Achim F. Himmelmann* (January 2011)

Transcription

1 SYSTEMATIC RISK DYNAMICS FOLLOWING R&D INCREASES Achim F. Himmelmann* (January 2011) ABSTRACT We analyze systematic risk dynamics subsequent to changes in corporate R&D investment policy. Findings are based on a sample of 1,170 cases between 1987 and 2004 when firms significantly increase their reported R&D expenditures. On average, systematic risk declines following R&D increases. After conditioning on the initial level of systematic risk, only firms with high original levels of beta exhibit a decline in systematic risk subsequent to the R&D investments, whereas low-risk firms, in fact, show an increase in systematic risk. Our findings motivate consideration of growth options' moderating role when analyzing R&D investments. We propose a differentiation between research and development activities both consolidated as R&D expenses: Research activities represent the creation of growth options whereas development actions proxy for the exercise of growth options. KEYWORDS R&D expenditures; Systematic Risk; Real Options JEL CLASSIFICATION G32; G14; M21 * Tech University Darmstadt Department of Business Administration, Economics, and Law Hochschulstrasse Darmstadt Germany Phone: +49 (0) Fax: +49 (0) himmelmann@bwl.tu-darmstadt.de 1

2 I. INTRODUCTION A company's systematic risk (beta) is the main driver of expected returns in equity capital markets. Investors account for beta in determining portfolio composition and performing risk management. Cost of capital and capital budgeting requires information on the systematic risk of the company as well. However, corporate decisions affect the systematic risk of a firm. Wellknown studies show that financing choices influence beta (e.g., Hamada (1972) and Rubinstein (1973)). Investment decisions are likely to impact the systematic risk of the firm, too. Companies face risky investment opportunity sets over time. In choosing to commit resources to a particular investment alternative for future growth, the company decides simultaneously on a specific cash-flow profile that, in turn, affects the systematic risk of the firm as a whole (Beaver et al. (1970), Miles (1986)). This paper investigates corporate investments decisions and their impact on systematic risk. Specifically, we focus on large investments in research and development (hereafter R&D) and their influence on the systematic risk of the firm. Wealth effects of corporate R&D activities are extensively analyzed in the literature. Investors seem to view R&D outlays as beneficial. Numerous studies reveal a positive market response to the overall level as well as changes in R&D expenditures (e.g., Bernstein and Nadiri (1988), Chan et al. (1990), Doukas and Switzer (1992), Chauvin and Hirschey (1993), Szewczyk et al. (1996), Lev and Sougiannis (1996), Chan et al. (2001), Chambers et al. (2002), Ballester et al. (2003), Eberhart et al. (2004)). Despite vast evidence on persistent high stock returns, risk dynamics associated with corporate R&D activities remain largely unexplored. This study contributes to the literature on corporate investment policy. We provide theoretical reasoning and empirical evidence of R&D investments and associated changes in systematic risk. Ho et al. (2004) present some support that firms with a high level of R&Dintensity also carry higher systematic risk than firms less exposed to R&D. In this paper, we investigate changes in R&D investment outlay. Thereby, we focus on alterations in the firm's investment policy and its implications for the systematic risk of the company. More precisely, we investigate systematic risk dynamics subsequent to a large increase in a firm's reported R&D expenditures. We understand R&D outlays as multi-stage investments and analyze them within a real options context (McDonald and Siegel (1986), Pindyck (1988), McGrath (1997)). Investments in R&D today not necessarily aim for immediate proceeds but intend to extract future economic benefits from discretionary follow-up investments. Real option features allow for managerial flexibility in a way that adjustments to predetermined decisions can be made upon arrival of new information. Thereby, the corporate decision to invest in R&D reflects a two-stage process. First, the firm chooses an initial investment outlay. After- 2

3 wards, the firm owes the flexibility, conditioning on the resolution of (systematic and unsystematic) uncertainty, to further pursue the activities or discard the venture. 1 Regularly, R&D is referred to as one single, indistinguishable item. We argue that R&D, in fact, summarizes two distinct corporate activities with respect to the initiation of an internal investment project. Research describes the process of generating new knowledge and technology. On the contrary, development refers to the process of translating this new knowledge into commercial use. Both activities are obviously different in nature. International Financial Reporting Standards (IFRS) account explicitly for the singular characteristics of both corporate activities. 2 The differentiation of research compared to development activities within R&D projects, leads to the differentiation between the creation and exercise of a growth option. Research activities are associated with the creation of a growth option. First-stage R&D investments are aimed for generating new knowledge, alternative products, or services. The firm explores novel alternatives subject to further evaluation and selection. Development activities represent the exercise of a growth option. The firm decides to advance basic groundwork in light of probable future economic benefits. In this second stage, the choice is made in favor of undertaking additional investments for commercialization in what proved to be promising basic exploration. Growth option creation and exercise, however, yield different implications for the systematic risk of the firm. A company can be described as a portfolio of assets already in place and growth options (Miller and Modigliani (1961), Myers (1977)). 3 It follows that the firm's beta is a weighted average of the beta of assets already in place and the beta of the growth options (Miles (1986)). Skinner (1993) and Bernardo et al. (2007) provide empirical evidence that growth options carry higher systematic risk than assets already in place. If a new growth option is created and added to the portfolio of the firm, the systematic risk of the firm will increase, ceteribus paribus. 4 However, if a growth option is exercised and converted into 1 Griliches (1981) points out that R&D investments create intangible assets associated with future albeit uncertain cash flows. Berk et al. (2004) develop a multistage model that captures features of R&D ventures and R&D-related uncertainty. They point out that R&D related future cash flow uncertainty exhibit both a purely idiosyncratic risk as well as a systematic risk component. Several authors study the effect of R&D investments on idiosyncratic risk. Empirical evidence favors a positive relationship between the level of growth options and idiosyncratic risk. Chan et al. (2001) find a positive relationship between firm return volatility (idiosyncratic risk) and R&D intensity. Schwert (2002) links growth options of large firms in high-tech industries with higher earnings and equity return volatility. Cao et al. (2008) establish a theoretic link between the level and variance of growth options and idiosyncratic risk of equity. The authors focus on the Galai and Masulis (1976) result that managers of levered firms have an incentive to select those investment projects from their set of growth opportunities that increase the idiosyncratic risk of the firm. 2 In IAS 38 the generation of an asset is split explicitly into a research phase and a development phase (IAS 38/50). Costs associated with the research phase must be expensed (IAS 38/52). During the research phase, projects are unlikely to demonstrate a sufficient probability for future economic benefits. Costs during the development phase can be capitalized because the probability to generate future benefits is assumed to be sufficiently high (IAS 38/55). 3 This differentiation is somewhat arbitrary. All assets exhibit a varying fraction of their value attributable to growth options from discretionary decisions as Jacquier et al. (2010) point out. 4 For simplicity, we assume that for every single firm new growth options carry the same risk as existing growth options. 3

4 assets in place, the systematic risk of the firm will decrease, ceteribus paribus. Carlson et al. (2006) develop a model with rational expectations and dynamically consistent corporate decisions in which a firm invests in expansion and thereby exercises growth options. The results show that the systematic risk of the firm decreases as growth options are converted into assets in place. When we investigate the impact of changes in R&D investments and associated shifts in systematic risk, we rely on reported R&D expenditures of the firm. U.S. GAAP consolidates research and development activities and requires both costs to be expensed as occurred. Because no distinction is made between the research and the development part, R&D investments measured as R&D expenditures are likely to have different implications for different firms. Based on the explanation above, we argue that firms carrying high systematic risk are likely to have lots of growth options available. If such firms undertake an additional, substantial investment in R&D, they most likely engage in development activities and thereby exercise parts of their existing growth options. As a results, systematic risk should decline. On the other hand, companies carrying low levels of systematic risk are probably less well endowed with growth options. If these companies engage in R&D investments, they most likely engage in research activities and thereby create growth options. Hence, systematic risk should increase. McGrath and Nerkar (2004) point out the incentives for a firm to invest in the competence to exercise real options increases when the level of opened options (and risk) is high. If a firm was engaged heavily in creating options, follow-up investments are rather aimed for capitalizing on and eventually exercising existing options than creating additional ones. The rational is that the value of the firm's option to defer exercise declines as the level of opened options inclines. The authors argue that investment in knowledge raises the odds of successful option exercise in general, thereby increasing the value of all opened options. The idea is placed within an option portfolio context where real options may not be additive but interact with respect to the overall underlying capacity of exercising them. Real options are often nested and valued less if internal resources are insufficient to provide adequate development capacity. Firms investing in the learning process increase capacity for option development and thereby decrease their exposure to systematic risk. The incentive to foster exploitation-oriented investment increases as the level of growth options and the exposure to systematic risk increases. 5 We seek to provide empirical evidence for this reasoning. Following the procedure in Eberhart et al. (2004), we identify a sample of 1,170 large increases of R&D expenditures by U.S.-listed companies between 1987 and Results suggest, on average, a significant decline in systematic risk following the 5 A different reasoning with focus on product-market effects is advocated by McAlister et al. (2007). The authors argue that R&D activities lead to increased dynamic efficiency and flexibility. The firm is able to more favorably adapt to environmental changes (price-, technology-, and customer-wise), thus insulating itself from market downturns and exhibiting a decline in systematic risk exposure. 4

5 investments. The findings remain unaffected when controlled for leverage and industry effects. Several robustness checks underline our results. Our sample firms come predominantly from technology-intensive industries and exhibit higher than average R&D intensities. By nature of their business, these firms carry lots of growth options. R&D increases for such firms, therefore, likely aim for developing activities and exercising growth options. Consequently, we would expect a decline in systematic risk in line with our empirical evidence. When clustering the sample firms according to the initial level of systematic risk prior to the R&D increase, we find evidence for a nonlinear relationship conditioning on the firm's initial asset composition and level of risk. Firms with high opening levels of systematic risk, in fact, exhibit a decline in beta subsequent to the investment in R&D. Conversely, firms with low levels of initial systematic risk exhibit an increase in beta. This result confirms our hypothesis and underlines the complex character of R&D investments. We conclude that firms with high risk levels have an incentive to invest in capacity to exercise growth options. To analyze whether the systematic risk development differs across various groups of companies, we refer to the classification of Cui and Mak (2002) and divide our sample firms into high- and low-tech subgroups. Moreover, we conduct a split into high- and low-growth firms (see Eberhart et al (2004)). We find little evidence that firms in different subgroups exhibit different systematic risk patterns. The remainder of the paper is organized as follows. In Section II, we discuss the sample selection procedure and present descriptive statistics. Section III contains empirical results. Our core focus lies on the analysis of systematic risk dynamics. We estimate pre- and post-r&d-increase assets. Robustness checks are conducted with respect to leverage, industry, and measurement interval effects. Finally, we conduct cross-sectional regression analysis. Concluding remarks are given in Section IV. II. DATA, EVENT SELECTION, AND METHODOLOGY Our goal is to analyze the systematic risk effects of changes in a firm's R&D investment policy. We proxy for R&D investments by the firm's reported R&D expenditures. R&D expenditure is an input related measure and therefore reflects the company's effort to put resources in the research and development process. To distinguish our findings from changes in R&D outlay as a result of firm growth and other confounding forces, we define events of interest as "large" increases in R&D expenditures. Throughout the analysis we use the terms R&D increase, announcement and event interchangeably. In accordance with Eberhart et al. (2004), we formalize a "large" R&D increase based upon two requirements: First, the company needs to exhibit an incline in absolute R&D expenditures of at least 5 percent as well as a rise in 5

6 R&D expenditures over assets of at least 5 percent. 6 Second, we reduce the sample to economically significant R&D increases. Therefore, companies need to show a sufficient R&D intensity (measured as the ratio of R&D expenses over assets and R&D expenses over sales) of at least 5 percent. We assume a 4- month gap until accounting data becomes publically available, as suggested by Chan et al. (2001). Based on these requirements, we search all firms listed either on NYSE, AMEX, or NASDAQ and with sufficient data coverage in Thomson Financial Datastream and Worldscope. We analyze fiscal year end data between 1987 and All firms report accounting data based on U.S. GAAP. Scholes and Williams (1977), Dimson (1979) and Cohen et al. (1980) show that beta estimates based on daily return data can be biased for stocks that are infrequently traded or that show frictions in the trading process. We therefore rely on weekly return data to overcome this concern as Brown and Kapadia (2007). The systematic risk of a company, measured as the firm's equity beta, is estimated using the Fama and French three-factor model. For robustness checks, we also estimate our results based on the fourfactor extension as suggested by Carhart (1997). 7 The Fama and French three-factor model is specified as: (1) R R = α + β R R ) + s SMB + h HML + ε. i, t f, t ( m, t f, t t t i, t R i,t is the return on firm i at period t, R f,t is the return on three-month Treasury bills, R m,t is the return on a value-weighted market index, SMB t is the return difference between a portfolio of small stocks and a portfolio of large stocks, HML t is return difference of a portfolio of stock with high book-to-market ratios and a portfolio of stock with low book-to-market ratios. Equity risk is also analyzed based on the Carhart (1997) four-factor model. 8 (2) R R = α + β ( Rm, t R ft ) + s SMBt + h HMLt + mumdt + ε i,. i, t f, t t 6 Eberhart et al. (2004) impose the restriction of 5 percent increase in the ratio of R&D expenses over assets when they argue that the market is likely to expect an increase in R&D as the firm grows over time. Therefore, R&D increases need to be adjusted for overall company growth expressed in the firm's assets. The authors point out that any potential "announcement period effects" are unlikely to distort the analysis of unexpected R&D increases because the "events" of interest are based on accounting data and not on any sort of formal R&D-related announcement. 7 Mitchell and Stafford (2000) argue that variations of stock returns for small-size firms are not well explained by the three-factor model. We neglect this criticism because less than five percent of our sample firms which increases their R&D expenditures come from the lowest size quintile. The data is retrieved from Kenneth French's website ( Fama and French (1993) provide a detailed discussion of the construction of the factors. Daily data on the momentum factor (up minus down, UMD) is converted into a weekly time-series. 8 Carhart (1997) extends the Fama and French three-factor model and includes an additional momentum factor which captures the price momentum anomaly (see Jegadeesh and Titman (1993)). 6

7 An additional momentum factor (UMD t ) is included which measures the return difference between a portfolio of stock with high prior period returns and a portfolio of stocks with low prior period returns. To investigate potential industry-effects, we refer to the classification of Cui and Mak (2002) to define sample firms as either high- or low-tech companies based on their primary SIC code. III. EMPIRICAL RESULTS A. Descriptive Statistics In total, we extract 1,170 "large" R&D increases by 468 firms according to our definition. Table I reports the distribution of the final sample over time and company type. It does not come as a surprise that most of our sample firms fall into the high-tech category. 84 percent of all R&D increases occur in the high-tech sector. A similar picture is drawn when we split our observations according to high-growth and lowgrowth industries. 90 percent of our observations are classified as high-growth where high-growth firms are defined to exhibit a market-to-book ratio greater than one. More than half of the identified R&D increases take place in the business equipment industry. 28 percent of the observations occur in the healthcare business. The distribution across industries remains largely stable over time. Looking at the chronological allocation of R&D increases we find that most observations take place after Our sample shows similar characteristics as in Eberhart et al. (2004). <<<<<< Table I about here >>>>>> Table II displays the observations and sample firms grouped by size and R&D intensity. Size is measured as the average market value of equity one year prior to the R&D increase. Quintiles are defined using breakpoints of all NYSE, AMEX and NASDAQ listed companies. Absolute numbers are reported as well as percentages of totals in parentheses below. Most observations are collected from the two largest categories of firms. 61 percent of all observations are attributable to the two largest firm categories (Panel A). Similarly, the two largest cells account for 58 percent of all firms that exhibited an R&D increase (Panel B). <<<<< Table II about here >>>>>> 7

8 B. Changes in Systematic Risk Following R&D Increases Figure I summarizes our results on the systematic risk development over time. One-year rollingwindow regression estimates of equity betas based on the Fama and French three-factor model specification are shown. The graph indicates that, on average, systematic risk declines subsequent to an increase in R&D investments. <<<<< Figure I about here >>>>> Following the graphical intuition provide in Figure I, we examine the shifts in systematic risk associated with R&D increases and analyze statistically the changes in stock beta for our sample firms over a 1, 3, and 5 year post-announcement horizon. Beta estimates are retrieved based on the Fama and French three-factor as well as the Carhart (1997) four-factor model outlined above. A pre-announcement beta is calculated over 1 year period prior to the R&D increase. Pre- and post-announcement betas are tested for significance using standard t-tests for means. The Wilcoxon signed-rank test is applied to test whether the median changes are significantly different from zero. We split our sample to capture any industry-related effects. Table III provides results on the systematic risk dynamics of our sample. Panel A summarizes findings for the full sample. One year after the R&D increase, changes in risk are negative albeit insignificant. However, for the 3- and 5-years post-announcement period risk changes are significantly negative at least at the 5 percent level for both factor model specifications. Relying on the three-factor model, 5 years after the R&D increase we find a mean risk reduction of percent (18.20 percent decline in medians). Panel B through D present shifts in beta for our subsamples of high-tech, low-tech, high-growth, and low-growth firms. Changes in beta throughout the first year after the R&D increase are insignificant for all subsamples. A comparison of betas thereafter shows declining measures of systematic risk which are highly significant for the high-tech and high-growth subsamples. Results are less clear-cut for the lowsubsample. Low-tech sample firms show only week statistical significance for a beta decline based on a five-year comparison. On the contrary, low-growth sample firms exhibit no significant beta dynamics according to both factor model specifications. <<<<< Table III about here >>>>> C. Robustness Check: Beta Changes and Changes in Leverage After R&D Increases 8

9 We are interested in the effects of R&D investment on the underlying business risk of the firm. Financial leverage, however, affects the beta of firm and could potentially bias our conclusion (Hamada (1972)). We follow the procedure in Denis and Kadlec (1994) to disentangle changes in systematic risk as a consequence of R&D increases from financial leverage-induced shifts in beta. In accordance with Hamada's model (1972) we assume a debt beta of zero. Then, the change in a firm's equity beta is related to its capital structure as follows: (3) β = β ( D / E). E A β E (β A ) refers to the firm's equity beta (asset beta), D and E to its debt (book value of long-term debt) and equity (market value of equity) level respectively. We make use of the model predictions to determine the implied effect of changes in the firm's capital structure on its systematic risk. In a first step, we unlever each sample firm's equity beta based on its total debt level and market value of equity. After calculating the changes in the firm's debt-to-equity ratio for the bandwidth of 1, 3, and 5 years subsequent to the R&D increase, we determine the predicted change in beta by multiplying the firm's preannouncement equity beta by the change of its debt-to-equity ratio. <<<<< Table IV about here >>>>> Results are reported in Table IV. Panel A summarizes (mean) predicted and actual beta changes. Predicted beta changes are close to zero for all post-announcement periods. Actual beta changes are negative ranging from for the 1-year comparison to for the 5-year comparison. The mean as well as median differences between predicted and actual beta changes are significant for the 3- to 5-year horizons. Findings suggest a significant reduction in systematic risk subsequent to R&D increases over and above capital structure-induced risk shifts. A similar conclusion is supported by our univariate regression results in Panel B. Actual and predicted beta changes are linearly related and parameters are estimated via OLS regression. Predicted beta changes exhibit an insignificant effect on actual beta changes for the 1- to 5-year horizon. However, the intercept in our regression models is statistically different from zero. This finding suggest that shifts in systematic risk after R&D increases are not only attributable to altered capital structures. 9 For consistency, we use unlevered equity betas (asset betas) for the rest of our analysis. Thereby, we can rule any distorting effects associated with changes in financial leverage. We derive asset betas by unlevering the equity beta 9 We note that the F-statistic suggests insignificance of the model. Since in a simple linear regression model the F- statistic is equal to the squared t-statistic of the coefficient estimate, this result does not come as a surprise. 9

10 using the ratio of market value of equity and book value of long-term debt under the assumption that the debt beta is zero. D. Robustness Checks: Conditional Betas, Estimation Interval, and Return Measurement We estimate conditional versions of the multi-factor asset pricing models described above to track time-wandering levels of risk. Factor-loadings are specified as linear functions of predetermined lagged instrument variables (e.g., Shanken (1990), Fama and French (1997)). Conditional regressions allow slope coefficients to vary with firm characteristics, thus, capturing temporal variation in risk factor sensitivities. We describe the excess return series in general as follows: (4) R R = α + β R + e M i, t f, t i i, F F, t i, t F = 1. R i,t - R f,t is the excess return of firm i at time t over risk-free three-months Treasury bills, α i is a constant, β i,f is factor sensitivities, R F,t represents the asset pricing factors. Pricing factors include (R m -R f ), SMB, HML, and UMD for the Fama and French three-factor and Carhart four-factor model, respectively. Similar to Levellen (1999), we describe the factor sensitivities as a linear function of a state variable z: (5) β b + b z. i, F, t = o, i, F 1, i, F t 1 Substituting equation (5) in expression (4) yields the conditional versions of our factor pricing models: (6) M R i, t R f, t = α i + ( bo, i, F + b1, i, F zt 1 ) RF, t + ei, t. F = 1 Within the conditional regression framework, we measure the level of systematic risk as the sum of the coefficient estimates ˆb 0 and ˆb 1, both capturing the effects of the market factor (R m - R f ) and the interaction term of the state variable and market factor (see Franzoni (2002)). We use the ratio of MTBV as state variable See Lewellen (1999) for a detailed discussion on the relationship between MTBV and stock returns. 10

11 As a robustness check for the time-varying nature of risk (Engle (1982), we also condition on time-wandering variance. Building upon Bollerslev's (1986) generalized autoregressive conditional heteroskedasticity framework (GARCH), a conditional model is estimated via maximum likelihood with a GARCH (1,1) specification of the variance term. 11 More specifically, the error term is modeled as 2, t i, t ε i ~ N(0, σ lagged variance: ) with the error term's variance depending linearly on previous period squared error and (7) σ ϖ + ϖ ε ϖ σ. i, t = 1, i 2, i i, t 1+ 3, i i, t 1 Findings are reported in Table V. Systematic risk declines significantly after one year subsequent to the R&D increase. When comparing the shift in beta over a three year post-announcement period, we find a decrease in systematic risk of 47.5% estimated within the Fama and French three-factor conditional framework (53.38% based on the Carhart four-factor model). Measuring the change in risk over a five year post-announcement period, systematic risk declines by 56.83% (61.98%). Similar results are obtained by employing a GARCH model specification in Panel B. We also investigate the effects of measuring changes in systematic risk relative to different preannouncement intervals. In additional to the one year pre-announcement interval (Panel A), Table V also reports systematic risk measured over a two year period prior to the R&D increase (Panel C). 12 Declines in systematic risk are robust to different pre-announcement measurement intervals. Systematic risk measured over a three and five year post-announcement period is significantly lower compared to pre-announcement levels. Conditional regression results are similar to unconditional regression results. Therefore, we continue our investigation within the unconditional Fama and French three-factor and Carhart four-factor framework. In Panel D, we estimate changes in systematic risk based on monthly return measures. Our major conclusion remains unaffected by different return measurement intervals. E. Growth Option Creation vs. Growth Option Exercise So far, we find empirical evidence, that systematic risk, on average, declines subsequent to large investments in R&D. To better understand the relationship between the systematic risk and the level of growth options, we investigate the correlation between the level of pre-announcement asset beta and vari- 11 The next period variance is routinely updated conditional on the most recent past information set. 12 It is possible that the R&D increase occurred sequentially during the fiscal year (e.g., quarterly financials releases). Therefore, risk dynamics may have started earlier than at the date of public release of fiscal year end data. Measuring pre-announcement beta over increased time-intervals could capture these effects. 11

12 ous growth option proxies. Growth options are not directly observable. We include different growth option measures as described in Cao et al. (2008) to control for the impact of alternative definitions. 13 MTBV refers to the market-to-book ratio. MABA refers to the ratio of (total assets - common equity + market value of equity) to total assets. Tobin's Q is defined as (market value of equity + preferred stocks + current liabilities - current asset + long-term debt) over total assets. Measures are determined for the year prior to the R&D increase. We calculate Spearman and Pearson correlations between the level of systematic risk and the growth option proxies. Results are summarized in Table VI. Findings are in favor of our hypothesis that high initial levels of systematic risk are associated with high levels of growth options. All growth option measures employed are positively related to the level of systematic risk and highly significant. The Pearson (Spearman) correlations between these measures range from to 0.17 (0.053 to 0.129). We also note that all three growth option proxies are highly correlated. <<<<< Table VI about here >>>>> We investigate the changes of beta subsequent to R&D increases further and split our sample according to the initial level of systematic risk. Companies with a high level of systematic risk also carry lots of growth options. As we argued above, such firms predominantly conduct development activities and thereby converting existing growth options into assets in place. Systematic risk then decreases. Firms with low levels of systematic risk aim for research activities and the creation of growth options for future economic benefits. An increase in R&D expenditures will lead to an increase in systematic risk for these firms. Conditional changes in systematic risk are presented in Table VII. Firms are sorted into quintiles according to their pre-r&d-increase level of systematic risk. Consistent with our hypothesis, companies in the bottom and top quintiles exhibit opposing systematic risk developments. Firms with high initial levels of systematic risk show a decline in risk subsequent to investments in R&D compared to firms with moderate levels of initial systematic risk which exhibit an increase in beta. The results are unaffected by industry-affiliation. Panel A to Panel D in Table VII report the systematic risk development for firms in the high-tech, low-tech, high-growth, and low-growth subsamples. The general conclusion is supported. Systematic risk decreases for firms in the highest quintile whereas beta increases for firms with low levels of pre-r&d-increase systematic risk. In the high-tech sample, beta 13 Measures are adjusted for data availability. We dismiss measures based on capital expenditures since the vast majority of sample firms come from technological-intensive industries in which capital expenditures is a lessinformative measure. Measures based on the firms' capital structures are omitted because we measure systematic risk as asset beta - net of capital structure effects. 12

13 declines from 1.24 in the highest quintile to 0.84 (0.63) measured over a one-year (five-year) postannouncement period. In the lowest quintile, systematic risk increases from to 0.52 (0.38) measured over a one-year (five-year) post-announcement period. The effects are similar albeit less pronounced for the low-tech and low-growth subsample, respectively. Small sample sizes for these sub-categories hinder us from drawing more distinct conclusions. <<<<< Table VII about here >>>>> Up to now, we analyzed absolute risk dynamics of R&D increasing firms. A potential problem in isolating the effect of R&D increases on systematic risk is that general factors such as industry cycles themselves could cause changes in beta and may coincide with our events of interest. For this reason, we calculate industry-adjusted measures of systematic risk. Thereby, we focus more closely on R&D investment related effects on beta and avoid spurious inferences. We adjust a R&D-increasing firm's level of systematic risk by comparing it to the beta estimate of a matched non-r&d increasing firm. Matching firms are selected to mirror best risk attributes of R&D-increasing firms. Therefore, we proceed similarly to Lewis et al. (2002) and García-Feijóo et al. (2010) and select comparable firms with respect to industry affiliation, firm size and operating income. To facilitate comparison, we contrast the sample firm's return series to the return development of a matching company (see Fee and Thomas (2004)). We match companies based on R&D intensity. For each R&D-increasing firm in our sample, we identify a set of companies in the same industry (according the two-digit SIC code) that have not themselves experienced a significant R&D increase one year before and five years after. Commencing with this initial set, we identify those companies with total asset size between 70% and 130% of the sample firm. From the remaining companies, we choose the firm as comparable company with the R&D intensity (asset-based) closest to that of our sample firm. If no firm meets these criteria, we extend our scope and redefine industry-matching based on the one-digit SIC code. If no comparable company meets the relaxation, we dismiss the industry affiliation requirement and match a company based on total asset size and the operating performance (measured as EBITDA over total assets) closest to the sample firm (see Lewis et al. (2002)). We employ additional matching algorithms to construct equally-weighted reference portfolios for comparison. The benchmarks are assembled based on the market value of equity as well as the market-tobook ratio criterion (see Lyon et al. (1999)), assuming non-rebalanced investment in each company in the reference portfolio. For each sample firm, we identify a set of no-event control companies in the same industry (two-digit SIC code) and a market value of equity (market-to-book ratio) in the range of 70% and 13

14 130% of the R&D increasing firm as a representative match. 14 The reference portfolio then comprises the companies with market value of equity (market-to-book ratio) closest to the sample firm. For each sample firm, we require at least five but no more than fifty comparable companies. The industry-adjusted excess (ex) beta is then defined as the difference between the asset beta of the R&D-increasing firm and the asset beta of the comparable company: (8) β = β β. ex i, t A i, t A C, t ex β i,t denotes the excess beta of R&D-increasing firm i, measured over the time period t. A β C,t represent the asset betas of R&D-increasing firm i and its comparable company C, both measured over time period t. Asset betas are derived by unlevering the equity beta using the debt-to-asset ratio under the assumption that the debt beta is zero. The debt-to-asset ratio is defined as the ratio of the book value of total debt divided by the sum of the book value of total debt and the market value of equity. Excess betas are calculated one year prior and up to five years subsequent to the abnormal R&D increase. The number of observations declines due to data unavailability. After deriving the excess betas for our R&D increasing firms, we split the sample into quintiles according to the firms' one year pre-announcement level of excess beta. We also divide our sample according to industry affiliation. Table VIII summarizes the results on industry-adjusted excess betas derived within a Fama and French three-factor model framework. Results based on Carhart's four factor model are similar and reported in Appendix 1. The findings are similar to the unadjusted beta analysis above. Post-announcement excess beta development is conditional on the initial level of systematic risk. Firms with high pre-announcement excess beta exhibit, on average, a decline in excess beta during the years thereafter, whereas firms with relatively low initial surplus systematic risk show an incline. The results also reveal a tendency of firms to revert their risk exposure back to industry mean levels. Throughout all model specifications we see that the excess betas are closer to zero compared to preannouncement levels. Firms could be interpreted to conduct investment policies that adjust their risk baring to industry average. The sample is split into high- and low-tech as well as high- and low-growth firms in Panel A to D. We find that results are most pronounced for firms in the high-tech and high-growth subsample, respectively. Low-tech and low-growth firms exhibit less definite results. Again, this may be attributable to the small sample size for these subgroups. β A i,t and 14 Bernardo et al. (2007) favor MTBV-matching because it proxies for the level of real options. When assessing the impact of additional real options on the firm's beta, such matching criterion seems reasonable. 14

15 F. Long-term Abnormal Security Returns Following R&D Increases So far, we focused on the systematic risk effects of R&D increases. However, changes in systematic risk are associated with changes in stock returns within a basic risk-return framework. For a more complete picture, we thus investigate shareholder wealth effects of R&D increases. We examine the long-term impact of R&D increases on stock returns. Similar to Eberhart et al. (2004), we use calendar-time returns (equal- and value-weighted based on the firm's market value of equity) to assess abnormal share performance subsequent to R&D increases. For every period t the calendartime portfolio includes sample firms that increased their R&D expenses during the preceding one, three, and five years, respectively. The intercept α is interpreted as the abnormal stock return of the event portfolio across various bandwidths. 15 Heteroscedasticity potentially effects our estimation results since the number of R&D increasing firms in the calendar-time portfolio changes over time. Hence, we conduct weighted least squares (WLS) regression in addition to ordinary least squares (OLS) to estimate model parameters. The number of firms in the portfolio at time t serves as weighting factor (see Lyon et al. (1999)). There is much debate on the appropriateness of measures for long-term abnormal stock returns. We therefore follow Lyon et al. (1999) and include buy-and-hold abnormal returns as an additional robustness check to identify long-run abnormal stock returns. 16 Stock returns of firms exhibiting a large R&D increase are compared to the return development of a suitable benchmark. We thereby account for possibly missing risk dimensions in the factor pricing models. 17 After matching R&D increasing firms with a benchmark, we calculate long-run buy-and-hold abnormal returns for each sample firm as the difference between the sample firm's return and the return of the benchmark: (9) R + 1) ( R + BHAR 1). T i, [ t, T ] = t ( i, t B, t t T 15 Under the null hypothesis of complete expected return description, the model includes all pricing factors relevant to fully determine future returns. Then, the intercept alpha measures the average periodic abnormal performance (mispricing). The joint test problem (Fama (1970)) occurs if the model provides only an imperfect description of expected returns. In this case, the intercept measures the combined effects of both mispricing and model misspecification. 16 The calendar-time approach is less sensitive to cross-sectional dependency among sample firms because returns are aggregated into a single portfolio. In contrast, buy-and-hold methodology more accurately reflects investor experience (see Barber and Lyon (1997), and Lyon et al. (1999)). 17 First, we compare the sample firm's return development to the return series of the market. We use the S&P 500 index as benchmark. Similar results are obtained when choosing the same market index employed in constructing the Fama and French factors (see Fama and French (1993) for a description) and the local index provided by DataStream (see Appendix 2). 15

16 R i,t represents the event firm's i return over the period from t to T, R B,t is the return on the benchmark from period t to T. We derive the buy-and-hold average abnormal returns (BHAR) over different holding periods. Statistical inference is determined using different test statistics in light of positively skewed buyand-hold abnormal returns for increased time horizons (e.g., Barber and Lyon (1997), Cowan and Sergeant (2001)). For each bandwidth, we calculate conventional t-statistics. As robustness check we included skewness-adjusted t-statistics as suggested by Lyon et al. (1999). Moreover, we bootstrap the test statistic by drawing 10,000 bootstrapped re-samples of size n/4 with n representing the original sample size of buy-and-hold returns. Table IX provides results on the long-term abnormal security returns for companies increasing their R&D spending within the sample period. Using both equally- and value-weighted calendar-time returns yield significantly positive intercept estimates. OLS and WLS model specifications report similar results which are robust to different factor pricing models. The findings are consistent with Eberhart et al. (2004) who attribute the positive stock performance to mispricing that takes years to correct. <<<<< Table IX about here >>>>> Relying on the three-factor model, the authors report 5-year alphas of (equally weighted) and (value weighted) in their basic model which are generally consistent with our results of 0.35 (equally-weighted, OLS) and 0.29 (value-weighted, OLS). The four-factor Carhart extension supports the picture. In Table X, we display buy-and-hold returns. Results are in line with our calendar-time analysis. BHAR are significantly positive across almost all post-announcement periods. BHAR for sample firms 3 years after their R&D increase range from 21% (size-matched control firms) to 22% (market-to-book ratio matched control firms) to 48% (R&D matched control company) to 71% (index-matched) and seem robust to different matching criteria. All testing procedures show significance at least at the 5 percent level. <<<<< Table X about here >>>>> In sum, the analysis of on abnormal stock performance subsequent to a large R&D increases confirms findings from previous studies. Similar to Eberhart et al. (2004) we find significant positive abnormal stock performance after based on calendar-time portfolios. R&D-increasing firms earn persistently abnormal returns. We also conduct BHAR-analysis and add evidence that R&D increasing firms outperform non-event industry matches. Overall, we conclude that R&D increasing firms show an aboveaverage long-term stock performance. These findings, on average, are in line with decline in systematic 16

17 risk we report above. If beta decreases, the required cost of equity also declines leading to an increase in stock prices and positive stock returns. G. Cross-Sectional Regression Analysis Variable Descriptions We conduct regression analyses to investigate the specific determinants of long-term abnormal stock returns and changes in systematic risk. In a first step, we relate the market-adjusted BHAR return risk subsequent to R&D increases to a set of explanatory and control variables. Then, we analyze the factors influencing changes in systematic risk. All regressors are measured one year prior to the R&D increase unless stated otherwise. We employ control variables that account for Size (measured as the natural logarithm of the market value of equity), Leverage (defined as the ratio of long-term debt to the market value of equity), return on equity capital (ROE), and the level of free cash-flow (FCF). 18 Moreover, we consider additional firm characteristics that were found to moderate the magnitude of the market valuation of R&D investments. These characteristics could potentially explain the long-term abnormal returns that we observe after R&D increases. Pindado et al. (2010) provide an thorough overview over various firm characteristics influencing R&D valuation. We follow the authors' reasoning and include the sample firms' external financial dependency (EFD) in our regressions. Firms characterized by a high level of EFD rely heavily on external funds to finance R&D investments. However, since R&D investments create intangible assets, such financing is more costly given intangible assets' reduced usability in the case of financial distress. Therefore, high EFD should lower the benefits from R&D investments to shareholders, ceteribus paribus. EFD i,t for firm i at time t is defined as CAPEX CF i, t i, t CF i, t, the percentage difference between capital expenditures (CAPEX) and cash-flow (CF). EFD_D is a binary variable which equals one for firms with EFD larger than the sample firms' mean, zero otherwise. 18 See Pindado et al. (2010) for an excellent summary on various firm characteristics and their presumed influence on the market reaction to R&D effort. Larger firms are assumed to benefit from easier capital market access, R&D cost spreading and economies of scale in R&D which grant them advantages in technological competition compared to smaller firms. With respect to free cash-flow one can argue in the spirit of Jensen (1986) that firms with high free cash-flow levels (excess funds) are prone to engage themselves in negative net present value projects. Thus, R&D investments should have less definite return effects on firms with high free cash-flow levels. 17

18 To account for the potential impact of human capital on the profitability of R&D investments, our analysis includes a measure of labor intensity. Efficient firms are able to capitalize on R&D investments with a small and efficient workforce. R&D-related profits are less diluted amongst employees leaving a larger share to stockholders. So, we define labor intensity (Labor_D) as a dummy variable that takes the value one if a firm's labor intensity, defined at the ratio of total sales to the number of full-time employees, is higher than the average of all sample firms, zero otherwise. We also include the firm's market share (Market_Share), proxied by the sample firm's total sales in percent of the total sales of the industry, when industry-affiliation is determined by the primary twodigit SIC code. Empirical evidence suggest that firms with high market shares innovate more. High market share creates entry barriers for competitors allowing the innovating firm to extract vast benefits from R&D investments. In a basic risk-return framework, changes in the firm's risk profile lead to altered market valuation of the firm. More precisely, if the systematic risk of the sample firm declines, cost of equity decreases leading to an increase in stock price. An inclining stock price means positive stock returns and positive long-term BHAR. Hence, we include the change in asset beta (ΔBeta) calculated as the percentage difference of the post-announcement asset beta and one year pre-announcement asset beta. Previous univariate results suggest a conditional relationship between pre-announcement levels of systematic risk and future risk development. Therefore, we include an interaction term (ΔBeta x Beta) and relate the change to the initial level of asset beta. R&D investment per se is an input-related measure and does not allow for immediate conclusion on output-related indications such as profitability. A company's ability to transform R&D investments into tangible future benefits is likely to influence the stock markets' perception and valuation of R&D investments. Similarly, McGrath and Nerkar (2004) and Ioulianou et al. (2010) point out the importance of a firm's growth option capacity when engaging in innovation. To proxy for the transformation capacity, we also include a firm's pre-event R&D intensity in our analysis. We argue that firms characterized by above average initial R&D intensity created sufficient knowledge and experience in the past to benefit more by large R&D increases than less R&D intensive firms. Hence, we define R&D-Intensity_D as a dummy variable taking the value one if the firm's R&D intensity, measured as the ration of R&D expenses to total assets, exceeds the average R&D intensity in its industry. We also include the absolute level of preannouncement R&D-intensity. To examine the potential effect of real option capacity, we also proxy for the concentration of our sample firms' business activities. Firms that focus their business activities presumably become specialists in their field of actions. Most likely, they build a knowledge advantage which should increase their capability of successfully undertake new ventures. Therefore, such firms should engage in option exercise when increasing their R&D and exhibit advantageous odds in doing so. We meas- 18