Endogenous Litigation Costs: An Empirical Analysis of Patent Disputes

Transcription

1 Endogenous Litigation Costs: An Empirical Analysis of Patent Disputes Christopher A. Cotropia, Jay P. Kesan, Kyle Rozema, and David L. Schwartz * December 14, 2016 A theoretical literature explains how parties can strategically sink litigation costs to maximize their prospects. We explore this endogenous litigation costs hypothesis empirically. To do so, we introduce measures of case progression, establish and validate new proxies for party and court costs, and parse the rich information about litigation contained in novel administrative data from court dockets. As a case study, we use the universe of 16,661 patent cases from 2005 to 2010, containing roughly 1.5 million docket entries. We document the costs of litigants and courts throughout litigation, and find that litigation costs cluster around events that must occur in order for litigation to proceed. Finally, we present evidence that costs around stages of litigation are higher for cases that eventually reach later stages of litigation, which we argue is consistent with the endogenous litigation costs hypothesis.. * Cotropia: Professor of Law and Austin Owen Research Fellow, University of Richmond School of Law, ccotropi@richmond.edu. Kesan: Professor and H. Ross & Helen Workman Research Scholar, University of Illinois College of Law, kesan@illinois.edu. Rozema: Post-Doctoral Fellow, Northwestern University Pritzker School of Law, kyle.rozema@law.northwestern.edu, website: Schwartz: Professor of Law, Northwestern University Pritzker School of Law, david.schwartz@law.northwestern.edu. For helpful comments we thank Colleen Chien, James Pfander, Greg Reilly, Michael Risch, and Neel Sukhatme, and participants at ETH Zurich's Workshop and Lecture Series on the Law & Economics of Innovation. We would like to thank Docket Navigator for providing us with its data relating to patent claim construction.

2 1. Introduction Strategic litigants maximize their prospects in litigation not only by choosing if and when to settle and how much to settle for, but also by choosing when and how much to expend on litigation (the endogenous litigation costs hypothesis ). Expending resources on litigation can directly influence the expected value of the case and can also serve as an important signal about the litigant s position. There is a theoretical literature related to the timing of settlement and the endogeneity of litigation costs. 1 However, unlike the rich empirical literature that studies the determinants and timing of settlement, 2 little empirical evidence has been brought to bear on the endogeneity and timing of litigation costs. We provide suggestive evidence of the endogenous litigation costs hypothesis by exploiting the rules of civil procedure that force costs to accumulate nonlinearly throughout litigation. As a case study, we use the universe of 16,116 patent infringement disputes in the U.S. federal district courts filed between 2005 and We find evidence that costs around early stages of litigation are higher for cases that eventually reach later stages of litigation, which we argue is consistent with the endogenous litigation costs hypothesis. The article proceeds in four parts. First, we introduce a framework for assessing the progression of litigation. Second, we introduce new measures of litigation costs, and validate the new and existing cost measures against a sample of attorneys fees. Third, we document how the 1 See, e.g., Priest and Klein, 1984; Rosenberg and Shavell, 1985; Katz, 1988; Hay, 1995; Bebchuk, 1996; Croson and Mnookin, 1996; Farmer and Pecorino, 2013; MacKenzie, 2014; Hubbard, We have learned that timing of settlement is affected by institutional delays in courts (Kessler, 1996), rules governing discovery (Huang, 2007), whether parties are represented by an attorney (Huang, 2008; Hyman et al., 2016), motion practice of attorneys (Boyd and Hoffman, 2012), frequent interactions between lawyers (Johnston and Waldfogel, 2002), rules for awarding attorneys fees to prevailing parties (Fournier and Zuehlke, 1996), and the macroeconomy (Marco, Miller, and Sichelman, 2015). 1

3 costs of litigation accumulate throughout the litigation process. Finally, we explore the endogeneity of litigation costs. In the first part of the article, we exploit novel administrative data on litigation docket entries textual descriptions of each item in the court file to establish important litigation milestones. We use a very formal definition of milestone: a litigation milestone is an event that must occur in order for litigation to proceed. We chose this definition because voluntary milestones including motions to dismiss, motions for summary judgment, and alternative dispute resolution are likely to suffer from endogeneity concerns and therefore do not offer robust measures of case progression. 3 The progression of patent litigation is sequential with four main milestones. After a lawsuit has been initiated in court, the first milestone of litigation is for a defendant to answer, a formal legal document where the defendant responds to each allegation in the plaintiff s complaint by admitting, denying, or asserting that it does not have sufficient information to admit or deny. Of the patent infringement cases filed between 2005 and 2010, an answer is filed in 79% of cases an average of 2.6 months after the filing date (i.e., 21% of cases either settle or are otherwise voluntarily dismissed before an answer is filed). The second milestone is a case management conference. At that conference, the judge, often with input from the parties, schedules deadlines for the parties. Until the conference, no formal litigation discovery the exchange of information is permitted. After an average of about 1 month from the filing of an answer in which another 20% of cases are resolved, the remaining 59% of filed cases convene for case management. The next milestone, which is unique to patent litigation, is called a claim construction or Markman ruling in which the court decides the scope of the patent claims involved in the litigation. After an 3 See Section II.B for a discussion as to why motions to dismiss under Rule 12(b)(6) are unlikely to influence our study. 2

4 average of a year and a half following the case management conference, another 49% of filed cases are resolved, and the remaining 10% of filed cases proceed to a claim construction ruling. The final major milestone is trial. Trial begins for 3% of cases on average about one year following the claim construction ruling, and are resolved on average about a year and a half later. In the second part of this article, we build a panel of litigation cost measures for each lawsuit and assess the validity of such measures as proxies for actual costs. We first investigate and empirically confirm the validity of existing measures of litigation costs total docket entries (Kesan and Ball, 2006) and motions filed in a case (Boyd and Hoffman, 2012) as proxies for costs using novel data on attorneys fees awarded in patent cases in our data. We find that a 1% increase in docket entries and party motions are associated with a 0.9% and 0.7% increase in attorneys fees in cases in which attorneys fees were awarded, respectively. Next, we introduce two measures for public costs of litigation. Whereas total docket entries and party motions capture important private costs of litigation, we argue that (1) judicial orders and opinions, and (2) hearings between the judge and the parties capture important public costs of litigation. Finally, it is worth emphasizing that, unlike previous research, a novel aspect of our cost measures is that we are able to track them over the course of litigation (at the daily level). In the third part of this article, we document the accumulation of litigation costs throughout the progression of patent cases. Using event studies of how litigation costs accumulate around each litigation milestone, we find that private and public costs spike around each milestone. The evidence confirms the well-known conception that the rules of civil procedure force litigants to spend nonlinearly over litigation. We also find that each milestone differs in the extent to which private and public costs accumulate. In particular, relative to the makeup of costs accumulating around answer and case management, the costs of parties compared to courts are higher for claim 3

5 construction and trial. This is consistent with how parties exert considerable effort for claim construction and trial with discovery disputes and substantive disagreements on the merits. In the final part of this article, we use the panel of litigation cost measures and exploit the rules of civil procedure to offer novel evidence on the endogenous litigation costs hypothesis. In particular, we investigate how litigation costs differentially accumulate around milestones for cases that are resolved at various future stages of litigation, thereby using whether later stages were reached to distinguish between different case types. Our exploration of the hypothesis therefore assumes that unobservable differences between cases are reflected in what stage the case reaches. Under this assumption, differences in costs between case types following a milestone would provide suggestive evidence that litigants are strategic in choosing when and how much to expend on litigation, where litigants who have the most to benefit from sinking costs spend more after an event in attempts to improve their prospects or signal their position to the other side. We find that costs around a milestone differ between cases that will or will not reach future milestones, providing evidence that the timing of litigation costs are endogenous. The rest of this article is as follows. Section II describes the data and the litigation milestones. Section III introduces the new measures of litigation costs and explores the validity of the measures as proxies for costs. Section IV provides descriptive evidence on the timing of each litigation milestone. Section V sheds light on the endogenous litigation costs hypothesis. Section VI discusses limitations and concludes. II. Litigation Costs and Progression The federal courts maintain dockets for each lawsuit. We use data containing each of the docket entries from the universe of patent infringement lawsuits filed in the U.S. district courts 4

6 from 2005 to These data were obtained from the Academic Expert Group (AEG) and are based on U.S. federal administrative data from Public Access to Online Court Electronic Records (PACER). The docket entries are the textual description of all papers in the court files. The data contains information on the 16,611 lawsuits with a Nature of Suit of Patent that were filed from 2005 to From the date each case was filed until when we obtained the data in late 2015, 1,517,980 docket entries have been recorded. 4 The number of dockets per case range from as few as 1 to as high as 7,891, with the median number of dockets at 35. Figure 1 plots the number of cases pending in a given month after the initiation of the lawsuit and the number of dockets filed within the month. Both the number of cases and the number of docket entries show a rapid decrease after the filing of the lawsuit. [Insert Figure 1 about here] Notably, while patent litigation is governed by a complex set of legal doctrines relating to patents, it is also governed the federal rules of civil procedure. Civil procedure rules have some uniformity but also permit judicial flexibility as to the timing of milestones. This results in a process that differs from court-to-court and even from case-to-case within a court, with milestones in patent litigation occurring at different times. The milestones may even occur in a different order in some cases. Essentially, based on our expertise in patent litigation, we present the framework that we feel encapsulates the benchmark process and note common deviations from the benchmark where applicable. This benchmark is largely supported by the data. In particular, conditional on a 4 We excluded docket entries for pro hac vice, entries of appearance, and extension of time motions before counting the number of docket entries. See Section II.C. Many of these documents are quite simple forms, often prepared by paralegals, at very low cost. 5

7 party reaching a stage, the milestones occurring before the stage in the benchmark process were met in the vast majority of all the cases. An important reason for assessing the litigation milestones stems from the fact that parties in litigation can settle at any stage of the litigation. Courts have explicit mechanisms that require the parties to discuss settlement such as judge-ordered mandatory mediation, as well as implicit mechanisms such as judges forcing the case to proceed. Settlements are typically accompanied with voluntary dismissals of the case by the court under Rule 41 of the Federal Rules of Civil Procedure ( FRCP ). In almost all of these situations, none of the aspects of the settlement are publicly available from the court s docket and parties filings. Settlements may be more explicitly entered into the public record by the court in the form of consent judgment, which can include some details of who prevailed in the settlement. In the vast majority of cases, it is not possible to capture from the public record why the parties settled and the terms of the settlement. For this reason, the more precise term for what we call settlement is resolution because a party could have dismissed the lawsuit without receiving a transfer from the other party. We do not consider motions to dismiss under Rule 12(b)(6), which are used to dismiss complaints which fail to plead a short and plain statement of the claim showing the pleader is entitled to relief. Over the course of our sample, two important civil procedure cases changed the rules governing pleadings (Bell Atl. Corp v. Twombly, 550 U.S. 544 (2007) & Ashcroft v. Iqbal, 556 U.S. 662 (2009)). Hubbard (2015) finds no evidence that Twombly and Iqbal had major effects on dismissal rates with prejudice, on the use of motions to dismiss, or on settlement and filing patterns. This is consistent with a sizeable literature that finds no detectable effects of both Twombly and Iqbal on dismissal rates with prejudice (e.g., Hannon, 2008; Hatamyar, 2010; Hatamyar Moore, 2012; Seiner 2009, 2010; Cecil et al., 2011a,b; Brescia, 2012). However, both 6

8 Hubbard (2015) and Clermont and Eisenberg (2014) find effects on pro se litigants. Also see Gelbach (2012), who finds some evidence of a small increase in the rate that defendants file motions to dismiss, although they were granted at the same rate. To understand the extent that motions to dismiss might influence our results, for one year of our data (2010), we use hand coded data on whether a case was dismissed as a result of a motion to dismiss, and obtained data on the party attorneys, from which we determined whether each plaintiff was a pro se plaintiff. Of the roughly 3,300 cases in 2010, we found that 0.6% of cases (19) were dismissed pursuant a motion to dismiss, and that 0.4% of cases (14 cases) were pro se. Of the granted motions to dismiss, only 4 cases were pro se plaintiffs. Combined with the evidence that around only 6% of cases have a motion to dismiss filed from 1988 to 2010 (Willging, 1989; Cecil et al., 2011a,b), we therefore do not expect Twombly and Iqbal to influence our results. Moreover, the U.S. Court of Appeals for the Federal Circuit, the appellate court hearing all patent cases, has ruled that Twombly and Iqbal do not apply in patent cases. (In re Bill of Lading Transmission and Processing Sys. Patent, 681 F.3d 1323 (Fed. Cir. 2012).) We parse docket entries to identify the dates in which litigation milestones occur. Table 1 provides a breakdown of the percent of cases reaching the milestones, and the average and median time from filing to the milestone and the average and median time from the milestone to resolution. [Insert Table 1 about here] Answer. A party initiates litigation through filing a formal legal document known as a complaint. Complaints alleging patent infringement must meet the notice pleading requirements under Rule 8 of the FRCP, which require the identification of the alleged infringer, what patents 7

9 are allegedly infringed, and what activity of the alleged infringer is accused of conducting. As discussed above, during the time period of our study, the level of specificity required in the answer is not high, given it is merely notice pleading (and Iqbal/Twombly do not apply). The FRCP was amended in December 2015 to increase the level of specificity needed in many patent cases; however, the amendment did not affect the cases in our study, which were all filed between 2005 and Pursuant to FRCP Rule 12(a), an alleged infringer must file an answer confirming or denying each paragraph of the complaint within 21 days of service. However, under FRCP Rule 12, the parties can agree to an extension of this time, and if no agreement can be reached, an alleged infringer can ask the court for an extension. Denial of most elements of the complaint, absent background information such as the state of incorporation of the defendant, is typical. It is not uncommon for parties to settle before the filing of an answer. We find that 19% of all cases settle before an answer is filed. For the cases where an answer is filed, 5 an answer is filed on average after about three months, which includes extensions and other delays in the litigation process (e.g., the court granting a stay for the parties to attend mediation or arbitration). The first panel of Figure 2 plots the distribution of time that lapses between filing date and answer for the 78.7% of cases in which an answer was filed. The time to answer varies somewhat, presumably because of discretionary extensions. Ninety-five percent of answers were filed within 8 months of the filing of the lawsuit, with the modal observation being at three months. [Insert Figure 2 about here] 5 We count a case as having an answer, if any defendant answers. 8

10 Case Management. Pursuant to the FRCP, after an answer is filed, all cases must have a case management conference to schedule significant dates in the lawsuit. Case management, governed in part by FRCP 16, identifies when fact discovery begins and ends, when expert discovery begins and ends, when the parties must disclose their positions on significant aspects of patent litigation such as claim construction (discussed below), infringement, and validity. Case management also details how and when parties can file summary judgment motions with the court, and often sets the trial date. We identify 62.3% of cases making it to case management. The average length of time from the filing date to the date of the case management conference for these cases is about four months. The second panel of Figure 2 plots the distribution of time that lapses between filing date and the case management conference and from the case management conference to resolution. More than ninety percent of cases that reach case management do so within ten months on an answer. In fact, case management frequently occurs within six months of answer. Similar to the panel for answers, many cases are resolved within a few months after case management. As with answers, the case management milestone is broadly relevant to all civil litigation. Claim Construction Ruling. In almost every patent infringement suit, the parties dispute the scope of the patent. To resolve this dispute, the judge must determine the meaning of the patent claims. This determination is known as claim construction and its result is often outcome determinative on liability. Thus, claim meaning is vigorously contested by the litigants. Claim construction occurs in a so-called Markman hearing named after the U.S. Supreme Court case that ruled that judges, not juries, must determine claim meaning, Markman v. Westview Instruments, Inc., 516 U.S. 370 (1996). These hearings can occur at any time during the case, but in some districts, claim construction occurs during the fact discovery period. Alternatively, these 9

11 hearings are sometimes scheduled after fact discovery closes but before summary judgment motion filings and trial begins. In. Experts are often engaged by both sides in patent litigation cases and sometimes testify during the claim construction hearing. We identify 9.7% of cases having a claim construction ruling. The average time from the filing date to the date of the claim construction ruling is 21.1 months. The third panel of Figure 2 plots the distribution of time that lapses between filing date and the claim construction ruling and from the claim construction ruling to resolution. The case duration before claim construction varies more than the duration before answer or case management. There is a large density of cases where a claim construction ruling occurs between one and two years after case filing. While the date for answer (before any extensions) is set forth in the FRCP, and case management is also required at the onset of cases, the timing of claim construction is more at the discretion of the court. It also requires substantially more court involvement than the previous stages of litigation. Answer requires no court involvement (it s done entirely by the defendant), and case management merely requires selecting dates for the milestones. Claim construction, in contrast, requires the court to study and attempt to understand the patent-in-suit. On average, the claim construction ruling occurs 17.2 months after case management and 11.7 months before the case is resolved. As for resolution after claim construction, the figure shows a large density of cases resolving within one year. Consistent with the view that parties settle as the likely judicial outcome becomes clearer, increased settlement after claim construction ruling is expected because, as the scope of protection accorded to the asserted patent claims is determined by the court, the parties have substantially more information about the prospects (i.e., likelihood of patent infringement and invalidity) of the underlying patent dispute. 10

12 Trial. Trial is a formal examination of evidence in order to decide liability. We identify 3.3% of cases as reaching trial. Trial starts on average 35 months from the filing date, and the average case reaching trial is resolved 19 months after the start of trial. The final panel of Figure 2 plots the distribution of time that lapses between filing date and the start of trial and from the start of trial to resolution. The time from lawsuit filing to trial varies the most of all of the milestones, with over 5% of cases reaching trial in about a year, and 39% of cases that go to trial reaching trial after three years. After trial begins, most cases are resolved within one year. III. Litigation Costs The dockets include a record of all papers filed with the court. Each entry on the docket includes a written description of the underlying document. For instance, most lawsuits commence with the filing of a formal legal document titled a Complaint. Consequently, the first entry on the docket is typically labeled Complaint. We parse the docket entries to classify each into 3 broad categories of litigation costs: (1) party motion or legal memoranda, (2) judicial orders and opinions, and (3) hearings. There are many other types of docket entries, including notations of in-person hearings, transcripts of hearings, trial briefs, jury verdicts, and notices of party dismissals from cases. There are several docket entries that are frequently present in the case files, but we believe are not indicative of real party or court costs. These include: (1) formal appearances, notations of which lawyers are handling the case; (2) pro hac vice applications, which are requests from out of town lawyers for permission to appear in a case; and (3) motions and orders relating to extensions of time. These three types of entries require almost no attorney or court time or attention, and yet are present in 11

13 the dockets. We therefore exclude these three types of docket entries from our analysis, including the total number of docket entries. A. Measures of Litigation Costs Table 2 presents the mean and standard deviation number of each measure per case, and are described below. The first panel of Figure 3 shows the distribution of docket entries within a case, where the solid line is the total number of docket entries and the dashed line is the total number of docket entries per year. The distribution of docket entries per case is skewed to the right. The average case has 96 docket entries, but 50% of cases have less than 36 docket entries and 10% of cases have more than 238 docket entries. [Insert Table 2 and Figure 3 about here] Motions and Memoranda. Motions are filings by a party during litigation requesting the court to make a ruling on a specific issue. The possible motions range from the very simple and uncontested, such as a motion to compel, to the dispositive, such as a party requesting dismissal on the pleadings under FRCP Rule 12(b)(6), or summary judgment on liability infringement under FRCP Rule 56(a), prior to trial. While the often-uncontested motions such as pro hac vice motions occur early and regularly in patent litigation, more substantive motion are significant for many reasons. First, such motions are expensive to draft and file often taking significant attorney time, research, and in some cases fact discovery to support their filing. Second, a party is forced to take a position and risk losing on an issue prior to trial when filing a motion. The second panel of Figure 3 shows the distribution of motions/memos within a case, where the solid line is the total number of motions/memos and the dashed line is the total number of motions/memos per year. A 12

14 relatively small proportion of cases have a significant number of motions and memoranda. The average case has 8.9 motions/memos. The distribution is skewed to the right, with 50% of cases having less than 4 memos and 10% of cases having more than 35. Judicial Orders and Opinions. We use a combined measure of judicial orders/opinions that consists of all orders and opinions issued by the court. An order is typically a minor ruling by the court, ranging from simply granting a motion to extend time without explanation to explaining why a party needs to produce more documents. An opinion, in contrast, is a more substantive ruling. For instance, an opinion granting summary judgment would set forth the applicable law and facts and analyze why summary judgment is appropriate. The third panel of Figure 3 shows the distribution of judicial orders/opinions within a case, where the solid line is the total number of judicial orders/opinions and the dashed line is the total number of judicial orders/opinions per year. The average case has 13.7 orders/opinions. The distribution is right-skewed, with many cases having almost no judicial orders or opinions. Fifty percent of cases have less than 6 orders/opinions, and 10% have over 34. These cases presumably do not require significant judicial resources. This is very similar to the evidence on party motions/memos with most of the cases having very few filed. Hearings. During the course of a lawsuit, the parties often appear before the court for a hearing. A hearing is a formal event, typically conducted in the judge s courtroom. The lawyers for each side and the judge participate, and a court reporter transcribes everything that is said. Some hearings are status conferences, in which the parties explain to the judge where the parties are with respect to various litigation milestones. Other hearings permit the court to hear oral arguments on motions or other disputes that the parties are having. The final panel of Figure 3 shows the distribution of hearings within a case, where the solid line is the total number of hearings 13

15 and the dashed line is the total number of hearings per year. The average case has 3.3 hearings. The distribution is right-skewed, which is consistent with the view that most cases do not demand significant judicial resources or attorney work. In fact, more than 50% of cases have no hearings. For the cases with hearings, 10% have at least 8 hearings, 5% have at least 15 hearings, and 1% have at least 42 hearings. In the next section, we explore how these cost measures are distributed over the litigation milestones. B. Validation We explore the validity of the motions/memos, judicial orders/opinions, and hearings as proxies for litigation costs. A full validation exercise for the use of these measures as proxies for costs would utilize information on attorneys fees throughout the course of litigation and information on the amount of time that judges and court employees spend on a case. To our knowledge, such information is currently not available to researchers. However, in exception patent cases, the court is permitted to award a successful litigant its attorneys fees. (35 U.S.C. 285.) In these somewhat rare cases, the attorneys fee amounts are public. We use information on all attorneys fee awards in patent cases from 2005 to 2010 that we obtained from Docket Navigator. There were 194 cases in which attorneys fees were awarded. We reviewed the details of cases in which had awarded attorneys fees below $50,000, and determined that these relatively low awards are awarded for one or a few specific litigation related disputes. Because these costs are not representative of the total costs for the case, we exclude these cases. This leaves us with 142 cases in which attorneys fees were awarded, which we use in combination with motions/memos, judicial orders/opinions, and hearings in our validation exercise. 14

16 The average award was roughly $1.2 million, and the median award was $207,213. Figure 4 plots the distribution of awarded attorneys fees. The distribution is skewed far to the right (the x axis is in natural log scale). For the validation exercise, we regress attorneys fees in a case on each of the case level measures. Because both attorneys fees and the cost measures are skewed to the right, we use the natural log of both. This also gives us a percent change interpretation in attorneys fees from a percent change in the cost measures. Table 3 provides the results. Columns 1 to 4 include separate regressions for each of the measures, and Column 5 estimates a joint model. The results in Column 1 suggest that a 1% increase in docket entries is associated with a 0.9% increase in attorneys fees. It is also worth emphasizing that the number of docket entries explains almost 30% of the variation in attorneys fees (R 2 of 0.292). Combined, we feel that the 0.9:1 association in docket entries to attorneys' fees along with the high degree of variation explained by docket entries provides evidence that docket entries are a valid proxy for litigation costs to the parties. Columns 2 to 4 provide evidence that party motions/memos, judicial orders, and hearings also contain significant information as to private litigation costs. The association with attorneys fees of each of the measures in Columns 2 to 4 is lower than for docket entries, and these measures explain less of the variation than that explained by docket entries. However, there are important differences between the measures. Party motions/memos explain the most variation in attorneys fees (nearly 25% of the variation), and have the strongest relationship (0.7:1). Hearings perform the worst in explaining the variation (around 12%) and have the smallest association (0.4:1). Judicial orders and opinions perform at a level between party motions/memos and hearings. Column 5 jointly estimates the effect of the measures on attorneys' fees. Note that Column 5 does not include total docket entries because it is partly determined by the other measures. The 15

17 magnitude and precision of the point estimates for each of the measures in the joint model decrease relative to the independent model, and only party motions/memos maintain statistical significance. We find the results very convincing: because judicial orders/opinions are likely a proxy for public costs (not private costs to the litigants), holding party motions/memos and hearings fixed one expects there to be no relationship between judicial orders/opinions and attorneys fees. This is precisely what we find. The magnitude of the point estimates for judicial orders/opinions in the joint model decreases the most relative to the separately estimated models. Moreover, the magnitude of the point estimate on hearings actually becomes larger than that on judicial orders/opinions, although the differences are not statistically significant. In summary, Table 3 provides strong evidence that the measures provide important information as to the private costs of litigation for which attorneys' fees were awarded. While we expect the small sample of attorneys fees available to us to be a non-random sample of attorneys fees in patent disputes, differences in attorneys' fees within the cases in which attorneys' fees were awarded (and thus observable) provide at least some validation of the measures as proxies for costs. IV. How are Litigation Costs Distributed Over Case Progression? In this section, we study how litigation costs are distributed over case progression. To do so, we employ event studies for litigation costs around each litigation milestone. For instance, for cases in which an answer is filed, we define event time 0 as the month in which the answer was filed. This aligns cases according to event time. We then calculate the aggregate number of cost measures that takes place in each event time at the month level. It is important to note that the event studies here show overall cost of parties and courts for all parties in which are pending in 16

18 the months before and after the litigation milestone. This implies that changes in cost in the event studies capture lower cost resulting from cases being resolved. This is precisely the hypothesis we study in the next section, where we examine how cost around the litigation milestones differs for cases which are resolved at different stages of litigation. Here we simply note that the event studies in this section present important aggregate cost measures around the milestones and abstract from heterogeneity in costs between cases. The first panel of Figure 5 shows how litigation costs are distributed around answer for the 79% of cases in which an answer is filed. The left hand y-axis indicates the number of docket entries in the case, and the right hand y-axis indicates the number of other three cost measures in the case. This same convention is maintained in the other panels described below. Consistent with the startup work that a court must do on the case, before an answer parties do not appear to have much cost, and the court does not hold many hearings. Costs increase the month of the answer and in the months following, and then decrease over time. Overall, each of the cost measures follow the same basic pattern. One interpretation of the spike in costs is that an answer forces the parties to engage the case, albeit for a short period of time, and make a decision whether to proceed. The party motion/memorandum filings decrease more slowly than the judicial order/opinion decrease, but faster than hearings. This is consistent with the court entering a variety of ministerial orders at the beginning of the case. The slight rise in costs in the months following an answer is likely due to the coordination of the case management conference, which occurs after an average of about one month later for cases which have a case management order (more than 50% of cases that progress to case management reach it less than one month after answer). [Insert Figure 5 about here] 17

19 The second panel of Figure 5 plots the event study for cases that reach case management. Similar to answer, there are still many more judicial orders/opinions than party motions/memos around case management, and many more party motions/memos than hearings. This is also consistent with the court still having a lot of startup work to do on the case. Because discovery does not begin until after case management, the parties are not filing many motions or memos. The same basic cost pattern exists following case management as with answers, which is to be expected because the answer and case management are typically somewhat close in time. There is one exception, however. Instead of the short continued increase in costs after case management, there is a substantial decrease in judicial orders/opinions and hearings in the month after, on the magnitude of one third. This is consistent with the court having completed the initial ministerial aspects of the case. One possible explanation is that case management triggers the beginning of the formal discovery process. Formal discovery is both expensive and likely to lead to many arguments between the parties as to production of documents. That party motion/memos increase around case management and decreases only slightly over the next 3 years is consistent with a period of continuous disputes and arguments about the case. From this view, case management does not signal a drop off in party work. Instead, it signals the beginning of a robust period of party work that is spread out over time. The third panel of Figure 5 plots the event study for cases that reached a claim construction ruling. For cases that reach a claim construction ruling, the relative costs by party and court has changed from earlier stages. Relative to the makeup of the cost measures around answer and case management, the costs of parties compared to courts is higher. This is true both in the months before and after a claim construction ruling, which is consistent with how parties exert 18

20 considerable effort with discovery disputes and substantive disagreements on the merits. Another interesting aspect of the panel is that the distributions appear to be almost normal. Considering that claim construction occurs late in the case, far past the median case duration of 11 months, one might have expected a left skewed distribution. On the other hand, selection of cases which reach claim construction (10%) means that the sample of cases in Figure 5 is very different than the cases which do not reach claim construction. The final panel of Figure 5 plots the event study for cases that began trial. The relative party costs are high, presumably because the parties are bringing last minute arguments to the court (e.g., requests to exclude evidence) and organizing for trial (e.g., providing trial exhibits, witness lists). There are many hearings in the month trial begins, which is consistent with the court ruling on a variety of evidentiary issues and gearing up for trial. As expected, costs decrease rapidly following trial. The synchronous nature of party and judicial costs at trial is consistent with the court ruling more expeditiously on party motions during trial. IV. The Endogenous Sequence of Litigation Cost Hypothesis In this section, we exploit the litigation milestones and the panel of litigation costs to offer novel evidence on the endogenous litigation costs hypothesis. In particular, we investigate how litigation costs differentially accumulate around milestones for cases that reach various future stages of litigation, therefore using whether later stages are reached to distinguish between different case types. To do so, we stratify the event studies for docket entries by cases that reach different milestones. With the goal of comparing cost around milestones for different types of cases, we construct a normalized cost measure such that the cost occurring in the month of the milestone is 19

21 equal to one for each of the different types of cases. This allows one to compare the cost of different case types before and after a milestone. Figure 6 presents the results. The first panel plots the normalized docket entry measure for cases where an answer was filed, separately by cases which only reached answer, cases which reached case management but no further, cases which reached claim construction but not trial, and cases that began trial. For two to three months after answer, all four types of cases follow the same path and appear relatively indistinguishable. Thereafter, cost in the types of cases differs. Cost for cases that resolve after answer but before case management smoothly decrease to zero activity; cost for cases that resolve after case management but before claim construction decrease more slowly but still continuously. In contrast, the cases that reach claim construction but resolve before trial have increased activity for approximately a year before tapering off. The cases that reach trial have increased costs for several years, and costs are higher over the whole case than that of cases that resolve after claim construction. It appears that, even in the early stages of litigation, there are endogenous differences between cases that will and will not proceed to trial. The differences in cost in the few months after answer between cases that only reach claim construction and cases that begin trial provide evidence that the sequence of litigation costs is endogenous. [Insert Figure 6 about here] The second panel of Figure 6 plots the normalized docket entry measure for cases which reached case management, separately for cases which only reached answer and case management, cases which reached claim construction but not trial, and cases that began trial. In this plot, the distribution of cases that reached only case management continuously decreases to zero over time. 20

22 Cases that reached later milestones appear indistinguishable from each other for six to nine months. At that point, costs for cases that proceed to trial increase relative to the cases that merely reach case construction. The third panel of Figure 6 plots the normalized docket entry measure for cases which reached claim construction, separately for cases that did and did not begin trial. Cost in the cases that go to trial are higher in the months following the claim construction ruling, but the difference in cost after a claim construction ruling is much smaller than differences in cost in different case types following answer and case management. Thus, the endogenous differences between cases that will and will not proceed to trial are more muted once a case has reached claim construction. That costs around a milestone differ between cases that will or will not reach future milestones provides evidence that the timing of litigation costs is endogenous. Moreover, the endogenous differences in the early stages of litigation also provides some evidence that the milestones we set forth are valid metrics of case progression and contain important information regarding case type. VI. Conclusion In this article, we have documented the costs of litigants and courts throughout the course of legal disputes, and investigated the endogenous litigation cost hypothesis. In particular, we have documented the progression and costs of litigation for recent patent disputes in the U.S. courts using novel administrative data on around 1.5 million litigation docket entries for the 16,116 cases filed between 2005 and We established measures for important litigation milestones and a panel of litigation cost measures. We used the milestones and cost measures to provide a comprehensive assessment of the litigation process. We found that private and public litigation 21

23 costs increase around litigation milestones. Finally, we explored the endogenous litigation cost hypothesis, and found that costs around litigation milestones differ based on the final milestone a particular case reaches. It is worth noting a few limitations of the results. First, cases in which there are higher potential levels of liability and more uncertain outcomes are likely to have more and a different makeup of cost. Even though our validation exercise suggested that the measures are a good proxy for cost overall, we did not observe the timing of the costs and therefore cannot rule out differences in costs between the measures at different points of litigation. For instance, we cannot rule out whether a motion is more costly during trial than before discovery. Second, the proxies for party costs that we employ are limited to items that appear in the court docket. We recognize that there are other important costs, including much of those related to discovery, will not necessarily be reflected in the docket. However, the number of discovery related disputes that we observe in the docket are likely correlated with the scope of discovery, and are therefore reflected in our cost measures. This is consistent with the results of our validation exercise, which used total attorneys' fees including the costs related to discovery. A third limitation is that the results might be limited to patent cases. Patent litigation is likely to differ from other types of litigation, especially less complex areas of litigation. However, patent litigation is important in its own right, and one might expect that the general findings are not unique to patent litigation. Moreover, the novel cost measures we established here can be used in investigations of other types of civil litigation as well. A final limitation is the applicability of patent cases from 2005 to 2010 to current patent litigation. In many areas of law, one may expect lawsuits filed today to be resolved similarly to lawsuits filed five to ten years ago. However, patent litigation may be different. Several major 22

24 changes have occurred in the last five years, including the rise of Inter Partes Review that is concurrent with much patent litigation, and the Supreme Court decision in Alice Corp. v. CLS Bank, 134 S.Ct (2014). Furthermore, the America Invents Act requires that lawsuits filed against multiple unrelated parties are filed separately. 35 U.S.C. 299 (2012). For example, in 2010, while a patentee could sue three defendants in one patent lawsuit in some venues, after the implementation of the American Invents Act, the same patentee may have to sue each defendant separately, resulting in three patent lawsuits. With these caveats, our findings shed new light on the litigation process and the endogenous litigation costs hypothesis. This was possible because we exploited a novel data source, and introduced new measures of litigation costs and case progression. We feel that our use of the cost measures and litigation milestones that we established here broke ground for these measures, and hope that these measures are employed in future work. We also hope future research will build on our effects of parsing the administrative records of lawsuits from federal courts to answer a wide array of questions about different types of disputes and about the litigation process more generally. 23

25 References Bebchuk, L. A. (1996). A new theory concerning the credibility and success of threats to sue. Journal of Legal Studies 25(1), Boyd, C. L. and D. A. Hoffman (2013). Litigating Toward Settlement. Journal of Law, Economics, and Organization 29(4), Cecil, J., G. Cort, M. Williams, and J. Bataillon (2011). Motions To Dismiss For Failure To State A Claim After Iqbal. Federal Judicial Center. Cecil, J., G. Cort, M. Williams, J. Bataillon, and J. Campbell (2011). Update On Resolution Of Rule 12(B)(6) Motions Granted With Leave To Amend. Federal Judicial Center. Clermont, K. and T. Eisenberg (2014). Plaintiphobia in the Supreme Court. Cornell Law Review 100, 193. Farmer, A. and P. Pecorino (2013). Discovery and Disclosure with Asymmetric Information and Endogenous Expenditure at Trial. Journal of Legal Studies 42(1), Fournier, G. M. and T. W. Zuehlke (1996). The Timing of Out-of-Court Settlements. RAND Journal of Economics 27(2), Galanter, M. (2004). The Vanishing Trial: An Examination of Trials and Related Matters in Federal and State Courts. Journal of Empirical Legal Studies 1(3), Gelbach, J. (2012). Locking the Doors to Discovery? Assessing the Effects of Twombly and Iqbal on Access to Discovery. Yale Law Journal 121, Hannon, K. (2008). Note, Much Ado About Twombly? A Study of the Impact of Bell Atlantic Corp. v. Twombly on 12(b)(6) Motions. Notre Dame Law Review 83, Hatamyar, P. (2010). The Tao of Pleading: Do Twombly and Iqbal Matter Empirically? American University Law Review 59, 553. Hatamyar Moore, P. (2012). An Updated Quantitative Study of Iqbal s Impact on 12(b)(6) Motions. University of Richmond Law Review 46, 603. Hay, B. (1995). Effort, Information, Settlement, Trial. Journal of Legal Studies 24(1), Henry, M. and J. Turner (2006). The Court of Appeals for the Federal Circuit s Impact on Patent Litigation. Journal of Legal Studies 35(1), Hoffman, D. A., A. Izenman, and J. Lidicker (2008). Docketology, District Courts, and Doctrine. Washington University Law Review 85(4), Huang, K.-C. (2008). How Legal Representation Affects Case Outcomes: An Empirical Perspective from Taiwan. Journal of Empirical Legal Studies 5(2),

26 Huang, K.-C. (2009). Does Discovery Promote Settlement? An Empirical Answer. Journal of Empirical Legal Studies 6(2), Hubbard, W. H. (2015). The Empirical Effects of Twombly and Iqbal. Working Paper. Hubbard, W. H. (2016). Sinking Costs to Force or Deter Settlement. Journal of Law, Economics, and Organization. Hyman, D., M. Rahmati, B. Black, and C. Silver (2016). Medical Malpractice Litigation and the Market for Plaintiff-Side Representation: Evidence from Illinois. Journal of Empirical Legal Studies 13, Isaacharoff, S. and G. Miller (2013). An Information-Forcing Approach to the Motion to Dismiss. Journal of Legal Analysis 5, 439. Johnston, J. S. and J. Waldfogel (2002). Does Repeat Play Elicit Cooperation? Evidence from Federal Civil Litigation. Journal of Legal Studies 31(1), Katz, A. (1988). Judicial Decisionmaking and Litigation Expenditure. International Review of Law and Economics 8, Kessler, D. (1996). Institutional Causes of Delay in the Settlement of Legal Disputes. Journal of Law, Economics, and Organization 12(2), Landes, W. M. (1971). An Economic Analysis of the Courts. Journal of Law and Economics 14(1), MacKenzie, I. (2016). Precaution with Endogenous Litigation Choices. Working Paper. Marco, A., S. Miller, and T. Sichelman (2015). Do Economic Downturns Dampen Patent Litigation? Journal of Empirical Legal Studies 12(3), Paik, M., B. Black, D. Hyman, W. Sage, and C. Silver (2012). How Do the Elderly Fare in Medical Malpractice Litigation, Before and After Tort Reform? Evidence from Texas. American Law and Economics Review 14(2), Posner, R. A. (1973). An Economic Approach to Legal Procedure and Judicial Administration. Journal of Legal Studies 2(2), Priest, G. and B. Klein (1984). The Selection of Disputes for Litigation. Journal of Legal Studies 13(1), Robert D. Cooter, D. L. R. (1989). Economic Analysis of Legal Disputes and Their Resolution. Journal of Economic Literature 27(3), Seiner, J. (2009). The Trouble with Twombly: A Proposed Plead-ing Standard for Employment Discrimination Cases. pp Shavell, S. (1982). Suit, Settlement, and Trial: A Theoretical Analysis under Alternative Methods for the Allocation of Legal Costs. Journal of Legal Studies 11(1),