Innovation and Firm Value: An Investigation of the Changing Role of Patents and Firm Publications

Innovation and Firm Value: An Investigation of the Changing Role of Patents and Firm Publications Sharon Belenzon 1, Andrea Patacconi 2 1 Fuqua School of Business, Duke University 2 University of Aberdeen Business School Abstract How to measure R&D capabilities is a key concern for policymakers and practitioners alike. This paper examines the changing role of patents and firm publications as a signal of scientific and technical competence. We argue that recent patent reforms might have reduced the informational value of USPTO patents relative to EPO patents. Moreover, the global trend toward stronger patent enforcement may have raised the cost of freely disclosing research results in the scientific domain. Using data from more than 33,000 mergers and acquisitions deals between 1985 and, we provide support for these conjectures. Controlling for patent citations or restricting attention to specialized or within-country deals reduces as expected the relative importance of EPO patents. Our results do not appear to be driven by changes in the importance of patenting in multiple jurisdictions. JEL Classification: O31, O32, O16 Keywords: firm value, patents, scientific publications 1. Introduction In fast-paced industries where intellectual property is key, patents are increasingly becoming a currency that allows technology transactions to take place and that ultimately determines firm value (Jaffe and Lerner, 2004). Patents are best positioned to fulfill the role of medium of exchange in technology transactions for two main reasons. First, by giving patent holders the right to exclude others from using an invention, patents protect inventors from the risk of expropriation. This benefits especially small firms which often have no other way to protect their ideas than to rely on the legal system. Second, patents disclose information a detailed description of the invention which is often important to mitigate problems of asymmetric information. Uncertainty may in fact lead to higher perceived risk and thus lower valuation for the technology under exam (Arrow, 1962; Akerlof, 1970; Ritter and Welch, 2002). 1

Recent patent and administrative reforms have most likely affected both the legal and the informational value of patents in technology markets. 1 Starting from the early 1980 s, patent protection has been considerably strengthened in most OECD countries. Important steps in this direction have been the creation in 1982 of a pro-patent Court of Appeals in the U.S., the subsequent upward harmonization of intellectual property protection standards via international agreements, and the extension of patentability to new subject matter such as biotechnology and software (in the U.S., also business methods). Unfortunately, there is evidence that, as patents gained importance, examination standards began to decline. This decline has been attributed at least in part to the huge increase in patent applications in the last thirty years and is believed by some to represent a real danger to the innovation process. Indeed, if standards decline so much that applicants can get patents for obvious or existing ideas, then patents might be used to harass, seek compensation from, or even shut down legitimate businesses (Jaffe and Lerner, 2004). Firms may even refuse to enter some technology sectors for fear of litigation, with possible negative long-run implications for the pace of technological progress (Lerner, 1995). While the strengthening of patent protection is arguably a global phenomenon (Lerner, 2001), the deterioration of examination standards and the consequent decline in patent quality appear to be concentrated mainly in the U.S. Reasons include the low patenting costs at the USPTO, which encourage marginal applications, and motivational and psychological factors that give USPTO examiners strong incentives to accept rather than reject applications (Lemley, 2001; Jaffe and Lerner, 2004; Guellec and van Pottelsberghe de la Potterie, ). Motivated by these considerations, the present paper explores how the correlation between firm value and multiple indicators of innovative activity (EPO and USPTO patents and firm publications) has changed over time. 2 Our analysis focuses on more than 33,000 mergers and acquisitions deals that took place between 1985 and. We hypothesize that a decline in the average quality of USPTO patents (relative to EPO patents) might have reduced the value of USPTO patents as a signal of a firm's ability to perform R&D. Previous research has shown that patents 1 For a discussion of the evolution of the patent system in the U.S., see Kortum and Lerner (1999), Jaffe (2000) and Gallini (2002), among others. The books by Jaffe and Lerner (2004) (from a U.S. perspective) and Guellec and van Pottelsberghe de la Potterie () (from a European perspective) provide detailed analyses of the causes and consequences of recent patent reforms. 2 Firm publications are articles published in "hard science" journals where one of the co-authors is a company employee. These articles may provide a useful signal of the technical and scientific competence of the sponsoring firm (see, e.g., Henderson and Cockburn, 1994; Cockburn and Henderson, 1998; Belenzon and Patacconi, 2009). 2

provide a useful measure of a firm's inventive efforts and stock market value (e.g., Griliches, 1990; Bloom and Van Reenen, 2002; Belenzon, 2009), especially when weighted by citations (Trajtenberg, 1990). Our main result is that over time EPO patents have become the dominant indicator of innovative activity, while USPTO patents have no effect on firm value in the late sample period. This is true also when patents are weighted by citations, although citations-weighting generally increases the importance of USPTO patents. 3 Moreover, consistent with the view that the strengthening of patent protection has increased the cost of disclosing research in the scientific domain, we find that, controlling for patents, the correlation between firm value and publications declines over time and is insignificant toward the end of the sample period. The robustness of these results is probed in several ways. An important finding is that EPO patents appear to be most closely associated with firm value when the acquiring and target firms belong to different industries or countries. We interpret this piece of evidence as suggesting that acquiring firms may differ in their ability to evaluate potential targets and that the uncertainty associated with the target s technology is large especially in cross-industry and cross-country deals. If this is true, in fact, then EPO patents may be perceived as less risky (and thus more valuable) than USPTO patents, especially in that type of transactions (an EPO certification effect). We also find substantial variation across industries. Our results hold most strongly in drugs and chemicals, where patents are generally very important, but much less so in electronics and communications, where patenting is not strongly related with value in any of the sample periods. 4 Lastly, our findings do not appear to be driven by changes in the importance of patenting in multiple jurisdictions. The remainder of the paper is organized as follows. The next section discusses recent trends in patent policy and spells out our main predictions. Section 3 describes the data. The econometric specification is discussed in Section 4 where we also highlight the main econometric problems we face. Section 5 reports the empirical results. Section 6 concludes. 3 This is not surprising since citations proxy for patent quality and patent quality may at least to some extent be observed by the acquiring firm (Hall et al., 2005). In that case, we would expect the coefficient on the less reliable indicator to rise, relative to the coefficient on the more reliable indicator. 4 Information technology is another interesting industry. There we find that EPO patents have no effect on value, while USPTO patents remain the dominant technology indicator throughout the whole estimation sample. This supports the common wisdom that the U.S. enjoys a strong technological leadership in information technology. 3

2. Background and predictions 2.1. The changing institutional environment 5 There is widespread consensus on the idea that from the early 1980's, various patent and administrative reforms have not only led to a strengthening the legal value of patents, but also to a decline in patent examination standards. In the U.S., a key step in the direction toward stronger IP protection was the creation in 1982 of a specialized Court of Appeals of the Federal Circuit. Although the stated objective of Congress was to provide uniformity in patent litigation cases, the court soon demonstrated a pro-patent disposition (Merges, 1992; Mazzoleni and Nelson, 1998). The court has been found, for instance, to uphold in appeal a larger proportion of decisions favorable to the patent holder, to reverse a larger proportion of decisions unfavorable to patent holder, and to substantially increase the rate of preliminary injunctions, compared to the previous system (Dunner 1988; Gallini, 2002; Lanjouw and Lerner, 2001). The definition of patentable subject matter was also broadened to include emerging new fields such as biotechnology and software, and patent duration was lengthened. Several countries followed the U.S. example in a process of upward harmonization of intellectual property protection standards (Lerner, 2001; Guellec and van Pottelsberghe de la Potterie, ). 6 The strengthening of the legal value of patents has given firms powerful incentives to seek patent protection (Kortum and Lerner, 1999; Hall and Ziedonis). Patent applications nearly tripled between 1980 and 2000 in the U.S., with Europe and Japan following suit. This huge increase in patent applications most likely contributed to the decline of examination standards that has been the focus of much debate recently. Jaffe and Lerner (2004) provide many examples of patents that should never have been awarded because they were based on obvious or not-novel ideas. These include a patent on a "Sealed Crustless Sandwich", option pricing formulas, and, perhaps more controversially, one-click business methods. Most importantly for our purposes, the decline in examination standards and the consequent fall in patent quality appear to be concentrated mainly in the U.S. Three possible explanations have been put forward for this. First, the USPTO is severely understaffed relative to other major patent offices. Guellec and van Pottelsberghe de la Potterie () estimate that "both the incoming 5 This subsection is based mainly on Gallini (2002), Jaffe and Lerner (2004) and Guellec and van Pottelsberghe de la Potterie (). 6 The 1994 trade-related intellectual property rights (TRIPS) agreement is a good case in point. 4

workload of examiners (number of claims filed per examiner) and their output (number of claims granted per examiner) is three to four times higher at the USPTO than at the EPO" (p. 201). Thus USPTO examiners spend considerably less time than EPO examiners on each claim. Guellec and van Pottelsberghe de la Potterie conclude that decisions at the EPO tend to be based on a more careful assessment of the prior art than at the USPTO, which might explain the much higher rejection rates in Europe. 7 Another possible explanation for the decline of U.S. examination standards relates to the pressures that USPTO examiners face to accept rather than reject applications (Lemley, 2001; Jaffe and Lerner, 2004). Many observers have pointed out that by transforming the USPTO in a user-fee funded entity, the 1990 Omnibus Act might have promoted a customer-friendly mentality. Jaffe and Lerner (2004) provide several anecdotes illustrating the dysfunctional effect of such a mentality. More concretely, Jaffe and Lerner also emphasize that USPTO examiners are rewarded according to how fast they process applications. However, since applicants can modify and appeal initially rejected patents, rejections tend to be very time-consuming. As a result, examiners have strong incentives to accept rather than reject applications, no matter how weak these applications might be. Last but not least, patenting costs are much lower in the U.S. than in Europe or Japan. van Pottelsberghe de la Potterie and François (2009) estimate that a EPO patent that is renewed for 20 (10) years in 13 member states costs almost 9 (4.6) times as much as a USPTO patent. An obvious effect of such low patenting costs is to encourage low-quality applications. This is likely to worsen the patent quality distribution and increase its variance. Perverse feedback effects may also result, with low patenting costs exacerbating existing workload problems. 8 7 See Guellec and van Pottelsberghe de la Potterie (, pp. 202-203) and the references therein. Jaffe and Lerner (2004) also show that successful USPTO applications have risen in the period between 1987 and 1998 twice as much as `important' inventions that were granted in all three of the world's major patent jurisdictions (the USPTO, EPO, and JPO). They argue that this finding is "hard to explain in any manner other than declining standards in the U.S. PTO, producing an ever-growing proportion of U.S. patents the patent-holders themselves did not think merited patenting elsewhere" (p. 143). Finally, Gallini (2002) argues that if the time series of Kortum and Lerner (1999) is extended to include the second half of the 1990's as well, Kortum and Lerner's conclusion that the U.S. did not become an increasingly attractive destination for foreign inventors should be tempered, thus "lending some support for the friendly court hypothesis" (p. 138). 8 How fee structures should be changed to improve the quantity-quality tradeoff is an important topic in current debates, especially after the USPTO raised its fees in 2004. See Guellec and van Pottelsberghe de la Potterie (, pp. 212-213) for more on this. 5

2.2. The informational role of patents Patent reforms such as those discussed above can affect the competitiveness of nations because technology markets typically under-provide incentives for innovation. For instance, at least since Arrow (1962) it has been clear that inventors may lack sufficient incentives to disclose their inventions and thus innovate for fear that buyers may steal their ideas. Patents offer a way to get around this problem by providing legal protection against expropriation (Arora et al., 2001). Moreover, because they are publicly available, they can provide information to prospective investors. The signaling value of patents often goes beyond the intrinsic value of the invention as it helps evaluate a firm s knowledge base. These considerations are relevant especially for small firms since they often have few other assets except their intellectual capital. Consistent with this view, survey evidence indicates that especially small firms patent for reputational reasons (Cohen et al., 2000) and to attract venture capital (Hall and Ziedonis, 2001). 9 The above remarks make clear why many firms, and especially small ones, may have strong incentives to patent. Thus, it is not surprising that patent statistics have been found to be good indicators of a firm s inventive efforts and value (Griliches, 1990). On the other hand, our previous discussion suggests that average patent quality might have dropped, especially in the U.S. This leads to our main empirical hypothesis, namely that over time EPO patents might have become a more reliable indicator of innovative activity, compared to USPTO patents. Moreover, this trend should be more pronounced for small firms. H1. Over time, firm value should become more strongly correlated with EPO patents than USPTO patents, especially for small firms. Despite their usefulness, a serious problem with patents is that they vary tremendously in their quality and economic significance. Prospective investors, in particular, may be expected to collect much more information about the target company s technology than simply patent counts. The most natural albeit imperfect way to proxy for investors' private information is to use patent 9 For instance, Cohen et al. find that smaller firms are significantly (at.01 confidence level) more likely to report the motive "to enhance the reputation of the firm or its employees" as a reason to patent. They argue that this could be rationalized "by the need of smaller firms in some high technology industries to hold patents in order to acquire financing or alliance partners" (p. 24). 6

citations. 10 Since the landmark work of Trajtenberg (1990), citation-weighted patent indexes have been shown to be strongly correlated with the social value of innovations (Trajtenberg, 1990), peer evaluation of their technical importance (Albert et al., 1991), renewal decisions (Thomas, 1999; Harhoff et al., 1999), and firm value (Deng et al., 1999; Bloom and Van Reenen, 2002; Hirschey and Richardson, 2004; Hall et al., 2005; Belenzon, 2009). Particularly interesting is Hall et al.'s (2005) finding that investors are able to accurately forecast the expected value of patented inventions, as it is later confirmed by future citations. This finding suggests that once we control for the intrinsic importance of the invention (as measured by patent citations), the predictive power of patents in firm value regressions should increase. This is likely to be true especially for USPTO patents since their quality should be very heterogeneous. By contrast, if the EPO label already provides quality certification, then citations might only marginally affect the informational value of European patents. These contrasting patterns should become more pronounced over time if USPTO patents have become more heterogeneous in terms of quality. H2. When patents are weighted by citations, firm value should become more strongly correlated with USPTO patents and less with EPO patents, especially in the late sample period. Finally, although all firms might attempt to gather as much information as possible, substantial heterogeneity in their ability to evaluate potential targets might still remain. It has long been recognized that learning about new technologies is a costly activity that requires substantial investment in absorptive capacity (Cohen and Levinthal, 1989; Gambardella, 1995). This is true also when evaluating potential alliance partners and acquisition targets (Liebeskind et al., 1996; Arora and Gambardella, 1990). A natural way to control for the amount of private information acquiring firms possess is to distinguish between deals that involve firms operating in the same industry, and deals that do not. The idea is of course that firms operating in the same industry might possess specialized knowledge that helps them to evaluate technology in that industry. Similarly, investors involved in within-country acquisitions may possess superior information about the target firm s technology thanks to interactions with local supplies, banks, and business partners. Based on these ideas, we conjecture that firms involved in between-industry or -country deals (and thus likely 10 An alternative way would be to estimate the private value of patents using annual renewal fees as in Schankerman and Pakes (1986). 7

to have poor information about a target firm s technology) might perceive USPTO patents as risky, at least compared to EPO patents. This might result in a discount for USPTO patents, effectively paying a premium for the certification provided by the EPO. H3. EPO patents should be more strongly related with firm value in between-industry and betweencountry acquisitions. 2.2. The informational role of firm publications Another way in which firms can demonstrate their scientific and technical competence is by publishing in academic journals (Henderson and Cockburn, 1994; Cockburn and Henderson, 1998; Belenzon and Patacconi, 2009). This could be because scientific publications capture at a high level of abstraction the scientific base of patented inventions. Many scientific ideas are in fact first published in academic journals and then patented (Murray, 2002; Murray and Stern, 2006). Alternatively, scientific publications may proxy for investments in basic research carried out to evaluate, monitor, and exploit research conducted in the public sector (Cohen and Levinthal, 1989; Cockburn and Henderson, 1998). Publications can benefit the sponsoring firm in several ways. One is to improve corporate image, or prestige (Hicks, 1995). Prestige, in turn, may help certify the quality of a firm's products to sophisticated buyers (Nelson, 1990; Azoulay, 2002), make it easier to raise external funds or win government contracts (Audretsch and Stephan, 1996; Higgins et al., 2008; Lichtenberg, 1986, 1988), and attract talented employees (Nelson, 1990; Audretsch and Stephan, 1996; Stern, 2004). Publications may also help company employees to remain better connected to the wider scientific community, thus fostering organizational learning (Rosenberg, 1990). These benefits, however, must be weighed against the potential cost. The most important drawback of any policy of disclosure is arguably that competitors may use the disclosed information to gain competitive advantage (Bhattacharya and Ritter, 1983). It is precisely for this reason that many companies systematically screen their publications (Koenig, 1983). We conjecture that the recent trend toward stronger patent protection might have increased the cost of freely disclosing information since competitors may use this information to secure patents themselves. 8

H4. Over time and controlling for patents, the correlation between firm value and publications should decline. On the other hand, strong patent protection may mitigate the risk of expropriation and thus increase the propensity to publish if the same piece of knowledge is both published and patented (Gambardella, 1995). Belenzon and Patacconi (2009) provide evidence consistent with the view that patent-publication pairs are more prevalent in BioMed and Chemicals. To control for this possibility, we will distinguish between different technology fields in the empirical part. 3. Data Our paper combines data from several sources: (i) M&A data from Thomson SDC Platinum, (ii) information on patents from the United States Patent and Trademark Office (USPTO), (iii) information on patents from the European Patent Office (EPO), and (iv) scientific publications from Thomson Web of Knowledge. Data on M&A deals is from Thomson SDC Platinum. We drop deals that do not report information on value, net total assets, and acquired stakes. This leaves us with 33,920 deals. We match these firms to our patent and publication datasets. Prior to the acquisition year, 5,471 firms have at least one patent in the USPTO, 4,108 firms have at least one patent in the EPO, and 1,557 firms have at least one scientific publication. Table 1 summarizes main descriptive statistics for our sample. Firm value is computed as the ratio between transaction value and acquired stakes. The average firm is valued at $182 million, has $91 million in total assets, and generates $137 million in annual sales and $23 million in profits. For each innovating firm we construct the patent and publication stocks prior to the acquisition year. Stocks are computed using the perpetual inventory method with a depreciation rate of 15 percent. To control for quality of patents and publications, we construct citations-weighted stocks. These are constructed by weighting each patent and publication by the number of citations received (excluding self-citations) divided by the average number of citations received by all other patents or publications in the same year. The average innovating firm has a stock of 12 USPTO patents (a median of 2), 9 EPO patents, and 3.6 publications. Insert Table 1 here 9

Table 2 examines the distribution of innovating firms across years. Among innovating firms, 89.2% have at least one USPTO patent, 67.0% have at least one EPO patent, and about 20% publish at least one scientific article. Over time, the share of firms that patent in the USPTO remains quite stable while the share of firms that patent in the EPO increases substantially from a low of 50% in 1985 to around 75% in. The share of publishing firms, on the other hand, declines from a high of nearly 30% in 1985 to around to 11% in. 85% of the firms that have at least one patent in the EPO also patent in the USPTO. This percentage is higher in the earlier periods of our sample. In the period 1990, 87% of the firms that patent in the EPO also patent in the USPTO, whereas in the post-2000 period only 77% of the firms that patent in the EPO also patent in the USPTO. A similar pattern holds for publishing firms. In 1990, 20% of the patenting firms also publish at least one scientific article, compared to 15% in the post-2000 period. The share of non-american acquisitions changes substantially over time. In the whole sample just over 40% of the acquisitions involve an American target, yet this figures falls from around 50% in 1990 to less than 20% in. Another important feature is the extent to which firms patent in more than one jurisdiction. 10% of all firms and 53% of the innovating firms patent in both USPTO and EPO. These figures remain stable over time (54% of the innovating firms patent in both jurisdictions in 1990 as compared to 55% in ). Intangible assets do not appear to become more important for firm value over time. The average value of Tobin's Q (value over total assets) pre- is 9.2 (a median of 1.9), and 9.0 (a median of 2.1) in the post- period. Insert Tables 2 and 3 here 4. Econometric analysis and potential biases We estimate the following specification for the value of target firms, for the whole sample and for different time periods. 11 lnvalueit = β 0 ln Assetsit + β1 ln(1 + PatUSit 1) + β 2 ln(1 + PatEUit 1) + β3 ln(1 + Pubit 1) + τ t + ηi + ε i 11 Lerner (1995) estimates a similar specification. 10

where Value it is the value of the target firm i at the acquisition completion year t and is computed as transaction value over acquired equity, Assets it is total net assets, and PatUS it 1, PatEU it 1 and Pub it 1 are respectively the target firm s USPTO patent stock, EPO patent stock, and publication stock prior to the deal completion year. τ t and η i denote complete sets of dummies for the transaction completion years and two digit target industry SIC codes. ε i is an iid error term. We proxy patent quality using citations. Each patent is weighted by the ratio of the number of citations it receives (excluding self-citations) and the average number of citations received by all patents that were at the same year. We also control for the quality of publications using the number of citations they receive. Our main interest is the way the coefficients β 1, β 2 and β 3 change over time. The key problem we face is that the firm sample changes from one year to another. This leaves our results sensitive to unobserved heterogeneity bias. A concern might be, for instance, that firms that choose to patent in the EPO in the late sample period may be, on average, of higher (unobserved) quality than firms that patent in the EPO in the beginning of the sample period. This means that later in the sample period the correlation between EPO patents and unobserved firm quality is positive, causing an upward bias in β 2. For instance, over time the payoff from patenting in the EPO might have increased (say because of globalization ), but only for firms with specific skills (a talented management, better distribution channels, etc.). While some of these high-quality firms may have chosen not to patent in the EPO in the earlier periods of our sample, most of them may do that in later periods. Alternatively, globalization might have made "important" patents even more valuable over time (important patents being those granted in more than one major jurisdiction). This effect might be captured by EPO patents since most firms in our sample patent in the USPTO. Clearly both stories may explain the increasing correlation between firm value and EPO patents over time. However, it is not clear how they could also explain the fading impact of USPTO patents on firm value that we observe. Another potential source of sample selection is industry composition. For instance, in the later period of the sample we are likely to observe more biotechnology and information technology acquisitions than in earlier periods. If the impact of patents on firm valuation varies across industries, then our key estimates could be biased. To deal with this concern, we report our main estimation results separately for the main innovative industries in our sample. Finally, a basic premise of our analysis is that patent citations are a useful proxy for patent quality. However, because the citations a patent receives accumulate over time, patent quality is 11

likely to be measured with error toward the end of the estimation sample. If EPO certification provides a measure of quality because few "bad" patents are granted in the EPO, then β 2 is likely to be higher (and β 1 lower) in the last years of the sample period even when patents are weighted by citations, since USPTO patents might still be extremely noisy indicators of innovative activity. Given our data, there is no reliable way to control for firm unobserved heterogeneity, and we are not likely to find a better measure of patent quality than patent citations. Our strategy, thus, is to examine different sub-samples and show that the EPO effect is particularly strong for firms and transactions with specific sets of characteristics. For example, we distinguish between within- and between-industry transactions. EPO certification is more likely to matter in between-industry transactions, since potential investors there arguably face substantial uncertainty about the target firm's technology. Finding evidence consistent with certification in between-industry transactions, but not within, is hard to explain by EPO selection or the citations-truncation argument alone. 5. Findings 5.1. Horse-race estimation Table 4 reports the estimation results for the complete sample period. We start with the unweighted patent measures. Columns 1-3 include separately each of the innovation indicators. are strongly related to firm value. The coefficients on USPTO patents, EPO patents, and firm publications are 0.121 (a standard error of 0.008), 0.158 (a standard error of 0.010), and 0.186 (a standard error of 0.029), respectively. Columns 4 to 6 provide horse-race estimations where each of the innovation indicators is matched with another. Column 7 reports the results when all three indicators are included in a single regression. While all three coefficients are highly significant, EPO patents appear to have the strongest effect on value with a coefficient of 0.106, as compared to coefficients of 0.049 and 0.069 on USPTO and publication stocks. This pattern is even more evident when we focus on innovating firms (i.e., those with at least one USPTO or EPO patent, or at least one scientific publication). In that regression, in fact, the value-epo relation remains strong, while the value-uspto relation is no longer significant (see column 8). Columns 9-14 control for the quality of patents and publications using the number of citations they receive. Column 9 includes USPTO and EPO patents. Both coefficients are positive and highly significant (0.081 with a standard error of 0.011, and 0.093 with a standard error of 12

0.014, for USPTO and EPO patents respectively). Column 10 includes USPTO patents and publications. Interestingly, the coefficient on USPTO patents is effectively identical to the unweighted coefficient (0.123 versus 0.121). On the other hand, the coefficient on the weighted publications stock is substantially lower than the unweighted one. This finding is consistent with the idea that publications might capture some of the patent quality heterogeneity that is observable only to firms. Column 12 includes all three indicators. Compared to the unweighted regressions, the coefficient on USPTO patents is much closer in magnitude to the coefficient on EPO patents. The same pattern of results continues to hold when including only innovating firms, or high-tech transactions. Finding that USPTO patents matter more for value when controlling for citations suggests that citations indeed capture information that firms possess and use when bidding for targets. Next, we explore how the value-innovation relation varies over time. Insert Table 4 here 5.2. Variation over time Table 5 reports the estimation results when the sample is split into four time periods. Several striking findings emerge. Looking at columns 2, 4, 6 and 8 (unweighted regressions, all innovation indicators included), it is clear that the importance of EPO patents increases over time, with β 2 rising from a low of 0.066 in 1990 to a high of 0.159 in. By contrast, the coefficient on USPTO patents shows no clear trend: β 1 first rises from 0.046 in 1990 to 0.077 in 1996-2000, and then sharply declines to a low of 0.013 in. 12 Also, the decline in the importance of USPTO patents seems to be driven by the inclusion of EPO patents. In the first sample period, controlling for EPO patents reduces the coefficient on USPTO patents from 0.076 to 0.046 (both coefficients are highly significant); in the last sample period, the coefficient drops from 0.119 to essentially zero. These findings are broadly consistent with Hypothesis 1: over time, EPO patents become the dominant innovation indicator, while USPTO patents are not informative about firm value in the late sample period. Columns 9-16 control for patent and publication quality. The same pattern of results continues to hold, with the coefficient on EPO patents nearly tripling in magnitude over the sample 12 A possible explanation for such a pattern is the dotcom bubble, which reached its apex in 2000. It is interesting to note that the coefficient on EPO patents appears not to have been affected by the bursting of the bubble in 2000. 13

period. A key comparison is that between weighted and unweighted regressions. Controlling for citations raises the coefficient on USPTO patents, while the coefficient on EPO patents generally drops. This is consistent with Hypothesis 2: citations raise the coefficient on the low-quality indicator much more than the coefficient on the more precise indicator. However, we also find that the gap between β 1 and β 2 does not seem to narrow toward the end of the sample period. 13 This is inconsistent with Hypothesis 2, but it could be explained by the limited power of citations in the lateperiod sample (a truncation problem). Finally, Table 5 provides evidence on the changing role of scientific publications as a signal of technical and scientific competence. We argued in Section 2 that the global trend toward stronger patent protection might have increased the cost of freely disclosing research results. Consistent with this view, the value-publications relation is strong only at the beginning of the sample period. Controlling for USPTO and EPO patents, the coefficient on publications stock is 0.115 and highly significant (a standard error of 0.038). Estimating the same specification for the period yields a completely different result: the coefficient is effectively zero (a coefficient of -0.001 and a standard error of 0.041). Interestingly, the decline in the value-publications relation does not seem to be driven by the strengthening of the value-epo relation. The coefficient on firm publications becomes insignificant in 1996-2000 also when not controlling for EPO patents (column 5). Insert Table 5 here 5.3. Target firm size Survey evidence indicates that especially small firms patent for signaling and reputational reasons (Cohen et al., 2000; Hall and Ziedonis, 2001). Table 6 studies whether our results are driven by differences between large and small target firms (we measure firm size by sales). Columns 1-8 examine firms in the fourth quartile of target firm size (the largest ones). On average, these firms have $409 million in annual sales and a value of $367.4 million. Neither USPTO nor EPO patents seem to matter particularly for these firms. By contrast, the pattern of results that emerges by looking at the firms in the lowest sales quartile (columns 9-16) is completely different. These firms have 13 Despite this, note that the coefficient on USPTO patents, while not significant in column 8, is marginally significant in column 16. 14

average annual sales of $6.3 million and average firm value of $44.8 million. At the beginning of the sample period, both USPTO and EPO patents are strongly correlated with firm value. For instance, in column 14 (weighted regression, period), the coefficients on USPTO and EPO patents are respectively 0.210 and 0.175, with standard errors of 0.045 and 0.056. By contrast, in (column 16), the coefficient on USPTO patents drops to 0.111 (with standard error of 0.089) and is no longer significant. EPO patents instead remain strongly associated with value (a coefficient of 0.376 with standard error of 0.100). Thus, consistent with our expectations, Hypothesis 1 appears to be confirmed especially for small firms. Finally (and perhaps surprisingly), we find no large difference between the coefficients on scientific publications for large and small firms. A possible explanation could be that although signaling is more important for small firms, many publications by start-ups are motivated by the desire of the scientist/entrepreneur to foster his or her academic career. As a result, these publications might be only loosely related to the activities of the publishing firm. 14 Insert Table 6 here 5.4. Within- and between-industry acquisitions This subsection explores the possibility that firms might be heterogeneous in their ability to evaluate acquisition targets. We distinguish between within- and between-industry acquisitions using the twodigit SIC codes of the acquiring and target firms. We expect acquiring firms to possess more detailed information about the target firm's technology in within-industry acquisitions due to their superior knowledge of the industry. Thus, controlling for citations, the EPO label should not add significant information on firm value in the case of specialized (within-industry) acquisitions (a small EPO certification effect). By contrast, firms that expand through diversification may not possess a deep understanding of the target's technological environment. As a result, USPTO patents might be perceived as risky, and thus discounted, relative to EPO patents (a strong EPO certification effect). Table 6 reports the estimation results. Columns 1-8 examine how the value-patent relation varies over time for within-industry acquisitions. When citations are not controlled for, the coefficient on USPTO patents is effectively zero in the later sample period (a coefficient of 0.023 14 Alternatively, publications might be the sign of poor organizational practices (an excessive focus on basic research), which may affect especially young, small firms. See also Belenzon and Patacconi (2009). 15

and a standard error of 0.041), while EPO patents appear to be strongly related to firm value (a coefficient of 0.110 and a standard error of 0.048). This pattern of results changes when citations are accounted for. Column 8 exhibits a significant positive correlation between firm value and USPTO patents (a coefficient of 0.077 and a standard error of 0.037). By contrast, the coefficient on EPO patents, albeit similar in magnitude (0.072), is insignificant (the standard error is 0.045). Thus the EPO certification effect appears to be small in within-industry acquisitions, at least when citations are controlled for. Columns 9-16 focus on between-industry acquisitions. A different pattern emerges. In the later sample period, USPTO patents have no effect on firm value in both weighted and unweighted regressions (coefficients of -0.003 and 0.017 for the unweighted and weighed regressions, respectively). The coefficient on EPO patents instead increases over time and is large especially at the end of the sample period. These findings suggest that the EPO might provide quality certification when the acquiring firm has limited information about the target firm s technology. Indeed, if firms cannot assess the quality of USPTO patents, then they might heavily discount them. As a result, EPO patents would become the dominant indicator of the target firm's technology in between-industry acquisitions (Hypothesis 3). Insert Table 7 here 5.5. Within- and between-country acquisitions The quality of information about the technology of a potential target may also be enhanced by physical proximity. Motivated by this observation, Table 8 examines the sensitivity of our results to cross-country transactions and to the nationality of the target company. In all regressions we control for patent and publication quality using citations. Columns 1-8 distinguish between transactions where the acquiring and target firms belong to the same country or to different countries. The main finding is that the importance of USPTO patents remains high later in the sample period for within-country deals (a coefficient of 0.077 and a standard error of 0.028), but not for cross-country deals (a coefficient of -0.039 with a standard error of 0.055). On the other hand, EPO patents are highly important both for within-country and betweencountry deals. This pattern is again consistent with Hypothesis 3: USPTO patents get heavily 16

discounted relative to EPO patents when the acquiring firm has limited information about the target firm. Columns 9-16 investigate the robustness of the results to the nationality of target firms. As shown in columns 13-16, our key findings are driven by non-american firms. It appears that the EPO premium exists near the end of the sample period especially for cross-country transactions involving non-american firms. This result is interesting because it suggests that the value-epo relation is not likely to be driven by American companies that over time became more global, thus putting more emphasis on EU intellectual property protection. Insert Table 8 here 5.6. Industry variation Patent protection varies considerably across industries. This variation can be driven by different legal patentability requirement or by differences in the nature of technology. To examine how the value-patents relation varies across industries, we separately estimate our main specification for the following industries: drugs and chemicals, telecommunications, computers and electronics, and information technologies. Drugs and chemicals are examples of industries where patent protection is extremely important. The fundamental nature of innovation and its long life cycle are some of the reasons. Telecommunications and computers and electronics are industries where innovation tends to require the integration of many different pieces of knowledge and is often incremental, with followup developments often rapidly displacing existing products. Information technology is a particularly interesting industry because patentability requirements vary across the United States and Europe. While in the U.S. advances in IT can be easily patented, the patentability requirements for such inventions are much stricter in Europe, regardless of the quality of the invention. The estimation results are summarized in Table 9. Our previous findings are confirmed in drugs and chemicals. In these industries, EPO patents become more dominant over time, while the effect of USPTO patents disappears. In telecommunications, patents do not seem to be an important determinant of firm value. In computers and electronics and information technologies, USPTO patents remain the dominant technology indicator also in the late sample period. A possible explanation is that the United States is the technological leader in these industries, with the USPTO 17

arguably granting the vast majority of the technical advances in these industries, typically by American firms. Finally, while the correlation between scientific publications and firm value declines sharply over time in all industry classes, substantial heterogeneity remains. The coefficients on publications are often negative in telecommunications and computer and electronics, reflecting perhaps growing skepticism toward basic research. However, they always remain positive (albeit sometimes insignificant) in drugs and chemicals, where absorptive capacity is deemed to be particularly important (Cohen and Levinthal, 1989; Gambardella, 1995; Cockburn and Henderson, 1998) and published knowledge can be effectively protected by patents (Gambardella, 1995). Insert Table 9 here 5.7. Patenting in multiple jurisdictions This subsection investigates the possibility that our results may be driven by changes in the importance of patenting in multiple jurisdictions. We conjecture that if adopting a global approach to business becomes more important over time, then firms that patent both in the United States and Europe might become increasingly attractive as targets over time, compared to firms that only patent in one jurisdiction the globalization hypothesis. This might help explain our findings because most innovating firms in our sample have patents in the USPTO and the main source of variation in dual jurisdiction comes from EPO patenting. 15 Table 8 examines whether patenting in both the USPTO and the EPO has become more important over time. We create a globalization dummy which receives the value of one for firms that patent in both EPO and USPTO, and zero for all other firms. Throughout we control for patent and publication quality. The results are not consistent with the globalization hypothesis. The coefficient on the globalization dummy is significant in the first half of the sample period but not in the second. The same pattern of results holds when we look at American and non-american firms separately (columns 5-12), and at firms that have at least one patent in the EPO (columns 13-16). A possible explanation for this finding is that if operating across countries becomes easier over time, then more 15 Indeed, Table 2 shows that the share of innovating firms that patent in the EPO rises from about 50% at the beginning of the sample to around 75% towards the end of the sample. 18

low-quality firms might apply for patents in multiple jurisdictions. Thus the average quality of global firms may decline over time. At any rate, it is clear that even when controlling for patenting in multiple jurisdictions, our main results are robust. Insert Table 10 here 6. Conclusions Motivated by recent debates on the evolution of the patent system in the United States and Europe, and in particular by a book by Jaffe and Lerner (2004), the present paper examines the effect of multiple indicators of inventive activity on firm value in a dataset of nearly 34,000 mergers and acquisitions deals between 1985 and. Consistent with the hypothesis that examination standards and hence patent quality might have declined especially in the U.S., we find that, even controlling for citations, EPO patents become over time the dominant technology indicator, while USPTO patents have little effect on firm value in the last few years of the sample period. The results are particularly strong in between-industry and between-country deals, suggesting that the EPO may play a certification role in these transactions. Lastly, consistent with the idea that stronger patent protection might have increased the cost of freely disclosing research, we find that controlling for patents, the effect of scientific publications on firm value declines over time and is insignificant in the late sample period. An important limitation of the present analysis is that since the firm sample changes over time, our results might be sensitive to unobserved heterogeneity bias. In particular, we cannot rule out the possibility that EPO patents have become more important over time because more and more high-quality firms have chosen to patent in the EPO towards the end of the sample period. While it is not clear how this explanation could rationalize the industry and country variation we document, an important direction for future research would certainly be to gather additional data to mitigate such concerns. Acknowledgements: We thank Hadar Gafni for excellent research assistance. We also thank Ashish Arora, Tim Barmby and Julian Williams for helpful comments. remaining errors are our own. 19

BIBLIOGRAPHY Akerlof, G.A., 1970. The Market for "Lemons": Quality Uncertainty and the Market Mechanism. Quarterly Journal of Economics, Vol. 84, pp. 488-500. Albert, M.B., D. Avery, F. Narin and P. Mcister, 1991. Direct validation of citation counts as indicators of industrially important patents. Research Policy, 20, 251-259. Arora, A., A. Fosfuri and A. Gambardella, 2001. Markets for technology: Economics of Innovation and Corporate Strategy, MIT Press, Cambridge, MA. Arora, A. and A. Gambardella, 1990. Complementarity and External Linkages: The Strategies of the Large Firms in Biotechnology. Journal of Industrial Economics, Vol. 38, pp. 361-379. Arrow, K., 1962. Economic Welfare and the ocation of Resources of Inventions. In R.R. Nelson, ed., The Rate and Direction of Inventive Activity, Princeton, NJ: Princeton University Press. Audretsch, D.B. and P.E. Stephan, 1996. Company-Scientist Locational Links: The Case of Biotechnology. American Economic Review, Vol. 86, pp. 641-652. Azoulay, P., 2002. Do Pharmaceutical Sales Respond to Scientific Evidence? Journal of Economics and Management Strategy, Vol. 11, pp. 551-594. Bhattacharya, S. and J.R. Ritter, 1983. Innovation and Communication: Signalling with Partial Disclosure. Review of Economic Studies, Vol. 50, pp. 331-46. Belenzon, S., 2009. Cumulative Innovation and Market Value: Evidence from Patent Citations. Working Paper. Belenzon, S. and A. Patacconi, 2009. Performance, Firm Size, and the Basicness of Research. Working Paper. Bloom, N. and J. Van Reenen, 2002. Patents, Real Options, and Firm Performance. Economic Journal, Vol. 112, pp. 97-116. Cockburn, I.M. and R. Henderson, 1998. Absorptive Capacity, Coauthoring Behavior, and the Organization of Research in Drug Discovery, Journal of Industrial Economics, Vol. 46, pp.157-182. Cohen, W.M. and D.A. Levinthal, 1990. Innovation and Learning: The Two Faces of R&D, Economic Journal, Vol. 99, pp. 569-596. Cohen, W.M., R.R. Nelson and J.P. Walsh, 2000. Protecting their Intellectual Assets: Appropriability Conditions and Why U.S. Manufacturing Firms Patent (or Not). NBER Working Paper No. 7552. 20