LITIGANT REPUTATION AND THE ECONOMICS OF PATENT LITIGATION IN THE UNITED STATES:

Transcription

1 LITIGANT REPUTATION AND THE ECONOMICS OF PATENT LITIGATION IN THE UNITED STATES: JAVAD ESKANDARIKHOEE * UNIVERSITY OF DELAWARE JANUARY 2014 JOB MARKET PAPER ABSTRACT Patent litigation in the United States has increased significantly since the mid-1990s. I use data on patent litigation from the Federal Judicial center and on firm characteristics from Compustat to investigate the economics of patent litigation. Even though, about three-quarters of patent litigation cases in U.S. district courts are settled or probably settled the cost of litigation are still huge, often costing millions of dollars. The average number of days for case that go to the trial was 991 which is twice as long as in cases settled prior to trial. Litigants strategic behaviors in building their reputations have a significant influence on their decisions to file suit and on settlement outcomes. Well-reputed litigants settle quickly to reduce uncertainty for shareholders. The dominant factors driving settlements for litigants are financial scale, R&D spending, financial performance, leverage, and industry composition. Results are much stronger for defendants than for plaintiffs. * PhD candidate, Department of Economics, University of Delaware, javad@udel.edu 1

2 Patent litigation in the United States has increased significantly in the last two decades. The number of patent cases filed in US federal courts has approximately more than doubled since the mid-1990s. There has been a similar increase in the number of patent settlements for the same period. Figure1 shows the trends in patent cases filed, and in patent settlements. Understanding patent litigation outcomes has attracted the interest and comments of a number of researchers. (E.g., Somaya 2003; Cook, 2007; Lanjouw and Schankerman, 2001a, 2001b, 2004; Schankerman and Scotchmer, 2001; Galasso and Schankerman, 2010; Bessen and Meurer, 2005; and Lerner, 1995, 2010). Patent litigation and the threats it poses have an important effect on innovation. Lerner (1995) asserts small firms avoid investing in R&D when the threat of litigation from larger firms is high. Similarly, Lanjouw and Lerner (2001) show that the use of preliminary injunctions by larger firms can inversely affect R&D investment by small firms. Lanjouw and Schankerman (2004) argue that information on post suit outcomes can help firms assess ex-ante litigation risk. As a result, the threat of costly enforcements can affect R&D investment and patenting strategies. An effective environment for innovation requires certainty and efficient court outcomes (e.g. early settlements). Uncertainty about court outcomes increases the duration of disputes and causes higher transaction costs for both parties engaged in a lawsuit. Litigation costs prevent litigants form doing effective R&D which is detrimental to technological progress. While going to court and trial can protect the patent holder, it, however, creates a threat for small firms for doing effective R&D investment. This would be a concern for small firms and policymakers. Moreover, understanding the role of litigant reputation in patent litigation is valuable because this allows us to identify an important part of litigants strategic behaviors in protecting their reputations. For plaintiffs, a decision to settle rather than litigate or a decision to settle after a suit being filed rather than going to trial plays a substantial role in constructing reputation. Similarly, for defendants, a prior settlement or post suit settlement at each stage of the litigation process have an influential effect on reputation. Well-reputed firms try to settle quickly since they are concerned about the reactions of investors and shareholders. Early settlement negotiations have an influential effect on building reputation. Firms that distinguish themselves as reputation builder usually have a lower litigation risk and higher profitability and better financial performance. Corporate reputation encourages shareholders to invest in a company, and 2

3 Settlement and probabale settlement cases Patent Cases Filed CEOs see reputation as valuable intangible assets (Institute of Directors, 1999). Firm reputation is assessed from financial value of firm or from the short-term financial performance (Chun, 2005). While reputation is endogenous to factors such as financial performance and leverage measures, building reputation is exogenous to decisions during the patent litigation process. Valuable patents are more litigated than other patents. One way to estimate the value of patent is assessing the litigation risk. We need two kinds of information to assess the effect of litigation on value of patents: the likelihood of filing lawsuits and the likelihood of certain court outcomes (settlement, injunction, trial, and winning at trial). This information can assist firms in assessing patent litigation risk and IP insurance strategies. Figure1: Patent Cases and Patent Settlements outcomes in US District Court Year Settlement and probable settlement cases Patent Cases filed Settlement cases 3

4 In this paper, I use a large data set that permits identification of well-reputed litigants versus notorious litigants for certain court outcomes. The paper first documents trends in patent litigation from 1996 to 2010 and shows how these outcomes differ by industry groups. Secondly, it identifies the relationship between determinants of litigant reputation, as an exogenous variable, and settlement outcomes. This paper is organized as follows. Section 2 reviews the literature. Section 3 explains conceptual models and hypotheses. Section 4 describes our data. Section 5 reports our empirical results, and Section 6 concludes. 2. LITERATURE REVIEW A. THEORETICAL FRAMEWORK Cooter and Rubinfeld (1989) develop the chronology of typical legal disputes and match up stages in legal disputes with the economic modeling. In the first stage of a dispute, one person (injurer) harms another (the victim). The frequency of harm is affected by decisions that people make to take greater precaution to lower the social cost of the harm. As a result, economic efficiency requires balancing the cost of harm against the cost of staying away from it. In the second stage of a dispute, the party that suffered harm decides whether or not to assert a legal claim. A rational self-interested person makes this decision by comparing the expected future benefit of filing a lawsuit versus expected costs of it. After a legal claim is asserted, in the third stage, parties attend preliminary hearings with judge and engage in pretrial discovery. The court objective is to encourage parties to bargain to settle their disputes. The result of bargaining game can be either cooperative solution, which leads to the settlement or non-cooperative solution which leads to the trial. Another feature of bargaining is the negotiators who are lawyers and sometimes their interests are not identical to their clients interest. This leads to the principleagent problem. The law encourages parties to resolve their disputes by bargaining and when the negotiations fail, the court dictates a resolution in the fourth stage of a legal dispute which is trials. Cooter and Rubinfeld (1989) assert that parties view trials as negative-sum games in which the sum of winnings (positives) and losses (negatives) is negative. This supports the fact that trials are costly. They mention two products of adjudication: dispute resolution and rule making. From the private viewpoint, trials are a method of resolving disputes between parties. However, from a social viewpoint, trials are a collective choice mechanism for creating laws to regulate society. 4

5 Eisenberg and Farber (1997) assert that a potential claimant s decision to file a lawsuit depends on the monetary expected value of the claim and the pecuniary and non-pecuniary costs of litigation. They not only consider the pecuniary cost and risk aversion cost but also psychological and emotional costs of confrontation. They build a model in which the expected value of filing a suit is a function of (1) the likelihood that the defendant would be found liable at trial, (2) the expected damages that would be awarded at trial conditioned on a finding of liability, and (3) litigation costs to both the plaintiff and defendant. A potential claimant will file a lawsuit if the expected value of filing a suit is positive. They mention several properties of their litigation model. First, a potential claimant will file a lawsuit if the costs of litigation are low or if the expected value of litigation is positive. Second, there will be more trials when the costs of litigation are lower, conditioned on a lawsuit being filed. Third, they mentioned that where litigation costs are lower, potential claimants will be more likely to file claims in which they have a lower probability of prevailing." Many scholars state that a rational, self-interested person will initiate a lawsuit if the initial costs of asserting a legal claim is less than the expected benefit of litigation. A rational decision maker will file a lawsuit if he expects a possibility of settlement or a favorable court judgment (Eisenberg and Farber, 1997; Cooter and Rubinfeld, 1989; Shavell, 1982; Posner, 1986). Most of the recent literature on the economics of settlement has motivated toward a game-theoretic framework in which there are information asymmetries and a variety of sequences by which settlement offers are made by the parties. Both parties have expected gains or losses about the size of transfer in trial and cost of trial. These expected gains and losses represent the extent of the threat which would result in cooperative solution (settlement) or noncooperative solution (trial). Recent theoretical work also takes a more sophisticated view of what happens within the case by modeling both the costs and the revelation of information that occur during the discovery phase preceding trial. These bargaining models have been used to explore how differences in stakes, the cost of litigation, and legal rules influence the choice between settling the dispute or going to trial. Waldfogel (1998) explains two models: the asymmetric information (AI) described in Bebchuk (1984) and the divergent expectation (DE) described in Priest and Klein (1984). In DE models, each party estimates the quality of his case with error and cases go to trial when one 5

6 party is sufficiently more optimistic than the other. In the DE models, uncertainty arises because both parties are simply unsure about the prospect of victory. This occurs most often when true case quality is near the court s decision standard, and this selection mechanism drives plaintiff win rates toward 50 percent. In AI models, the probability that the plaintiff will win is private information. One party has more information than the other, and his settlement offers are intended to serve as a signal of his bargaining strength or as a mechanism for determining the strength of his opponent. An uninformed party makes a settlement offer that is accepted by the informed party only when he has a low probability of winning at trial. Trials arise in (separating) equilibria because settlement offers have some probability of failing owing to the information asymmetry. This one-sided selection mechanism predicts that the win rate for the informed party should tend toward 100 percent. Lanjouw and Schankerman (2004) provide an evidence strongly favor the DE model for patent infringement suits. The literature (Lanjouw and Schankerman, 2004; Bessen and Meurer, 2005; Lerner, 2010; Hall et al., 2005; Hall and MacGarvie, 2010; Lai and Che, 2009; Harhoff et al., 1999; Harhoff et al., 2003; Jaffe and Trajtenberg, 2002; and Lanjouw et al., 1998) suggests that we can indirectly measure the value of patents. Theory suggests that more valuable patents will be more frequently litigated. Valuable patents have higher expected gains in court and therefore will be more litigated. B. PATENT LITIGATION AND POST-SUIT SETTLEMENT Cook (2007) reports that the number of patent cases filed in U.S. federal courts has approximately doubled during the period. Cook 2007 examines the friendly court hypothesis and the hypothesis of an increase in research productivity. He states that the increased application of computers led to increases in research productivity and therefore more patent grants and subsequently more patent litigation. Under the friendly court hypothesis, he expects that trial court outcomes affected by the establishment of the Court of Appeals for the Federal Circuit (CAFC) as the sole court of appeal for patent cases in late He mentioned that It could be that a court more friendly to the patent holder led to an increase (either directly or indirectly) in patent suits, particularly infringement suits, by increasing firms incentives to file for patents. Cook (2007) shows a significant relationship between court outcomes and the amount of litigation controlling for the increase in patents granted and if the friendly 6

7 hypothesis were a true assumption. However, Kortum and Lerner (1998) reject the friendly court hypothesis and express that the increase in patenting grants can be resulted from the technological opportunity." Lanjouw and Schankerman (2004) study the determinants of patent suits and post-suit settlement cases. Their findings suggest that litigation risk is much higher for patents that are owned by individuals and firms with small patent portfolios. Having a larger portfolio of patents reduces the probability of filing a suit. They predict that the domestic patents have lower costs of detecting and prosecuting infringements in the United Sates relative to the cost of settlement. As a result, domestic patent owners have higher litigation rates than foreign patentees. They also find that firms operating in the more concentrated technology area, where patenting is dominated by fewer companies, are much less likely to be involved in patent infringement suits. These firms most likely have greater incentives for settlement. They state two main mechanisms that patentees can settle without resorting to litigation. The first mechanism is by trading intellectual property in the different form such as cross-licensing agreements, patent exchanges, balancing cash payments. The second mechanism is by the expectation of repeated interaction among patentees. The repeated interaction in the theory of super games increases both the ability and the incentive to cooperatively settle a dispute without filing suits. They state that patent owners who are relatively larger than disputants, are more likely less resort to the court. Lanjouw and Schankerman (2001b) state probabilities of litigation differ very substantially among technology fields such as chemicals, software, biotechnology, drugs and non-drug health patents, and are systematically related to patent characteristics and characteristics of their owners. They believe that this heterogeneity of patents, and their owners, is a central issue for the enforcement of intellectual property rights and therefore the process of enforcing patent rights is sorting among patent disputes. This sorting can occur at each stage of the legal process, beginning with the decision to file a suit and ending either with a post-suit settlement or adjudication at trial. Their findings suggest: first, most settlements occur soon after the suit is filed, before the pre-trial hearing is held and, second, post-suit settlements are high, about 95 percent. One of the earliest studies of the probability of a patent being involved in litigation was done by Lerner (1999). Based on a sample of 530 biotechnology firms, he calculates the number 7

8 of Massachusetts patent suits in which the sampled firms were involved during January 1990 to June 1994 and compares that number with the number of patents they were awarded during that period. He finds that six cases per hundred patents held by those firms will be litigated. He concludes that patents in new technologies, such as biotechnology, are more likely to be litigated than those in mature fields because there is more uncertainty about case outcomes. Eisenberg and Farber (1997) empirically model the frequency of trials and plaintiff wins and examine data on these outcomes in a larger number of civil suits filed in federal courts. They conclude that case selection that leads to lawsuits depends on the monetary expected value of the claim and also on the pecuniary and non pecuniary costs of litigation. The lower litigation cost will imply higher trial rates. They argue that plaintiffs with lower litigation costs will be willing to file cases that have a smaller probability of prevailing at trial. They also argue that plaintiff win rates will be negatively related to the variation in the distribution of plaintiffs litigation costs in the population of potential claims. They also present predictions about the identity of the plaintiff which indicates that trial rates will be higher for individual plaintiff rather than corporation plaintiff. They also conclude lower plaintiff costs imply higher trial rates and lower plaintiff win rates. In their findings, plaintiff win rate is lower for individual plaintiff compared to corporation plaintiff. They find that high trial rates are associated with low plaintiff win rates. Galasso and Schankerman (2010) investigate how fragmentation of patent rights and the establishment of the CAFC in 1982 affected the length of patent infringement disputes. They state that licensing negotiation are affected both by the characteristics of the patents and disputants, and by the legal environment within which negotiations take place. Their empirical findings suggest that patent disputes in U.S. district courts are settled more quickly when infringers required access to fragmented external rights. They interpret the number of required patents, for a given technology, as a measure of the degree of fragmentation of patent right. C. INJUCTIONS One of the studies of the preliminary injunctive relief model was done by Lanjouw and Lerner (1996). They investigate how the availability of preliminary injunctive relief affects the probability of cases going to trial and the impact of this legal remedy on high and low cost plaintiffs and defendants. Their findings indicate that preliminary injunction requests are more common in cases where the plaintiffs had greater sales than the defendant. They expect that 8

9 patent awards in new area of technology such as software and biotechnology, with few prior patents, are more likely to be characterized by greater uncertainty. They assert that patents in the subclass where awards are frequently reexamined are likely to be in areas with substantial legal uncertainty. They conclude that litigation is more frequent when there is more uncertainty. Lerner (2010) investigates the litigation of patents relating to financial products and services. He finds that financial patents are litigated at a rate times greater than the rate of patents as a whole. He mentioned four criteria that can increase the probability of a trial: (1) the likelihood that the offense is detected by the potential plaintiff, (2) the size of the stake under dispute, (3) the uncertainty about the outcome of the controversy between the two parties, and (4) the cost of settlement relative to that of trial. These criteria are consistent with the Lanjouw and Lerner s (1996) findings that the probability of a tria l increases when there is more uncertainty. D. PATENT CHARACTERISTICS The literature (Lanjouw and Schankerman, 2004; Bessen and Meurer, 2005; Lerner, 2010; Hall et al., 2005; Hall and MacGarvie, 2010; Lai and Che, 2009; Harhoff et al., 1999; Harhoff et al., 2003; Jaffe and Trajtenberg, 2002; and Lanjouw et al., 1998) suggests that we can indirectly measure the value of patents. Patent value can be captured through the number of claims made in the issued patent, the number of forward citations (future citations received by a patent), and the number of backward citations (the number of prior patents cited per claim in patent documents). Theory suggests that valuable patents have a higher expected benefit of litigation and therefore will be more frequently litigated. Hall et al. (2005) confirm that patent citations, R&D intensity, and patent yield contain significant information on the market value of the firm. They find that an extra citation per patent boosting a market value by 3%. Lanjouw and Schankerman (2004) also use citations, along with other measures such as the number of claims, nationality of patent owner(s), technology field, patent portfolio size, relative size of potential disputants and ownership to determine the probability of litigation. They conclude that the probability of litigation increases with respect to the number of claims and forward citation. Their findings suggest that the likelihood of a suit falls with the respect to the number of backward citations per claim. This result is consistent with 9

10 the view that backward citations are an indication that the patent is in an already well developed technology area and so it's less likely to cause disputes. Hall and Ziedonis (2007) investigate the litigation of patent lawsuits in the 136 semiconductor firms. They explore the relationship between litigation probability and patent portfolio size and firm level characteristics such as R&D spending, size, and patent propensity. Their findings suggest that the probability of being target (defendant in an infringement suit or the plaintiff in a validity suit) increases more rapidly with size and R&D intensity for semiconductor firms than for other firms. One of the studies of the patent valuation model was done by Lai and Che (2009). They study patent infringement lawsuits in the U.S. district courts and propose an integrated evaluator for patent management. They set the damage award as the endogenous variable and the 17 patent indicators as the exogenous variables. Exogenous variables describe the quantitative features of a patent. These indicators are: number of assignees and number of inventors for each patent, number of independent and dependent claims for each patent, US patent references, foreign patent references, non-patent references, forward citation, international patent classifications, US patent classifications, Worldwide patent families, US patent families, number of office opinions by the examiner of USPTO for each patent, number of responses to USPTO by the assignee for each patent, examination period, number of drawings for each patent, and patent Life-span. The authors state that a linear relationship between the damage award and the patent indicators could not be modeled as a simple linear equation. Hence, they construct Back- Propagation Neural Network model to evaluate patents. Their results are somehow different with other scholars (Hirschey and Richardson, 2001; Hereof et al., 2003; Hirschey and Richardson, 2004; Von Wartburg et al, 2005; and Silverberg and Verspagenb, 2007) 3. MODEL A. ECONOMIC MODEL In this section, I identify factors that affect a decision to initiate litigation, a decision to settle a dispute prior to trial, and a decision to proceed to trial. Lerner (2010) states that factors affect the decision to settle a dispute prior to trial should drive the decision to settle prior to the filing a lawsuit. The same factors, which are probability of success, the extent of uncertainty, and the expected rewards at trial if successful, also shape the decision to file a lawsuit and proceed to 10

11 trial. Bessen and Meurer (2005) assert that the expected cost of patent disputes varies proportionally with the firm s hazard rate of entering disputes. This expected cost reduces a firm's incentive to invest in R&D. They believe that trends in the hazard of filing reflect trends in the total hazard of disputes. I identify the main factors influencing post-suit settlement during the patent litigation for both plaintiffs and alleged defendants. This gives us an estimate of the contribution of different factors to the litigation outcomes at different stages of the litigation process. I present a decision model that tests hypotheses about litigious behavior of litigants during the patent litigation process. My model motivated by Eisenberg and Farber (1997). The suit has some expected value to each parties (l), plaintiff or defendant, as a function of the likelihood that other party would be found liable at trial,, the expected damage at trial,, litigant reputation,, and the costs of litigation,. A potential plaintiff will decide to continue a dispute if the expected value of proceeding litigation is greater than the expected value of settlement. The expected value for each party at court is:, l: plaintiff or defendant The condition for proceeding with litigation rather than settle (s) for each litigant at court is: I employed two proxies for litigants' potential costs: firms size ( employment and total asset) and proxies for litigant reputation: financial performance (current ratio, profit margin for plaintiff and return on equity for defendant), innovation investment ( R&D per employee and market value per employee) and leverage (sum of long-term debt and short-term debt divided by sum of long-term debt, short-term debt and total stockholder's equity). The literature on the patent litigation pays too much attention to firm characteristics such as R&D, employment, asset, market value and there is less attention to the firm financial performance such as profitability ratios, solvency ratios, liquidity ratios, etc. and their effects on court outcomes during the patent litigation process. Lanjouw and Schankerman s (2001b) findings suggest that most settlements occur soon after the suit is filed, before the pre-trial hearing is held (about 80 percent). They indicate that 11

12 post-suit settlement rate is high (about 95 percent). They also state that the larger firms in the new technology area are more likely to settle in the early stage of litigation. Lanjouw and Schankerman (2004) state two main mechanisms, which are trading intellectual property and expectation of repeated interaction among patentees, that promotes settlement of disputes between parties. Repeated interaction in game theory increases incentives to settle disputes cooperatively. Lerner (2010) investigates the characteristics of defendants. He asserts that large firms should have lower litigation costs because of learning curve effects while they are more vulnerable to damage and reputation from an adverse judgment. He employed several proxies for measuring litigation costs such as firms experience, firm s assets, financial condition, leverage, location of headquarters, the extent of innovations by a firm, the extent of other innovations in the firm s ZIP code, and academic connectedness all as exogenous variables and the number of filings in all patent lawsuits as an endogenous variable for the firm as a defendant. His findings suggest that strongest determinant of a firm being a defendant is its financial scale. H1a: The probability of settlement between two firms increases with the size of the stakes, all else equal H1b: The probability of settlement between two firms increases when reputation is at stake, all else equal Indirect evidence is provided by Bhagat, Brickley, and Coles s (1994) event study suggests that thinly capitalized defendants have significantly larger negative stock price reactions to the filing of lawsuits than do other firms. Lanjouw and Lerner (2001) argue that the use of preliminarily injunctions by large firms can discourage R&D by small firms. Hall and Ziedonis s (2007) findings suggest that probability of being target (defendant in an infringement suit or the plaintiff in a validity suit) increases more rapidly with size and R&D intensity for semiconductor firms than for other firms. They also find that doubling a firm s size increases the probability of litigation involvement by about 6%. Bessen and Meurer (2005) hypothesize that firms (defendants) that simply imitate without bearing any R&D spending to invent will be more likely to be sued. 12

13 H2: The probability of settlement between two firms decreases with innovation investment by litigants if the litigants trial payoff is greater than a joint settlement payoff, all else equal Patents work differently in different industries. However, in the pharmaceutical, chemical and biotechnology industries the patent equals the final product. The pharmaceutical industry has an important characteristic that sets it apart from other industries that rely on patent protection. Most importantly, unlike industries which produce products requiring expensive manufacturing infrastructures, the patented products of pharmaceutical companies can be easily and cheaply replicated by copiers with little investment. Since investment in the pharmaceutical industry disproportionately is directed to laboratory research and clinical trials rather than the manufacture of the final product, patent exclusivity is the only effective way to protect and receive a return on that investment (Wegner, 1994). H3: The probability of settlement decreases between two firms competing in technology areas where a patent is close to final product, all else equal Lanjouw and Lerner (1996) provide evidence that financially strong plaintiffs use preliminary injunctive relief to predate on weaker firms by driving up their cost. They state that the probability of winning an injunction may be improved with greater expenditure on legal services, and larger firms with good financial strength may spend more on such services. Small firms and individual are less sophisticated in intellectual property disputes, and therefore they have lower probabilities of winning in court. H4: Notorious plaintiffs target financially strong defendants and seek non-cooperative solution by nonsettlement mechanism 4. DATA A. CONSTRUCTION OF THE DATA SET I matched records from two databases: lawsuit filings from the Federal Judiciary Center (FJC), and firm financial data from Compustat. The data on the outcomes of patent litigation is from the Federal Judiciary Center (FJC), which includes observations for each case filed with information on awards, filing and termination dates, the parties involved in a case, whether an injunctive relief was granted, and court outcomes. This data set spans over 15 years from 1996 to 2010 and consist of 45,814 observations. Then, to explore characteristics of firms involved in lawsuits, I matched the litigant's name with the Compustat database of U.S. firms using the names of the parties and their gvkey numbers. The Federal Judicial Center gathers data directly 13

14 from the administration of the courts and they consistently report a larger number of filings. I removed duplicate records involving the same lawsuit, with the same docket number, same section numbers and in the same filing year. I use various samples in this paper. Figure 2 explains the classification of our sample at each stage of my analysis. Initially, our patent sample consists of 45,814 patent lawsuits. I delete cases that are transferred to other districts, remanded, or statistically closed. The resulting sample reduces to 40,678 observations. The second database comes from the Compustate which consists of firms financial information from 1995 to I matched FJC database with Compustat and identified the name of companies for at least one of the litigants. The resulting sample size turned out to 11,583 of which 5,471 is plaintiff parties and 6,112 is defendant parties. I also generate the sample from identified defendant and plaintiff, which contains both defendant and plaintiff s financial information together with court outcomes for each lawsuit. This sample consists of 1311 observations of pairs of firms for each case involved in patent infringement. To check the validity of this match, I randomly selected a number of parties involved in a lawsuit and manually checked them using databases including Bloomberg Law and LexMachina. The rate of falsely matched were not more than 3%, no more than 8 out of 300 parties were found to have been falsely matched. I also generated a matched sample of firms from the population of all U.S. public corporation randomly from Compustat. For each litigant, a firm (litigated or not litigated) was chosen randomly with the same SIC and same filing year code defined in variable section. The comparisons between litigated firms and matched firms helpto control both for technology and cohort effects. My analysis differs from previous literature (e.g. Lanjouw and Schankerman, 2004; Lerner, 2010) in that I use the litigant as the unit of analysis rather than patent as the unit of analysis. Similar to the Bessen and Meurer (2005), my aim is to discover how the plaintiff s choices at different stages of the litigation process affect litigation rates and how parties are affected by litigation hazards. 14

15 Figure 2: Sample description for patent lawsuits during Filter/Merge Number of patent lawsuits Data sources Tables/Figures Initial sample 45,814 Federal Judicial Center (FJC) Deleted cases if transferred to other districts, remanded, or statistically closed (disposition codes*: 0,1,10, 11,& 18) 40,678 FJC Figure 1, Tables 3,4,5,6,7,8,9 Identified companies after merged with Compustat 11,583 FJC & Compustat Table 10 database Identified plaintiffs 5,471 FJC & Compustat Tables 1, 12 Identified defendants 6,112 FJC & Compustat Tables 1, 13 Matched samples: Matched sample for plaintiffs 5471 FJC & Compustat Tables 1, 12 Matched sample for defendants 6112 FJC & Compustat Tables 1, 13 * Table 2 in appendix explains cases disposition codes B. Sample Characteristics Table 1 shows mean of several variables estimated for firm-years using identified samples and matched samples. The first three columns indicate all firm-years observations for plaintiffs. Plaintiff employment size, total asset size, and R&D spending are much larger than the matched sample. Patent litigation is very much dominated by large and R&D intensive firms. The last three columns show all firm-years observations for defendants. Defendants spend about the same R&D as plaintiffs do. Alleged infringers are also larger than plaintiffs in terms of employment size, total asset size, and market value of equity. Patent litigants, both defendants and plaintiffs, tend to have relatively larger current ratio than matched sample, and both almost have a same degree of leverage. The means of profit margin for plaintiffs and the mean of return on equity for defendants both are almost five times greater than their matched samples. Table 1: Sample characteristics for patent litigants during Means Plaintiffs Defendants All sample Identified litigated sample Matched sample All sample Identified litigated sample Matched sample Market Value (Millions) R&D (Millions) Leverage Current ratio Total Asset (Millions) Employee (Thousands) Profit Margin Return on Equity

16 C. Variables Similar to the Hall and Ziedonis (2007) and Bessen and Meurer (2005), I estimate the likelihood of certain court outcomes particularly settlement at the firm level. This allows me to explore the relationship between litigant characteristics and court outcomes at different stages of the litigation process. The unit of analysis in my research is "firm-year". The main variables of interest are as follows: Settlements or probable settlements per litigant. This is a dummy if a litigant settled or probably settled after filing a lawsuit, and it considered as a dependent variable in my models. In my analysis, consent judgments, stipulated dismissals with or without prejudice, and voluntary dismissals were all considered as the probable settlements. Industry group. I divide litigants into twelve industry groups according to their primary product category as identified by Compustat: SIC 28 (chemicals excluding drugs), SIC 283 (drugs, including pharmaceuticals), SIC 35 (machinery, excluding computers), SIC 357( computer and office equipment), SIC 36 (electronics), SIC 38 (instruments), other manufacturing (SIC 20-39, excluding the above), SIC (retail and wholesale), SIC (finance, insurance, and real estate), SIC:73 (Business services excluding SIC 737), SIC 737 (computer programming, data processing, and other computer-related services), and other nonmanufacturing excluding the above. These classifications use the SIC code assigned by Compustat for the primary line of business of the firm for the given year. Litigant financial variable. I include two firm size measures: employees in thousands and total asset in millions. I consider several proxies for litigant reputation: current ratio (current assets divided by current liability), profit margin (total net income/loss divided by total sale), and return on equity (total net income/loss divided by market value of equity). Other measures are innovation investment: R&D per employee and market value of equity per employee, and leverage (sum of long-term debt and short-term debt divided by sum of long-term debt, shortterm debt and total stockholder's equity). I also define time dummies in my models. 16

17 Empirical Results A. Outcomes of patent litigation In this section, I investigate the trends in patent litigation outcomes during Most settlements about 80 percent occurs soon after the suit is filed, before the pre-trial hearing is held (Lanjouw and Schankerman, 2001b). They indicate that post-suit settlement rate is high (about 95 percent). They also state that the larger firms in the new technology area are more likely to settle in the early stage of litigation. Lanjouw and Schankerman (2004) state two main mechanisms, which are trading intellectual property and expectation of repeated interaction among patentees, that promotes settlement of disputes between parties. Repeated interaction in game theory increases incentives to settle disputes cooperatively. Kesan and Ball (2006) raise question concerns the resolution of cases: are cases being adjudicated through to a final decision by the courts, or do parties settle their dispute without waiting for a final ruling by the courts? If the vast majority of cases are settled along the way, the courts may be fulfilling their role of protecting patent rights at relatively low cost. However, a small number of final rulings on the merits also means that very few patents are being inspected to determine the scope, validity, and infringement of patent rights. Some authors (e.g. Hall et all, 2003 and Lanjouw & Schankerman, 2004) have mentioned that 5% of cases are terminated through a trial and therefore 95% of cases are settled. Kesan and Ball (2006) assert that this figure ignores the fact that many cases are resolved through other pretrial terminations. Moreover, our results show that 10.4% of cases are terminated through final rulings granting a motion for summary judgment. The actual categories for case disposition reported in the Federal Judicial Center database are shown in Table 2 in appendix. As we can see, many of these categories e. g., dismissed: other ; judgment: other are quite ambiguous. The ambiguity of these definitions makes it a difficult task to precisely classify the manner in which cases are decided. For example, if the two parties reach an agreement and request that a consent judgment be entered, the final outcomse could be coded as either a settlement, a consent judgment, or judgment: other. Other settlements may be coded as dismissals with or without prejudice, voluntary or stipulated dismissals, or just as dismissals. 17

18 There is also some ambiguity about the coding of summary judgment rulings they can also be classified in the other judgment category, as could a consent judgment, which is more likely to be a settlement (Kesan and Ball, 2006). In my analysis, consent judgments that were not explicitly named as a settlement, stipulated dismissals with or without prejudice that were not named as settlement, and consent judgments were all considered probable settlements. Voluntary dismissals were also classified as probable settlements. The classification categories are listed in Table 3 in appendix. The results of the classification for cases during are shown in Tables 4. To avoid double counting, I eliminated cases that were transferred to other districts, cases that are remanded, or statistically closed during the litigation process. About 10 percent of the cases terminated in grants of summary judgment which coded as a motion before trial. About 3.3 percent of cases terminated in final trials include cases disposed of by jury trial, bench trial, and directed verdict. As a consequence, about 21.4 percent of all cases terminated by a court decision and through some sort of court ruling on the merits. Previous studies (e.g. Hall et all, 2003 and Lanjouw & Schankerman, 2004) have stated that settlement rate is about 95%, but this figure applies to cases that terminated without a trial, whereas the grant for summary judgment motions appear to be equally influential in this result. Similar to the Kesan and Ball (2006), I conclude that a much larger share of cases is adjudicated to a final resolution compared to the suggested literature. However, I still believe that most cases settle. Our results show that almost threequarters of federal civil cases are dismissed or settled, about half of the cases that reach the final judgment stage were disposed of via summary judgment and only 15% of cases terminated in final trials includes cases disposed of by jury trial, bench trial, and directed verdict. Table 5 shows the number of cases disposed from 1996 to Nearly half of all dismissed cases are definitely settlements. Cases that are terminated in a consent judgment have a high probability of being settled since the parties request a consent agreement to formalize the settlement. A voluntary dismissal can also denote that the case has settled. If these cases are included, approximately 70 percent of all patent cases terminate in a settlement, of which 39% were direct settlements, 24.5% were voluntary dismissals and 6.3% were consent judgments. Voluntary dismissals and consent judgments both suggest the probable settlements. 18

19 Table 4: Patent Litigation Outcomes* Non-Merit Disposition Settlement and Probable Ruling and Verdict Settlement Outcome Number of Outcome Number of Outcome Number of Cases Cases Cases Lack of Jurisdiction 384 1% Identified % Summary Judgments** % Settlements Want of Prosecution % Consent % Judgment on Jury Verdicts % Judgments Default Judgment % Voluntary % Judgment on Bench Trials 428 1% Dismissals Judgment as a Matter of % Law (Directed Verdict) Arbitration 11 0% Subtotals % % % Total of Dismissals 31,924 Total of Judgments 8,754 Total 40,678 * Patent litigation outcomes were produced by the Federal Judicial Center which keeps statistics on the number of civil cases commenced in federal courts annually and the nature of the suit. ** Includes a motion before trial, and other judgment excluding judgment on jury verdicts, judgment on bench trials, and judgment as a matter of law. Table 5: Patent Litigation Cases Concluded In U.S. District Courts, By Disposition*, Year Number of complaints disposed** Percent of cases disposed Dismissed (%) Judgment (%) Settled Voluntary Lack of Want of Other Total Consent Trial*** Other**** Total Jurisdiction prosecution ,580 38% 20% 1% 2% 13% 73% 10% 5% 12% 27% ,667 39% 21% 1% 2% 13% 75% 10% 4% 11% 25% ,875 38% 20% 1% 2% 13% 74% 10% 5% 12% 26% ,005 41% 21% 1% 2% 11% 76% 8% 4% 12% 24% ,040 39% 23% 1% 2% 12% 77% 8% 3% 13% 23% ,292 40% 22% 1% 2% 11% 77% 8% 3% 12% 23% ,283 40% 23% 1% 2% 12% 78% 8% 3% 11% 22% ,368 41% 24% 1% 2% 14% 81% 5% 2% 11% 19% ,918 38% 25% 1% 2% 14% 80% 6% 3% 11% 20% ,435 40% 24% 1% 2% 13% 80% 6% 3% 12% 20% ,502 42% 24% 1% 1% 12% 81% 5% 3% 11% 19% ,362 39% 25% 1% 1% 11% 77% 6% 3% 14% 23% ,980 41% 26% 1% 1% 12% 80% 5% 3% 11% 20% ,449 40% 27% 1% 1% 11% 80% 5% 3% 12% 20% ,922 36% 29% 1% 1% 14% 80% 4% 4% 12% 20% Total/ Average % 40,678 39% 24% 1% 1% 12% 78% 6% 3% 12% 22% * Patent litigation outcomes and disposition codes were produced by the Federal Judicial Center which keeps statistics on the number of civil cases commenced in federal courts annually and the nature of the suit. **Excludes transfers, remand, and statistical closures. ***Trials includes cases disposed of by jury trial, bench trial, and directed verdict, the parties may have settled before the competition of the trial. ****Includes judgments by default, consent, a motion before trial, judgment of arbitrator or by some other final judgment met hods. 19

20 Table 6 describes the number of cases that received monetary awards and average for each year from 1996 to Given the small number of rulings of infringement, damages is only awarded in a small number of cases. Damages were awarded in only 2.7% cases (1096 cases out of 40678). The average awarded in these cases is $287 thousands. This figure in biased since Federal Judicial Center data set recorded any damage awards greater than $1 million as 999,900. The number of cases that received more than $1 million is 230 cases out of 1096 cases. As a result, the statistics on average awards size are all biased. Table 6: Damage awards to plaintiff in U.S. District Courts, Damage Awards* Year Total 6 Number of Cases with monetary awards Trial** Consent Judgment Other Rulin gs Total Average Award Size(in thousands $) 4 * Information on damage awards were produced by the Federal Judicial Center which keeps statistics on the number of civil cases commenced in federal courts annually, the nature of judgment, and the nature of the suit. **Trials includes cases disposed of by jury trial, bench trial, and directed verdict. As shown in Table 7, injunctions are most commonly found in consent judgments and even a motion before trial is a mechanism for formalizing the agreement. Injunctions sometime employed in a default judgment as a way of controlling an infringing party who has not presented himself in court. Table 7: Injunctions Number of injunctions* Year Total Motion before trial(summary judgment) Jury Verdict Consent Judgment Default Judgment Other Total Percent of All % Cases (%) * Information on number of injunctions were produced by the Federal Judicial Center which keeps statistics on the number of civil cases commenced in federal courts annually, disposition code, and the nature of the suit. 20

21 Economic theory suggests that settlements are a mechanism used by the parties to avoid the high litigation costs of patent cases. Our results show that nearly 70 % of all patent cases settle and only 3% of cases go to trial due to litigation costs. If a ruling turns out to be expensive, the incentive to settle the dispute will be high. Individuals will balance their private benefits against the costs of continuing litigation. Thus, if rulings, particularly rulings of invalidity, are expensive, too few cases will be pursued to a final adjudication of validity or infringement on the merits. In this section, I measured the litigation costs associated with the case resolutions. Unfortunately, it is almost impossible to measure litigation costs for the general population of patent cases. However, I have developed two proxies for costs: length of time to termination and whether the cases reached the stage of filing a motion for summary judgment of cases during Time to termination is a traditional measure of the resources expended on a court case. However, while it has a strong intuitive appeal, this measure is also likely to be inaccurate. There can be long delays in scheduling court hearings and periods of inactivity that are not necessarily associated with higher costs. In addition, litigation costs begin to rise when cases reach the claim construction or summary judgment stage. Even if the case settles after that point, there will have been a considerable expenditure of resources. Therefore, reaching the stage of filing a motion of summary judgment can be an indication that the level of expenditure in the case was moderate to high. Table 8 reports our results for the duration of cases filed during The average number of cases terminated less than a quarter was 337 and the average number of cases terminated in eight quarters was 83. The average number of cases more than two years was 422 out of 2372 during About 50% of cases were resolved within nine months during The average number of days for termination of case was 425 during These results demonstrate that on the average expenditures in patent cases are not extremely high. 21