1
Accounting for intangibles: The value-relevance of R&D disclosures in the
technology sector.
A comparison between Germany, France and the UK.
Thomas Jobse
5991595
Universiteit van Amsterdam
Faculty of Economics and Business
Accountancy & Control
Master Thesis
1
stSupervisor: dr. A.J. Brouwer RA
2
ndSupervisor: dr. F.H.M. Verbeeten MBA
October 27
th, 2013
2
Accounting for intangibles: The value-relevance of R&D disclosures
IntroductionIntangible assets have become more and more important in the last decades. Intangibles represent a large part of the capital of many of today’s firms. Despite the increased importance of intangibles, current financial statements provide very little
information about these assets (Lev, 2003). Lev (2001) shows in his book that the market-to-book ratios of the S & P 500 companies have risen from approximately 150 % in the early 70’s to more than 700 % in the late 90’s (see fig. 1). Though the market-to-book ratio’s decreased to about 600 % by the year 2000, this implies that the current accounting values of assets do not capture all future benefits that investors expect them to generate.
Current Generally Accepted Accounting Principles do not allow firms to account for most intangibles on the balance sheet. For example, critics of the current accounting practice argue that ‘R&D outlays generate some of the most priced economic assets in the economy, and that accountants’ refusal to recognize these expenditures as assets seriously impairs the credibility and relevance of financial reporting (Kanodia, 2004).
The purpose of Financial Accounting is to provide users of financial statements with information that is useful for efficient decision making (Canibano, 2000). Financial
statements therefore need to contain information that is relevant for investors to make rational investment decisions. In order to make rational decisions, investors will need to know what the actual value of the assets of a firm is. Investors value a firm’s assets, based on the
expected future revenue the assets will generate, so investors need to have information about the future performance of the assets. However, critics raise concerns that financial measures provide little insight into a company’s future performance (FASC, 2002).
Given the recent increasing importance of intangibles and the critics on accounting practices to have missed to address the value of these assets in the past, it is inevitable that investors seek to find information on all assets including intangibles, that helps them to make informed investment decisions. Investors want to benefit from the intangibles of companies, but need to have proper information about their value, as they do not want to run the risk of companies overstating the values of their assets. Overstatement of the value of companies in the capital markets results in significant losses for investors when stock prices revert to their fundamental values (Garcia-Ayuso, 2003). It is also in the interest of companies that make large investments in intangibles to provide investors with information on the value of their intangibles. Investors help companies to fulfil their capital needs. Undervaluation reduces the
3
ability of a firm to raise additional capital (Garcia-Ayuso, 2003). It is therefore to be expected that recent annual reports of companies contain more decision useful information on
intangibles. An important part of firms’ intangible assets is created directly or indirectly by means of research and development (R&D activities). This paper aims to examine the value-relevance of disclosures of information on R&D activities.
Fig 1. Historical market-to-book ratio’s
Prior literature
The definition of intangibles
To date, there are not many studies which have specifically focused on R&D and it’s value-relevance. There are however several prior studies available, which have focused on the valuation of other components of intangibles or the valuation of intangibles in general. In order to understand the relevant prior literature, it is therefore important to first gain a clear definition of intangibles, previous to determining the place R&D takes within this broader concept of intangibles.
Canibano et al (2000) are concerned with defining intangibles. They argue that the definition of intangibles provided by the main regulatory bodies are rather similar, as they
4
usually characterize intangible assets as non-physical and non-monetary sources of probable future economic profits accruing to the firm as a result of past events or transactions. Further, Canibano et al (2000) note that intangibles are often identified with goodwill and understood as the excess cost of an acquired company over the value of its net tangible assets. Garcia-Ayuso (2002) however, argues that it is not reasonable to state that the difference between the market value and the book value of equity is entirely due to the existence of intangibles not reflected by the accounting model for three reasons. First, the book value of equity may be negatively biased due to the undervaluation of tangible and financial assets. Second, even under prudent GAAP accounting, it is possible that shareholders’ equity does not include all intangible liabilities. Third, stock prices may not always be considered as unbiased estimates of the value of companies as there is an overwhelming amount of evidence for the existence of market anomalies such as the under- or overreaction to earnings announcements followed by a post-announcement drift. Society’s interests in the correct valuation of intangibles are significant. The significant increase in the stock prices of the so-called new economy firms during the late 1990s and the subsequent burst of the technological bubble have not only resulted in enormous losses for investors, but also resulted in dramatic job slashes.
Classification of intangibles and the outlining of R&D
The literature has identified many different categorisations of intangibles. However, there is no generally accepted classification (Canibano et al, 2002). For this study, we will follow the categorisation that Verbeeten (2008) and many other researchers have adopted. This means that intangibles are divided into four categories. The first category is
infrastructure assets. This category includes the technologies, methodologies and processes which enable the organization to function, like corporate culture and corporate spirit. The second category is intellectual property assets. This category includes research and development, discovery, patents, copyrights and innovation. These intellectual asset
components are usually created by the employees or are acquired. Decisions can be made to invest in, or replace these intangibles. The third category is market, relational or external assets. This category includes brands, customer or supplier relations and strategic partners. The fourth category is human capital assets. This category includes personell and
compensation policies that create specific cultures and/ or competences.
It has now become clear that R&D assets make part of intellectual property assets, a category of intangible assets. This knowledge also assists in understanding what R&D assets
5
are not. R&D assets are no infrastructure assets, no market assets, neither human capital assets.
Valuation of intangibles
Kanodia et al (2004) are concerned with a very fundamental question, namely: Should intangibles be measured ? They conclude that intangibles should be measured only when their relative importance in constituting the firm’s capital stock is high and when they can be measured with sufficiently high precision. In all other cases, attempts to separate intangible investments from operating expenses would be counterproductive. However, the research of Kanodia et al (2004) is not based on empirical testing. They base their conclusions on the outcomes of a number of statistical and mathematical equations, relying on a number of extreme assumptions. If one or more of these assumptions are incorrect, the conclusions of Kanodia et al (2004) will have to be revisited. One of the assumptions they have employed, is that the firm’s manager makes all decisions in the best interest of the firm’s current
shareholders. This contradicts with the outcomes of many empirical studies on the consequences of agency conflicts within firms (e.g Heath, 2009).Moreover, many well-known accounting scandals in the past have learned us that managers often act to their own self-interest, rather than in the company’s best interest (Satava et al, 2006).
Garcia-Ayuso (2002) tries to identify why intangibles are currently valued so inefficiently, and finds 4 impediments to the efficient valuation of intangibles. First, the quality of information i.e. ‘a general lack of meaningful and useful disclosures about intangible assets.’ Second, market imperfections i.e. there are no markets for intangibles. Third, limited capability of financial analysts. Fourth, ethics i.e. inefficient valuations may be caused by misleading management practices such as disclosure of false information. Possible solutions that Garcia-Ayuso (2002) proposes are changes in regulation in order to improve the market imperfections and improvement of the current accounting model.
If a revision of the accounting models is to be expected on the short term is however questionable. The FASC (2002) has presented a paper in which they examine academic research papers on nonfinancial performance measures, and concludes that nonfinancial performance measures are relevant for predicting future financial performance and valuing corporate equity. However, according to the FASC (2002) the type of measure that is relevant for equity valuation is context dependent, which becomes problematic if standard-setters’ goal is to mandate a consistent set of required disclosures for all companies. Given these
6
example, customer satisfaction, quality, and the like would not best serve investors. Rather, the Committee believes that companies should be encouraged to provide such disclosures voluntarily.
Prior research on the value (relevance) of R and D investments
Lev and Sougiannis (1996) are concerned with the value-relevance of R&D capital. They make an estimate of the R&D capital of a large sample of public companies. They find that adjustments on net earnings and book values for R&D capitalization are value-relevant to investors. Subsequently, Lev and Sougiannis (1996) analyse the relationship between firms’ R&D capital and subsequent stock returns. They reported a significant relationship between these variables, suggesting either a systematic mispricing of the shares of R&D intensive companies, or a compensation of market risk associated with R&D. Chan et al (2001) have also scrutinized the relationship between R&D intensity and subsequent stock returns. They argue that this relationship would not exist in an efficient market, as stock prices in efficient markets share prices would reflect the value of R&D assets along with all other assets. Based on a sample of data from 1970, 1975, 1980, 1985,1990 and 1995 covering all firms listed on NYSE, AMEX, and Nasdaq, the study confirms that there is indeed no relationship between R & D intensity and stock returns. This contradicts with the results of the study of Lev and Sougiannis (1996). In addition, the results of the study of Chan et al (2001) reveals that there is a positive relationship between R&D intensity and return volatility.
The studies of Chan et al (2001) and Lev et al (1996) do not take possible scale effects of R&D investments into account. Ciftci and Cready (2011) saw this as an opportunity to do a similar study as Chan et al (2001) conducted. Motivated by Schumpeter’s (1942) proposition that larger firms are better positioned than smaller firms to implement and
successfully exploit R&D efforts, they scrutinise the relationship between R&D intensity and future earnings and divide their sample into four quintiles of size. From their results, Ciftci and Cready (2011) conclude that both future earnings and earnings volatility impacts vary with scale. More specifically, they show that scaled larger firms’ R&D investment is associated with substantially higher future earnings realisations and substantially lower
earnings variability than is comparable scaled R&D investment by smaller firms. The research does not explicitly examine where in the R&D process scale effects arise. However, the analysis does suggest the underlying advantages stem primarily from post-innovation development and marketing stages in the D portion of R&D, where size-related attributes
7
such as organisational stability, market presence and reputational capital are likely to be key factors in maximising innovation value.
Prior research on disclosures regarding intangibles
Just the fact that a firm has invested a certain amount in R&D, does not mean that this firm also succeeds to develop the new products or innovations this firm was hoping to
develop when making the investments. Therefore, investors can benefit from additional information to assess the probability that profits will ensue from the expenses. To date, no research papers are known that focus solely on (the value-relevance of) R&D disclosures. However, there are papers known that examined (the value-relevance of) related theoretical constructs, like intangibles in general and intellectual capital. Intellectual capital is closely related to research and development, as intellectual capital is an important resource for R&D activities.
Bozzolan et al (2006) examine whether there is a relationship between company’s industry type, i.e. knowledge intensive or traditional and the quantity of disclosures on intellectual capital. Based on a sample of companies in both Italy and the UK, they find that that companies in the knowledge intensive sector disclose more information on intellectual capital in their annual accounts, than companies in the traditional sector. A secondary finding of this study is that there is a positive relationship between firm size and the level of
disclosures on intellectual capital. Bozzolan et al (2006) also expected to find significant differences in the level of disclosures on intellectual capital between the Italian and the UK companies, based on differences in their legal system (civil/ code vs common law), stock market size (small vs large), ownership structures and the influence of regulatory and professional bodies. However, no statistically significant differences in the quantity of
disclosures on intellectual capital between Italian and UK companies were found. Vergauwen and Alem (2005) also studied differences in the levels of disclosures on intellectual capital between countries, namely The Netherlands, France and Germany. Drawing upon
classifications of national accounting systems as optimistic and transparent (The Netherlands) as opposed to conservative and secretive (France) according to the Gray scale, they expected more intensive disclosure in The Netherlands compared to Germany and France. This was however not confirmed by the results, as the level of disclosures in the Netherlands was relatively low. Based on the proposition that audit conservatism i.e. choosing an international audit firm leads to less disclosures on intellectual capital, because international audit firms
8
seek maximum risk avoidance, Vergauwen and Alem (2005) expected more extensive disclosures in France. This was confirmed by the results.
Verbeeten (2008) investigates the value-relevance of intangible assets for companies in the service sector in Belgium, France, the UK and the US in the years 2003-2005. He found that all 62 companies disclose some information on intangible assets. Most of the disclosures are about market assets (e.g. brands, partnerships). Other information is about human capital assets (e.g. competences), infrastructure assets (company culture, IT) and intellectual property. Verbeeten (2008) constructed his own Intangible Asset Disclosure Index (IADI), which indicates the level of disclosures of information on intangible assets. The value of the index is determined by the number of hits that searches of words that are
associated with intangibles in the annual reports of the companies in the sample produce. The relationship between the change of the IADI in sequential years and changes in the share price that cannot be explained by changes in net income per share, book value of equity per share or earnings quality is examined. The conclusion of this research is that an increase in intangible asset disclosure is related to the market value of a company in the service sector in US firms, yet not for European firms.
Motivation & Research question
In the first sections of this paper, the importance of intangibles for investors and society have been demonstrated. Intangible assets represent a large part of today’s firms’ assets, and are therefore important to investors. More accurate valuation of firms’ intangible assets may help to prevent bubbles in market values from being created. The bursting of such bubbles have resulted in dramatic job slashes in the recent past.
R&D activities have been identified as one important element that contributes to the creation of intangibles. Prior studies have revealed that research and development
expenditures are value-relevant. However, as R&D activities can either result in new products and innovations or in failure, it is expected that additional information can be value-relevant to investors. No prior studies have focused on the value-relevance of R&D disclosures specifically. The research question of this paper is therefore formulated as follows: Are R&D disclosures relevant for investors ?
9
Relevance of disclosures: Signalling theory
In the above sections of this paper, references to research regarding the value-relevance of disclosures have been used to introduce the research question of this paper, which is also about the value-relevance of disclosures. Disclosures are only ‘value-relevant’, if investors respond to these disclosures by making decisions on whether to buy or sell stocks, or consolidate their current positions. The assumption that investors may or may not respond to disclosures cannot be made without theoretical support. This support can be found in
signalling theory.
Signalling theory originated in the early 1970s. Signalling theory is based on the concept of asymmetric information. Information affects decision-making processes. Individuals (e.g. investors) make decisions based on public information, which is freely available, and private information, which is only available to certain individuals or groups of individuals. Before the introduction of the concept of information asymmetry, economic models assumed perfect information. Perfect information means that all individuals have access to all available information. According to most countries laws, all investors should have access to the same information at the same time, to create equal chances for stock traders. In practice, there will be some exceptions e.g. where large investors have private conversations with representatives of the investees. This research paper is specifically looking into the response of investors to information which is published in the annual reports of companies. Annual reports are public information, which is available to all investors. This does however not imply perfect information. There is still information asymmetry between investors and investees. Disclosures about R&D activities are in most cases voluntary, which means that companies can choose which information about their R&D activities they publish in their annual reports or not.
Signalling theory is useful for describing behaviour when two parties have access to different information (Connelly et al, 2011). Typically, one party, the sender or signaller, must choose whether and how to communicate (i.e. signal) that information. The other party, the receiver, must choose how to interpret the signal and subsequently, how to respond to that signal. Typically, the signaller communicates information regarding its own unobservable qualities. Michael Spence (1973), being one of the founders of signalling theory, presented a classic example of signalling theory in his article ‘Job Market signalling’. In this article, Spence explains the signalling function of education in the job labour market. Employers do not have sufficient information about the qualities of job candidates. Potential candidates for
10
the jobs are aware of this and try to convince the employers of their qualities by obtaining educational degrees. The educational degrees are interpreted as reliable signals by the employers, as people who lack the qualities needed would be unable to complete the educational programs. For the purpose of this research, disclosures in annual reports are considered signals to convince investors of their unobservable qualities to transform R&D expenses into profitable products. This research assumes that companies will in most cases choose to only communicate positive news about R&D projects to let investors know that their investments are in good hands. Moreover, it is assumed that investors will interpret these signals as being reliable. It is not assumed that companies are always honest in their
communications and that investors always believe in the messages sent by companies. However, misleading signals would proliferate until receivers learn to ignore them. Thus, to maintain their effectiveness, signal will be structured in such a way that dishonest signals do not pay (Connelly et al, 2011). For the purpose of this research, it will therefore be assumed that investors have some trust in the information companies disclose in their annual reports.
Hypotheses
Once more, prior research has led to the conclusion that R&D expenditures are value-relevant. However, in order to carry out this research rigourously, before starting with testing the value-relevance of R&D disclosures, the value-relevance of R&D expenditures
themselves will be tested first. The outcome of this test can be used to support analysis of the results of the other hypothetical tests. Consistent with previous research in this area
(Verbeeten 2008; Barth et al, 2001) price levels is considered an adequate indicators for value-relevance. The first hypothesis is therefore stated as follows:
H1: R&D expenditures are positively associated with share prices
Recently, Ciftci and Cready (2011) have demonstrated significant scale effects of R&D investments. More specifically, they find that the positive association between the level of R&D intensity and future earnings increases with firm size. A possible explanation for this can be found in the research of Cohen and Klepper (1996) who suggest that despite the fact that smaller firms account for a disproportionally large number of patents, large firms benefit from cost spreading advantages, leading to a higher output per R &D dollar. In order to test if the distinction between the effectiveness of R&D investments also apply to the sample of this research, the second hypotheses is as follows:
11
H2: R&D expenditures are more positively associated with share prices of large firms compared to smaller firms
As most disclosures on R&D are voluntary, it can be expected that companies will disclose more information on R&D activities when these activities are more successful. Moreover, disclosures regarding the intellectual assets (which include R&D assets)
disclosures are often value-relevant to investors (Wyatt, 2008). Further, as successful R&D activities are likely to generate future earnings (Ciftci and Cready, 2011), it is expected that more R&D disclosures lead to higher share prices. Therefore, the next hypothesis is
formulated as follows:
H3: R&D disclosures are positively associated with share prices
As R&D disclosures are describing R&D activities that lead to R&D expenses, it expected that there is interaction between R&D disclosures and R&D expenses. It is expected that an extensively described R&D activity is perceived more positively by investors than a more briefly described R&D activity. Likewise, the following hypothesis is added to this research:
H4: R&D expenses are more positively associated with share prices when the activities related to these expenses are more extensively described
Prior research has in some cases revealed significant geographical differences regarding intangibles disclosures (Vergauwen & Van Alem, 2005; Verbeeten, 2008).
Although clear explanations for these differences have not yet been provided by the literature, suggestions for the causes of these differences have been made, and are often referred to as ‘institutional’ or ‘cultural’ differences’. For example, Vergauwen & Alem (2005) suggest that the level of conservatism can explain differences in quantities of information disclosed in different companies. Using the Gray scale, they explain that the national accounting system of Germany is considered moderately conservative and highly secretive by reputation, whereas the national accounting system of France is considered highly conservative and low secretive. On this basis, less extensive disclosure may be expected in Germany, relative to France.
Another difference in the level of disclosures may be caused by the differences in information needs of the investors in various countries, as explained by Joos & Lang (1994).
12
Whereas in Germany, firms have relied more heavily on debt financing by a relatively small number of banks, with concentrated equity ownership, capital in the UK is provided by numerous small shareholders. France has historically been more similar to Germany, with a strong focus on reporting to creditors. When companies are financed by a select group of investors, it is likely that there is more emphasis on information exchanged with the investors directly, than through the official annual report. Therefore, based on the common resources of capital in these countries, it is likely that more information is disclosed in the annual reports of UK companies compared to companies in France and Germany.
As prior research has led to mixed results, the next hypothesis is stated in the null-form:
H5: The value-relevance of R&D disclosures does not differ amongst countries.
As mentioned before, Ciftci and Cready (2011) have demonstrated significant scale effects of R&D investments. As the actual positive effect of R&D investment has been
demonstrated to be larger in larger firms, it is expected that a higher level of R&D disclosures of larger firms also has a higher value to investors:
H6: R&D disclosures are more positively associated with share prices of larger firms
Methodology
Research method
The research method used in this paper can be described as content analysis. Though this method is not undisputed, it is assumed that content analysis is the best possible method to examine the research question of this paper. For related research questions , content analysis has also been successfully used (e.g. Bozzolan et al. 2006; Vergauwen and Van Alem, 2005; Verbeeten, 2008). Guthrie et al (2004) state that content analysis is the most popular research method in the related area of intellectual capital reporting. This is supported by the proposition that content analysis is a good instrument to measure comparative positions and trends in reporting. This type of research is also often used in the related area of Social and Environmental Reporting (SER). Guthrie et al (2004) explain the similarities between SER and R&D reporting through the use of legitimacy theory. Like SER, R&D reporting is voluntary. R&D expenses cannot be fully capitalized. Therefore, it is impossible for
13
corporate success. Voluntary, narrative disclosures are therefore used by companies to legitimise their innovative status.
Unit of analysis
Like Verbeeten (2008) and Vergauwen and Alem (2005), words will be used as the unit of analysis for this study. More specifically, a self-constructed index will be developed based on frameworks available in the existing literature (e.g. Abdolmohammadi, 2005). The advantage of a self-constructed measure is that it is likely to capture what it is intended to capture (Healy and Palepu, 2001; Verbeeten, 2008). The advantage of using words is that there is no problem with reliability, as the research is easily replicated. The disadvantage of the usage of a self-constructed index is the subjectivity involved in the composition of the index. The index will be computed by counting the number of hits in the annual reports of the sample companies on the search terms listed below.
To most of the words or combinations of words, a weight of 1 will be assigned. Other words or combinations of words will be weighted by a factor larger or smaller than 1. Some words or combinations of words are more likely to indicate disclosures on R&D than others. For example, a hit on ‘R and D’ is obviously a clear indication for a disclosure on R&D. A hit on the combination of words ‘new technology’ is also likely to indicate a disclosure on R&D, as new technologies are developed through R&D activities. However, there is a reasonable possibility that this is a hit on a disclosure regarding e.g. a new technology used in the
company’s existing production process which has been purchased from a supplier rather than a new technology which has been developed by the company itself. Therefore, a hit on ‘R and D’ will be assigned a weight factor of 3, which is significantly higher than the weight factor of 2,0 which will be assigned to a hit on ‘new technology’.
The RDDI also takes into account that a hit on certain words automatically lead to a hit on combinations of words. As explained above, this research intends to assign a weight of 3 to a hit on ‘R and D’. However, there is obviously no difference between for example the combination of words ‘research and development’ and it’s abbreviation ‘R and D’. Therefore, a hit on both each of the words ‘research’ and ‘development’ as well as the combination
‘research and development’ are assigned a weight of 1. This leads to a total weight of 3 for the combination ‘research and the development’, which is the same as the weight of 3 assigned to just ‘R and D’.
14
The complete RDDI is composed as follows:
Search terms Weight
Research and development 1
Research 1,5 Development 0,5 R&D 3 R&D 3 Innovation 1,5 Innovative 1,5 New product 1,5
New developed product 1,5 Newly developed product 1,5 Product application 0,5 Technology 1 Technologies 1 New technology 1 New technologies 1 Research project 1,5 Patent 2
Table 1: Composition RDDI
Sample
The sample will be limited to companies in the technology sector. The reason to limit this research to a specific sector is to avoid biases caused by differences in relevance of the subject of R&D within the sample. The choice for the technology sector is made because R&D is considered highly relevant in this sector. This study assumes that if R&D disclosures are not relevant in the technology sector, R&D disclosures are not relevant in other sectors either.
In order to test the hypotheses, basic data is be collected from the Compustat database. The Compustat database is chosen as a data source, because this database contains a large set of data items from a large number of companies in Germany, France and the UK. To the best of the knowledge of the researcher, there is no database available which contains the data items needed for this research from a larger amount of companies than Compustat. This research is meant to be as well-underpinned as possible. Therefore, data from all companies that fit the selection criteria (i.e. based in Germany, France or the UK and technology sector) will be included in the sample for the purpose of H1 and H2. As the work in order to derive the RDDI’s is rather time consuming, not all companies included in the selection for H3-H6
15
are included in the sample. At the other hand, in order to have sufficient data to be able to draw conclusions about the population, the sample cannot be too small. A sample of N=30 is often considered large enough to draw conclusions by statisticians, which is the reason 30 companies per country will be selected for the sample to test H3-H6.
In order to be able to compare the data (e.g. share prices) between years, the same data items will be collected for 3 years. This research looks at relationships in the recent past. By now (2013), it is expected that the 2011 annual reports of most companies will be available. The years selected are therefore 2009-2011.
Data source
As much data as possible, will be gathered from the Compustat databases, as Compustat is known to be a reliable database, containing financial information items regarding a large set of companies. The annual reports needed to obtain the word counts for the RDDI calculations, will be downloaded from the investor relations pages of the selected companies.
Introduction to the regression models
The regression analysis models that are used to test the hypotheses, are based on the assumption that share prices are determined by investors, based on their expected future cash cash-flows. This research will analyse the significance of the R&D expenses (H1 and H2) and RDDI (H3-H6) variable in the models. The research will make use of a share price model. For the most part, valuation models that form the basis for tests in the value-relevance literature are developed in terms of the level of firm value, whereas examining changes in share prices or returns is an alternative approach (Barth et al, 2001). According to Wyatt (2008) trade-offs exist between both models. Wyatt (2008) states that there may be little change in the variable of interest in a narrow return interval (e.g. change in stock price over a year). For example, R&D intensity (or disclosure) levels may be highly value-relevant in a stock price levels regression but the value might change very little on an annual basis. Barth et al (2001) state that the key distinction between value relevance studies examining price levels and those examining price changes, is that the former are interested in determining what is reflected in firm value and the latter are interested in determining what is reflected in changes in value over a specific period of time. Thus, if the research question involves determining whether the accounting amount is timely, examining changes in value is the appropriate design is the appropriate research design choice. For this study, it is assumed that the choice of the appropriate dependent variable for examining the relationship between disclosure amounts
16
(rather than accounting amounts) and equity market values also depends on whether the research is interested in what is reflected in firm value or what is reflected in changes in value over a certain period of time. Further, Barth et al (2001) note that selection of which approach to use depends on econometric assumptions. Given the share price model is the common method to determine value-relevance, the share price model will be used for the research, rather than the alternative share returns approach.
For the first part of this research project (H1 and H2), the following share price model is estimated:
SPt,i = β1t,i+ β2YRt,i + β3CRt,i + β2VARNIt,i + β3SIZEt,i + β4EPSEXRD,i + β5CEQPSt,i + β6RDEPSt,i + εi
Where:
SP = Share Price at fiscal year-end,
YR = YeaR dummies for 2010 and 2011 respectively ( 2009 = benchmark),
CR = CountRy dummies for France and the UK respectively (Germany = benchmark), VARNI = variance in net income (a proxy for earnings quality),
SIZE = company SIZE, measured as net income in millions, EPSEXRD= Earnings Per Share Excluding R&D expenses , CEQPS = Common / ordinary EQuity Per Share,
RDEPS= R&D Expenses Per Share .
Before proceeding with the second part of this research project (H3-H6), the same regression model will be run for the sample used for H3-H6, to compare this smaller sample to the sample used for the first part of this research. As all companies in the sample for H3-H6 are included in the larger sample for H1 & H2, and these companies are selected at random, no large differences are expected.
For the second part of this research project (H3-H6), the following share price model is estimated:
SP1204t,i = β1t,i+ β2YRt,i + β3CRt,i + β2VARNIt,i + β3SIZEt,i + β4EPSEXRD,i + β5CEQPSt,I + β6RDEPSt,i + β7RDDIt,i + β8RDDIRDEPSt,i +εi
17
Where:
SP1204 = Share Price at April 12th, following fiscal year-end, RDDI = Research & Development Disclosure Index,
RDDIRDEPS = RDDI * RDEPS = Interaction variable to capture possible enhancing effects of R&D disclosures on R&D investments and finally, share price.
The choice to measure the effects of R&D disclosures in annual reports on share prices per April 12th, is based on the publication dates of the annual reports in the sample. The
researcher has registered the publication dates of 54 of the annual reports in the sample (20%). The researcher found that most annual reports are published in the second half of March. It is expected that the information in the annual reports is fully absorbed by the investors, a few weeks later. April 12th is a date that falls outside the weekend in all of the 3 years 2010, 2011 and 2012.
Further, to enable the researcher to analyse the results by year and by country, the year- and country dummies can also be used as split variables to carry out regressions on subsamples, i.e. specific years and/ or countries.
Finally, to be able to compare between ‘large’ and ‘small’ companies, this distinction will be made based on the average revenue of the companies selected. Companies (cases) with total revenue (REVT) above the median are classed as large. A dummy variable for large companies will be added, where ‘small’ is the benchmark:
LARGE = Dummy for large companies (cases)
This model can be interpreted using Ohlson’s (1995) model, where the disclosure index is ‘the other information variable’ (Verbeeten, 2008). The Ohlson (1995) model represents firm value as a linear function of book value of equity and the present value of expected future abnormal earnings. The attractiveness of the Ohlson model is, that it provides a testable pricing equation that identifies the roles of accounting and non-accounting
information (Lo and Lys, 2000). To understand this model, it is important to be aware of the underlying assumption of information asymmetry. The model would not hold if perfect information would be assumed. In perfect markets there is no substantive role for accounting or other information as there are no information asymmetries. If stock price already
incorporates all available information, financial statements provide no added value (Lo and Lys, 2000).
18
In order to assess the data, first the descriptive statistics will be calculated and analysed (i.e. mean, median, 10%, 25%, 75% etc). Moreover, the data will be assessed, by analysing one on one relationships of the ‘basic independent variables’ (i.e. net income, equity) as well as the coefficients of interest i.e. R&D expenses per share and the RDDI with the dependent variable (i.e. share prices). Then, to assess the effects of our control variables (i.e. size and variance in net income) their one on one relationships with the other variables will be evaluated. Finally, the linear regression analysis will be performed on both of the models. The outcomes will be first be used to test the models by calculation and analysis of the F-statistic and the R2. If the data and the models are considered correct, the coefficients and their significance will be used to assess the hypotheses.
Possible biases and limitations of the research method
A general limitation of a regression methodology is that it measures association, rather than causation (FASB, 2002). From the results of this research, it will be possible to tell whether the amount of R&D disclosure is associated with stock-prices or returns. This does not guarantee that investors actually use the R&D disclosures. Instead, investors may use other information that is correlated with R&D disclosures. However, the above
argumentations do provide reasons to expect certain relationships, which is sufficient to at least provide suggestions for the possible causations. Further confirmation of these causations may involve additional research. For instance, interviews with investors may reveal more detail about the way R&D disclosures are used in practice.
Wyatt (2008) notes that a factor adversely impacting the inferences available from value-relevance studies is the problem of omitted correlated variables. A common problem is the effect of differences in firm size on the test results. Size differences can produce
significant results that have little to do with the intrinsic attributes of the item of interest. To eliminate this problem, a control variable for firm size is included in the models. Apart from its function as a control variable firm size will also be used to test hypotheses 4 and 5.
Another possible bias that is to be prevented relates to earnings quality. Francis et al (2008) suggest that firms with poor (good) earnings quality will issue more (less) expansive disclosures because information asymmetry between the firm and investors is higher (lower) in such firms. Like Verbeeten (2008), variance in net income (VARNI) will be used as a proxy for earnings quality, to follow up on the suggestion of Francis et al (2008).
19
The final anticipated bias is related to the countries where the firms in the sample reside. Vergauwen and Alem (2005) suggest that country specific attributes like regulatory barriers may significantly influence the amount of intellectual capital disclosures, which was partly confirmed by their research results. Moreover, the research results of Verbeeten (2008), suggest that intangibles disclosures are value-relevant for investors of US companies, yet not for European companies. Therefore, a country dummy is incorporated in the models that will be used for this research. Distinguishing between countries is also necessary in order to test hypothesis 4.
Selections & data collection
Collection of company information from Compustat
As discussed before, the Compustat database will be used as the data source for the basic data of this research. To be more specific, the ‘Compustat Global – Fundamentals Annual’ and ‘Compustat Global – Securities Daily database are selected as the relevant data source for this research.
Before filtering out only the companies within the technology sector, for each of the relevant countries, data of all companies available in the database was downloaded from the Compustat database. This means that for each country, one download was executed. The
screening variable ‘country code’ has been used to select each country, i.e. DEU for
Germany, FRA for France and GBR for the UK. Data items classed as identifying information like company name and city and SIC code were included in the download to give the
researcher the opportunity to quick scan the data e.g. to establish that the data is indeed data from companies in Germany. A date range from January 2007 until December 2011 has been chosen for each of the downloads. The reason to also include data related to the year before and the year after the selected years is just to give the researcher the opportunity to do some extra checks on the data e.g. to check if companies have been included in the download which did not yet exist in the year before the selected period or which did not exist anymore in the year after the selected period. Finally, data items needed to perform the regression models have been selected in the Compustat query.
Selection of companies and defining high-tech
In order to be able to filter out only the companies in the technology sector, certain criteria that identify companies as being technology based or ´high-tech’ companies need to
20
be available. Moreover, a definition of high-tech in terms of one or more of the available company identifying items within the database is needed to enable the researcher to filter out the companies that are relevant to this research. The Bureau of Labor Statistics (BLS) in the United States developed a list of high-technology industries based on Standard Industrial Classification (SIC) codes (Hecker, 1999). The list is based on measures of industry employment in both R&D and technology-oriented occupations, using Occupational Employment Statistics surveys from 1993 to 1995 in which employers were asked to
explicitly report the number of R&D workers and technology-oriented occupations accounted for a proportion of employment was at least twice the average for all industries surveyed. These industries have at least 6 R&D and 76 technology-oriented workers per 1.000 workers. The Office of Technology Policy converted the BLS list of SIC codes to the 1997 edition of the North American Industrial Classification System (NAICS) codes using the concordance between the two classification systems. The NAICS codes were updated in 2002, and this revised coding system was used for this research. Companies which are classed in Compustat according to one of the NAICS codes shown in appendix 1 are not excluded from the
selection of companies to be used for this research.
Filtering out only the companies in one of the above high-tech industries from the Compustat download, lead to a first selection of 325 companies in Germany, 270 in France and 544 in the UK. Filtering out only the companies that reported R&D expenses in all of the years between 2007 and 2011, significantly decreased the size of the selection; 136 high-tech companies in Germany, 78 in France and 196 in the UK, were left. In order for the selections to be consistent, all companies that have a financial year that differs from the calendar year have been removed from the list of selected companies. This has especially shortened the list of selected companies from the UK. This is not considered a problem, as there is a sufficient number of companies from the UK left. Further, companies for which too much data (e.g. net income) was missing, were also removed from the selection. For companies of which a limited number of data items was missing, the missing data was entered manually based on annual reports downloaded from the internet.
The first major problem with the data the researcher ran into so far, was a large
amount of missing data in the data cells for the earnings per share (EPSINCON) data item. To solve this problem, numbers of shares outstanding were downloaded from the ‘Compustat Global – Securities Daily’ database. Shares outstanding (as well as share prices) were linked to the right companies, using a vertical lookup basis the International Security Identification Number (ISIN). Earnings per share were then calculated by dividing net income (NICON,
21
derived from the Fundamentals Annual database) by the number of shares outstanding. For some companies share prices and the numbers of shares outstanding were not available in Securities Daily database. These companies were also removed from the selection. The final list of selected companies for the assessment of H1 and H2 consists of 112 companies in Germany, 68 companies in France and 83 companies in the UK.
In order to test hypotheses 3 and further, the data of 30 companies needs to be enriched with RDDI’s. For this purpose, random numbers were generated using the ASELECT function in Excel and linked to the selection of companies for H1 and H2. For each country, the 30 companies with the lowest random numbers were selected for the RDDI calculations. Search terms containing the company name and ‘annual report’ were put into the Google search engine to find the PDF versions of the annual reports needed for the research. Most annual reports were easy to find. In some cases links to de PDF’s of the annual reports were found directly, in other cases they were found on the investor relations pages of the companies. For some companies in Germany and France, no English versions of the annual reports were found. If no English annual report was found, the company was excluded from the selection for H3-H6, and the company with the next lowest random number was added to the selection. An alternative would be to translate the search terms into French and German. However, this could lead to comparability problems. All downloaded annual reports were then searched by means of the search function in Adobe Acrobat reader. All word counts were registered in an Excel sheet. The word counts were then multiplied by the factors as stated in table 1, to end up with a complete list of companies and their RDDI points. Subsequently, the average RDDI points for 2009 was calculated (297) and set to 100. To arrive at the final RDDI’s, the RDDI points of all companies were divided by 297 and multiplied by 100. See appendix 2 for a complete list of selected companies, word counts and RDDI’s.
Finally, to test H2 and H6, the selections of companies (cases) need to be divided into ‘small’ and ‘large’. The average yearly revenue in the sample was calculated at 370 million. All cases where revenue exceeds 370 million EURO or 320 million GBP are classed large. In total, 115 companies are classed large.
Data analysis
Initial data analysis; checking the downloaded data
Before uploading the data into SPSS, the data downloaded from the Compustat databases was checked thoroughly. First, the data was checked for zero-values. The number of zero-values in the earnings per share (EPSINCON) column in the ‘fundamentals annual’
22
database appeared unacceptably high. For this reason, the researcher decided not to use this data item as input for the regression analyses. Rather, the amounts in the column for net income (NICON) were divided by the numbers of shares outstanding, which were pulled from the ‘securities daily’ database. The data in the NICON column also showed some zero-
values. The missing company profits (or losses) were derived from the annual reports of the companies. After adding all the missing data, the data was checked for extremely low values. As the companies in the Compustat database are expected to be large, stock listed companies, no profits, costs or losses below 1.000 EURO or GBP should be expected. However, several values < 1.000 were identified in the NICON column, as well as in the column for R&D expenses (XRD). All low values were checked with the annual reports of the companies. In most cases, the low numbers needed to be multiplied by 10.000 to get to the right numbers. All low numbers which appeared incorrect were amended manually. After amending all incorrect missing and low values, sense checks were carried out. E.g. it was checked if dividing earnings per share by the share price resulted in realistic returns. Where this was not the case, the data was checked with other sources. Share prices were checked with Yahoo! Finance for example. Where needed, the data was amended.
Descriptive statistics R&D expenses versus share price sample (for H1 & H2)
After completing the initial data checks, descriptive statistics were calculated for all variables in the model for the first part of the research. The descriptive statistics for most of the variables are shown in table 2 below. The descriptive statistics for the year dummies for 2010 and 2011 (2009 = benchmark) are not included in the table below. For every company, 3 years of data is included in the research. Therefore, the mean of both of the year dummies is obviously 0,33.
N is 789, which equals the number of companies selected (263) times 3 years. The mean of variance in net income is 3,89. This seems to indicate a highly varying net income, but the mean is in fact strongly influenced by a relatively small amount of companies reporting strongly varying net income figures during 2009-2011. This is confirmed by the figures shown in the percentiles statistics (being 0,26, 0,67 and 1,6, respectively).
The dispersion of size indicates that there is a small group of companies that are extremely big compared to the rest of the companies in the sample. This is confirmed by e.g. a mean which is 18 times higher than the median, despite company size is scaled logassets. The size variable shows a minimum of 0. This is due to 1 company reporting no revenues in all of
23
the 3 years (Verona Pharma PLC). This has been checked with the annual reports of the company.
Mean as well as median for earnings per share excluding R&D expenses (EPSEXRD) are positive, and are still positive when R&D expenses are deducted. This means that most companies in the sample are profitable. Earnings per share (including R&D expenses) are negative for the lowest quartile, which means that at least 25% of the companies in the sample are loss-making.
In accordance with expectations, the mean and median for common equity per share are positive. The minimum for this item shows a slightly negative figure. This is correct, as there are 13 companies which report a negative common / ordinary equity in one or more years.
The mean of R&D expenses per share is 2,07, which means that the companies spent on average 2,07 EURO per share on R&D. This is a relatively high amount as compared to earnings per share excluding R&D expenses, which is 3,77 EURO on average. R&D expenses per share shows a minimum of 0. This minimum is in fact not 0, but very close to zero. There are some companies which have reported low amounts of R&D expenses against a high amount of shares outstanding.
Share price at year end (SPYE) shows a minimum of 0,002 (0,2 cent). British
companies often have high amounts of shares outstanding and therefore a low share price. As per 31/12/2011, the share price of Biome Technologies PLC was 0,17 cent. The highest share price in the sample is related to a French company. The share price of Dassault Aviation SA per 31/12/2010 was 601 EURO.
24
*LARGE = Dummy variable for large companies, CRDFR = Country dummy for France (Germany = benchmark), CRDUK = Country
dummy for the UK, VARNI = Variance in net income, SIZE = Net turnover in millions / logassets, CEQPS = Common equity per share, EPSEXRD = Earnings per share excluding R&D expenses, RDEPS = R&D expenses per share, SPYE = Share price at year end.
Table 2. Descriptive statistics R&D expenses versus share price sample
Descriptive statistics RDDI versus share price sample (for H3-H6)
Descriptive statistics were calculated for all variables in the model. See table 3 below for the results. The results for the country and year dummy variables have not been included in table 3, as the results for these items are self-evident. N = 270, which confirms the sample of 30 companies times 3 countries times 3 years. Naturally, the mean of both the year- and country dummies, is 0,33.
The pattern of the percentiles of the size variable (5, 34, 224) is roughly comparable to the pattern seen in the first sample (3, 19, 121; see table 2 above). As in the first sample, the average is much higher than the median, indicating that the mean is strongly influenced by a relatively small amount of companies that are extremely big. In this sample, the mean is 13 times higher than the median, compared to 18 times in the first sample.
As in the first sample, mean and median for earnings per share including R&D expenses are positive, which means that most companies in the sample are profitable. A difference with the first sample, is that earnings per share in the lowest quartile is also positive. This indicates that at most 25% of the companies in this sample is loss-making (rather than at least, as in the first sample).
As expected, common equity per share is positive for all quartiles. The minimum for common equity per share is -3,41, which is the same as the minimum of the first sample. The
25
explanation for this is that Skyepharma PLC was selected out of the first, larger sample. Skyepharma reported a negative common equity of 81,7 mln GBP in 2011.
The mean of the RDDI is 106, which indicates that companies have on average disclosed more information about R&D in their annual reports in 2010-2011, compared to 2009. This could be due to the increasing importance of technology and product innovation. The median of the RDDI is 73, which indicates that most companies disclose a less than average amount of information, whereas a small group of companies discloses an extremely high amount of information. This is confirmed by the relatively high RDDI score of 143 in the 75% quartile. This score indicates that 25% of the companies in the sample at least 43% more information on R&D, compared to the 2009 average.
LARGE = Dummy variable for large companies, VARNI = Variance in net income, SIZE = Net turnover in millions / logassets, CEQPS = Common equity per share, EPSEXRD = Earnings per share excluding R&D expenses, RDEPS = R&D expenses per share, RDDI = R&D
disclosure index, SP1204 = Share price at April 12th, following year-end.
Table 3. Descriptive statistics R& D Disclosure Index (RDDI) versus share price sample
Correlations R&D expenses versus share price sample (for H1 & H2)
Correlations were calculated for all variables in the model. See table 4 below. The basic variables (i.e. equity, net income) yield positive and significant correlations with share prices as expected.
The R&D expenses variable also shows a positive and significant correlation with share prices, which can be a sign that this relationship will be confirmed in the regression analysis as well.
The size control variable (SIZE) as well as the variable for splitting the sample into small and large companies (LARGE) are showing positive and significant correlations with all basic as well as the dependent variable, which confirms influences of company size. The
26
correlation coefficients also show that compared to Germany, the sample contains more large classed companies in France and less in the UK.
The variance in net income control (VARNI) variable does not show any significant correlations with any of the variables. However, to follow up the suggestions of Francis (2008), this variable will still be part of the analysis.
There are no significant correlations related to the year dummies, which indicates that there are no significant differences between years. There are significant correlations between the country variables and the most important other variables, i.e. common equity per share, earnings per share, R&D expenses per share and share price. Based on these correlations, differences between countries are expected.
27
LARGE = Dummy variable for large companies, YRD10 = year dummy for 2010 (2009 = benchmark), YRD11 = year dummy for 2011, CRDFR = Country dummy for France, CRDUK = Country dummy for the UK , VARNI = Variance in net income, SIZE = Net turnover in millions / logassets, CEQPS = Common equity per share, EPSEXRD = Earnings per share excluding R&D expenses, RDEPS = R&D expenses per share, SPYE = Share price at year end.
Table 4. Correlations R&D expenses versus share prices selection Correlations R&D expenses versus share price sample (for H3-H6)
Correlations were calculated for all variables in the model. See table 5 below. Similar to the results in the previous analysis, the basic variables, i.e. earnings per share and equity per share show a positive and significant correlation with share prices.
The control variable for variance in net income (VARNI) does not yield significant correlations with any of the other variables, except for the country dummy for France. As in the first part of this research, this control variable will only be left in the analysis to follow up the recommendations of Francis (2008). The control variable for company size (SIZE) also yields no significant correlations with other variables than the country dummies (and
28
naturally, LARGE). However, the variable to split the sample into small and large companies (LARGE) does yield positive and significant correlations with both the basic variables (equity, earnings per share) and the ‘variables of interest’ (R&D expenses, R&D disclosure and share price). This indicates that differences between large and small companies can be expected. Also, this confirms the finding of Bozzolan et al (2006), that the disclosure level is positively correlated with firm size. To follow-up on the recommendations of Wyatt (2008), the SIZE variable is also maintained (rather than removed) in the analysis.
Like in the first part of this research, the R&D expenses per share (RDEPS) variable shows a positive and significant correlation with share price. No direct correlation between R&D disclosure (RDDI) and share price is found. However, the correlations matrix does show there is a direct correlation between the interaction variable (which combines R&D expenses and R&D disclosures) and share prices. This could mean that R&D disclosure enhances the market-value of R&D expenses. However, this is to be confirmed by the regression analysis, in which all of the variables are included to determine the complete model.
29
LARGE = Dummy variable for large companies, YRD10 = year dummy for 2010 (2009 = benchmark), YRD11 = year dummy for 2011, CRDFR = Country dummy for France, CRDUK = Country dummy for the UK, VARNI = Variance in net income, SIZE = Net turnover in millions / logassets, CEQPS = Common equity per share, EPSEXRD = Earnings per share excluding R&D expenses, RDEPS = R&D
expenses per share, RDDI = R&D disclosure index, SP1204 = Share price at April 12th, following year-end.
Table 5. Correlations R&D disclosure index versus share prices selection Regression analysis R&D expenses versus share prices for H1
A linear regression analysis on the first model was run, using SPSS. The F-statistic of the model is highly significant (,000), which means that the data fits the model well. The adjusted R2 for the model is 0,720, which means that 72,0% of the variance in share price can be explained by the independent variables in the model. The coefficients of the model are shown in table 6 below. Based on a 95% confidence interval, the coefficients of the control variables VARNI and SIZE are not significant. However, to follow-up on suggestions made in previous research (Francis et al, 2008; Wyatt, 2008), these variables will not be removed from the model. The country dummy variable for the UK shows a negative and significant
relationship with share price. The unstandardized beta is -9,69, which means that the market pays almost 10 EURO’s less for shares of UK based companies. An underlying reason for this may be found in the results of the correlations analysis (see table 4 above). There are less
30
large classed companies based in the UK and large classed companies are positively correlated with both earnings per share and share price.
As expected, the basic variables, being earnings per share (excluding R&D expenses and equity per share, show positive and significant relationships with share prices. The R&D expenses per share variable shows a positive relationship with share prices, but it is not significant (,361). Unfortunately, this result means there is insufficient evidence that R&D expenses are positively associated with share prices. Based on this analysis, H1 is not adopted but not rejected either, as there is also no evidence that R&D expenses are not positively associated with share prices.
YRD10 = year dummy for 2010 (2009 = benchmark), YRD11 = year dummy for 2011, CRDFR = Country dummy for France, CRDUK = Country dummy for the UK, VARNI = Variance in net income, SIZE = Net turnover in millions / logassets, CEQPS = Common equity per share, EPSEXRD = Earnings per share excluding R&D expenses, RDEPS = R&D expenses per share.
Table 6. Regression analysis R&D expenses versus share prices
Regression analysis R&D expenses versus share prices for H2 (small versus large companies)
To test the hypothesis that R&D expenses are more positively associated with share prices of large firms compared to smaller firms, a distinction between small and large companies has been made. Companies (cases) with total revenue (REVT) above the median are classed as large. The main sample was divided into one sample containing all firms classed large and one sample containing all firms not classed large. Separate regressions were run on both of the samples. The results are combined in table 7 below. Regressions on both samples resulted in acceptable R-squared values (,715 and ,656) and highly significant F-statistics (both ,000). The coefficient for the R&D expenses variable is negative for large companies versus positive for small companies, whereas the coefficient is only significant for small companies. Based on the correlations analysis, differences between large and small
31
companies were expected. The correlations analysis yields significant correlations between the dummy variable for large companies (LARGE) and both the basic variables (earnings per share, common equity) and the ‘variables of interest’ (R&D expenses per share, share price). However, based on previous literature of Ciftci and Cready (2011), a (more) positive
association between R&D expenses and share prices was expected for large companies (H2), rather than small companies. Based on the below regression results, H2 is rejected.
YRD10 = year dummy for 2010 (2009 = benchmark), YRD11 = year dummy for 2011, CRDFR = Country dummy for France (Germany = benchmark), CRDUK = Country dummy for the UK, VARNI = Variance in net income, SIZE = Net turnover in millions / logassets, CEQPS = Common equity per share, EPSEXRD = Earnings per share excluding R&D expenses, RDEPS = R&D expenses per share.
Table 7. Regression analysis R&D expenses versus share prices; small vs. large companies
Regression analysis R&D expenses versus share prices, using the sample to test H3-H6 As announced under ‘methodology’, an additional regression analysis is run over the smaller sample meant to test the relationship between the RDDI and the share price as per April 12th. This means this smaller sample is now used to run the same regression model as in the above. This is to compare the smaller sample to the larger sample (meant to test the relationship between R&D expenses and share prices). The results of this additional regression also show a highly significant F-statistic (,000). Moreover, the adjusted R2 of 0,783 is higher than the R2 for the regression of the larger sample (being 0,720). As in the first analysis, the R&D
32
expenses variable is positive but not significant. However, against expectations, not all results of this regression (see table 8 below) are comparable to the results of the first regression (see table 6 above).
YRD10 = year dummy for 2010 (2009 = benchmark), YRD11 = year dummy for 2011, CRDFR = Country dummy for France (Germany = benchmark), CRDUK = Country dummy for the UK, VARNI = Variance in net income, SIZE = Net turnover in millions / logassets, CEQPS = Common equity per share, EPSEXRD = Earnings per share excluding R&D expenses, RDEPS = R&D expenses per share.
Table 8. Regression analysis R&D expenses versus share prices; sample for H3-H6
There are some minor issues with the results shown above, as there are differences between the significance and values of some of the control variables. E.g. the country dummy for France has gained significance, whereas the country dummy for the UK has lost some significance. However, there is one important issue that needs to be addressed. That is, the variable for earnings per share excluding R&D expenses (EPSEXRD) has a negative sign. This is an important issue, as it is violating the Ohlson (1995) model, which is the basis for this research. According to the Ohlson (1995) model, firm value is a (positive) linear function of both equity and net earnings. Some investigation has been carried out to find the cause of the unexpected sign. The scope of this research has been limited to the high-tech sector, to avoid incorporating companies in the samples, for which R&D is not relevant. It is however not impossible that there are companies included in the sample that do meet the definition of high-tech based on their NAICS code, but at the same time have low R&D expenses. E.g. companies may purchase technology externally. If there are too many companies for which R&D is relatively irrelevant, this could disturb the model. An additional regression was run, with the cases within the lowest quartile of companies R&D expenses as a percentage of turnover removed. See table 9 below.
33
YRD10 = year dummy for 2010 (2009 = benchmark), YRD11 = year dummy for 2011, CRDFR = Country dummy for France (Germany = benchmark), CRDUK = Country dummy for the UK, VARNI = Variance in net income, SIZE = Net turnover in millions / logassets, CEQPS = Common equity per share, EPSEXRD = Earnings per share excluding R&D expenses, RDEPS = R&D expenses per share.
Table 9. Regression analysis R&D expenses versus share prices; sample for H3-H6; cases lowest quartile R&D expenses removed
Table 9 above shows that removing the cases with the lowest relative R&D expenses does change the sign of earnings per share coefficient into a positive one. The researcher has looked into the cases that were removed, and found that in 34% of these cases a year-on-year decrease of net earnings coincides with a year-on-year increase of share price (or the other way around. The negative sign on earnings per share is likely to be caused by these cases, rather than by irrelevance of R&D. The data of these cases has been checked thoroughly and proved correct. Moreover, the sign of earnings per share in the regression on the larger sample was positive as expected. The researcher is therefore confident to proceed with the current data.
Regression analysis disclosure index versus share prices to test H3 and H4
As stated before, the model for the regression to test the influence of the disclosure index (RDDI) on share prices is the same as the model to test the influence of actual R&D expenses on share prices, except the RDDI is added as an additional independent variable and share price per April 12th in stead of December 31st is used as the dependent variable. The adjusted R2 of the model is 0,783, which is a bit higher than the R2 of the regression on the model to test the influence of R&D expenses (being 0,720) and exactly the same as the R2 of the ‘control’ regression above. The F-statistic is again highly significant (,000). These
statistics indicate that the data fits the model well and a high portion of the variance in share prices is explained by the model. See table 10 below for the coefficients of the regression. R&D expenses per share has a positive sign and is significant. The sign of R&D expenses was
