Empirical study of the RandD-Return relationship across Europe

UNIVERSITEIT VAN AMSTERDAM

Empirical study of the R&D-Return

relationship across Europe

Master’s Thesis

Bas Michaël van Stigt

Student ID: 10653368

Thesis Supervisor: Prof. Dr. Enrico Perotti

Msc Business Economics: Finance Track

9/30/2014

This thesis will analyze whether R&D intensive firms are correctly priced by investors across Europe. Several tests are performed whether R&D intensive firms earn excess returns compared to lower R&D intensive firms in the short-run, this is done by comparing firms by matching R&D intensity and book-to-market ratios. My results suggest that R&D intensive firms earn higher returns and are prone to underpricing. No evidence is found however that this mispricing is caused by systematic sources of risk. The sample analyzed prove that R&D intensive stocks are prone to underpricing under the Fama and French (1992) model after controlling for size and the book-to-market ratio. Overall results confirm that R&D intensive stocks earn excess returns and that investors should be more aware of firms disclosing R&D in order to help them value a certain firm in the future as my results suggest that the R&D-return relationship seems to be caused by a mispricing.

1 Statement of Originality

This document is written by Student Bas Michaël van Stigt who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

2

Table of Contents

5.1. Distinguishing stock returns using portfolio formation ... 19

5.2. Addition of an R&D variable to the Fama and French (1992) model. ... 20

5.3. Under or over-pricing estimation through the Fama and French (1993) three factor model and through the Carhart (1997) four factor model. ... 22

5.3.1. Fama French three-factor model ... 22

5.3.2. Carhart four factor model ... 23

5.4. R&D and stock return volatility ... 27

5.5. Additional Analysis ... 30

6. Conclusion ... 32

3

1. Introduction

The development of IT and technology has revolutionized the economy, this has also led to the growth of intellectual property and the subsequent decrease of capital-intensive work. Intangibles are of crucial importance in today’s economy and can be argued to be the driver of many firm’s future growth options. Research and Development (R&D) expenses are crucial in the creation of intangibles and have been a larger portion of a firm’s expenses over the past twenty years. My study will comprise whether R&D intensive firms (firms with a high R&D/Market Capitalization ratio) earn excess stock returns compared to firms with lower R&D intensity. Although R&D expenditures and intangibles have received greater attention among researchers as of late, there is still scarce research data on the European markets due to the vast differences among countries and the difficulties faced with the collection of data.

Prior research has documented a positive R&D-return relationship (Lev and Sougiannis 1996; Chambers, Jennings, and Thompson II 2002, Duqi et al 2013). The R&D-return relationship has been attributed to an underestimation of systematic risk when studying R&D intensive firms or simply due to a mispricing of firms with high R&D expenses, although this has only been documented on data in the US. Eberhart, Maxwell and Siddique (2004) find that an unexpected increase in R&D can lead to excess stock returns which can persist for up to five years, their results argue that this is due to investors systematically underreacting to the unexpected R&D increase.

Several complexities are faced when studying R&D expenditures, one is the accounting rules which apply to R&D expenditures in Europe. “Under IASB standards, capitalization is permitted for development expenditures only when a clear connection can be demonstrated between these costs and a hypothetical future product” Duqi et al (2013, page 2). This rule is difficult to apply, hence firms usually completely expense R&D in the year it is incurred, this is not an accurate view however, as it causes significant changes of R&D disclosures across years. Another issue surrounding the assessment of R&D expenditures is the fact that highly R&D intensive firms generally speaking have a low amount of tangible assets and are uncertain as to what the future might hold. This is due to the fact that future cash flows which could be generated through R&D expenditures are highly uncertain. What the data sample shows is that it is usually the large firms that disclose their R&D expenses.

4 Looking at European data as opposed to data from the United States might offer some new interesting insights. I am combining the six countries analyzed into one pool of data to see whether there is a common effect across Europe. This thesis will aim to test whether there is a link between a firm’s level of R&D expenditures and excess stock returns. The reason as to why R&D intensive firms may be prone to underpricing is due to the uncertainty of future R&D benefits. Studies have suggested that a significant increase in R&D causes a firm to earn excess stock returns, which suggests that an increase in R&D expenditures signals to the markets that this particular firm will carry more risk which might be beneficial in the long run, as suggested by Eberhart, Maxwell and Siddique (2004). Fama and French’s methodology will be used a vast amount in this thesis as it has proven to be an efficient methodology to use in the context of R&D and trying to explain excess returns after controlling for certain risk characteristics. Lev and Sougiannis (1996) show that the inclusion of an R&D variable increases the overall predictive power of the Fama and French (1992) Model. As their results suggest that the addition of an R&D variable to the Fama and French model is positive and statistically significant which would suggest R&D intensive firms generate excess returns. As opposed to current research on R&D intensity in Europe, Duqi et al (2013), my study will use the Carhart four-factor model to take into account the fourth factor, momentum, as previous studies have shown that this increases the predictive power of the Fama and French (1993) regression. Momentum takes account of the fact that previous winning stocks have been argued to keep winning in the short-term. Carhart (1997)introduced this fourth factor by analyzing mutual fund performance and it has been a crucial factor ever since, which Fama and French have acknowledged themselves and have created a similar variable. This regression will allow me to determine whether a mispricing is present and whether this is caused by systematic sources of risk not captured by the Fama and French model. Fama and French (1992, 1993) suggest that the capital asset pricing model (hereafter CAPM) fails to capture the risk of a firm and that the inclusion of the book-to-market ratio and a firm’s size is able to explain risk that the CAPM fails to do. Portfolio formation will be used for the Fama and French (1993) three factor model, this is essential as it will allow me to compare results with previous analysis and will allow me to suggest whether these portfolios have different systematic sources of risk or not.

What this thesis aims to find is whether R&D is prone to mispricing across Europe. Do firms, with high R&D expenditures, earn excess stock returns? Does having a higher R&D intensity signal a higher risk to investors? And can this be explained solely by the fact that they have higher R&D

5 expenditures or are there other reasons. If R&D intensive firms do earn excess stock returns, how can this be explained? Perhaps R&D intensive firms carry higher systematic risk which is not captured by investors? Another explanation is that these firms are underpriced, and that their excess returns cannot be explained by its risk characteristics. I also examine whether firms that disclose R&D signal higher business risk to the market.

The main issue faced with studying R&D, especially with European data, is the challenge faced with the collection of data. In the US for example “R&D Costs are generally expensed as they are incurred” (PWC , March 2010)1, this makes the collection of data easier for US firms as all firms provide an accurate view of their actual R&D expenses on a yearly basis. Several databases have to be used to collect the necessary data for R&D expenditures in the EU, lots of firms however did not expense any of their R&D as they are not obliged to, this might make some of the results unreliable due to the large number of firms not expensing their R&D expenses when in reality they may have been spending money on R&D.

Previous research has tried to document the effect of R&D on a firm’s stock price, the main focus has been on data in the USA. This thesis will differ in certain aspects due to the inclusion of new countries that have not been researched before and the inclusion of the Carhart fourth factor in the Fama and French three factor model. The sample period used is 2000-2013, where various European countries are analyzed which are pooled together, the countries analyzed include; Belgium, Germany, Italy, the Netherlands, Spain and the UK.

The hypotheses tested are as follows:

Hypothesis 1: R&D intensive firms earn excess stock returns relative to firms with lower R&D intensity.

Hypothesis 2: The inclusion of an R&D variable in the Fama and French three factor model proves that R&D is significant at explaining future stock returns.

Hypothesis 3: Portfolios consisting of R&D intensive firms are underpriced when estimated through the Carhart Four factor model and through the Fama French three factor model and these portfolios differ in systematic sources of risk.

1

6 Hypothesis 4: R&D increases stock return variability.

Besides covering the hypotheses mentioned above, the remainder of this thesis will also explain descriptive statistics that will analyze the differences between R&D firms and non-R&D firms across the dataset and assimilate results with comparable research papers. Section 2 will mention previous research papers that focus on R&D, how they assimilate with my thesis and what is done differently. Section 3 will explain the research methodology used to test the hypotheses mentioned previously. Section 4 will briefly explain the data collection process and report the descriptive statistics. Section 5 will exhibit the empirical results for the four hypotheses and also includes some additional analysis. Section 6 contains a summary of results and will provide a conclusion to this thesis.

2. Literature Review

Prior research has tried to document the link between R&D and a firm’s market value, whether by looking at a firm’s stock returns or through earnings. Lev and Sougiannis (1996) is one of the first research papers that addresses R&D’s effect on stock returns. They use the Fama and French (1993) three factor model in their regressions. Lev and Sougiannis “document a significant inter temporal association between firms R&D capital and subsequent stock returns” (Lev and Sougiannis 1996, page 107)2. This paper was a catalyst to future research on R&D, considering lots of future research used the same or similar methodologies as Lev and Sougiannis did. What caused Lev and Sougiannis (1996) to turn to the Fama and French three factor model was the fact that Lev and Sougiannis found empirical proof which suggested that investors might not fully recognize the importance of R&D whenever expensed by a firm under GAAP. They compared this observation with the ‘earnings announcement drift’, which was first documented by Ball and Brown (1968). The post-earnings announcement drift is an anomaly whereby a positive post-earnings announcement seems to not have a direct impact on a firm’s stock price but that it takes several months before the announcement is really captured in the stock price itself. It is argued that investors initially underreact to this earnings announcement, whereby the actual reaction depends on the efficiency of the market and the information available to investors. Similarly, investors might not know how to

2

7 react to certain R&D disclosures and what effect it might have on a firms stock price in the long run. Using the Fama and French (1992) model, which uses lagged fundamentals, adjusting their model by adding R&D as an extra independent variable in their model will make it possible to analyze the effect R&D has on a firms stock price. Lev and Sougiannis (1996) found that the inclusion of the R&D variable into the Fama and French (1992) model was statistically significant and had a positive coefficient, which would suggest an excess return. Lev and Sougiannis (1999) found that by using the Fama and French (1993) three factor model, the positive R&D-return relationship can be explained by measures of systematic risk.

Earlier research focused on the accounting aspect of R&D and whether R&D had a positive relationship with any future benefits. Earlier research such as Milburn (1971) tried to estimate this through cross-sectional regressions to see whether there was any correlation between R&D and future benefits but failed to find any. The fact that firms whom expense R&D benefit from tax-shields has also been analyzed, as theory would suggest that this should incentivize firms to spend more money on R&D. This matter was proven by Scholes and Wolfson (1992) who found that R&D tax shields must be value-relevant.

The Fama and French (1992) model is extensively used in most papers mentioned in this literature review and will also be used in my thesis. Fama and French initially identified the fact that the capital asset pricing model was not very significant in identifying risk factors, as they showed empirical evidence that during the period 1969-1990, a firm’s specific beta explained very little concerning future stock returns. Fama and French identified that lagged fundamentals such as a firm’s size and it’s book-to-market ratio are better at capturing risk patterns of a specific firm than the CAPM. Fama and French also identified that the earnings-to-price ratio and a firm’s leverage were efficient proxies at explaining stock returns. All papers that use this specific model argue that by including R&D as an independent variable will improve the predictive power of this regression, hence, it helps explain future stock returns.

Chan, Lakonishok and Sougiannis (2001) examine R&D expenses to see whether markets value intangibles properly by looking at R&D expenses from a US accounting standards perspective. They find that the average historical stocks returns of firms doing R&D have a similar average stock returns to firms who do not have/report any R&D expenses. They however do find that firms with high R&D intensity earn large excess returns, this may be due to the fact that investors underestimate the long run effects of R&D expenditures for firms which invest a lot in R&D as

8 compared to firms who invest very little. They were able to estimate this effect through the use of portfolio formation.

Lakonishok, Shleifer, and Vishny (1994) found that glamour stocks experience positive long term returns following portfolio formations. This paper is based on the argument that value stocks tend to outperform the market, where value stocks are stocks that have relatively low prices compared to any proxy used to measure a firm’s fundamental value. They compared the relative riskiness between investing in a value or growth portfolio of stocks. The fact that one portfolio outperforms the other is not explained by bearing extra fundamental risk, but simply due to the fact that value stocks are undervalued compared to its fundamental value, so they earn abnormal returns. This paper correlates with findings in Fama and French (1992,1993) whom also use portfolio formation in order to calculate their factors which is also done in this thesis by replacing their size proxy with an R&D variable.

Chambers, Jennings, and Thompson II (2002) show that more innovative firms tend to gain abnormal returns that may persist for up to 10 years. They were one of the first researchers who argued that R&D intensive firms can be mispriced for other reasons besides accounting standards, but for instance simply due to the fact that investors underestimate the risk posed by R&D, as theory states, the higher a firm’s risk the more investors demands in return (i.e. risk-return trade off). Mispricing a stock is a whole different concept than under-estimating a firm’s risk which in turn leads to a wrong price which therefore results in these specific firms earning large excess returns due to investors underestimating the risks posed by R&D.

An essential issue is to differentiate R&D from other expenses, this is done by differentiating between tangible and intangible information, where R&D consists of intangible information. Daniel and Titman (2006) examined this difference and aimed at decomposing the information that moves a stock price and defined intangible information as “information that is orthogonal to the tangible information that appears on a firm’s accounting statements“ (Daniel and Titman 2006, page 1640). Contrary to Fama and French (1992) they do not find any relationship between past performance and future stock returns, whereby it was argued that poor past performance would usually be followed by higher future stock returns. They argue that “the book-to-market and reversal effects arise because future returns are cross-sectionally related to past realizations of intangible information” (Daniel and Titman 2006, page 1638). This paper provides evidence as to why this particular intangible information might not follow the efficient market hypothesis and in turn causes

9 R&D intensive firm’s to earn stock returns which are not coherent with lesser R&D intensive firms. This paper suggests that momentum is an abundant variable.

Dedman, Mouselli, Shen and Stark (2009) consider the UK treatment of intangible assets, from a disclosure point of view. They also find that the ratio of R&D to market value is positively associated with positive average returns. Ciftci and Cready (2011) explore how R&D related earnings performance and earnings variability depend upon firm size and find that the association between the volatility of future earnings and R&D intensity decreases with firm size. This can be associated with the regression I will use to measure R&D’s impact on stock return variability. Eberhart, Maxwell and Siddique’s (2004) find consistent evidence that whenever a firm announces a significant increase in R&D expenditures, investors underreact, and these firms eventually earn abnormal stock returns due to the fact that this intangible information is undervalued. Their results suggest that R&D information is affected by limited attention, that investors do not

Duqi, Jaafar and Torluccio (2013) use a European dataset to see whether R&D intensive firms earn superior stock returns and are one of the first to do so. They do find that innovative firm’s stock prices are undervalued by investors. As they find that firms with high R&D expenses can earn substantial extra returns in the short run but find that this is not caused by systematic sources of risk. This paper is especially useful as it mentions the relevant issues when using European data and how to correct for these issues.

Hou and Robinson (2004) have found a link by using industry concentration, and found that highly concentrated industries earn positive average stock returns. Gu (2012) used this research and found a link between competitive industries and the positive R&D-return relation. Gu (2012) found that R&D firms operating in a competitive industry earn higher stock returns than firms in concentrated industry. An interesting paper by Jiang, Qian and Yao (2013) examines the R&D-spillover effect, which argues that if an R&D leader firm within a specific industry significantly increases its R&D expenditures, it’s peer firms whom do not adjust their R&D expenditures will also experience abnormal returns. This suggests that firms within an industry are interconnected and are affected by each other’s financial decisions. They suggest that this is due to limited investor attention. This paper follows from Griliches (1998) who was one of the first researches to identify the R&D-spillover effect. Although this research paper will not be used, it does offer potential insights as to why

10 current research postulates that R&D intensive portfolios earn higher stock returns when compared to lower R&D intensive portfolios and why this occurrence is difficult to explain by asset pricing model theories.

Although none of these papers have coherent findings, most of these papers do come to the conclusion that highly R&D intensive firms seem to earn excess returns. The explanation as to why this seems to be true is not coherent among research papers. Arguments as to why this ‘anomaly’ occurs is that R&D might carry systematic risk which is not captured fully by investors which in turn leads to undervaluation of these R&D intensive stocks or that these stocks are simply mispriced by investors. What is captured in almost all of these research papers is that by paying more attention to R&D as an investors will decrease his or hers forecast error in the potential future price or future stock price movement. This anomaly as to why R&D intensive firms seem to earn higher returns can be associated with limited investor attention. My research relates to several of these papers as regressions are used from Lev and Sougiannis (1996,1999) and from Chan, Lakonishok and Sougiannis (2001). All my proposed hypotheses can be related to research mentioned in this literature review.

3. Research Methodology

One basic method to distinguish the stock returns of firms with different R&D intensity is to use portfolio formation. I use two different methods of portfolio formation to find a link between R&D intensity and stock returns. I initially use portfolio formation to create six portfolios, where one portfolio consists of firms with no R&D expenses and the other five do, whereby the higher portfolio represents firms with a higher proportion of R&D/ME. This allows me to distinguish stock returns between firms with higher and lower R&D intensity. Initially the firms are elected to one of the portfolios by its respective R&D/ME ratio, which creates six portfolios. Following this, the monthly returns are cumulated on a firm basis to get the cumulative annual return for a specific year, whereby the year starts from July till the end of June the following year. Finally, the mean one-year ahead return is calculated per portfolio during the years 2000-2013, the sample of firms is pooled over the years. These results will allow me to distinguish the portfolio returns of portfolios consisting of firms with high R&D intensity as compared to portfolios consisting of lower R&D intensity.

11 Another method to examine whether R&D intensive firms earn excess returns over less R&D intensive firms is to follow a method introduced by Lev and Sougiannis (1999) which follows from Fama and French (1992). Lev and Sougiannis examine the effect between the book-to-market ratio and R&D as a proportion of its market value of equity. The sample of firms is initially formed into five book-to-market portfolios each year, whereby the five portfolios represent different quintiles, hence portfolio 5 includes firms with the highest book-to-market ratio in that specific year . From these five book-to-market portfolios another five portfolios are formed based on the R&D/ME ratio, which are also re-calculated yearly. Of these 25 cross-sectional portfolios, the mean one-year ahead returns are calculated. The full sample period 2000-2013 is used whereby the data is pooled over the years. This analysis will show whether firms with high R&D intensity and high book-to-market ratios earn higher stock returns relative to portfolios with lower R&D intensity and lower book-to-market ratios. The two methods explained above will allow me analyze hypothesis 1 as both methods test whether investors correctly price firm’s shares, and in particular focus on the effect that R&D might have. Results could potentially suggest that R&D stocks are simply value stocks due to their low book-to-market ratios. For hypothesis 1 to be proven, my results should suggest that higher R&D intensive portfolio earn higher portfolio returns than lower R&D intensive portfolios.

Following Lev and Sougiannis (1999), who used the Fama and French (1992) model but amended it by adding an R&D variable to examine whether R&D could help explain future stock returns. Fama and French (1992) introduced this time-series regression, having proven that using lagged values of certain firm fundamentals are significant at explaining future stock returns. This regression came to existence as Fama and French found beta to be insignificant at explaining stock returns and used other proxies that did explain future stock returns. Lev and Sougiannis(1996,1999) regressed monthly returns on beta, Ln(ME) and Ln(BM) and then ran another regression, which added Ln(RD/ME) as an additional independent variable. As a robustness check, I will also run the regression using a different denominator to identify the R&D variable. The regression is also run using total assets and operating profit as a denominator for the fourth independent variable. Each country is cross-sectionally regressed on future stock returns after controlling for the previously mentioned variables using the Fama and Macbeth (1973) methodology. The regression is as follows:

𝑅_𝑖,𝑡+1= 𝑎₀+ 𝑎₁𝛽_𝑖,𝑡+ 𝑎₂𝐿𝑛(𝑀𝐸)_𝑖,𝑡+ 𝑎₃𝐿𝑛(𝐵𝑀)_𝑖,𝑡+ 𝑎₄𝐿𝑛(𝑅𝐷

𝑀𝐸)𝑖,𝑡+ 𝜀𝑖,𝑡 (1)

12 𝑅_𝑖.𝑡+1 = returns: monthly stock returns, starting in July.

𝛽_𝑖,𝑡 = beta: CAPM based beta, based on 𝐶𝑂𝑉(𝑅_𝜎2 𝑖,𝑅𝑚)

(𝑅𝑖)

.

𝐿𝑛(𝑀𝐸)_𝑖,𝑡 = size: Market value of firm (i) in month (t). calculated as price times number of shares outstanding.

𝐿𝑛(𝐵𝑀)_𝑖,𝑡 = book-to-market ratio: calculated as deferred taxes plus common equity divided by the number of shares outstanding multiplied by the share price at the end of the fiscal year.

𝐿𝑛(𝑅𝐷

𝑀𝐸)𝑖,𝑡 = R&D: R&D as a proportion of the market value of equity.

All independent variables besides beta, are calculated as a natural logarithm. Monthly regression are run from 2005 to 2013, 2005 is the starting date as opposed to 2000 due to the calculation of beta. What these two different regressions will tell me is the impact of adding the R&D variable will do to the model. What can be analyzed is the coefficient it will have, whether it is negative (positive), which would induce lower (higher) stock returns depending on whether the variable is statistically significant or not. Analyzing the difference between the four regressions will allow me to determine the validity of hypothesis 2.

To measure whether R&D induces some sort of stock mispricing the Fama and French (1993) three factor model is used. Fama and French (1993) have captured the fact that beta is not statistically significant in capturing systematic sources of risk and concluded that the book-to-market ratio and size are better proxies, they thus proposed a three factor methodology that does capture this. Fama and French (1993) included three factors to capture systematic sources of risk, which includes a Small Minus Big (SMB) portfolio which captures the book-to-market ratio effect, a High Minus Low portfolio (HML) which captures the size effect and beta, which takes account of the return of the market minus the risk free rate.

In order to calculate the SMB variable and the HML variable, for year (t), book value of year (t-1) June and market value of year (t-1) December is used for the book-to-market ratio. Where the book-to-market ratio is defined as common equity plus deferred taxes divided by the number of

13 shares outstanding multiplied by the current market price. For the size proxy, for year (t), the market value of equity of year (t-1) December is used. Both these proxies are then used to calculate SMB and HML for year (t) to year (t+1), which is done from July year (t) till June year (t+1), where-after the portfolios are formed again to recalculate the variables for the following year.

Six portfolios are formed using the previously mentioned proxies. Two groups are formed based on size, Big (B) and Small (S), where the median is the cutoff point between the two. For the Book-to-Market ratio, three groups are formed, where the top and bottom 30% quintiles represent two groups and the ‘middle’40% represents the third group, called High (H), Neutral (N) and Low (L). The cross-section of these groups create six portfolios, where the monthly value-weighted return of these six portfolios for 12 months is calculated, after which the portfolios are formed again. Negative Book-to-Market ratios and negative Book Value of equity are rejected in the creation of these six portfolios. All independent variables included in the regressions performed in this paper are winsorized by the top and bottom 1%. Robustness and clustering checks are also performed in the regressions to ensure all coefficient are robust to cross-sectional correlation and heteroscedasticity. These six portfolios create the HML and SMB variables which are used in the regressions. They are formed as follows: 𝐻𝑀𝐿 =(𝐵𝐻 + 𝑆𝐻) 2 − (𝐵𝐿 − 𝑆𝐿) 2 𝑆𝑀𝐵 =𝑆𝐿 + 𝑆𝑁 + 𝑆𝐻 3 − 𝐵𝐻 + 𝐵𝑁 + 𝐵𝐿 3

All the data on HML and SMB variables was gathered from the Fama and French website, whom calculate these variables using a large number of countries from the EU. These variables were used as creating these variables on a country by country basis would most likely be unreliable due to the fact that investors have a large geographical scope when investing and are unlikely to invest in one specific country.

The dataset will be divided into several quintiles, which are formed based on the respective R&D/ME ratio. This ratio divides the dataset into six groups, where one group consists of firms with no R&D expenditures and the other five do. These portfolios are created in June every year, and are recalculated every year. The following time series regression is then run for each portfolio:

14 Where 𝑟𝑖,𝑡 is the monthly average return for portfolio i, where portfolio (i) represents a portfolio

based on the R&D/ME quintile ratio, 𝑟𝑓,𝑡 is the risk-free rate of return, and 𝑅𝑚,𝑡 is the

value-weighted market return from the full sample. The risk-free rate is used from the Fama and French website3. What this regression will explain is that whenever an intercept is significantly different from zero, it might induce that R&D provides an incorrect valuation of a firm’s stock return due to a potential risk which might not have been captured by the regression above. This could provide evidence that R&D intensive portfolios carry a different amount of systematic risk compared to lower R&D intensive portfolios and that R&D expenditures could be mispriced by investors. In total there are a 168 monthly observations for each firm.

The addition of my thesis to current research is the fact that I will be using a momentum variable in my regression. Carhart (1997) showed the importance of momentum in expected return measures. Momentum accounts for the fact that previous winning stocks are likely to remain winning, this issue was also accounted for by Jegadeesh and Titman(2001). Momentum is an anomaly which can be associated with the limited attention of investors. This variable can be associated with R&D in the aspect that not enough attention is given to R&D when analyzing a firm’s stock price. Limited attention has been argued to be a cause of why investors underreact to value relevant information such as R&D.

By constructing five groups, two based on size, with a median breakoff point and three on prior returns, where the breakpoints are the 30th and 70th percentiles of a stock’s cumulative returns from t-1 to t-11, allows me to create six cross-sectional portfolios. The previous eleven cumulative months are taken as to avoid using the sorting month. The momentum variable accounts for the difference between the two high past return portfolios and the two low past return portfolios, where every year the portfolios are recalculated and the respective monthly average stock returns are calculated again for the respective year.

𝑀𝑂𝑀 = (𝑆𝐻 + 𝐵𝐻)

2 −

(𝑆𝐿 + 𝐵𝐿) 2

This variable will then be included as an extra independent variable in the previously mentioned regression as follows:

3

Monthly factors on SMB, HML and MOM are available and Kenneth French’s website: “http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html”

15 𝑟𝑖,𝑡− 𝑟𝑓,𝑡 = 𝑎𝑖+ 𝛽𝑖(𝑅𝑚,𝑡− 𝑅𝑓,𝑡) + 𝛾𝑖𝑆𝑀𝐵𝑡+ 𝛿𝑖𝐻𝑀𝐿𝑡+ 𝜃𝑖𝑀𝑂𝑀𝑡+ 𝜀𝑖,𝑡 (3)

The addition of this variable should improve the regression, as the adjusted R-squared among the two different regressions should show. The addition of this variable can also have an impact on the other remaining independent variables.

The HML factor calculates whether the portfolio in the dataset of the respective country relates to growth or value stocks. A positive coefficient means the returns can be attributed to a value stock premium, due to the high book-to-market ratio. Whereas a negative coefficient means the returns can be attributed to a growth-stock premium, where the low book-to-market insinuates that the respective stocks are priced lower on the stock exchange compared to the value shown on the balance sheet. The SMB factor calculates the size of the firm in the dataset, and whether this size attribute shows any signs of a small-stock premium or whether the dataset mainly consists of large stocks. Empirical research suggests a small-stock (size) premium, where small stocks seem to earn a significant higher stock return compared to larger stocks. The MOM factor calculates whether a portfolio of stocks which in the previous year experienced a higher stock return compared to a portfolio which experienced a lower stock return in the previous year, earn higher stock returns in the following year. This attribute has also been called “Hot Hands”, and is well documented in mutual funds. Regression (2) and (3) will allow me to analyze hypothesis three and will tell me whether portfolios consistent of R&D stocks have a potential mispricing from a systematic source of risk that is not captured by the Carhart four factor model.

All these methodologies so far are assuming and proposing that R&D carries risk, whereby the more a firm invests in R&D proportional to its market value of equity, the more riskier the firm should be. Arguments as to why a R&D intensive firm should be riskier revolve around the fact that higher R&D does not guarantee that a firm should have higher future returns in the future. R&D also carries a significant amount of information asymmetry risk as shareholders might not be aware of how the R&D expenses of a firm are carried out and whether they are going to be successful or not, successful R&D will only be announced once this information is shared by the firm itself.

Following Chan, Lakonishok, and Sougiannis(2001), estimating the effect of R&D on stock return variability should show that R&D intensive firms carry more business risk. Size and the BM ratio are included as independent variables in the following regression as they are good proxies to use as proven by Fama and French (1993), both variables are expected to have a negative coefficient. This

16 regression will be run three times as a robustness check, the R&D variable will be divided by market capitalization, total assets and revenue. The Fama and Macbeth (1973) methodology is used for running this regression:

𝜎_𝑖,𝑡+6 = 𝑎₀+ 𝑎₁𝛽₁+ 𝑎₂𝐿𝑁(𝑀𝐸)_𝑖,𝑡+ 𝑎₃𝐿𝑁(𝐵𝑀)_𝑖,𝑡+ 𝑎₄𝐿𝑁(𝑅&𝐷

𝑀𝐸)𝑖,𝑡+ 𝜀𝑖,𝑡+6 (4)

At the end of June each year, each firm’s stock return standard deviation is calculated over the next 12 months. The dependent variable is the stocks standard deviation, which is regressed on the firm’s respective market equity, book-to-market ratio and R&D’s ratio as of market equity. It has to be noted that natural logarithms are taken for all variables, due to the large differences in size among firms in the dataset. Chan, Lakonishok, and Sougiannis (2001) however include a dummy variable to account for the age of firm, I have not done this, Datastream does have this variable but lacks sufficient data. For this regression to prove the methodology it is expected that the R&D independent variable will have a positive statistically significant coefficient whereby we can conclude that R&D does impose risk on a firm’s stock return.

4. Sample Description

The sample includes all listed firms in the following EU countries; Belgium, Germany, Italy, the Netherlands, Spain and the UK for the years 2000-2013. These EU countries represent a high percentage of the R&D-intensive firms in Europe and includes countries which have not been analyzed before in this setting. The data was tackled country by country using Datastream where after the data was merged. I deleted any ‘dead’ firms or firms that have been taken over or merged during the sample period. In order to get a reliable estimate, I deleted any firms whom operate in the financial industry (insurance, banks, mutual funds etc.), due to their differences in R&D expenditures and R&D disclosures among my dataset. I am using data from 2000-2013, this period is used, as before 2001, no or close to no R&D data is available from databases besides Thomson Datastream.

The tricky part of this data collection has to do with the R&D data. I initially use Thomson Datastream to gather data, which however lacks data, this made me use Bureau van Dijk’s Amadeus

17 Database to complement my dataset. I also used the EU Industrial R&D Investment Scoreboard4 by the European Commission to complement my dataset which is a scoreboard of the companies with the highest R&D expenditures across Europe. However, due to accounting rules under the IASB, firms are not obliged to expense their R&D every year on the accounts, they can choose to expense their R&D expenses in the year where a product is actually ‘created’ or put on the market. This causes my dataset to have a lot of firms with ‘Not Available’ (NA) result for R&D expenses, the closer to the end of the dataset, the more information is available on R&D. Although firms have the option to disclose R&D whenever it ‘suits them best’, this is not apparent across my dataset, as firms usually always disclose their R&D expenses or don’t, which would suggest these firms simply do not invest any money in R&D.

In order to calculate the Book-to-Market ratio, I used common equity plus deferred taxes as book equity and divided it by share price multiplied by the amount of common shares outstanding. When this is not available, Datastream offers a market-to-book ratio, so I replace this (the inverse) if the initial calculation is not available. What the descriptive statistics suggests in tables 1 and 2 below is the fact that non-R&D firms have a larger book-to-market ratio than firms who do have R&D expenditures, this already indicates that R&D firms are ‘underpriced’ relative to its market value or it can be argued that they are value stocks. Non-R&D firms in this context are firms that do not report any R&D or when no information is available about a firms R&D disclosures. What is also observed is the fact that R&D firms are on average larger than non-R&D firms across all countries. Duqi et al (2013) also found this across Europe and it has been documented by Hall and Oriani (2006) whom found that R&D disclosures is related to size.

Across all countries, R&D disclosures is below 50%, which is low and therefore does not provide an accurate view of firms. As explained previously, firms are not obliged to document their R&D expenses yearly. What has to be noted however is the fact that R&D disclosures are increasing significantly.

4

Data from the EU Industrial R&D Investment Scoreboard was gathered from the following website: “http://iri.jrc.ec.europa.eu/scoreboard.html”

18 Country

R&D Non R&D Non R&D R&D Non R&D R&D Non R&D R&D

BEL 1279 4080 11.93 11.95 0.84 0.69 12.53 12.32 GER 7392 30882 10.88 11.27 0.73 0.62 11.12 11.57 ITA 3720 6370 12.17 12.25 0.82 0.83 13.04 12.87 NED 1836 5535 12.04 12.78 0.64 0.6 12.56 13.01 SPA 3621 1901 13.26 12.72 0.9 1.13 13.85 13.06 UK 19192 47211 10.87 11.02 0.67 0.76 10.72 10.87

Table reports mean values. N is the total amount of monthly observations per country. Non-R&D observations refer to firms who did not have any Non-R&D in that specific month or did not report any R&D expenses.BM is defined as deferred taxes plus common equity divided by the number of shares outstanding multiplied by its market price at the end of the fiscal year. Natural logarithms are used for market capitalization and total assets. The sample period is 2000-2013.

N LN(ME) BM LN(Total Assets)

Table 1: Descriptive Statistics

Country mean median Std.Dev. Min Max

BEL 0.061 0.025 0.105 0.000 0.730 GER 0.061 0.026 0.095 0.000 0.572 ITA 0.041 0.005 0.073 0.000 0.378 NED 0.043 0.017 0.068 0.000 0.405 SPA 0.006 0.000 0.019 0.000 0.128 UK 0.046 0.012 0.091 0.000 0.574

Table reports R&D/ME per country. Non-R&D observations are not included. The sample period is 2000-2013

19

5. Empirical Analysis

5.1. Distinguishing stock returns using portfolio formation

One method to distinguish the difference of stock returns between firms with higher and lower R&D intensity is to sort the dataset into portfolios as mentioned in the research methodology. Distinguishing the mean one-year ahead returns of the six portfolios will allow me to analyze whether portfolios consisting of firms with high R&D/ME ratios earn higher returns compared to firms with lower R&D intensity. The mean-one year ahead returns for the portfolios is calculated by initially calculating the cumulative return per firm per year, where after they are sorted into portfolios in order to calculate the mean annual-return per portfolio.

Table 3 below shows that portfolios containing R&D experience a monotonous increase in portfolio returns as the R&D portfolio becomes higher. The portfolio containing non-R&D firms, which in this case are firms that have zero R&D expenditures experience higher return than the bottom two R&D intensive portfolios. Portfolio five experiences a significantly higher return than the other portfolios, on average they experience a stock return of 19.76% per year whereas the lowest R&D intensive portfolio on average experiences a stock return of 3.72%.

Following the method proposed by Lev and Sougiannis (1999), forming the dataset into 25 portfolios, 5 based on the book-to-market ratio and then within these 5 book-to-market portfolios, another 5 portfolios are formed based on the R&D/ME ratio, this allows me to examine whether R&D intensive firms earn excess stock returns after controlling for the book-to-market ratio. This method neglects firms which do not disclose any R&D. This method also allows me to examine whether the book-to-market ratio is an efficient proxy to estimate the R&D effect. What has to be noted is that due to the presence of a lot of firms with low R&D intensity, as was noted with the high standard deviation in Table 2, the book-to-market effect analyzed in this table can be driven by other reasons which are unrelated to the R&D activities of this particular firm.

R&D/ME Low High 1 2 3 4 5 Full Sample 4.99% 3.72% 4.81% 6.12% 9.44% 19.76% Portfolio non-R&D

Table 3: One-year mean returns following portfolio formation

Portfolios are formed based on firm's proportion of R&D/ME. For these firms, the one-year mean returns are calculated, which are shown above in percentage. Sample period is 2000-2013.

20 What can be seen from Table 4 is that in most cases an increase of the book-to-market portfolio is followed by a higher portfolio return. Within each R&D portfolio, a monotonous increase in returns can be seen as the book-to-market ratio portfolio becomes higher, although there are a few exceptions. What is odd to see for example is that within the highest R&D intensive portfolio, the highest book-to-market portfolio does not earn the highest average return as portfolio Q2 does. The average return of all portfolios that fall within a particular R&D/ME portfolio indicates that the highest R&D intensive portfolios earn the largest portfolio returns. Results somewhat coincide with Lev and Sougiannis (1999) although they have a perfect monotonous increase within each book-to-market ratio. These two methods of portfolio formation allows me to examine whether portfolios consistent of R&D intensive stocks earn larger portfolio returns compared to portfolios with lower R&D intensive firms. My results suggest that this is true, as both methods suggest that the higher a portfolio’s R&D intensity, the higher its return.

5.2.

Addition of an R&D variable to the Fama and French (1992) model.

R&D has never been observed as a variable that could help explain a firms fundamentals value or that it could help explain a firm’s performance or a firm’s stock returns. Lev and Sougiannis (1999) have shown that if R&D is in fact a statistical significant variable when included in the Fama and French (1992) model, it would indicate that it does help explain a firm’s future stock return. Running this regression using lagged fundamentals will show whether these independent variables help

low R&D/ME high

Book-to-Market 1 2 3 4 5 low 1 0.51% 1.41% 3.66% 1.24% 5.70% 2 6.02% 8.02% 9.25% 11.97% 23.32% 3 8.11% -0.42% 0.17% 3.81% 15.60% 4 5.11% 0.28% 5.65% 7.31% 17.06% high 5 13.54% 12.54% 8.31% 15.88% 23.24% 6.66% 4.37% 5.41% 8.04% 16.98%

Avg. Return per R&D Portfolio

Table 4: Mean one-year ahead returns following portfolio formation based on the book-to-market ratio and R&D/ME ratio.

Twenty-five portfolios are formed,five based on the BM ratio, then another five on the R&D/ME ratio. Average one-year returns are then calculated per portfolio whereby the years are pooled together over time. The average return per R&D portfolio calculates the average return of the five portfolios within one specific R&D portfolio.The total sample period is 2000-2013.

21 explain future stock returns. Four regressions are run where three regressions include an R&D variable, whereby the three regressions differ in the aspect of what R&D is divided by. The remaining regression omits the R&D variable. The impact of this R&D variable among the three regressions will allow me to analyze hypothesis two.

Table 5 below shows the results of these four regressions, what can be seen right away is that beta is statistically insignificant across all the regressions, this proves Fama and French’s work who identified beta to be statistically insignificant. The book-to-market ratio is positive and statistically significant at the 5% level when run for the full sample, its statistical significance disappears however whenever an R&D variable is included. Size carries a negative coefficient for the full sample regression, as expected, it however does not carry any statistical significance however. The same can be accounted for the three remaining regressions, the size coefficient becomes positive but remains statistically insignificant. The inclusion of an R&D variable, which was included in three different regressions improves the predictive power of the Fama and French (1992) model, due to it being statistically significant for all three regressions at the 1% level and the increase of the adjusted r-squared. Considering the R&D variable carries a positive coefficient for all three regressions indicates a higher annual return for firm’s carrying out R&D. Considering the fact that the statistical significance disappears for the book-to-market ratio once the R&D variable is included would indicate that the R&D variable subsumes the book-to-market effect (Lev and Sougiannis 1999). My results coincide with what Lev and Sougiannis (1999) found and to a certain extent with what Duqi et al (2013) found, considering they also found the R&D variable to be positive and statistically significant, although Duqi et al did not find it for all of their countries analyzed. My results therefore prove that R&D is a powerful proxy for identifying future stock returns for the countries analyzed and also suggests that firms carrying out R&D experience higher stock returns than firms who don’t.

22

5.3.

Under or over-pricing estimation through the Fama and French (1993) three

factor model and through the Carhart (1997) four factor model.

5.3.1. Fama French three-factor model

The main question this thesis attempts to answer is whether R&D intensive firms earn abnormal returns and whether this might be due to an incorrect valuation. Previous analysis so far has shown that R&D intensive stocks seem to earn excess returns compared to lower R&D intensive stocks following two different methods of portfolio formation. This was also captured by using the Fama and French (1992) model which proved R&D intensive stocks were underpriced for the full European sample after controlling for the book-to-market ratio and size. The multi-factor model by Fama and French (1993) is used, as portfolio formation could indicate whether R&D mispricing is caused by sources of additional risk that are not captured by the Fama and French (1993) three factor model. The full sample is regressed on different portfolios, these portfolios reflect R&D intensity. Six portfolios are created, one for non-R&D firms and the remaining portfolios reflect lower and higher R&D intensity. The regression follows from Fama and French (1993) and is as follows:

Full Sample

Beta 0.004 Beta 0.004 Beta 0.001 Beta 0.004 (0.81) (0.65) (0.27) (0.69) LN(BM) 0.001** LN(BM) 0.001 LN(BM) 0.001 LN(BM) 0.002 (2.38) (0.83) (1.46) (1.42) LN(Size) -0.0004 LN(Size) 0.0002 LN(Size) 0.0002 LN(Size) 0.0002

(-1.12) (0.64) (0.46) (0.66) LN(RD/ME) - LN(RD/ME) 0.001*** LN(RD/OP) 0.001*** LN(RD/TA) 0.001***

(-) (2.83) (2.94) (2.79) Intercept 0.005 Intercept 0.006 Intercept 0.005 Intercept 0.004 (1.06) (0.94) (0.76) (0.67) N 155037 N 53281 N 41634 N 53350 Adj. R² 0.0385 Adj. R² 0.048 Adj. R² 0.0564 Adj. R² 0.0482

** Significance at 5% level * Significance at 10% level

Firms with R&D

Table 5: R&D Capital and subsequent stock returns

Regressions are run using the Fama and Macbeth (1973) methodology. Beta is estimated using past data, 60

previous months are used to calculate the current beta which is then rolled forward for future periods (24 months is the minimum required period). Book-to-market ratio is defined as deferred taxes plus common equity divided by the number of shares outstanding multiplied by the market price, the BM ratio of the last period before fiscal year end is used. Size is the market capitilization of month t and RD/ME is the respective R&D of a firm that month divided by market capitilization. RD/TA is the respective R&D of a firm that month divided by the total assets of that firm. RD/OP is the respective R&D of a firm that month divided by the total operating profit of that firm.

23 𝑟𝑖,𝑡− 𝑟𝑓,𝑡 = 𝑎𝑖+ 𝛽𝑖(𝑅𝑚,𝑡− 𝑅𝑓,𝑡) + 𝛾𝑖𝑆𝑀𝐵𝑡+ 𝛿𝑖𝐻𝑀𝐿𝑡+ 𝜀𝑖,𝑡 (1)

Each country’s dataset uses the time period 2000-2013, which are a 168 monthly observations. What this regression will tell me, is that whenever an intercept is significantly away from zero could reflect the fact that R&D might cause an incorrect valuation of a portfolio’s stock returns due systematic sources of risk not captured by this regression. As the other factors should already capture the risk that R&D might induce, a significant intercept will tell otherwise. Firms are only used if the full coverage of data is available. Robustness and clustering checks are performed on each regression to ensure coefficients are robust to heteroscedasticity and cross-sectional correlation. All independent variables are winsorized by the top and bottom 1% to ensure outliers are removed.

Results in Table 6 offer some interesting insights compared to current research when looking at the intercept. The Fama French three factor model has a very high predictive power, all regressions run for the different portfolios have an adjusted r-squared above 75%. The beta, in this case the market factor is highly significant for all portfolios at the 1% level, as expected. The interesting insight is the negative and statistically significant intercept for the top R&D intensive portfolio which would indicate an overpricing. This would suggest that the high stock returns which the high portfolios seem to earn as seen by earlier analysis is unjustified by the Fama and French three factor model risk characteristics. An intercept significantly different from zero also suggests that the Fama and French three factor model is not efficient at explaining future stock returns. An intercept significantly different from zero suggests that there is a compensation for risk. What is also seen is that the Sharpe ratio, beta, increases as it moves to higher R&D intensive portfolios. This suggests that R&D intensive portfolios suffer from larger market exposure compared to lesser R&D intensive portfolios.

5.3.2. Carhart four factor model

The Carhart four factor model uses an additional variable, momentum, which has proven to explain excess stock returns such as in Carhart (1997) and also when in use for R&D studies such as in Eberhart, Maxwell and Siddique (2004) paper. Previous winning stocks seem to earn positive stock returns in the nearby future, this anomaly is corrected for by the inclusion of this variable. The momentum variable has been argued to take account of the lack of attention by investors. It has been argued that R&D is overlooked by investors, hence this proxy could adjust for this issue. This variable should also improve the adjusted R-squared compared to the Fama and French (1993) three factor model.

24 Table 6 confirms the fact that the adjusted R-squared has very slightly increased for most of the regressions, hence the predicting power of the regression has improved very marginally. The momentum variable does not seem to have a conform impact, considering it does not carry any statistical significance whatsoever. This somewhat coincides with Eberhart, Maxwell and Siddique (2004) who found little statistical significance for the momentum variable itself, but were able to interpret difference between the two regressions. What they found was a higher abnormal return when using the Carhart four factor model as opposed to the Fama French three factor model. There seems to be no difference however between the Fama and French three factor model and the Carhart four factor model. All coefficient have the same sign, and the statistical significance remains the same across the two regressions.

It can therefore be argued that the Carhart four-factor model has a slightly higher predictive power of explaining future stocks returns as compared to the Fama and French three factor model. Results between the two regressions remains the same however. Both regressions suggest that there is a significant negative risk-based excess return for the top R&D intensive portfolios and that they suffer from overpricing. This would suggest that the high returns these portfolios seem to earn as suggested by previous analysis in unjustified by the Fama and French risk characteristics.

These results do not coincide with previous research, as for example Lev and Sougiannis (1999) found that the top R&D intensive portfolios have significantly positive alphas which suggests that these portfolios earn excess returns after controlling for the Fama and French risk characteristics. These results would suggest that the higher returns associated with the top R&D intensive portfolios suggested by earlier analysis is due to a risk factor associated with R&D and not due to a certain mispricing.

As an additional analysis I altered the way portfolios are formed to see whether results would differ at all. Initially portfolios were formed on the basis of R&D intensity, I then formed portfolio on the basis of R&D divided by total assets and revenue. I then ran the Carhart four-factor model using these two methods of portfolio formation. As can be seen in table 7, results do alter quite significantly, what is most interesting to see is that the intercept is statistically insignificant for all the portfolios regressed, this would state that systematic sources of risk captured by the Carhart four factor model are similar for all the portfolios. Hence, earlier analysis, which suggested that R&D intensive portfolios earn excess returns, are not due to an extra risk compensation as Table 7 suggests, but rather due to a mispricing.

25 Fa ma F re nc h T hr ee F ac tor M od el Ca rh ar t F ou r F ac tor M od el R& D R& D N on -R & D Q 1 ( lo w ) Q2 Q3 Q 4 Q 5 ( H ig h) N on -R & D Q 1 ( lo w ) Q2 Q3 Q 4 Q 5 ( H ig h) Rm-Rf 0.9 48 ** * 0.8 62 ** * 1.0 33 ** * 1.1 31 ** * 1.2 30 1.2 98 ** * Rm-Rf 0.9 48 ** * 0.8 64 ** * 1.0 38 ** * 1.1 41 ** * 1.2 32 ** * 1.3 02 ** * (1 3.1 7) (1 9.2 8) (2 6.5 5) (2 5.8 9) (2 5.4 7) (2 3.6 2) (1 3.1 6) (1 8.9 1) (2 5.9 3) (2 6.0 1) (2 6) (2 3.4 ) SM B 0.1 41 0.0 46 0.0 82 0.0 28 0.0 68 0.0 44 SM B 0.1 40 0.0 43 0.0 74 0.0 07 0.0 62 0.0 31 (1 .2 5) (0 .4 8) (0 .7 5) (0 .2 7) (0 .5 6) (0 .3 8) (1 .2 4) (0 .4 4) (0 .6 9) (0 .0 6) (0 .5 1) (0 .2 7) H M L 0.0 92 0.0 33 0.0 29 -0 .0 25 0.0 38 -0 .0 65 H M L 0.0 86 0.0 35 0.0 34 -0 .0 11 0.0 41 -0 .0 59 (1 .5 3) (0 .4 7) (0 .3 9) (-0 .3 4) (0 .4 9) (-0 .8 7) (1 .3 6) (0 .5 ) (0 .4 4) (-0 .1 6) (0 .5 4) (-0 .7 7) M O M -M O M 0.0 00 0.0 00 0.0 00 0.0 01 0.0 00 0.0 00 2 -(-0 .2 1) (0 .2 6) (0 .5 1) (1 .4 7) (0 .4 1) (0 .5 2) In te rc ep t 0.0 02 0.0 05 ** * 0.0 06 ** * 0.0 04 ** * 0.0 01 -0 .0 11 ** * In te rc ep t 0.0 03 0.0 05 ** * 0.0 06 ** * 0.0 03 * 0.0 01 -0 .0 12 ** * (1 .4 5) (2 .9 8) (3 .6 5) (2 .4 2) (0 .6 8) (-5 .8 3) (1 .4 8) (2 .8 1) (3 .2 8) (2 .0 3) (0 .5 9) (-5 .7 7) N 13249 11078 10986 10998 11000 10950 N 13243 11078 10986 10987 10999 10950 A dj. R ² 0.7 63 9 0.7 63 4 0.8 42 7 0.8 56 4 0.8 69 2 0.8 64 7 A dj. R ² 0.7 63 9 0.7 63 4 0.8 43 0 0.8 56 6 0.8 69 2 0.8 64 9 * S ig nif ica nc e a t 1 0% le ve l ** S ig nif ica nc e a t 5 % le ve l ** * S ig nif ica nc e a t 1 % le ve l Fo r e ac h c ou ntr y, fiv e p or tf oli os a re cr ea te d, w he re by o ne p or tf oli o r ep re se nts stoc ks w ith n on -R & D fir ms , th e o th er fo ur re pr es en ts fir ms w ith R & D , w he re th e hig he r th e p or tf oli o, th e h ig he r th e r ati o o f R & D /M E. Po rtf oli os a re cr ea te d e ac h y ea r in Ju ne , w he re th ey a re re ba la nc ed y ea rly . R m-Rf is th e mo nth ly e xc es s r etu rn of th e s pe cif ic p or tf oli o i n q ue sti on o ve r th e r isk fr ee ra te ; S M B, H M L, M O M a re th e mo nth ly re tu rn s o f p or tf oli os w hic h a cc ou nt fo r s ize , B M ris k a nd mo me ntu m. D elta R -s qu ar ed a cc ou nts fo r th e d iff er en ce o f R -s qu ar ed b etw ee n th e tw o r eg re ss io n d ue to th e i nc lu sio n o f th e mo me ntu m va ria ble . T -s ta ti sti cs a re g iv en in b old te xt. Ta_ble 6 : R & D so rte d p or tfo lio s usi ng th e F am a F re nc h t hr ee fa cto r m od el an d t he C ar ha rt f ou r f ac to r m od el

26 R& D R& D N on -R & D Q 1 ( lo w ) Q2 Q3 Q 4 Q 5 ( H ig h) N on -R & D Q 1 ( lo w ) Q2 Q3 Q 4 Q 5 ( H ig h) Rm-Rf 0.9 48 ** * 0.8 70 ** * 1.0 87 ** * 1.0 90 ** * 1.2 31 ** * 1.3 18 ** * Rm-Rf 0.9 48 ** * 0.8 71 ** * 1.0 48 ** * 1.1 26 ** * 1.2 18 ** * 1.3 39 ** * (1 3.1 5) (1 9.6 6) (2 5.3 7) (2 5.6 3) (2 6.9 2) (2 3.5 6) (1 3.1 5) (2 0.1 9) (2 4.0 3) (2 5.8 8) (2 5.4 4) (2 3.3 9) SM B 0.1 40 0.0 79 -0 .0 03 0.0 93 0.0 77 -0 .0 28 SM B 0.1 40 0.0 06 0.0 40 0.0 24 0.1 26 0.0 18 (1 .2 3) (0 .8 3) (-0 .0 3) (0 .8 3) (0 .6 3) (-0 .2 4) (1 .2 3) (0 .0 7) (0 .3 9) (0 .2 ) (0 .9 6) (0 .1 6) H M L 0.0 86 0.0 80 0.1 13 -0 .0 45 0.0 05 -0 .1 15 H M L 0.0 86 0.0 85 0.1 22 0.0 34 -0 .0 07 -0 .1 53 (1 .3 5) (1 .1 1) (1 .5 4) (-0 .6 2) (0 .0 6) (-1 .3 7) (1 .3 5) (1 .2 2) (1 .7 6) (0 .4 6) (-0 .0 8) (-1 .8 ) M O M -0 .0 00 1 0.0 01 ** 0.0 01 0.0 01 -0 .0 00 2 -0 .0 00 4 M O M -0 .0 00 1 0.0 01 ** 0.0 01 0.0 01 ** * -0 .0 00 1 -0 .0 01 (-0 .2 1) (2 .3 ) (1 .8 7) (1 .7 6) (-0 .6 2) (-0 .9 5) (-0 .2 1) (2 .4 5) (1 .5 9) (2 .8 3) (-0 .3 5) (-1 .4 1) In te rc ep t 0.0 03 0.0 00 1 -0 .0 01 0.0 02 0.0 01 0.0 00 2 In te rc ep t 0.0 03 0.0 01 0.0 01 0.0 01 0.0 00 4 -0 .0 01 (1 .4 7) (0 .0 9) (-0 .3 5) (0 .9 6) (0 .3 4) (0 .0 9) (1 .4 7) (0 .3 2) (0 .4 1) (0 .5 5) (0 .2 1) (-0 .5 4) N 13229 11026 10960 10995 10981 10940 N 13229 10986 10942 10957 10953 10897 A dj. R ² 0.7 64 2 0.7 96 4 0.8 51 9 0.8 58 2 0.8 60 4 0.8 27 8 A dj. R ² 0.7 64 2 0.7 98 9 0.8 43 3 0.8 52 4 0.8 60 2 0.8 18 8 * S ig nif ica nc e a t 1 0% le ve l ** S ig nif ica nc e a t 5 % le ve l ** * S ig nif ica nc e a t 1 % le ve l Ta ble 7 : R & D so rte d p or tfo lio s usi ng th e Ca rh ar t f ou r f ac to r m od el Fo r e ac h c ou ntr y, fiv e p or tf oli os a re cr ea te d, w he re by o ne p or tf oli o r ep re se nts stoc ks w ith n on -R & D fir ms , th e o th er fo ur re pr es en ts fir ms w ith R & D , w he re th e hig he r th e p or tf oli o, th e h ig he r th e r ati o o f R & D /M E. Po rtf oli os a re cr ea te d e ac h y ea r in Ju ne , w he re th ey a re re ba la nc ed y ea rly . R m-Rf is th e mo nth ly e xc es s r etu rn of th e s pe cif ic p or tf oli o i n q ue sti on o ve r th e r isk fr ee ra te ; S M B, H M L, M O M a re th e mo nth ly re tu rn s o f p or tf oli os w hic h a cc ou nt fo r s ize , B M ris k a nd mo me ntu m. D elta R -s qu ar ed a cc ou nts fo r th e d iff er en ce o f R -s qu ar ed b etw ee n th e tw o r eg re ss io n d ue to th e i nc lu sio n o f th e mo me ntu m va ria ble . T -s ta ti sti cs a re g iv en in b old te xt. Ca rh ar t F ou r F ac tor M od el, b as ed o n R & D /T A Ca rh ar t F ou r F ac tor M od el, b as ed o n R & D /R EV

27

5.4.

R&D and stock return volatility

As mentioned previously, higher R&D could impose more risk on a firm, due to the fact that higher R&D does not ensure higher future returns due to the uncertainty that R&D carries. R&D also carries a significant amount of information asymmetry and a significant amount of costs which adds even more risk. Following Chan, Lakonishok, and Sougiannis(2001), estimating the effect of R&D on stock return variability might suggest that R&D intensive firms carry more business risk. Size and the book-to-market ratio in the following regression are used as risk characteristics, both variables are expected to have a negative coefficient. The Fama and Macbeth (1973) methodology is used for this regression.

𝜎_𝑖,𝑡+12= 𝑎₀+ 𝑎₁𝐿𝑁(𝑀𝐸)_𝑖,𝑡+ 𝑎₂𝐿𝑁(𝐵𝑀)_𝑖,𝑡+ 𝑎₃𝐿𝑁(𝑅&𝐷

𝑀𝐸)𝑖,𝑡+ 𝜀𝑖,𝑡+12 (3)

At the end of June each year, I calculate the stock return standard deviation over the next 12 months. The sample includes all stocks. The dependent variable is the stocks standard deviation, which is regressed on the firm’s respective market equity, book-to-market ratio and R&D’s intensity ratio. It has to be noted that natural logarithms are taken for all variables, due to the large differences in size among firms in the dataset. Three different regressions are run as a robustness check. The only difference among the three regressions is the denominator of the third independent variable. R&D is divided by market capitalization, total assets and revenue. Results in Table 8 below proves the methodology that R&D imposes more business risk on a firm due to its positive coefficient and statistical significance at the 1% level among all the regressions. Size is significantly negative on all countries, as expected, as larger firms tend to be less volatile

The book-to-market ratio is significantly negative for all three regressions. The higher stock return variability partially caused by R&D can be the fact that a high commitment to R&D induces a lot of costs which will negatively affect short term earnings, as R&D expenses can take years to generate revenue or any profitability. However, when R&D does create a new product with potential high revenues, it could lead to significant positive stock returns, as indicated.

This regression can somewhat explain previous regressions. Although the R&D variable carries a positive coefficient, although it is just slightly positive, could explain the fact that firms with high R&D intensity earn higher stock returns than firms with lower R&D intensity. Although it can be

28 concluded that on average, R&D could induce higher stock volatility, it cannot be seen as a guarantee that investing more money in R&D will guarantee higher stock returns. As can be seen from the adjusted r-squared, there are lots of omitted variables in this regression which could perhaps change the statistical significance or coefficient of the R&D variable. Chan, Lakonishok, and Sougiannis(2001) document similar results and argue that “as the limited disclosure of R&D contributes to higher return volatility, there may be a cost associated with the present accounting treatment of R&D” (Chan, Lakonishok, and Sougiannis 2001, page 2453). Results somehow coincide with what was found in the Carhart four factor model, which indicated that the higher R&D intensive portfolios had higher sharpe ratios, which suggest more market exposure.

29 Ln (M E) Ln (M E) Ln (M E) -0 .0 04 ** * (-2 6.4 9) Ln (B M ) Ln (B M ) Ln (B M ) -0 .0 01 ** * (-8 .5 8) Ln (R D /M E) Ln (R D /T A ) Ln (R D /R EV ) 0.0 03 ** * (1 8.7 2) In te rc e p t In te rc e p t In te rc e p t 0.0 73 ** * (2 8.6 8) N N N 84423 A vg R ² A vg R ² A vg R ² 0.1 32 4 ** * S ig n ifi ca n ce a t 1% le ve l ** S ig n ifi ca n ce a t 5% le ve l * S ig n ifi ca n ce a t 10 % le ve l 7C : R & D sc a le d b y r ev en u e 7A : R & D sc a le d b y m a rk et va lu e o f e q u ity 7B : R & D sc a le d b y to ta l a ss ets Ta b le 8 : T h e e ffe ct o f R& D o n st o ck v o la til ity 0.0 74 ** * 85778 (-2 8.9 8) -0 .0 01 ** * (-1 2.9 2) 0.0 01 ** * (9 .2 2) 0.0 71 ** * (2 9.5 2) Stoc k v o la ti lity is th e d e p e n d e n t va ria b le s w h ic h me as u re s th e sta n d ar d d e via ti o n o f 1 2 mo n th s o f s toc ks re tu rn s. LN (M E) is th e lo ga rith m o f ma rk e t va lu e o f e q u ity ; L N (B M ) i s th e lo ga rith m o f th e b o o k- to-ma rk e t ra ti o ; a n d w h e re R D is sc ale d b y th re e d iff e re n t p ro xie s, ma rk e t va lu e o f e q u ity , tota l a ss e ts a n d re ve n u e . -0 .0 04 ** * -0 .0 04 ** * (-2 8.1 ) -0 .0 01 ** * 0.1 12 4 (2 9.2 9) 85778 0.1 17 1 (-9 .6 6) 0.0 02 ** * (1 5.3 5)