Patent Trolls and Small Business Employment Ian Appel† , Joan Farre-Mensa‡ and Elena Simintzi∗ July 21, 2017

Abstract We analyze how frivolous patent-infringement claims made by non-practicing entities (NPEs, or “patent trolls”) affect small firms’ ability to grow, create jobs, and raise capital. Our identification strategy exploits the adoption of anti-troll laws in 32 US states. We find the adoption of these laws leads to a 2% increase in employment at small high-tech firms—an increase driven by IT firms, a frequent target of NPEs. By contrast, the laws have no significant impact on employment at larger or non-high-tech firms. Financing is a key channel driving our findings: In states with an established VC presence, antitroll laws increase the number of firms receiving VC funding by 14%. Our findings suggest measures aimed at curbing the threat posed by NPEs can help reduce both the real and financing frictions faced by small firms.

Keywords: employment, patent trolls, NPEs, venture capital. Affiliations: † Carroll School of Management, Boston College; ‡ Cornerstone Research; ∗ Sauder School of Business, University of British Columbia. e-mails: [email protected], [email protected], [email protected]. Acknowledgments: We would like to thank Manuel Adelino, Shai Bernstein, Alan Crane, Slava Fos, Will Gornall, Kai Li, William Mann, Jordan Nickerson as well as audiences at the ASU Sonoran Winter Finance Conference, FMA Napa Conference on Financial Markets Research, USC Private Equity Conference, Western Finance Association, SFS Cavalcade, Boston College, Georgia State University, and UBC for helpful comments.

I

Introduction

Small businesses are a key economic engine, but they are hindered by a number of frictions that limit their ability to grow, create jobs, and innovate (Himmelberg and Petersen 1994; Carpenter and Petersen 2002; Kerr and Nanda 2011). One friction that has received considerable attention in recent years is non-practicing entities (NPEs, also known as “patent trolls”), organizations that own patents but do not create or use the patented technology. Patent litigation has increased tenfold since 2000, and in 2015 NPEs accounted for 69% of these lawsuits (RPX 2015). Yet, survey evidence suggests that those cases that end up in court may be a small fraction of the instances in which NPEs target firms, particularly small ones, with infringement claims (Chien 2014). Indeed, a common practice used by NPEs is to send “demand letters” with often dubious patent-infringement claims to thousands of mostly small firms, offering them licensing deals to avoid litigation (AIPLA 2013).1 Examples abound of the potential negative consequences of NPEs on employment and firm growth. For instance, in 2013, the Washington Post quoted the CFO of a Vermont technology company as saying: “We had two projects that we were close to doing. Then our clients contacted us saying, ‘We’re being threatened by patent trolls’.” As a result, “both projects were cancelled, nixing the firm’s plans to hire several new employees.”2 Motivated by these concerns, several bills have been introduced in Congress since 2012 to limit the activities of NPEs. However, as of today, none of them have become law. As Cohen, Gurun, and Kominers (2015) note, a common concern among those opposing restrictions on NPEs is that while some NPEs may use abusive tactics, others “serve a key financial intermediary 1

For example, one of the most notorious NPEs, MPHJ, sent demand letters to over 16,000 small firms between 2012 and 2013, but never filed a single lawsuit (FTC 2015). 2

“How Vermont could save the nation from patent trolls” (The Washington Post, Aug. 1, 2013).

–1–

role, policing infringement by well-funded firms that could otherwise infringe upon small inventors’ intellectual property at will” (p. 1). In response to the lack of congressional action, state legislatures, beginning with Vermont in 2013, have passed laws to limit NPEs’ ability to target local firms with “bad faith assertions of patent infringement” via demand letters (Vermont Act 44). Through the first half of 2016, 32 states had passed such laws. In this paper, we exploit the adoption of the laws in a difference-in-differences framework to analyze how bad-faith patent infringement claims made by NPEs affect small firms’ ability to grow, create jobs, and raise capital.3 We show that the adoption of state anti-troll legislation leads to an average 2% increase in employment at small high-tech firms, which are precisely the firms that NPEs tend to target (AIPLA 2013). This increase is mainly driven by firms in the information technology (IT) industry, where NPEs tend to be most active (Chien 2014). By contrast, we find no effect for larger high-tech firms (those with more than 20 employees) or for non-hightech firms regardless of their size. The employment increase at small high-tech firms is strongest for college-educated workers, though we also find somewhat weaker evidence of an employment increase for workers without a college degree. These findings suggest that the benefits of curbing NPEs’ abusive activities can be felt across the skill distribution. While the recentness of the anti-troll laws limits our ability to capture their long-term effects, we find no evidence that their effect on employment is short-lived—if anything, it appears to increase over time. Empirically identifying the effects of policy changes such as the adoption of anti-troll legislation is plagued with endogeneity concerns. At the heart of these concerns is the 3

Henceforth, we refer to the laws as “anti-troll laws” to reflect the fact that their stated goal is to curb the most abusive activities of NPEs.

–2–

possibility that there could be confounding variables, such as macroeconomic shocks, that may affect both a state’s decision to pass anti-troll legislation and the hiring decisions of firms in the state. Our diff-in-diff identification strategy helps overcome this challenge by exploiting the fact that not all US states have adopted anti-troll laws; in addition, among those states that have adopted them, not all of them did so at the same time. We can thus construct a time-varying control group that provides a counterfactual for how employment would have evolved in the treated states had they not adopted anti-troll legislation. For our control group to be a valid counterfactual, the evolutions of employment for treated and control states need to share parallel trends. While the parallel-trends assumption is ultimately untestable, the following facts are consistent with it. First, we find that employment trends for treated and control states are indistinguishable prior to the adoption of the laws. Second, the fact that the effect of anti-troll laws in the treated states is concentrated among small high-tech firms in the IT industry, which tend to be the focus of NPEs’ frivolous demand letters, is consistent with us capturing the effect of the laws and not some other state-level shock that would affect a broader set of firms. Third, our analysis controls for standard state-level macroeconomic variables: state aggregate and per-capita income, unemployment, and patenting volume. These macroeconomic variables are in addition to the state and year-quarter fixed effects that we include in all specifications, which account for time-invariant differences across states and time-varying shocks affecting all states, respectively. Fourth, our results are robust to including interacted region and year-quarter fixed effects, which absorb regional macroeconomics shocks. Fifth, if we estimate placebo diff-in-diffs in neighboring states or pre-event years, the estimated placebo treatment effect is indistinguishable from zero. Another potential concern is that the passage of the anti-troll laws may have coincided

–3–

with other legislation or policies also aimed at promoting the growth of small high-tech firms. To address this concern, we collect information on contemporaneous state policies aimed at fostering high-tech employment. All our conclusions are robust to controlling for such policies, thereby assuaging concerns that our results are driven by these potential confounds. In addition, we obtain information on the motives that drove the adoption of the state anti-troll laws from a variety of sources, including conversations with several legislators that sponsored the bills. We do not find instances in which the laws were passed as part of a broader legislative package aimed at promoting the local economy. Finally, we present evidence that the adoption of the laws is associated with a significant reduction in NPE-related Google searches—a prediction that is likely unique to anti-troll laws and not shared by potential confounds. This finding further reinforces the notion that the state anti-troll laws were effective in shielding local firms from NPEs. What are the likely mechanisms driving the positive effect that state anti-troll laws have on small high-tech firms’ ability to create jobs and grow? By limiting the instances in which NPEs target small firms with frivolous patent infringement claims, anti-troll laws allow firms to avoid having to devote time and money to fight these claims. Perhaps even more importantly, the laws may also affect firms’ ability to raise capital by reducing the risk that any small high-tech firm in a state faces of being targeted by NPEs, thereby increasing the net present value (NPV) of investing in these firms. We test this financing channel by analyzing the effect of the laws on venture capital (VC) investment. In line with our hypothesis, we find that, in states with above-median VC presence at the beginning of our sample period, anti-troll laws lead to a 14% increase in the number of firms raising VC funding—an effect that again appears to be driven by IT

–4–

firms and is negligible for life sciences firms.4 Further reinforcing the importance of this financing channel, we find that the positive effect of anti-troll laws on employment for small high-tech firms is driven by states with above-median VC presence. We also find a positive effect on the number of high-tech establishments in these states, suggesting that the observed growth in employment is driven both by the growth of existing establishments and by the creation of new ones. Finally, we show that the effect of the laws is strongest for firms in equity-dependent industries (Rajan and Zingales 1998), which are those that are most likely to benefit from increased access to equity capital. Our study joins a small but growing literature that analyzes the economic consequences of NPEs. Cohen, Gurun, and Kominers (2015) show that, when suing public firms, NPEs tend to target those that are cash rich, leading to a substantial decrease in their innovative activities. Tucker (2014) and Smeets (2015) also find a decline in innovation at firms sued by NPEs. Our study differs from previous papers in two distinct ways. First, prior work has focused on analyzing the effects of being sued by an NPE. Lawsuits are indeed the main tool NPEs have when targeting large, public firms (as in in Cohen et al.’s and Smeets’s samples). Crucially, our findings indicate that the simple threat of litigation expressed via the frivolous demand letters that anti-troll laws aim to curtail can disrupt the operations of small high-tech firms. Second, prior studies examine the consequences for firms that have been directly targeted by NPEs. By contrast, our empirical setting captures both the realized and expected effects of frivolous NPE demands by analyzing legal changes that reduce the risk any firm in a state faces of being targeted by such demands. 4

By contrast, we find that the laws have no significant effect on VC investments in those states with low VC activity to begin with, which suggests that the adoption of anti-troll laws, on its own, is not sufficient to attract VCs to states with historically limited VC presence.

–5–

Our paper provides the first analysis of the effects of state anti-troll laws. The adoption of these laws has been surrounded by an intense debate on whether limiting the activities of NPEs may do more harm than good by hindering their ability to help small innovative firms monetize their inventions.5 Moreover, a number of commentators and scholars have expressed doubts regarding whether states should legislate in an area, intellectual property protection, that has long been considered a federal matter (e.g., DeSisto 2015). Our results indicate that state anti-troll laws have had a net positive effect for small firms in high-tech industries, helping them create jobs and making them more attractive to VC investors. These findings suggest that the laws can have a multiplier effect by not only decreasing the resources that small firms need to spend responding to NPE demands, but by also facilitating their access to the funding, monitoring, and networks provided by VCs (Hellmann and Puri 2002; Hochberg, Ljungqvist, and Lu 2007; Bernstein, Giroud, and Townsend 2015). At the same time, the fact that state anti-troll laws target a particularly egregious type of behavior by NPEs—frivolous patent demand letters—suggests we should exercise caution in interpreting our findings as applying to the whole NPE class. Not all NPEs engage in the type of large-scale, “spray and pray” campaigns that the laws target. In particular, the anti-troll laws we study are not intended to curb the activities of those NPEs that follow a “big fish” business model focused on assembling large patent portfolios to sue big, deep-pocketed firms, as courts are unlikely to rule that these claims are made in bad faith. Our paper also contributes to the literature in finance and economics studying the effects of frictions that hinder firms’ ability to create jobs. Several papers have analyzed how 5

See, e.g., “Not So Scary, After All: In Defense Of Patent Trolls” (Forbes, Feb. 1, 2013), or “The Myth of the Wicked Patent Troll” (The Wall Street Journal, June 30, 2014). As we explain in the next section, this debate is also a key reason why some large states like California have not passed an anti-troll law.

–6–

frictions in the capital markets affect firm employment growth (e.g., Benmelech, Bergman, and Seru 2011; Chodorow-Reich 2014; Adelino, Schoar, and Severino 2015; Giroud and Mueller 2015, 2016; Benmelech, Frydman, and Papanikolaou 2016; Michaels, Page, and Whited 2016). Our study focuses on a different type of friction, showing that frivolous patent-infringement claims can limit small high-tech firms’ ability to grow and create jobs. The remainder of the paper is organized as follows. Section II discusses the institutional background of state anti-troll laws. Section III summarizes the data and describes our methodology. Section IV presents the baseline results as well as our identification tests. Section V discusses the mechanisms that may explain our findings. Section VI concludes.

II II.1

Institutional Background NPEs’ patent-infringement claims

Recent years have seen both an increase in patent litigation and in the share of patent cases brought by NPEs, which has increased from 35% in 2010 to 69% in 2015 (RPX 2015; CEA 2016). Litigation, however, is not the only way patent-infringement allegations can be made. Instead of directly suing an alleged infringer, a patent owner often first sends a “demand letter.” Common purposes of such letters include to inquire whether a product uses a particular patent, make a licensing offer, and/or threaten litigation unless the alleged infringement stops or a royalty is paid (AIPLA 2013). Demand letters are thus “essential enforcement tools for patent owners, and they allow many patent disputes to be resolved long before court intervention is necessary;” however, in recent years a number of NPEs appear to have abused this tool by “sending thousands of demand letters to unknowing end users of allegedly patented products, using the recipient’s lack of experience with the patent

–7–

system to coerce them into paying settlements” (AIPLA 2013; p. 18). One notorious case is MPHJ, which between September 2012 and May 2013 sent demand letters to 16,465 small businesses in an attempt to collect license fees on its scanner patents, threatening the firms with litigation that it actually never pursued (FTC 2015). Another example is Lodsys, which between 2011 and 2013 threatened to sue thousands of iOS application developers. While Lodsys did sue some large firms, most of its letters were directed to small app developers who lacked the resources to defend patent litigation.6 Systematic data on the number of businesses targeted by NPEs via demand letters are not available.7 However, survey evidence indicates that, particularly among small firms, the number of targets is substantially larger than the number of firms that are sued in court. In a survey of high-tech firms, Chien (2014) finds that lawsuits represented only 31% of patentinfringement demands received by companies with under $10 million in revenue, while they represented 67% of demands received by companies with over $10 million in revenue.8 Accordingly, Scott Morton and Shapiro (2014) write that patent lawsuits “are just the tip of the iceberg. Surely, there are far more patent assertions than actual litigations, and these assertions impose various costs on targets, including legal expenses, design-around costs, and settlement costs” (p. 469). Chien’s (2014) survey shows that these costs can be substantial, particularly for small 6

See Figure IA.1 in the Internet Appendix for a sample demand letter sent by Lodsys. The website https://trollingeffects.org/letters contains examples of a number of other seemingly abusive demand letters sent by NPEs. 7 Daniel Nazer from the Electronic Frontier Foundation wrote to us in an email: “We’ve been disappointed with the low number of submissions of letters to trollingeffects.org. We learned that companies are generally quite nervous about sharing information about demand letters.” 8

This evidence suggests that studies that use litigation data to analyze the effects of NPEs on publicly-listed firms (e.g., Cohen, Gurn, and Kominers 2015) capture the vast majority of instances in which NPEs target a firm of the size of a public firm. Yet litigation data provide a far less complete picture of the effects of NPEs on small and medium-sized firms.

–8–

firms. Among the high-tech firms in her sample targeted by an NPE, the most common response was to fight the infringement allegations out of court. This strategy, which was followed by 24% of the targeted firms, resulted in average expenses of $168,000, or 5% of annual revenue on average. For the 18% of firms that ended up entering into a settlement, the expenses increased to $340,000 (13% of annual revenue), while the 11% of firms that fought the demand in court faced average expenses of $857,000 (24% of annual revenue).9 In addition to monetary costs, 73% of high-tech firms reported that dealing with the NPEs’ demands had required founder time, and 89% acknowledged that doing so had been a distraction from their core business. The end result of these monetary and time costs was that 41% of firms targeted by NPEs reported that the demands had a significant impact on their operations, including delays in hiring or in meeting operational milestones, loss of firm value, and even having to exit a business line or the business altogether. Not surprisingly, the operational impacts were concentrated among the smaller firms. The potentially large costs associated with NPEs have led a growing number of scholars and public commentators to urge policymakers to act to curtail NPEs’ activities (e.g., Asay et al. 2015). In the US, patent law generally falls under the purview of the federal government. The latest patent reform law passed by Congress, the America Invents Act (AIA) of 2011, includes a provision intended to curb abusive patent litigation by making it more difficult to sue multiple defendants in the same patent-infringement suit (Bryant 2012). However, evidence suggests the AIA has had limited effect on NPE behavior (CEA 2016). As a result, Congress has subsequently considered several pieces of legislation aimed 9

The remaining firms did nothing (22% of targeted firms), conducted a product or business change (9%), or had some other response (17%).

–9–

at further restricting NPEs, and in particular their ability to send abusive demand letters.10 Yet, as of this writing, none of them has become law.

II.2

State anti-troll laws

In response to concerns that the AIA does not sufficiently protect firms from frivolous patent-infringement claims, a number of state legislatures have taken action. Beginning with Vermont in 2013, states have adopted patent reforms that protect local businesses from bad-faith infringement claims. These anti-troll laws have been framed as consumer protection laws, thereby sidestepping the fact that most aspects of patent law are considered a federal matter.11 The stated goal of these state-level anti-troll laws is to reduce the costs that abusive patent claims impose on the state economies. For example, the Vermont law notes:

Abusive patent litigation, and especially the assertion of bad faith infringement claims, can harm Vermont companies. A business that receives a letter asserting such claims faces the threat of expensive and protracted litigation and may feel that it has no choice but to settle and to pay a licensing fee, even if the claim is meritless. This is especially so for small and medium sized companies and nonprofits that lack the resources to investigate and defend themselves against 10

These include the Targeting Rogue and Opaque Letters (TROL) Act (H.R. 2045), the Patent Transparency and Improvements Act (S. 1720), the Saving High-tech Innovators from Egregious Legal Disputes (SHIELD) Act (H.R. 845), the Innovation Act (H.R. 3309), the Stopping the Offensive Use of Patents (STOP) Act (H.R. 2766), the Transparency in Assertion of Patents Act (S. 2049), and the Demand Letter Transparency Act (H.R. 1896). 11

DeSisto (2015) writes about the Vermont law: “The Act attempts to apply consumer protection law to the methods of patent trolls and to avoid an evaluation of the underlying patents by focusing on one of the patent trolls’ favorite and most effective weapons: demand letters. By avoiding the evaluation of patents completely, and instead simply laying out a number of factors courts may consider when deciding whether a demand letter to an alleged patent infringer was sent in bad faith, the Act successfully dodges intrusion upon Congress’s exclusive [patent] jurisdiction” (p. 126).

– 10 –

infringement claims. Not only do bad faith patent infringement claims impose a significant burden on individual Vermont businesses, they also undermine Vermont’s efforts to attract and nurture small and medium sized IT and other knowledge based companies. Funds used to avoid the threat of bad faith litigation are no longer available to invest, produce new products, expand, or hire new workers, thereby harming Vermont’s economy. The Vermont law “seeks to change the calculations of patent trolls in Vermont by increasing the potential costs of sending out mass demand letters.”12 To this end, courts can consider “whether the letter had the required information, requested an unreasonable license fee, or demanded payment in an unreasonably short period of time” in deciding whether a patent demand letter was send in bad faith (DeSisto 2015; p. 124-125). If such a determination is reached, the court can compel the NPE to post bond equal to the target’s expected litigation costs. In addition, the law establishes that if a court finds that a Vermont firm has been the target13 of bad-faith patent infringement assertions, then the court may award it the following remedies: “(1) equitable relief; (2) damages; (3) costs and fees, including reasonable attorney’s fees; and (4) exemplary damages in an amount equal to $50,000.00 or three times the total of damages, costs, and fees, whichever is greater.” In order to minimize the burden imposed on firms that are targeted by NPEs, the law allows 12

Some commentators have questioned whether the anti-troll laws may infringe on NPEs’ freedom of expression and thus be unconstitutional. Importantly, as long as the laws are in force, they increase the “expected costs of sending out mass demand letters,” even if there is a chance that someday the laws are found unconstitutional. 13

The law defines a “target” as a Vermont (physical or legal) person: “(A) who has received a demand letter or against whom an assertion or allegation of patent infringement has been made; (B) who has been threatened with litigation or against whom a lawsuit has been filed alleging patent infringement; or (C) whose customers have received a demand letter asserting that the person’s product, service, or technology has infringed a patent.”

– 11 –

the state’s Attorney General to initiate the legal actions against abusive NPEs. The Vermont law has served as a model for other states, and 32 states had passed anti-troll laws through the first half of 2016. Table IA.1 in the Internet Appendix lists the states that adopted anti-troll laws during our sample along with their signing dates. The non-partisan nature of the laws (the Vermont law was sponsored by a Democratic and a Republican state representative) has likely contributed to their rapid expansion. The statutes passed by these states share two critical components: (1) they aim to curtail badfaith patent-infringement assertions made via demand letters by allowing courts to impose penalties on the senders of such letters; and (2) they cover any target firm located in the state, regardless of where the firm is incorporated or where the sender of the letter is located.

II.3

Political economy of the anti-troll laws

Various constituencies lobbied for the passage of the anti-troll laws in the different states. In Vermont, the push for patent reform was led by a tech firm with around 100 employees that eventually started a grassroots coalition with other local firms.14 Similarly, in Alabama, the sponsoring legislator was initially approached by the owner of a small engineering firm who had received a demand letter for the use of software even though her firm had a license for it. Further investigation into the matter led legislators to the conclusion that more small businesses in the state were being targeted by trolls, and they also became aware of similar legislation passed in other states.15 In several other states, the push for anti-troll laws was initiated by financial institutions. For example, in Georgia the push for the law was led by the Georgia Bankers Association af14

“How Vermont could save the nation from patent trolls” (The Washington Post, Aug. 1, 2013).

15

Correspondence with State Senator Arthur Orr, sponsor of Alabama’s anti-troll law.

– 12 –

ter a number of its member banks were sent demand letters for alleged patent-infringement by their ATMs. This effort was later joined by other groups, including the Printing Industries Association and the Georgia Chamber of Commerce.16 Likewise, the Maryland Bankers Association played a key role in helping pass anti-troll legislation in that state.17 Overall, our investigation of the political economy surrounding the passage of the state anti-troll laws suggests that lobbying for the laws was initiated by either a single firm (as in Vermont and Alabama) or by a non-high-tech industry group (as in the case of the banking associations in Georgia and Maryland), thereby mitigating reverse-causality concerns. While the anti-troll laws have spread quickly, the states that have yet to pass anti-troll legislation include some of the largest and most innovative states, most notably California. In fact, an anti-troll law was introduced in the California State Senate in February 2015 (S.B. 681). However, the bill was not passed despite having the support of the Silicon Valley Leadership Group, which includes leading tech companies and investors such as Facebook, Google, HP, Kleiner Perkins Caufield & Bryers, Silicon Valley Bank, SV Angel, and Tesla Motors. The office of State Senator Hill, who introduced S.B. 681, wrote to us in an email that “the original bill had support and no opposition but the Chair of the Senate Judiciary Committee wanted amendments that our supporters didn’t like.” Ultimately, legislators could not agree on a text that all key senators thought would deter “fraudulent patent infringement claim letters” without having “a chilling effect on legitimate communications between small innovators trying to engage in coordination, development and licensing communications with larger businesses.” Similar disagreements have been at the root of Congress’s inability to pass analogous anti-troll legislation at the federal level. 16

Correspondence with State Representative Bruce Williamson.

17

Correspondence with the office of State Senator Middleton.

– 13 –

III III.1

Data and Methodology Data

We obtain state-level employment data from the Census’ Quarterly Workforce Indicators (QWI) database. Employment is measured at the end of each quarter (EmpEnd). For each state-quarter, the QWI reports employment data aggregated by firm size and 4-digit NAICS industry. The QWI reports data for five size categories (in employees): 0-19, 20-49, 50-249, 250-499, and 500+. For brevity, we collapse the three largest size categories into a single group, and report results for small (<20 employees), medium (20-50), and large (50+) firms (our results remain qualitatively the same if we instead use the original five categories). Because NPEs tend to target firms in high-tech industries (AIPLA 2013; Chien 2014), our analyses focus on employment in high-tech industries, which we identify following Kile and Phillips (2009).18 In addition, we use demographic information from the QWI to analyze how the effect of anti-troll laws on employment depends on the workers’ education level. Our employment analysis sample begins in 2012Q1, five quarters before the adoption of the first anti-troll law in Vermont, and ends in 2016Q2, the last quarter for which QWI data are available. We obtain establishment count data from the Quarterly Census of Employment and Wages (QCEW) database. The data are available at the industry-state-year-quarter level, and we again distinguish between high-tech and non-high-tech firms. Venture capital (VC) data are from Thomson’s VentureXpert database, and we focus on early-stage VC rounds raised by high-tech firms. Our samples for the establishment and venture capital analyses 18

Specifically, the following 4-digit NAICS industries are classified as high-tech: 3254, 3341, 3342, 3344, 3345, 3346, 3353, 3391, 5112, 5141, 5171, 5172, 5179, 5182, 5191, 5413, 5415, 5416, 5417. Firms in 3254, 3391, and 5417 are classified as life sciences firms, and the rest as IT.

– 14 –

also end in 2016Q2. We collect data on NPE-related internet searches using Google’s search volume index (SVI). Specifically, we focus on searches for the terms “patent trolls” or “patent attorney” at the state-year level over the 2012-2016 period. We work with annual-level data because the data required for our analysis are not available at higher frequencies. This variable is scaled from 0 to 100 based on the total search volume in each state. When an observation is missing due to the low number of searches for our focus terms, we set it to zero. We present our results with and without controlling for the following macroeconomic variables: state-level quarterly gross state product (GSP) and income per capita, both logtransformed, from the Bureau of Economic Analysis; state-level quarterly unemployment rate, from the Bureau of Labor Statistics; and state-level patenting activity, measured by the log number of patents granted in each state-year, from the United States Patent and Trademark Office. Table 1, Panel A reports summary statistics for our sample. The average state has approximately 2.3 million people employed. Just over 450 thousand of them are employed in establishments with fewer than 20 employees, and 169 thousand are in high-tech industries. The number of people employed in small high-tech firms in the average state is approximately 26 thousand, which aggregates to 1.3 million across all 50 states. The SVI for “patent trolls” and “patent attorney” averages 12.7 and 25.6, respectively. On average, there are 15 unique firms that receive VC funding in each state-quarter, of which 5.5 raise first rounds. However, there is substantial variation in VC investments across states, reflecting the high degree of concentration of VC activity in a few states (mainly, California and Massachusetts). Summary statistics for our macroeconomic controls are as follows: The mean state-level quarterly GSP is $309 billion; income per capita averages $44,800;

– 15 –

the unemployment rate during our sample period averages 6.2%; and the mean number of granted patents per state-year is nearly 3,000. Panel B compares the mean values of the outcome and control variables for states with and without anti-troll laws, measured during the four quarters prior to the adoption of the first anti-troll law in Vermont. Specifically, column 1 reports the mean of each variable over 2012Q2–2013Q1 for those states that pass an anti-troll law at some point during our sample period, while column 2 reports the mean for those states that pass no anti-troll legislation. The differences between these means and the corresponding p-values (accounting for clustering at the state level) are reported in columns 3 and 4, respectively. While the treated states tend to be smaller than their never-treated counterparts, this difference is not statistically significant. The never-treated states also concentrate an insignificantly higher level of VC activity and internet searches for NPE-related terms, mainly due to the presence of California and Massachusetts in the never-treated group. To ensure that these two states do not drive our conclusions, we will show that our findings are robust to excluding California and Massachusetts from the sample.

III.2

Methodology

To identify the effect of anti-troll laws on state-level outcomes such as employment and VC investment, we estimate the following difference-in-differences specification at the stateyear-quarter level:

ys t = δ · Anti-Troll Laws t + β · Xs t−1 + λs + αt + s t ,

(1)

where s denotes states and t denotes year-quarters. Anti-Troll Law is an indicator set equal to one if anti-troll legislation has been signed into law in state s in or before quarter t, and

– 16 –

zero otherwise. Xs t−1 is the vector of state-level control variables (GSP, income per capita, unemployment rate, and patenting activity); their inclusion alleviates concerns that changes in a state’s economic conditions could affect both the passage of anti-troll legislation and our measured outcomes. λs is a state fixed effect, which controls for state characteristics that do not vary over our sample period, such as the fact that California is larger and more high-tech oriented than Vermont; and αt is a year-quarter fixed effect, which absorbs aggregate shocks affecting all states. In all specifications, we report robust standard errors clustered at the state level (Bertrand, Duflo, and Mullainathan 2004). Our empirical setting allows the same state to be part of the treatment and control groups at different points in time. Specifically, at any year-quarter t, the control group includes both states that passed anti-troll legislation after year-quarter t (but before the end of our sample period) and so that will be treated eventually, and states that are never treated (either because they have not yet passed an anti-troll law or did so after the end of our sample period).

IV IV.1

Anti-Troll Laws and Employment Baseline results

We begin by examining how anti-troll laws affect employment at firms of different sizes and industries. Table 2 presents the results. Column 1 shows that the adoption of anti-troll legislation in a state leads to a 2% average increase in employment at small (0-19 employees) high-tech firms in the state (p=0.015). The estimated effect remains very similar (although less precisely estimated) at 1.8% when we exclude state-level controls in column 2. Columns 3 and 4 report the effect of anti-troll laws on employment at medium (20-49 employees) and

– 17 –

large (50+ employees) firms in high-tech industries. In contrast to the results in columns 1 and 2, we find no evidence that the laws are associated with economically or statistically significant changes in employment at medium or large high-tech firms. Nor do we find any effect for firms in non-high-tech industries, regardless of their size (columns 5-7). The results in Table 2 indicate that the effects of state anti-troll laws are confined to small businesses in high-tech industries. This finding is consistent with the notion that the frivolous patent demand letters that are the focus of these laws are particularly costly for small firms, which tend to have little experience with the patent system and are more likely to be coerced into settlements to avoid the threat of litigation (AIPLA 2013; Chien 2014).19

IV.2

Identification

In this section, we present evidence consistent with the parallel trends assumption necessary for identification in our setting. Omitted variables and reverse causality pose a potential challenge to this assumption. However, the fact that we only observe an increase in employment at small high-tech firms mitigates these concerns. To drive our findings, an omitted variable would not only need to coincide in time and space with the staggered adoption of the laws, but it would also need to differentially affect small high-tech firms. What is more, small firms are unlikely to be able to spend much on lobbying, alleviating concerns that our results may be driven by reverse causality or by some unobservable shock affecting both the growth opportunities and lobbying efforts of small high-tech firms. Figure 2 shows how employment in small high-tech firms changes around the passage of anti-troll legislation. The figure plots the estimated average difference in employment at 19

Large firms are more likely to behave in line with the public-firm executive quoted by Cohen, Gurun, and Kominers (2015), whose response to NPEs that send demand letters is: “If you have a truly viable case you will sue; otherwise don’t waste my time with this letter(!).”

– 18 –

treated states relative to control states in quarters t − 6 through t + 6 and beyond, where quarter t is the quarter when the anti-troll law is signed into law. Consistent with the parallel-trends assumption, we find no significant difference in the evolution of employment at treated and control states prior to the passage of anti-troll legislation (Roberts and Whited 2012). As expected given that employment adjustments are not immediate, we also find no significant difference in employment at the end of the quarter that the antitroll law is signed. However, we begin to observe a significant employment increase in the treated states relative to the control states the following quarter, and the size of this increase in employment grows over the following six quarters, becoming a highly significant 4.3% difference (p=0.007) seven or more quarters after the anti-troll law is signed. Table IA.2 in the Internet Appendix provides additional evidence in support of our identification assumption by showing that the adoption of anti-troll legislation in a state does not appear to be predicted by the state’s economic conditions one quarter or one year prior to the adoption the date. While the parallel-trends assumption is ultimately untestable, Table 3 reports a battery of tests aimed at further alleviating identification concerns; throughout, we focus on estimating the effect of anti-troll laws on employment at small high-tech firms. Columns 1 and 2 report the results of our baseline specification (1) with the only difference being that we include year-quarter fixed effects interacted with 4-Census-region and 9-Census-division fixed effects, respectively. The inclusion of these interacted fixed effects allows us to absorb regional economic shocks that might be contemporaneous with the passage of the laws; in particular, it helps account for the fact that employment dynamics in Vermont may be quite different from those in California, but are likely similar to dynamics in Maine or New Hampshire. Still, the results of this test are similar to those of our baseline analysis, both

– 19 –

in terms of magnitude and statistical significance. Column 3 reports the results of a placebo test that provides a complementary approach to examine whether our results are driven by regional economic shocks: We set Anti-Troll Law Neighbors t equal to one if state s has not passed an anti-troll law at time t but at least one neighboring state has, and include this placebo indicator in equation (1) alongside the actual Anti-Troll Law indicator. We exclude from the analysis five states that never pass an anti-troll law nor have a neighbor that does. As expected, we continue to find an increase in employment in states that pass an anti-troll law (p=0.030), but we find no significant employment change in placebo states that have not passed an anti-troll law but have a neighbor that has (p=0.624). Column 4 performs an alternative placebo test where we change the timing of passage of the anti-troll laws instead of their location. Specifically, we falsely assume that each state that passed an anti-troll law did so three years before the actual law was passed in the state. For example, we assume that Vermont passed its anti-troll law in May 2010 instead of in (the actual date of) May 2013. The placebo sample goes from 2009 to mid-2013 to be consistent with the panel duration of our baseline analysis. Contrary to our baseline results in Table 2, this placebo treatment effect is nether economically nor statistically significant (p=0.710), which supports the validity of our identification strategy. In columns 5 and 6, we estimate our baseline equation (1) excluding California and both California and Massachusetts, the two states with the highest per-capita employment in small high-tech firms. The estimated effect of anti-troll laws remains similar in magnitude and statistical significance (p=0.023 and p=0.012, respectively). We next address the concern that the anti-troll laws may have been adopted simultane-

– 20 –

ously with other measures intended to promote the growth of small high-tech firms; if this were the case, we may be wrongly attributing to the anti-troll laws effects that are actually driven by some other measures. To address this concern, column 7 controls for contemporaneous state laws and programs aimed to boost employment at small high-tech firms. We identify such measures using the Council for Community and Economic Research State Business Incentive (CCERSBI) database,20 which we supplement with our own searches of the Legiscan database. Table IA.3 in the Internet Appendix provides a complete list of these contemporaneous measures and their corresponding start dates. Analogously as our baseline Anti-Troll Law indicator, the variable Other State Laws in column 7 equals one after a state has adopted one such measure. The inclusion of this additional control in equation (1) leaves both the magnitude and statistical significance of the estimated coefficient for Anti-Troll Law unchanged. Table 4 provides a complementary test of the hypothesis that anti-troll laws are an effective shield against NPEs by examining how the adoption of anti-troll legislation in a state affects NPE-related Google searches in the state. To do so, we collect data on Google’s search volume index (SVI) for the terms “patent trolls” and “patent attorney” for each state-year. Google’s SVI has been shown to be a good proxy for attention in a variety of contexts (e.g., Da, Engelberg, and Gao 2011; Boguth, Gregoire, and Martineau 2017). In columns 1 and 2, we find that anti-troll laws are associated with a decrease in Google’s SVI for “patent trolls” of 9.3% when we include macroeconomic controls (p=0.062) and of 10.3% when they are excluded (p=0.037). This finding is consistent with the notion that state anti-troll laws are effective in curbing the activities of NPEs and, in particular, their ability 20

This database collects information on state business incentive programs using “state agency websites, statutes and codes, budget documents, and interviews with state agency representatives.”

– 21 –

to target small firms—the kind of firms that are most likely to initially resort to Google when receiving a frivolous demand letter. In columns 3 and 4, we also find that the laws are associated with a decrease in Google searches for “patent attorney,” although this effect is smaller and only marginally significant (p= 0.129 and 0.093, respectively). Importantly, the results in Table 4 reinforce the notion that our diff-in-diff employment analyses capture the effect of the state anti-troll laws and not of some other confound, as most such confounds are unlikely to also lead to a decrease in the attention firms pay to NPEs.

IV.3

Heterogeneity analysis

This section investigates the types of firms and employees most affected by anti-troll legislation. Specifically, we analyze the effect of anti-troll laws for firms in different high-tech industries (IT vs. life sciences) and employee education levels (college vs. non-college). Feng and Jaravel (2016) show that many of the patents held by NPEs are concentrated in the IT sector, a finding in line with the survey evidence reported by Chien (2014) and Feldman (2014). One potential driver of this preference of NPEs for IT patents is the fact that such patents, particularly those software-related, often have “overly broad or unclear claims or both” (GAO 2013). Indeed, Feng and Jaravel find that NPEs tend to purchase vaguely-worded patents, perhaps because these can be most easily leveraged in litigation. Of course, an alternative interpretation is that the IT sector has a larger concentration of the small, resource-constrained inventors that rely on NPEs to monetize their inventions. Regardless of what explains the preference of NPEs for the IT sector, we expect to find a stronger effect of anti-troll laws on small-firm employment in this sector—particularly given that the sector also concentrates a larger fraction of the vague patents that are most prone to be abused (GAO 2013). The results in columns 1-4 of Table 5 are consistent with

– 22 –

this prediction. Specifically, we find that the effect of anti-troll laws on small high-tech employment is driven by firms in the IT sector (p=0.012 in column 1 when we include state-level controls and p=0.062 in column 2 when we do not). By contrast, we find no economically or statistically significant effect for firms in the life sciences sector (p=0.881 and 0.673 in columns 3 and 4, respectively). Columns 5-8 in Table 5 shed further light on what type of employment increases in states that pass anti-troll laws by comparing workers with and without college education. We find that the increase in employment in small high-tech firms is most significant among workers with at least some college education. Specifically, anti-troll laws are associated with a 2%-1.8% increase in the employment of college-educated workers (p=0.008 and 0.063 with and without state-level controls, respectively). In addition, we also find somewhat noisier evidence of a similar increase in employment for workers without college degrees. These results thus suggest that the benefits of protecting small high-tech firms from being targeted by NPEs can be felt across the skill distribution.

IV.4

External validity

The fact that some of the states that have adopted anti-troll legislation are among the smallest in the nation (in population terms), such as Wyoming, Vermont, and the Dakotas, while some large states like California or New York have yet to pass an anti-troll law raises the following question: Are our findings driven by these small states? If so, our conclusions may be unlikely to extend to larger states should these states adopt similar laws. To investigate this question, column 1 in Table 6 estimates our baseline equation (1) using a weighted OLS regression with weights based on (the log of) each state’s employment in small high-tech firms at the beginning of the sample (2012Q1). The 1.9% employment

– 23 –

increase we estimate here is similar in terms of magnitude and statistical significance to the 2% increase we report in column 1 of Table 2, thus indicating that our baseline findings are not disproportionately driven by small states. In addition, the quantile regressions in columns 2-4 indicate that for states in the 75th and 50th percentiles of the employment distribution, anti-troll laws lead to a 1.9% and 2% increase in small high-tech employment (p=0.018 and 0.009, respectively); if anything, the effect in the 25th -percentile states is smaller, at 1.8% (p=0.051). The results in Table 6 thus suggest that the effects of anti-troll laws are not exclusive to small states. Our findings also raise the question of whether the increase in employment we capture in treated states is purely a zero-sum reallocation of workers from control to treated states. If this were the case, the effects of anti-troll laws should all but disappear if all states or Congress were to adopt similar legislation, as there would no longer be any states left for workers to reallocate from. Although our identification strategy does not allow a comprehensive assessment of general equilibrium effects, two pieces of evidence suggest that our findings are unlikely to be simply the result of small high-tech firms moving from control to treated states. First, recall from Table 3 that we find no effect of anti-troll laws on untreated neighboring states—and, in particular, no negative effect—unlike what would be expected if workers were reallocating from these states to their treated neighbors. Second, in Figure 3, we plot the evolution of average small-high-tech employment in treated states (top graph) and matched control states (bottom graph) around the passage of anti-troll legislation. Specifically, we match each treated state s to the state satisfying the following two conditions: (1) the state has either not passed an anti-troll law or passed it more than six quarters after state s adopted its law; and (2) at the beginning of our sample period, the state is closest to s among

– 24 –

those states that satisfy condition (1) based on a metric that is a linear combination of the geographical distance between the states’ population centroids and the distance between each of the four state-level controls we include in our regressions.21 The figure shows that the observed employment growth captured in Figure 2 is driven by an increase in employment in treated states, while employment in matched control states remains largely flat. Taken together, these results suggest that the employment increase at small high-tech firms in states that adopt anti-troll laws is more likely to be the result of an improvement in the viability of these firms than a zero-sum reallocation of workers from states without such laws.

V

How Do Anti-Troll Laws Affect Employment? Financing Channel

Firms that receive a demand letter from an NPE often devote considerable resources to fighting the NPE’s claims, even when they have little merit. Such expenses can be particularly disruptive for small firms, which often face important financing constraints (e.g., Kerr and Nanda 2011; Farre-Mensa and Ljungqvist 2016), and thus can limit their ability to grow and create jobs, as documented in Section IV. In addition to reducing the instances of frivolous patent-infringement claims and the direct costs that these claims impose on the targeted firms, anti-troll laws also reduce the risk that any firm in a state faces of being targeted with a frivolous claim. By reducing the risk posed by NPEs, state anti-troll laws may increase the net present value (NPV) of 21

In untabulated results, we find that estimating our baseline equation (1) in this matched sample—or an analogous sample where the control group is made up of synthetic matches–leaves our conclusions unchanged.

– 25 –

investing in the firms in a state, thereby making them more attractive to investors. In this section, we provide evidence of a financing channel through which protecting small firms from being the target of frivolous patent claims can affect firms’ ability to grow and create jobs. To test the financing channel, we focus on investments by venture capitalists (VCs). While venture capital is not the only source of funding for small firms, it has two distinct advantages for our purposes. First, comprehensive data on VC investments is available in commercial databases such as VentureXpert; the same is not true for other sources of private financing such as family offices or bank loans. Second, VCs are a particularly important source of capital for the high-tech firms that NPEs tend to target (e.g., Kortum and Lerner 2000). Indeed, anecdotal evidence suggests that VC-funded firms are often the recipients of demand letters: In a survey of 114 VCs, Chien (2013) found that 75% of surveyed VCs had been impacted by a demand from an NPE to one (or more) of their portfolio companies. Table 7 investigates whether the risk posed by NPEs affects the investment decisions of venture capitalists by examining the impact of anti-troll laws on VC investment. Column 1, which is estimated in all 50 US states, shows that anti-troll laws are associated with a 6.5% average increase in the number of firms raising venture capital, though this effect is not statistically significant (p=0.244). We continue to estimate a positive but insignificant effect in column 2, where we exclude state-level macroeconomic controls (p=0.357). Recall from Table 1 that in the median state-quarter, only three firms raise venture capital; thus, the analysis in columns 1-2 includes many states with persistently low levels of VC activity. To account for the fact that anti-troll laws may not be sufficient to attract VCs into states with historically low VC presence, the remainder of the table focuses on states with an established VC presence. Specifically, columns 3-4 of Table 7 estimate the

– 26 –

same regressions as columns 1-2 but excluding states with below-median VC activity in 2012 (the year before the first anti-troll law was passed). Column 3 shows that anti-troll laws lead to a 14.4% increase in the number of firms raising venture capital (p=0.044); this estimated effect is slightly larger when we exclude state-level macroeconomic controls in column 4 (p=0.024).22 Consistent with our finding in Table 5 that the effect of anti-troll laws is strongest in the IT sector, columns 5 and 6 show that the increase in VC investment associated with anti-troll laws is concentrated in IT firms (p=0.012 and 0.008), a finding that is robust to focusing only on first-round investments (columns 7-8). By contrast, and in line with Table 5, we continue to find no effect on VC investment in life sciences firms (columns 9-10). Table IA.4 in the Internet Appendix shows a battery of identification tests for our VC results analogous to those in Figure 2 and Table 3 for employment. (As in columns 5-8 of Table 7, we focus on IT firms in high VC-presence states.) Column 1 shows that VC investment displays no significant pre-trend in treated states. Columns 2 and 3 show the VC investment results are robust to including year-quarter fixed effects interacted with 4-Census-region and 9-Census-division fixed effects, respectively. In column 4, we find no evidence of a change in VC investment in placebo states that have not passed an anti-troll law but have a neighbor that has (p=0.998); in line with our discussion in Section IV.4, this finding suggests that the increase in VC investment in treated states is not at the expense of their untreated neighbors. Column 5 similarly shows no evidence of a placebo treatment effect when we falsely assume that each state that passed an anti-troll law did so three years before the actual law was passed. Columns 6 and 7 show that our VC results are robust to 22

In untabulated results, we find that the estimated effect is also larger but noisier if we examine the dollar amount of capital invested by VCs in the state (p=0.078 and 0.066 with and without state-level controls, respectively).

– 27 –

excluding California as well as both California and Massachusetts from the sample (p=0.013 and 0.016, respectively). Finally, column 8 shows that the VC investment results remain unchanged when we control for contemporaneous state laws and programs aimed to boost employment at small high-tech firms. Taken together, the evidence in Tables 7 and IA.4 supports the hypothesis that anti-troll laws reduce the threat posed by NPEs to small innovative firms in a state, thus making the firms more attractive to VC investors—although this effect does not appear to be sufficient to attract VCs to states with historically low VC presence. One plausible interpretation of our findings is that by reducing the risk that firms in a state receive bad-faith patent infringement claims, anti-troll laws increase the likelihood that VCs invest in those firms instead of in similar out-of-state firms that have been targeted by NPEs. This effect is likely magnified by the high levels of uncertainty and information asymmetry prevalent in the entrepreneurial finance market, where small differences—such as a decrease in litigation risk—can end up being the deciding factor VCs use to break ties among otherwise similar investment candidates. Further reinforcing the importance of the VC financing channel, columns 1-4 in Table 8 show that the positive effect of anti-troll laws on small high-tech employment is driven by states with a level of VC activity in 2012 at or above the median of all 50 states. Column 1 shows that in high VC-presence states, anti-troll laws lead to a 3.8% increase in the employment of small firms in high-tech industries (p<0.001); without macroeconomic controls in column 2, the estimated increase is 3.5% (p=0.015). By contrast, employment change in states with below-median VC presence is insignificant (columns 3 and 4; p=0.532 and 0.704, respectively). Similarly, columns 5 and 6 show that in high VC-presence states, the adoption of anti-

– 28 –

troll legislation leads to an increase in the number of high-tech establishments. In these states, the estimated establishment increase is 2.7% and 2.2% with and without macroeconomic controls, respectively (p=0.005 and 0.023). By contrast, the laws have no significant effect on the number of high-tech establishments in states with below-median VC presence (columns 7 and 8). These results suggest that the growth in small high-tech employment induced by anti-troll laws is driven both by the growth of existing establishments and by the creation of new ones (i.e., both intensive- and extensive-margin growth appear to be at play). Table 9 provides additional evidence of the importance of the financing channel by showing that the employment increase associated with anti-troll laws is largest in industries with high equity dependence, as defined by Rajan and Zingales (1998). Columns 1 and 2 show that in highly equity-dependent industries, the passage of anti-troll legislation leads to a 2.4%-2.5% increase in employment at small high-tech firms (p=0.008 and 0.034 with and without macroeconomic controls, respectively). By contrast, the employment increase in industries with low-equity dependence is less than half that size and statistically insignificant. Thus, anti-troll laws are particularly helpful to equity-dependent firms, which are those most likely to benefit from increased access to venture and other forms of equity capital.

VI

Conclusion

Recent years have seen considerable debate in both academic and policy circles regarding the effects of NPEs. While some commentators contend NPEs play a key financial intermediary role, others argue their infringement claims are often frivolous and ultimately impede the growth of small businesses. In this paper, we analyze how bad-faith patent-infringement

– 29 –

claims made by NPEs affect small firms’ ability to grow, create jobs, and raise capital. Our identification strategy exploits the adoption of anti-troll laws in 32 states that aim to limit NPEs’ use of demand letters to coerce small firms into quick settlements to avoid litigation. We show that, following the adoption of anti-troll legislation, employment at small hightech firms in a state increases by 2%—an increase that is driven by firms in the IT sector, where NPEs are most active. By contrast, we find no evidence of a significant effect on employment at larger high-tech firms or at non-high-tech firms of any size. We also show that, in states with an established VC presence, anti-troll laws are associated with a 14% increase in the number of firms raising early-stage VC funding—an increase that is again driven by IT firms. Thus, by reducing the risk posed by frivolous patent demands, state anti-troll laws appear to increase the NPV of investing in small high-tech firms, thereby improving the financing environment faced by these firms. Chien (2014) writes of the America Invents Act of 2011, the latest patent reform law passed by Congress: “Reforms to reduce the cost of litigation defense are laudable, and likely deter some suits from being brought in the first place, but do not reach small companies against whom litigation is threatened, but not brought” (p. 462). Our results suggest that measures aimed at curbing NPEs’ ability to target small firms with bad-faith patentinfringement claims can play an important role in reducing both the real and financing frictions faced by such firms. To the extent that small businesses are key engines of productivity and economic growth (Aghion and Howitt 1992; Akcigit and Kerr 2016), the benefits of such measures could be felt economy-wide.

– 30 –

References [1] Adelino, M., A. Schoar, and F. Severino, 2015, “House Prices, Collateral and SelfEmployment,” Journal of Financial Economics 117, 288-306. [2] Aghion, P., and P. Howitt, 1992, “A Model of Growth through Creative Destruction,” Econometrica 60, 323-351. [3] Akcigit U., and W.R. Kerr, 2016, “Growth through Heterogeneous Innovations,” Journal of Political Economy, forthcoming. [4] American Intellectual Property Law Association (AIPLA), 2013, Testimony of Todd Dickinson before the US Senate Judiciary Committee. Hearing on “Protecting Small Businesses and Promoting Innovation by Limiting Patent Troll Abuse.” [5] Asay, C. D., et al., 2015, Letter to Congress (March 3). Available at: http://www. utdallas.edu/~ugg041000/IPScholarsLettertoCongress_March_2_2015.pdf. [6] Benmelech E., N. Bergman, and A. Seru, 2011, “Financing Labor,” Working Paper. [7] Benmelech E., C. Frydman, and D. Papanikolaou, 2016, “Credit Market Disruptions and Employment during the Great Depression: Evidence from Firm-level Data,” Working Paper. [8] Bernstein, S., X. Giroud, and R. R. Townsend, 2015, “The Impact of Venture Capital Monitoring,” Journal of Finance 71, 1591-1622. [9] Bertrand, M., E. Duflo, and S. Mullainathan, 2004, “How Much Should We Trust Differences-in-Differences Estimates?,” Quarterly Journal of Economics 119, 249-275. [10] Boguth, O., V. Grégoire, and C. Martineau, 2017, “Shaping Expectations and Coordinating Attention: The Unintended Consequences of FOMC Press Conferences,” Working Paper. [11] Bryant, T., L., 2012, “The America Invents Act: Slaying Trolls, Limited Joinder,” Harvard Journal of Law & Technology 25, 674-695. [12] Carpenter R. E. and B. C. Petersen, 2002, “Is the Growth of Small Firms Constrained by Internal Finance?,” Review of Economics and Statistics 84, 298-309. [13] Chien, C. V., 2013, “Patent Assertion and Startup Innovation,” Working Paper. [14] Chien, C. V., 2014, “Startups and Patent Trolls,” Stanford Technology Law Review 17, 461-506. [15] Chodorow-Reich, G., 2014, “The Employment Effects of Credit Market Disruptions: Firm-level Evidence from the 2008-09 Financial Crisis,” Quarterly Journal of Economics 129, 1-59.

– 31 –

[16] Cohen, L., U. G. Gurun, and S. D. Kominers, 2015, “Patent Trolls: Evidence from Targeted Firms,” Working Paper. [17] Council of Economic Advisers (CEA), 2016, “The Patent Litigation Landscape: Recent Research and Developments.” [18] Da, Z., J. Engelberg, and P. Gao, 2011, “In Search of Attention,” Journal of Finance 66, 1461-1499. [19] DeSisto, R., 2015, “Vermont vs. the Patent Troll: Is State Action a Bridge Too Far?,” Suffolk University Law Review 109, 109-130. [20] Farre-Mensa, J., and A. Ljungqvist, 2016, “Do Measures of Financial Constraints Measure Financial Constraints?” Review of Financial Studies 29, 271-308. [21] Federal Trade Commission (FTC), 2015, Complaint Against MPHJ Technology Investments. Docket No. C-4513. [22] Feldman, R., 2014, “Patent Demands and Startup Companies: The View from the Venture Capital Community,” Yale Journal of Law and Technology 16, 236-284. [23] Feng, J., and X. Jaravel, 2016, “Who Feeds the Trolls? Patent Trolls and the Patent Examination Process,” Working Paper. [24] Government Accountability Office (GAO), 2013, “Assessing Factors That Affect Patent Infringement Litigation Could Help Improve Patent Quality”(Report to Congressional Committees). [25] Giroud, X., and H. Mueller, 2015, “Firm Leverage, Consumer Demand, and Employment Losses during the Great Recession,” Quarterly Journal of Economics, forthcoming. [26] Giroud, X., and H. Mueller, 2016, “Redistribution of Local Labor Market Shocks through Firms’ Internal Networks,” Working Paper. [27] Hellman, T., and M. Puri, 2002, “Venture Capital and the Professionalization of StartUp Firms: Empirical Evidence,” Journal of Finance 57, 169-197. [28] Himmelberg, C. P. and B. C. Petersen, 1994, “R&D and Internal Finance: A Panel Study of Small Firms in High-Tech Industries,” The Review of Economics and Statistics 76, 38-51. [29] Hochberg, Y. V., A. Ljungqvist, and Y. Lu, 2007, “Whom You Know Matters: Venture Capital Networks and Investment Performance,” Journal of Finance 62, 251-301. [30] Kerr, W. R., and R. Nanda, 2011, “Financing Constraints and Entrepreneurship,”in David Audretsch, Oliver Falck, and Stephan Heblich, eds. Handbook of Research on Innovation and Entrepreneurship (Cheltenham, U.K.: Edward Elgar Publishing).

– 32 –

[31] Kile, C. O., and M. E. Phillips, 2009, “Using Industry Classification Codes to Sample High-Technology Firms: Analysis and Recommendations,” Journal of Accounting, Auditing & Finance 24, 35-58. [32] Kortum, S., and J. Lerner, 2000, “Assessing the Contribution of Venture Capital to Innovation,” RAND Journal of Economics 31, 674-692. [33] Michaels, R., T. B. Page, T. M. Whited, 2016, “Labor and Capital Dynamics under Financing Frictions,” Working Paper. [34] Rajan, R. G., and L. Zingales, 1998, “Financial Dependence and Growth,” American Economic Review 88, 559-586. [35] Roberts, M. R., and T. M. Whited, 2012, “Endogeneity in Empirical Corporate Finance,”in George Constantinides, Milton Harris, and Rene Stulz, eds. Handbook of the Economics of Finance 2, 493-572. [36] RPX, 2015, “2015 Report: Patent Litigation, Patent Marketplace, and NPE Cost.” [37] Scott Morton, F., and C. Shapiro, 2014, “Strategic Patent Acquisitions,” Antitrust Law Journal 79, 463-499. [38] Smeets, R., 2015, “Does Patent Litigation Reduce Corporate R&D? An Analysis of US Public Firms,” Working Paper. [39] Tucker, C. E., 2014, “Patent Trolls and Technology Diffusion: The Case of Medical Imaging,” Working Paper.

– 33 –

Figure 1: States with Anti-Troll Laws This map shows the states that adopted an anti-troll law during our sample. Signing dates are provided in Table IA.1 in the Internet Appendix.

– 34 –

Figure 2: Evolution of Employment Around Signing of Anti-Troll Laws This figure shows the evolution of employment at small high-tech firms in treated states with antitroll laws relative to control states without such laws. We estimate our baseline employment regression (1) as in Table 2, column 1, except that we replace the Anti-Troll Law indicator with indicators that identify quarters t–6, t–5, ..., t, t+1, ..., t+6, and >t+6 for states that pass an anti-troll law, where quarter t is the quarter the anti-troll law is signed. The figure plots the point estimates associated with each of these indicators, alongside their 95% confidence interval. States that adopted an anti-troll law in the first quarter of 2015 or later, for which we observe fewer than six post-treatment observations, are excluded from the treatment group (these states can still be part of the control group before their treatment). Robust standard errors are clustered by state.

0.1 0.08 0.06 0.04 0.02 0 -0.02 -0.04 -0.06 t-6

t-5

t-4

t-3

t-2

t-1

t

– 35 –

t+1

t+2

t+3

t+4

t+5

t+6 >t+6

Figure 3: Evolution of Average Employment for Treated and Control States This figure shows the evolution of average employment at small high-tech firms in states that pass an anti-troll law (top graph) and in matched states that do not pass an anti-troll law (bottom graph), alongside their 95% confidence intervals. To construct the figure, we match each treated state s to the state satisfying the following two conditions: (1) the state has either not passed an anti-troll law or passed it more than six quarters after state s adopted its law; and (2) at the beginning of our sample period (2012Q1), the state is closest to s among those states that satisfy condition (1) based on a metric that is a linear combination of the geographical distance between the states’ population centroids (50% of the weight) and the distance between each of the four state-level controls we include in our regressions (50% of the weight combined). As in column 1 of Table 2, the underlying analysis includes macroeconomic controls as well as state and year-quarter fixed effects. States that adopted an anti-troll law in the first quarter of 2015 or later, for which we observe less than six post-treatment observations, are excluded from the treatment group (these states can still be part of the control group). The horizontal lines display the pre-treatment means of employment. Treated States: Average Employment at Small High-Tech Firms 13,900 13,700 13,500 13,300 13,100 12,900 12,700 12,500 t-6

t-5

t-4

t-3

t-2

t-1

t

t+1

t+2

t+3

t+4

t+5

t+6

Control States: Average Employment at Small High-Tech Firms 11,900 11,700 11,500 11,300 11,100 10,900 10,700 10,500 t-6

t-5

t-4

t-3

t-2

t-1

t

t+1

t+2

t+3

t+4

t+5

t+6

Table 1: Summary Statistics Panel A reports summary statistics for the full sample. All variables are quarterly except Google search volume index (SVI), which is annual. Panel B compares the mean values of all variables for states with and without anti-troll laws, measured prior to the adoption of the first anti-troll law. Column 1 reports the mean of each variable over 2012Q2–2013Q1 (2012 for SVI data) for those states that pass an anti-troll law, and column 2 reports the mean for those states that do not pass anti-troll legislation. The difference between these means and the corresponding p-values (clustered at the state level) are reported in columns 3 and 4, respectively. Panel A Observations

Mean

Median

SD

Total Employment, 000s

885

2301.0

1503.2

2514.7

Employment (<20 Emp.), 000s

885

451.7

293.3

529.1

Employment (High-Tech), 000s

885

168.6

94.9

225.1

Employment (<20 Emp. & High-Tech), 000s

885

25.8

14.2

38.0

Patent Trolls SVI

247

12.7

0

22.1

Patent Attorney SVI

247

25.6

0

29.4

VC (# Unique Firms)

900

15.1

3

47.7

VC (# First Rounds)

900

5.5

1

17.5

Establishments (High-Tech), 000s

900

14.7

8.7

16.6

GSP, billions

900

308.6

185.4

378.0

Income Per Capita, 000s

900

44.8

43.9

7.3

Unemployment Rate

900

6.2

6.1

1.8

Patents

900

2955.3

1088

5834.5

– 37 –

Panel B Eventually

Never

Treated

Treated

Difference

P-value

Ln(Total Employment, 000s)

14.07

14.21

-0.14

0.65

Ln(Employment (<20 Emp.), 000s)

12.52

12.60

-0.08

0.78

Ln(Employment (High-Tech), 000s)

11.28

11.40

-0.11

0.76

Ln(Employment (<20 Emp. & High-Tech), 000s)

9.47

9.53

-0.06

0.86

Ln(Patent Trolls SVI)

1.01

1.58

-0.58

0.32

Ln(Patent Attorney SVI)

1.77

1.97

-0.20

0.74

Ln(VC (# Unique Firms))

1.31

1.70

-0.40

0.35

Ln(VC (# First Rounds))

0.74

1.13

-0.39

0.24

Ln(Establishments (High-Tech), 000s)

9.04

9.07

-0.03

0.92

Ln(GSP, billions)

12.00

12.26

-0.26

0.39

Ln(Income Per Capita, 000s)

10.64

10.71

-0.06

0.19

Unemployment Rate

7.19

7.58

-0.39

0.44

Ln(Patents)

6.89

7.11

-0.23

0.62

– 38 –

Table 2: Employment by Firm Size and Sector This table reports the effect of state anti-troll laws on employment at firms of different sizes and sectors. The dependent variable in columns 1-4 is the natural logarithm of employment in the high-tech sector for firms with 0-19, 20-49, and 50+ employees. Similarly, the dependent variable in columns 5-7 is the natural logarithm of employment for firms of different sizes in non-high-tech sectors. Anti-Troll Law is an indicator equal to one if a state has passed anti-troll legislation in or before that quarter. State-level control variables are defined in the text. Each observation is a state-quarter. All specifications include state and year-quarter fixed effects. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively. Ln(High-Tech Emp.)

Dep. Var. = Firm Size =

Anti-Troll Law

Ln(GSP)

Ln(Inc. Per Capita)

Unemployment Rate

Ln(Patents)

Ln(Non-High-Tech Emp.)

0-19

0-19

20-49

50+

0-19

20-49

50+

(1)

(2)

(3)

(4)

(5)

(6)

(7)

0.020**

0.018*

-0.003

0.003

-0.001

0.001

0.006

(0.008)

(0.011)

(0.015)

(0.011)

(0.005)

(0.006)

(0.004)

0.655***

1.224**

0.219

0.225**

0.228

0.308***

(0.186)

(0.559)

(0.164)

(0.087)

(0.142)

(0.055)

0.312

-0.587

0.139

0.439***

0.435**

0.409***

(0.288)

(0.393)

(0.265)

(0.151)

(0.179)

(0.099)

-0.008*

-0.026***

-0.004

0.003

-0.005

-0.003

(0.005)

(0.007)

(0.006)

(0.005)

(0.005)

(0.003)

-0.014

0.073

0.180**

-0.003

-0.000

-0.000

(0.039)

(0.057)

(0.084)

(0.022)

(0.035)

(0.022)

State FE

yes

yes

yes

yes

yes

yes

yes

Year-Quarter FE

yes

yes

yes

yes

yes

yes

yes

No. Observations

885

885

885

885

885

885

885

Within R-Squared

0.212

0.021

0.176

0.098

0.057

0.122

0.230

– 39 –

Table 3: Identifying Assumption The dependent variable in this table is the logarithm of employment at high-tech firms with 0-19 employees. Anti-Troll Law is an indicator equal to one if a state has passed anti-troll legislation in or before that quarter. Columns 1 and 2 estimate our baseline equation (1) replacing year-quarter fixed effects with year-quarter × 4-Census-region and year-quarter × 9-Census-division fixed effects, respectively. In column 3, Anti-Troll Law Neighbor is a placebo indicator set equal to one if the state has not passed an anti-troll law but at least one neighboring state has. In column 4, Anti-Troll Lawt-12 is a placebo indicator that equals one 12 quarters prior to a state passing anti-troll legislation; this placebo sample goes from 2009Q1 to 2013Q2. Columns 5 and 6 exclude California and California and Massachusetts from the analysis, respectively. Column 7 controls for contemporaneous state laws and programs intended to promote the growth of small high-tech firms. Each observation is a state-quarter. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively. Ln(Small-Firm High-Tech Emp.)

Dep. Var. = Controlling for

Neighbor

Placebo

Regional Shocks

States

Timing

CA

CA + MA

Other State Laws

(3)

(4)

(5)

(6)

(7)

(1)

Anti-Troll Law

(2)

Excluding:

Controlling for

0.020** 0.018**

0.016**

0.018**

0.020**

0.020**

(0.008) (0.009)

(0.007)

(0.008)

(0.008)

(0.008)

Anti-Troll Law Neighbor

0.003 (0.006)

Anti-Troll Lawt-12

0.003 (0.009)

Other State Laws

-0.002 (0.007)

State Controls

yes

yes

yes

yes

yes

yes

yes

State FE

yes

yes

yes

yes

yes

yes

yes

Year-Quarter FE

no

no

yes

yes

yes

yes

yes

Year-Quarter-Region FE

yes

no

no

no

no

no

no

Year-Quarter-Division FE

no

yes

no

no

no

no

no

No. Observations

885

885

815

896

867

849

885

Within R-Squared

0.222

0.238

0.185

0.132

0.218

0.224

0.212

– 40 –

Table 4: Search Volume Index The dependent variable in this table is the natural logarithm of one plus Google’s search volume index (SVI) measure for “patent trolls” (columns 1-2) and “pattent attorney” (columns 3-4). SVI is scaled from 0 to 100, and missing observations (due to an insufficient number of searches) are replaced with zeros. Anti-Troll Law is an indicator equal to one if a state has passed anti-troll legislation in or before the current year. Each observation is a state-year. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively.

Dep. Var. =

Ln(1 + Patent Troll SVI)

Ln(1+Patent Attorney SVI)

(1)

(2)

(3)

(4)

-0.093*

-0.103**

-0.053

-0.052*

(0.049)

(0.048)

(0.034)

(0.031)

State Controls

yes

no

yes

no

State FE

yes

yes

yes

yes

Year FE

yes

yes

yes

yes

No. Observations

247

247

247

247

Within R-Squared

0.033

0.018

0.029

0.017

Anti-Troll Law

– 41 –

Table 5: Employment by Industry and Education The dependent variable in this table is the natural logarithm of employment at high-tech firms with 0-19 employees. Columns 1-4 break down the sample in IT vs. life sciences industries. In columns 5-8, we report the employment change for workers with at least some college vs. those with no college education. Anti-Troll Law is an indicator equal to one if a state has passed anti-troll legislation in or before that quarter. Each observation is a state-quarter. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively.

Ln(Small-Firm High-Tech Emp.)

Dep. Var. =

By Industry IT (1)

Anti-Troll Law

By Worker Education

Life Sciences (2)

Some college

(3)

(4)

(5)

(6)

0.022** 0.021*

-0.003

-0.011

0.020*** 0.018*

(0.008) (0.011)

(0.021) (0.025)

(0.007) (0.010)

No college (7)

(8)

0.019*

0.017

(0.011) (0.014)

State Controls

yes

no

yes

no

yes

no

yes

no

State FE

yes

yes

yes

yes

yes

yes

yes

yes

Year-Quarter FE

yes

yes

yes

yes

yes

yes

yes

yes

No. Observations

885

885

885

885

885

885

885

885

Within R-Squared

0.212

0.028

0.060

0.001

0.200

0.027

0.229

0.012

– 42 –

Table 6: External Validity The dependent variable in this table is the natural logarithm of employment at high-tech firms with 0-19 employees. In column 1, we estimate weighted OLS regressions with weights based on the natural logarithm of each state’s employment at the beginning of the sample (2012Q1). Columns 2-4 report the results of quantile regressions. Anti-Troll Law is an indicator equal to one if a state has passed anti-troll legislation in or before that quarter. Each observation is a state-quarter. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively.

Dep. Var. =

Ln(Small-Firm High-Tech Employment) Weighted

Quantile regression

Regression

75%

50%

25%

(1)

(2)

(3)

(4)

0.019**

0.019**

0.020***

0.018*

(0.008)

(0.008)

(0.008)

(0.009)

State Controls

yes

yes

yes

yes

State FE

yes

yes

yes

yes

Year-Quarter FE

yes

yes

yes

yes

No. Observations

885

885

885

885

Within R-Squared

0.2097

na

na

na

Anti-Troll Law

– 43 –

Table 7: VC Financing This table reports the effect of state anti-troll laws on venture capital (VC) investment. The dependent variables are the (log-transformed) number of unique firms raising VC funding in a state (columns 1-6 and 9-10) and of firms raising their first VC round (columns 7-8). Throughout, we focus on early-stage VC rounds raised by firms in high-tech industries. Specifically, we measure VC-fundraising by high-tech firms at the “startup/seed”, “early stage”, or “expansion” phase that were founded in 2005 or later. In columns 1-2, the sample includes all states, while in columns 3-10 we focus on states with an above-median level of VC activity in 2012. Columns 5-8 and 9-10 focus on IT and life-sciences firms, respectively. Anti-Troll Law is an indicator equal to one if a state has passed anti-troll legislation in or before that quarter. Each observation is a state-quarter. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively.

All States

– 44 –

Dep. Var. =

High VC Activity States

All Industries

All Industries

Ln(1 + # Firms)

Ln(1 + # Firms)

IT Ln(1 + # Firms)

Life Sciences

Ln(1 + # 1st Rounds)

Ln(1 + # Firms)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

0.065

0.076

0.144**

0.187**

0.176**

0.206***

0.185*

0.240**

-0.010

0.027

(0.055)

(0.057)

(0.068)

(0.078)

(0.065)

(0.072)

(0.105)

(0.094)

(0.079)

(0.086)

State Controls

yes

no

yes

no

yes

no

yes

no

yes

no

State FE

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

Year-Quarter FE

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

No. observations

900

900

486

486

486

486

486

486

486

486

0.010

0.003

0.042

0.019

0.033

0.020

0.052

0.019

0.019

0.000

Anti-Troll Law

Within R-Squared

Table 8: Employment and Number of Establishments by VC Activity The dependent variables in this table are the logarithm of employment at high-tech firms with 0-19 employees (columns 1-4) and the logarithm of the number of high-tech establishments in a state (columns 5-8). Columns 1-2 and 5-6 report results for states with VC activity in 2012 at or above the median of all states; columns 3-4 and 7-8 report results for states with below-median VC activity. Anti-Troll Law is an indicator equal to one if a state has passed anti-troll legislation in or before that quarter. Each observation is a state-quarter. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively.

Dep. Var.=

Ln(Small-Firm High-Tech Emp.) High VC States (1)

Anti-Troll Law

(2)

0.038*** 0.035**

Low VC States (3)

(4)

-0.006

-0.005

(0.009)

(0.015)

(0.010) (0.013)

State Controls

yes

no

yes

State FE

yes

yes

Year-Quarter FE

yes

No. Observations Within R-Squared

Ln(# High-Tech Establishments) High VC States (5)

(6)

0.027*** 0.022**

Low VC States (7)

(8)

-0.012

-0.001

(0.009)

(0.009)

no

yes

no

yes

no

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

511

511

374

374

486

486

414

414

0.302

0.067

0.158

0.003

0.147

0.042

0.317

0.010

– 45 –

(0.009) (0.012)

Table 9: Employment by Industry Equity Dependence The dependent variable in this table is the logarithm of employment at high-tech firms with 0-19 employees. Our measure of equity dependence follows Rajan and Zingales (1998). Specifically, we characterize as having high equity dependence those industries whose level of equity dependence is above the median of the high-tech sector in the universe of Compustat firms. Anti-Troll Law is an indicator equal to one if a state has passed anti-troll legislation in or before that quarter. Each observation is a state-quarter. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively.

Ln(Small-Firm High-Tech Emp.)

Dep. Var. =

High Equity Dependence

Low Equity Dependence

(1)

(2)

(3)

(4)

0.024***

0.025**

0.011

0.005

(0.009)

(0.012)

(0.011)

(0.014)

State Controls

yes

no

yes

no

State FE

yes

yes

yes

yes

Year-Quarter FE

yes

yes

yes

yes

Observations

885

885

885

885

0.208

0.030

0.129

0.001

Anti-Troll Law

Within R-Squared

– 46 –

Patent Trolls and Small Business Employment Ian Appel, Joan Farre-Mensa, and Elena Simintzi

INTERNET APPENDIX

–1–

Figure IA.1: Sample Demand Letter The following is an example of a patent demand letter sent by Lodsys LLC, a non-practicing entity, in 2011. The letter has been redacted to remove the name and address of the recipient. (Source: https://trollingeffects.org/letters.)

–2–

–3–

–4–

Table IA.1: Signing Dates of State Anti-Troll Laws This table lists the 32 states with anti-troll laws in our sample period along with corresponding signing dates. Connecticut and Michigan also adopted laws in 2017 after the end of our sample. State

Law Signed

AL

4/2/2014

AZ

3/24/2016

CO

6/5/2015

FL

6/2/2015

GA

4/15/2014

ID

3/26/2014

IL

8/26/2014

IN

5/5/2015

KS

5/20/2015

LA

5/28/2014

ME

4/14/2014

MD

5/5/2014

MN

4/29/2016

MS

3/28/2015

MO

7/8/2014

MT

4/2/2015

NH

7/11/2014

NC

8/6/2014

ND

3/26/2015

OK

5/16/2014

OR

3/3/2014

RI

6/4/2016

SC

6/9/2016

SD

3/26/2014

TN

5/1/2014

TX

6/17/2015

UT

4/1/2014

VT

5/22/2013

VA

5/23/2014

WA

4/25/2015

WI

4/24/2014

WY

3/11/2016

–5–

Table IA.2: Predictive Regressions This table examines whether a state’s economic conditions predict the adoption of anti-troll legislation. The dependent variable is an indicator equal to one if a state has passed anti-troll legislation in that quarter. Observations for states that adopt an anti-troll law are excluded from the sample after the law is passed. The control variables are lagged by one quarter in columns 1-5 and by four quarters in columns 6-10. Each observation is a state-quarter. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively.

Ln(Small-Firm High-Tech Employment)

Dep. Var. = (1)

Ln(GSP)t-1

(2)

(3)

(4)

(5)

0.487

0.817

(0.455)

(0.537)

Ln(Inc. Per Capita)t-1

-0.710

-1.318

(0.898)

(0.980)

Unemployment Ratet-1

0.001

0.003

(0.018)

(0.017)

Ln(Patents)t-1

(6)

(7)

(8)

(9)

(10)

0.127 0.144 (0.147) (0.152)

Ln(GSP)t-4

0.181

-0.097

(0.313)

(0.416)

Ln(Inc. Per Capita)t-4

0.568

0.750

(0.555)

(0.705)

Unemployment Ratet-4

0.005

0.004

(0.018)

(0.016)

Ln(Patents)t-4

-0.147 -0.164 (0.128) (0.129)

No. observations Within R-Squared

713

713

713

713

713

713

713

713

713

713

0.002 0.002 0.000 0.002 0.008 0.000 0.002 0.000 0.002 0.005

–6–

Table IA.3: Other State Initiatives to Promote Small High-Tech Firms This table lists other state laws and programs aimed to boost employment at small high-tech firms during our sample period. We identify such measures using the Council for Community and Economic Research State Business Incentive (CCERSBI) database. We specifically focus on programs targeting the information (NAICS 51) and Professional, Scientific, and Professional Services (NAICS 54) sectors, as well as those that involve equity investments (across any industry). If a start date for the program is not provided, we attempt to identify this information through internet searches. All programs identified from this database are assumed to start in the first quarter of the year and last through the end of the sample. We supplement the programs detailed above with searches for the passage of state legislation aimed at promoting entrepreneurship using our own searches of the Legiscan database. We find that intrastate crowdfunding laws, intended to promote small business creation, were adopted by a number of states during our sample period, and append those to the list of collected state business incentive programs. State

Law/Program

Date

AK

49th State Angel Fund (49SAF)

1/1/2012

AL

Alabama Innovation Fund

1/1/2012

AL

Crowdfunding Law

4/9/2014

AZ

Computer Data Center Program

1/1/2013

AZ

Crowdfunding Law

4/1/2015

CO

Advanced Industries Accelerator Programs

1/1/2013

CO

Crowdfunding Law

4/13/2015

CT

Service and Manufacturing Facilities Tax Credit

1/1/2012

CT

Connecticut Bioscience Innovation Fund (CBIF)

1/1/2013

CT

Regenerative Medicine Research Fund

1/1/2014

FL

Crowdfunding Law

6/16/2015

IA

Crowdfunding Law

7/2/2015

ID

Crowdfunding Law

1/20/2012

ID

Idaho Opportunity Fund

1/1/2013

IL

Crowdfunding Law

7/29/2015

IN

Crowdfunding Law

3/25/2014

KY

Crowdfunding Law

3/19/2015

MA

AmplifyMass

1/1/2014

MA

DeployMass

1/1/2014

MA

Crowdfunding Law

1/15/2015

MD

Maryland Venture Fund (MVF)

1/1/2012

MD

Propel Baltimore Fund

1/1/2012

MD

Veterans Opportunity Fund (VOF)

1/1/2012

MD

Cybersecurity Investment Incentive Tax Credit (CIITC)

1/1/2013

MD

Cyber Security Investment Fund (CIF)

1/1/2014

MD

Maryland E-Nnovation Initiative Fund (MEIF)

1/1/2015

–7–

Table IA.3 Cont. State

Law/Program

Date

MD

Crowdfunding Law

5/16/2016

ME

Maine Economic Development VC Investment Program

1/1/2012

ME

Crowdfunding Law

3/2/2014

MI

Crowdfunding Law

12/30/2013

MN

Crowdfunding Law

6/15/2015

MS

Crowdfunding Law

2/9/2015

MT

Crowdfunding Law

4/1/2015

ND

Research ND

1/1/2013

NE

Crowdfunding Law

5/27/2015

NJ

Angel Investor Tax Credit Program

1/1/2013

NJ

Crowdfunding Law

11/9/2015

NJ

Opportunity Partnership Grants

1/1/2016

NJ

Skills Partnership/Customized Training Grant

1/1/2016

NY

Innovate NY Fund

1/1/2012

NY

Start-up New York

1/1/2014

OR

Crowdfunding Law

10/15/2015

RI

Industry Cluster Grant

1/1/2015

SC

Technology Intensive Facility Sales Tax Exemption

1/1/2013

SC

Crowdfunding Law

6/26/2015

TN

Crowdfunding Law

5/19/2014

TX

Jobs for Texas

1/1/2013

TX

Franchise Tax Credit for Qualified R&D Activities

1/1/2014

TX

Crowdfunding Law

VA

Virginia Biosciences Health Research Grants

1/1/2013

VA

Crowdfunding Law

3/23/2015

VT

Crowdfunding Law

6/16/2014

WA

Crowdfunding Law

3/28/2014

WA

Data Center Tax Exemption

1/1/2015

WI

Crowdfunding Law

11/7/2013

WV

Crowdfunding Law

3/15/2016

WY

Seed Capital Network Program

1/1/2012

WY

Crowdfunding Law

3/3/2016

10/22/2014

–8–

Table IA.4: Identifying Assumption – VC Analysis The dependent variable is the natural logarithm of the number of unique IT firms raising VC funding in a state. Column 1 shows the trend of the Anti-Troll Law coefficient prior to the adoption of anti-troll legislation. The remainder of the table is analogous to Table 3. Throughout the table, the sample consists of all states with a level of VC activity in 2012 at or above the median of all states; the only exception is in column 5, where the placebo sample goes from 2009Q1 to 2013Q2 and so the level of initial VC activity is measured in 2009. Anti-Troll Law is an indicator equal to one if a state has passed anti-troll legislation in or before that quarter. Each observation is a state-quarter. Robust standard errors are clustered by state. We use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively.

Ln(1 + # IT Firms Raising VC)

Dep. Var. = Coefficient

Controlling for

Trend

Regional Shocks

(1)

Anti-Troll Lawt-4

(2)

(3)

Neighbor Placebo

Excluding:

States

Timing

CA

(4)

(5)

(6)

Controlling for

CA + MA Other State Laws (7)

(8)

0.010 (0.100)

Anti-Troll Lawt-3

0.109 (0.091)

Anti-Troll Lawt-2

0.060 (0.097)

Anti-Troll Lawt-1

0.108 (0.103)

Anti-Troll Law

0.230***

0.183** 0.230***

0.176*

0.171** 0.175**

0.179**

(0.083)

(0.085) (0.081)

(0.096)

(0.064)

(0.064)

Anti-Troll Law Neighbor

(0.067)

0.000 (0.084)

Anti-Troll Lawt-12

-0.071 (0.102)

Other State Laws

-0.081 (0.089)

State Controls

yes

yes

yes

yes

yes

yes

yes

yes

State FE

yes

yes

yes

yes

yes

yes

yes

yes

Year-Quarter FE

yes

no

no

yes

yes

yes

yes

yes

Year-Quarter-Region FE

no

yes

no

no

no

no

no

no

Year-Quarter-Division FE

no

no

yes

no

no

no

no

no

No. Observations

486

486

486

486

504

468

450

486

Within R-Squared

0.037

0.024

0.026

0.033

0.016

0.034

0.034

0.036