that autocorrelation is not an issue. However, as a result of heteroskedasticity in the sample there is sufficient reason to believe that the estimated standard errors in Model 1 are too low and thereby inflating the t-statistic

27 and p-values. When introducing White’s errors in Model 2, 3, 4 it can be inferred that the t-statistics are substantially lower due to increased standard errors for all variables. The earnings surprise loses its significance after correcting by White’s heteroskedasticity errors. The t-statistics using White’s period error (Model 3) seem to result in the most conservative values and therefore, although arbitrary, the rest of the models will be estimated using White’s period errors to correct for heteroskedasticity issues.

Main results

Model 1 in Table 6 (same as model 3 in Table 5) shows the results for the basic model. All the variables in the model have the expected coefficient sign. Book value of equity, book value of goodwill and earnings consensus are positively associated to the market value of equity and are statistically significant (p-value<0.0000). A priori it was expected that an earnings surprise would be positively related to the market value of equity as it provides information that the company has performed better than the analyst forecast consensus. Contrary to expectations, the model shows a negative sign on the coefficient indicating that a positive earnings surprise is perceived negatively by investors in the valuation of the equity. However, no formal conclusions can be drawn as the earnings surprise coefficient lacks significance. The variable that is of particular interest, namely the goodwill impairment, has a negative coefficient sign and is statistically significant (p-value<0.0000), which indicates that a goodwill impairment is value relevant to investors. In other words, it provides evidence that investors take into account a goodwill impairment negatively in their valuation assessment. Besides, it is an indication that goodwill impairments are measured reliably enough to be taken into account in the market value of equity, i.e. the share price of the company. Concluding, there is sufficient evidence to reject Hypothesis 1 that goodwill impairments are not value relevant to investors.

In model 2 of Table 6, the leverage of the firm is added to the regression to test for its significance and to observe any increase in the R-squared obtained. As can be inferred, the variables of the basic model remain statistically significant with their predicted signs except for the earnings surprise variable. Leverage is significant (p-value<0.01) and negatively associated with the market value of equity and adding the leverage variable to the model increases the adjusted R-squared by about 2.0%. Model 3 adds the Big 4 auditor to the basic model (p-value<0.025) and yields about the same results as adding the leverage variable, although this is only positive and significant at the 2.5% level. These control variables are thus associated with the market value of equity.

28 Table 6

Fixed Effects regression of the Total model*

Basic model Leverage model Big 4 model Total model

Model 1 Model 2 Model 3 Model 4

Variable Coefficient T-Stat. Coefficient T-Stat. Coefficient T-Stat. Coefficient T-Stat.

Intercept 0.58323 (0.0000) 5.409 0.66647 (0.0000) 7.003 0.41496 (0.0020) 3.092 0.50196 (0.0001) 3.914 Book value of equity 0.55169 (0.0000) 4.233 0.53236 (0.0000) 4.699 0.54701 (0.0000) 4.531 0.49800 (0.0000) 6.358 Book value of goodwill 0.54964 (0.0001) 3.847 0.51527 (0.0000) 4.140 0.55111 (0.0003) 3.645 0.50642 (0.0000) 4.401 Earnings consensus 0.75249 (0.0000) 4.189 0.58132 (0.0003) 3.596 0.76644 (0.0001) 4.056 0.58520 (0.0001) 3.832 Earnings surprise -0.12002 (0.6346) -0.475 -0.21083 (0.2725) -1.097 -0.05930 (0.8595) -0.177 -0.15778 (0.5267) -2.977 Goodwill impairment -2.09675 (0.0000) -4.668 -1.66275 (0.0004) -3.558 -1.96541 (0.0001) -3.937 -1.59254 (0.0029) -2.978 Leverage -0.02299 (0.0001) -3.987 -0.03411 (0.0124) -2.502 Big 4 Auditor 0.18788 (0.0129) 2.487 0.23225 (0.0244) 2.243 N 3,035 3,035 2,812 2,812 Adjusted R-Squared 59.41% 61.41% 61.75% 65.48% F-statistic 10.934 (0.000) 11.778 (0.000) 11.200 (0.000) 12.954 (0.000) DW Statistic 2.087 2.114 1.957 2.066

Fixed effects Cross sectional & period effects

Cross sectional & period effects

Cross sectional & period effects

Cross sectional & period effects *Fixed cross sectional effects and fixed period effects are added as the null hypotheses of random effects were rejected through the Hausmann test and counter possible intra-group correlation. White’s period standard errors are applied to correct for heteroskedasticy.

The total model (model 4) includes both the Big 4 Auditor and the leverage variables and shows that both variables are significant at the 2.5% level. The variables from the basic model have the same significance levels in the total model as they had in the basic model. Taken together, the explanatory power of the model increases by about 10% (or 6% in absolute terms) after the addition of the leverage and Big 4 Auditor as control variables.

Institutional differences

Potential differences in the institutional environment are measured through the classification of common law or civil law countries. The countries are split up in their respective groups. The results for the civil law countries are first estimated without the German or French origin separation and

29 Table 7

Fixed Effects regression of the Total model* with a separation between common law countries and civil law countries**

Total model Common law Civil law German civil law French civil law

Model 1 Model 2 Model 3 Model 4 Model 5

Variable Coefficient T-Stat. Coefficient T-Stat. Coefficient T-Stat. Coefficient T-Stat. Coefficient T-Stat.

Intercept 0.66647 (0.0000) 7.003 0.72949 (0.0000) 8.155 0.61915 (0.0000) 4.666 0.70295 (0.0000) 17.015 0.60468 (0.0000) 4.147

Book value of equity 0.53236 (0.0000) 4.699 0.44160 (0.0000) 4.775 0.60062 (0.0002) 3.777 0.32465 (0.0000) 4.726 0.60701 (0.0002) 3.703

Book value of goodwill 0.51527 (0.0000) 4.140 0.57403 (0.0000) 5.589 0.39255 (0.0216) 2.300 0.63105 (0.0000) 7.536 0.38520 (0.0285) 2.193 Earnings consensus 0.58132 (0.0003) 3.596 1.06330 (0.0540) 1.929 0.38945 (0.0084) 2.638 1.32443 (0.0004) 3.593 0.35968 (0.0128) 2.491 Earnings surprise -0.21083 (0.2725) -1.097 -0.55450 (0.0427) -2.029 0.09946 (0.5439) 0.607 0.56269 (0.0800) 1.759 0.08323 (0.6303) 0.481 Goodwill impairment -1.66275 (0.0004) -3.558 -4.26813 (0.0009) -3.340 -0.93133 (0.0508) -1.955 -1.80782 (0.0131) -2.503 -0.89670 (0.0655) -1.843 Leverage -0.02299 (0.0001) -3.987 -0.00627 (0.4212) -0.805 -0.02325 (0.0000) -4.218 -0.01801 (0.0006) -3.499 -0.02347 (0.0004) -3.569 N 3,035 1,270 1,765 253 1,512 Adjusted R-Squared 61.41% 58.42% 66.70% 58.38% 67.43% F-statistic 11.778 (0.000) 9.785 (0.000) 14.754 (0.000) 8.685 (0.000) 15.030 (0.000) DW Statistic 2.114 1.956 2.135 1.908 2.144

Fixed effects Cross sectional & period effects

Cross sectional & period effects

Cross sectional & period effects

Cross sectional & period effects

Cross sectional & period effects

*Fixed cross sectional effects and fixed period effects are added as the null hypotheses of random effects were rejected through the Hausmann test and counter possible intra-group correlation. White’s period standard errors are applied to correct for heteroskedasticy.

**Common law countries include the United Kingdom and the Netherlands. The civil law countries in model 3 include all other countries. German civil law includes only Germany and the French civil law countries include France, Belgium, Italy, Spain and Portugal in line with LLSV (1998). Please refer to Appendix A for the regression outputs per country.