ON TIME ON TIME ON TIME BUSINESS MANAGEMENT A+ WORK, ON TIME, NO PLAGARIZING; ON TIME

EBSCO-FullText-10_04_2026.pdf

Home >Business & Finance homework help >Management homework help > ON TIME ON TIME ON TIME BUSINESS MANAGEMENT A+ WORK, ON TIME, NO PLAGARIZING; ON TIME

Do Proprietary Costs Deter Insider Trading? Lyungmae Choi,a Lucile Faurel,b Stephen Hillegeistb,* a The University of Sydney Business School, University of Sydney, Darlington, New South Wales 2006, Australia; b W.P. Carey School of Business, Arizona State University, Tempe, Arizona 85287 *Corresponding author Contact: [email protected], https://orcid.org/0000-0003-3971-1275 (LC); [email protected],

https://orcid.org/0000-0002-7241-0995 (LF); [email protected], https://orcid.org/0000-0001-9324-4853 (SH)

Received: July 26, 2021 Revised: November 22, 2022; December 7, 2023 Accepted: December 24, 2023 Published Online in Articles in Advance: July 12, 2024

https://doi.org/10.1287/mnsc.2021.02469

Abstract. Insider trading conveys insiders’ private information to outsiders. This private information potentially benefits rival firms, which may reduce the competitive advantage of the insiders’ firms. Using a composite proprietary cost measure, we find proprietary costs are negatively associated with insiders’ purchases, especially when their trades are more likely to be informative to rivals. Consistent with proprietary costs increasing the costs of insider pur- chases and, hence, the expected benefits required to trade, insiders earn significantly higher abnormal profits when proprietary costs are higher. Exploiting settings with exogenous and event-driven variation in proprietary costs, we find insiders significantly reduce their pur- chases when noncompete agreement enforceability is high and before new product launches. Moreover, firms with higher proprietary costs are more likely to impose window-based insider trading restrictions and insiders with greater equity holdings reduce their purchases more strongly in the presence of proprietary costs. Finally, we provide evidence of real effects of insider trading on rivals’ investment decisions. We find that investments are associated with insiders’ purchases at rival firms, and these associations are stronger when proprietary costs at rivals are higher. Our findings indicate insiders and firms are aware of potential pro- prietary costs when insiders trade on private information and respond accordingly.

History: Accepted by Suraj Srinivasan, accounting. Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/

mnsc.2021.02469.

Keywords: proprietary costs • insider trading • competition • managerial learning • investment decisions

1. Introduction Insider trading is an important mechanism through which insiders’ private information is impounded into stock prices (Meulbroek 1992, Damodaran and Liu 1993, Aboody and Lev 2000). Prior research explores the deter- minants and consequences of insider trading, and largely focuses on its capital market implications. The voluntary disclosure literature focuses on the capital market benefits and product market costs of disclosures, such as management earnings forecasts (Verrecchia 1983, 2001; Dye 1986; Brown et al. 2004; Li 2010; Ali et al. 2014). An important reason why firms do not fully dis- close their private information is because it reveals pro- prietary information to rivals who may use this information to compete more effectively against disclos- ing firms. In this paper, we postulate that insider trading not only conveys insiders’ private information to capital market participants but also to rival firms. As such, a potential yet previously unanalyzed cost of insider trad- ing relates to its ability to harm the competitive position of insiders’ firms by conveying proprietary information to rival firms (i.e., insider trades incur proprietary costs). Thus, we hypothesize that insiders at high proprietary

cost firms limit their trading to avoid revealing proprie- tary information and undermining their firms’ competi- tive positions.

Although insider trading does not directly provide detailed proprietary information, such as the progress of research and development (R&D) activities or the cur- rent status of a drug trial, it conveys insiders’ views on their firms’ prospects. If, albeit noisy, inferences can be drawn from insiders’ trades, this information could prove valuable for rivals in revising product, invest- ment, and marketing strategies. Competitors do not need to rely directly on the observed insider trade, per se, to make inferences about the disclosing firm and/or the competitive landscape. Instead, observing the trade may spur rivals to increase or focus their competitive intelligence efforts on the insider’s firm more than they otherwise would, and base their strategic responses on the additional information they acquire.1 This process is similar to that documented in Chen et al. (2020, p. 119). They find insider trades prompt institutional investors to investigate the insider’s firm more closely in an effort to discover the private information the insider’s trade was based on. Furthermore, they conclude “insider

3185

MANAGEMENT SCIENCE Vol. 71, No. 4, April 2025, pp. 3185–3215

ISSN 0025-1909 (print), ISSN 1526-5501 (online) https://pubsonline.informs.org/journal/mnsc

mailto:[email protected]

https://orcid.org/0000-0003-3971-1275

mailto:[email protected]

https://orcid.org/0000-0002-7241-0995

mailto:[email protected]

https://orcid.org/0000-0001-9324-4853

https://doi.org/10.1287/mnsc.2021.02469

trades are likely amongst the most valuable [information signals] for unlocking a … comparative advantage for a given fund manager.”

We expect product market rivals react in a similar fashion and use insider trades (either directly or indi- rectly after subsequent information search activities) to make inferences about insiders’ private information. More specifically, insider purchases signal to rivals that the insiders possess private information about positive or promising developments at the insiders’ firms that have not yet been incorporated into stock prices. For instance, an insider might have private information about the success of R&D efforts. By signaling the exis- tence of this private information, insider purchases pro- vide early warnings to competitors.2 With this enhanced awareness, rival firms can respond more quickly and more effectively to the competitive threat posed by the successful R&D project at the insider’s firm. Beyond speeding up their own R&D efforts, rival firms can also acquire competing intellectual property through licens- ing agreements and acquisitions to compete more effec- tively with the insider’s firm. Rivals can also reduce their prices and lock customers in long-term contracts to limit the benefits the insider’s firm accrues with its suc- cessful R&D project. More generally, insider trading enables rivals to set their product, pricing, and market- ing strategies more effectively, adjust their production schedules and investments, and/or mimic successful business strategies (Botosan and Stanford 2005). Conse- quently, we hypothesize that insider trading imposes proprietary costs on firms, and we predict insiders reduce their purchases when their firms have higher proprietary costs.

To investigate the link between proprietary costs and insider trading, we examine whether insiders at firms with higher proprietary costs make fewer purchases. We focus on purchases because they are more likely to be based on, and hence convey, private information com- pared with sales. In addition, purchases are less likely to entail litigation and regulatory costs, which helps elimi- nate alternative explanations. Our primary measure of insider purchases is the total number of shares pur- chased by insiders during the year, scaled by total shares outstanding. We employ a composite measure, PropCost, comprised of R&D intensity (Hall and Ziedonis 2001, Wang 2007, Ellis et al. 2012, Albring et al. 2016), patent applications (King et al. 1990, Plumlee et al. 2015), and product similarity (Hoberg and Phillips 2016) to capture multiple dimensions of firms’ innovation efforts and competitive environments, and hence, proprietary costs.

Consistent with our hypothesis, we find a strong, neg- ative association between proprietary costs and insider purchases. The associations are economically significant. A one-standard-deviation increase in PropCost is associ- ated with a 10.56% decrease in insider purchases. The magnitude of the negative association increases when

insider trades are more likely to be informative to com- petitors. Specifically, our results are stronger for oppor- tunistic trades versus routine trades (Cohen et al. 2012), top executives versus other insiders (Peress 2010, Cheng et al. 2016), low versus high complexity firms (Frankel et al. 2006), small versus large firms (Bernard 2016), and high versus low media coverage firms (Dai et al. 2015).3 Overall, the results indicate that insiders take proprietary costs into account when making their purchase decisions.

Next, we investigate how trade profitability varies with proprietary costs. Insiders face a tradeoff between their own financial benefits and their firms’ proprietary costs. To trade, the expected benefits must be greater than the associated costs. Thus, when insiders at high proprietary cost firms trade, we hypothesize they earn relatively higher profits to offset the higher proprietary costs. The results are consistent with our hypothesis: Insider purchase profitability is positively associated with proprietary costs. A one-standard-deviation increase in PropCost is associated with annual abnormal stock returns that are 3.34%–4.96% higher on average. Thus, although insiders engage in fewer purchases when pro- prietary costs are higher, their purchases are more likely to be based on (more valuable) private information.

We provide additional evidence using two alternative identification strategies that enable us to exploit event- driven, and even exogenous, variation in proprietary costs. Doing so helps alleviate concerns that unobserva- ble and time-varying firm characteristics drive our results. First, we investigate variation in the enforce- ability of employee noncompete agreements (Garmaise 2011). Higher enforceability impedes rivals from dis- covering private information by hiring a firm’s employees (Aobdia 2018), thereby increasing the pro- prietary costs of insider trades. Using a difference-in- differences research design, we find insiders reduce (increase) their purchases when enforceability exoge- nously increases (decreases) in the state in which their firm is headquartered. Second, we analyze insider pur- chases prior to new product announcements when we expect proprietary costs are especially high. We find insiders significantly reduce their purchases during these proprietary periods relative to other periods, con- sistent with proprietary concerns discouraging insider trading. Together, our results indicate that proprietary costs are negatively associated with (and even causally reduce) insider purchases.

Although the financial gains solely accrue to insiders, proprietary costs are borne by shareholders. We examine two specific mechanisms that potentially drive the nega- tive association between proprietary costs and insider purchases. First, firms with high proprietary costs may adopt or more strictly enforce existing insider trading restrictions. Many firms use event-specific “blackout” periods such as prior to new product announcements,

clinical trials, or mergers and acquisitions (Bettis et al. 2000, Roulstone 2003). Using inferred measures of black- out policies (Roulstone 2003), we find firms are more likely to use insider trading restrictions and/or more strictly enforce existing policies when proprietary costs are higher. Second, insiders may voluntarily refrain from trading if their incentives are closely aligned with those of shareholders. Our results show the negative associa- tion between proprietary costs and insider purchases is significantly stronger when insiders have larger owner- ship stakes. We also find that the positive association between proprietary costs and trading profits is larger when insider ownership is higher, which further indi- cates that high ownership insiders internalize proprietary costs to a greater extent.

The mechanism that underlies our hypotheses is that rival firms (directly or indirectly) use the private infor- mation conveyed by insider purchases to undermine the firms’ competitive position. Similar to the voluntary dis- closure literature, we are unable to provide direct evi- dence that insider purchases reduce the value of high proprietary cost firms. Alternatively, we provide empiri- cal support for the underlying mechanism. Specifically, we show that the degree of managerial learning from rivals’ stock prices (Foucault and Fresard 2014) is higher when rival insiders make more purchases, and that this moderating effect is stronger when proprietary costs at rivals are higher. These findings are novel and identify real effects of insider purchases on rival firms. These results are consistent with the underlying mechanism whereby insiders’ trades convey proprietary informa- tion to rivals, and hence, impose proprietary costs on their firms. Moreover, we provide descriptive evidence that is also consistent with the underlying mechanism. First, we find firms regularly download information about their rivals’ insider trades from the Securities and Exchange Commission’s (SEC’s) EDGAR server. Second, we find the majority of insider trading policies explicitly prohibit trading based on specific events and/or types of private information that have high proprietary costs (e.g., related to new products, new services, and/or R&D projects).

This study makes at least three contributions to the lit- erature. First, it adds to the literature on the determi- nants of insider trading. We identify proprietary costs as a previously unexamined determinant of both insiders’ purchase decisions and trading profits. Prior studies examine the associations between insider trading and lit- igation risk (Johnson et al. 2007, Billings and Cedergren 2015, Cheng et al. 2016, Huang et al. 2024), information asymmetry (Aboody and Lev 2000, Huddart and Ke 2007, Wu 2021), and product market power (Peress 2010). As discussed in Section 4.2, our proprietary cost hypothesis generates a unique set of empirical predic- tions that allows us to distinguish our results from these alternative determinants of insider purchases and

profitability. Our results are likely to be of interest to regulators and directors in setting insider trading poli- cies and assist investors in interpreting insider trading signals.

Second, this paper complements the literature on the proprietary costs of voluntary disclosure (Botosan and Stanford 2005, Ellis et al. 2012, Bernard 2016, Huang et al. 2017). Prior studies indicate that proprietary costs inhibit voluntary disclosures as firms evaluate the costs and benefits. Insider trades can be viewed as a form of involuntary disclosure because insiders are not inten- tionally disclosing their private information when they trade. We provide the first empirical evidence on how firm-level proprietary costs are associated with insider trading decisions, based on insiders’ own costs and benefits.

Third, this study contributes to the literature that examines how managers learn investment-related infor- mation from rivals’ disclosures and stock prices. Prior literature indicates that firms can learn investment- related information directly from their rivals’ disclosures about earnings and restatements (Durnev and Mangen 2009, Badertscher et al. 2013) and indirectly from rivals’ stock prices (Foucault and Fresard 2014, Edmans et al. 2017). We find that firms’ investment decisions are influ- enced (directly and indirectly) by insider purchases at their rivals and that the degree of influence is increasing with rivals’ proprietary costs. These findings are impor- tant and novel because they indicate that insider pur- chases have real effects on rivals’ investment decisions and document a mechanism through which insider pur- chases impose proprietary costs.

2. Hypothesis Development Insider trading occurs when corporate insiders, includ- ing officers, directors, and large shareholders, trade in public financial markets. It is illegal for insiders to trade while in possession of material nonpublic information (Securities and Exchange Acts of 1933 and 1934; Insider Trading Sanctions Act of 1984; Insider Trading and Secu- rities Fraud Enforcement Act of 1988). Informally, mate- rial, nonpublic (i.e., “private”) information consists of any information that could reasonably be expected to affect the price of the security. Examples include infor- mation about upcoming accounting disclosures, capital market transactions (i.e., large equity or debt issuances, mergers and acquisitions, stock repurchases), R&D out- comes, and new product or service launches. Violations are punishable by imprisonment, forfeiture of gains, and financial penalties of up to three times the profits gained.

Despite its illegality, insiders frequently trade on their private information as they earn abnormal profits, on average (Seyhun 1986, Aboody and Lev 2000, Ravina and Sapienza 2010). Insiders profitably trade on private

information about corporate events such as mergers and acquisitions, modified audit reports, Chapter 11 bank- ruptcy filings, stock repurchases, earnings restatements, and SEC investigations (Gosnell et al. 1992, Agrawal and Jaffe 1995, Badertscher et al. 2011, Blackburne et al. 2021, Cziraki et al. 2021, Arif et al. 2022). Despite widespread evidence that insiders trade on private information, insi- ders are rarely prosecuted for insider trading violations (Del Guercio et al. 2017, Davidson and Pirinsky 2022).

The SEC has long required insiders to report their trading activities on Form 4 filings. As of June 30, 2003, the SEC also mandates electronic filings through the EDGAR server. Insider trading reports are widely dis- seminated by newspapers, business magazines, and websites, which enables easy access to this information (Dai et al. 2015). Investors and other capital market parti- cipants are aware that insiders trade on private informa- tion and react accordingly (see Bhattacharya (2014) for a review of this extensive literature). Chen et al. (2020) report that Form 4 insider trading filings are the most frequently downloaded form by institutional investors. They find that when institutional investors mimic tracked insider purchases, their trades are highly profit- able and generate abnormal returns of 12% per year, on average. Importantly, institutional investors are able to distinguish between informative and noninformative tracked insider trades. Thus, a tracked insider trade appears to motivate institutional investors to investigate the insider’s firm and base their trading decisions on the outcomes of these investigations. In this way, insider trading serves as a valuable source of private informa- tion to the capital markets.

Bettis et al. (2000) report that more than 90% of sur- veyed firms have adopted some type of insider trading policy (ITP). They argue ITPs are designed primarily to reduce the negative externalities that privately informed insider trading imposes on shareholders rather than for legal protection reasons. ITPs typically include “black- out” policies that establish restricted insider trading per- iods, especially during periods leading up to and shortly following quarterly earnings announcements (Bettis et al. 2000, Jagolinzer et al. 2011). In addition, most ITPs delin- eate several events (e.g., upcoming acquisitions, securi- ties issuances, product launches) and/or types of private information that insiders are specifically prohibited from trading on. To provide descriptive evidence, we review 166 ITPs from Jagolinzer et al. (2011) that includes at least one specific example of what constitutes material, nonpublic information. We search each ITP for prohibitions that are related to proprietary costs.4 We find 96 ITPs (58%) identify “new product(s)” or “new product development” or the equivalent, 27 (16%) iden- tify “new service(s)” or “new markets” or the equiva- lent, and 43 (26%) identify “R&D projects(s)” or “R&D outcomes” or the equivalent. Finally, 104 (63%) identify at least one of New Products, New Services, or R&D.

The fact that firms mention these specific items indicates that trading on them imposes negative externalities. We expect that revealing proprietary information is among the negative externalities associated with insider trading and firms prohibit trading based on these items, at least in part, because they are concerned about the associated proprietary costs.5

We posit that information conveyed by insider trades is also useful to product market rivals. We expect com- petitors find this information to be particularly relevant for two reasons. First, insider trading contains firm- specific information. Other sources of information, such as analyst reports, often convey industry-specific or mac- roeconomic information (Piotroski and Roulstone 2004), which is already accessible to rivals. Second, insider trading signals are credible because they involve insi- ders’ personal wealth, unlike management forecasts that are subject to “cheap talk” considerations (Stocken 2000).

Signaling private information to competitors repre- sents a potential cost to the insider’s firm because rivals can use the information to disadvantage the insider’s firm in the product market (Healy and Palepu 2001). Although insider purchases do not provide detailed pro- prietary information directly, they signal that the insider possesses private information about positive or promis- ing developments at the insider’s firm that have not yet been incorporated into stock prices. For instance, the insider might have private information about the firm’s financial prospects, expected sales growth, investment outcomes, new product launch, progress of R&D efforts, clinical trials results, upcoming Food and Drug Admin- istration (FDA) approval decisions, technological ad- vancements, and so on. We expect that observing insider trades causes rivals to enhance and/or focus their corpo- rate intelligence efforts on the insider’s firm.6 Such behavior is similar to how insider trades prompt institu- tional investors to search for the private information underlying insiders’ trades (Chen et al. 2020), as dis- cussed previously. Thus, insider purchases serve to increase rivals’ awareness of positive or promising developments at the insiders’ firms. Awareness of com- petitors’ actions is critical as it is one of the three key dri- vers of interfirm rivalry intensity (alongside motivation and capability), where greater awareness leads to more intense competition (Smith et al. 2001).

By providing early warnings, insider purchases allow rival firms to respond more quickly to the competitive threats posed by the insider’s firm than they would otherwise have. As an illustration, consider insider pur- chases prior to a new product launch (which we exam- ine empirically below). If competitors are unaware that a firm is developing a new product, then observing insider purchases may spur rivals to focus their competi- tive intelligence efforts on the firm, and thus, discover the existence of the new product. Alternatively, if

competitors were previously aware that the firm is developing a new product, then observing insider pur- chases before the product launch may signal the pro- gress and potential success of the product development efforts.

In either case, and likely after additional investigation, rivals could take actions that would reduce the value of the new product. Rivals frequently respond to new product introductions by reducing prices and increasing their marketing and branding efforts (Marion 1998, Deb- ruyne et al. 2002). Wang and Shaver (2016) find that firms reposition existing products and adjust their own product launch strategies in response to a competitor’s product launch. In addition, rivals could hire employees from the insiders’ firm who are developing the new product. Doing so would both slow the new product development efforts and provide valuable competitive intelligence. Rivals could also take actions designed to reduce demand for the new product, such as signing exclusivity contracts with distributors and/or retail outlets.

Responding quickly to new product introductions is crucial for rivals because delays allow the new product to become firmly entrenched (Edwards 2014).7 By quickly responding to a new product introduction, rivals can adversely affect the durability of the first mover advan- tages and reduce the potential profits from the new prod- uct. Lee et al. (2000) find that when rivals respond quickly to a new product introduction, the first mover firm experiences negative abnormal returns. Thus, either directly or indirectly, insider trading enables competitors to set their product, pricing, and marketing strategies more effectively, adjust their production schedules and investments, and/or mimic successful business strategies (Botosan and Stanford 2005). Consequently, we hypothe- size that insider trading imposes proprietary costs on firms and that they are higher when firms are more vul- nerable to competitors’ actions.

To provide additional insights into this plausible but relatively undocumented practice of firms obtaining riv- als’ insider trades, we analyze SEC downloads of a firm about its competitors using data from Bernard et al. (2020).8 We find that 93.4% of sample firms download at least one insider trading filing (Form 3, 4, or 5) of a com- petitor. Furthermore, firms download at least one com- petitor’s insider trading filing in 61.3% of the sample years. These percentages represent conservative esti- mates as the data collection method undercounts the true number of downloads (Bernard et al. 2020). This evidence indicates that firms believe they can extract valuable information from their rivals’ insider trading activities. If their beliefs are correct, then firms will incur proprietary costs when their insiders trade on pro- prietary private information. Accordingly, we expect insiders to reduce their trading activities to avoid propri- etary costs.

Another reason why insiders at high proprietary cost firms refrain from purchasing shares is because the asso- ciated proprietary costs will negatively affect the firm’s value. To avoid revealing proprietary information, insi- ders will reduce their purchase activities. In the extreme, if proprietary costs are so high that insider purchases result in lower firm values, then insiders will refrain from trading altogether as such trades would be unprofitable.

In summary, we expect competitors are able to par- tially infer, directly or indirectly, the private information underlying insider purchases. Hence, we predict insi- ders reduce their purchases when their firms have higher proprietary costs.9 Accordingly, we make the fol- lowing hypothesis.

Hypothesis 1. Insider purchases are negatively associated with proprietary costs.

Although decreasing their frequency, proprietary costs will increase the average profitability of the re- maining information-based purchases. Insiders weigh their expected costs and benefits when they make trad- ing decisions. In order for insiders to be willing to trade, the expected financial gains must exceed the expected costs. While insiders obtain financial gains from trading on their private information, they face potential costs from both internal sources (e.g., sanctions for violating the firm’s insider trading policy) and external sources (e.g., unfavorable publicity, civil liability, and criminal prosecution). We expect proprietary costs increase the potential costs to an insider for at least two reasons. First, the associated proprietary costs will negatively affect the value of the insider’s stock and option holdings by reducing the future value of the firm. Second, the firm may be more likely to discipline the insider, or impose stronger penalties, if the trade violates its insider trading policy when proprietary costs are higher. Therefore, the expected trading profits must be higher for insiders at high proprietary cost firms to offset the higher associ- ated costs. In which case, we expect that insiders’ trading profits are higher when proprietary costs are higher. Accordingly, we hypothesize the following.

Hypothesis 2. The profitability of insider purchases is pos- itively associated with proprietary costs.

Although the financial benefits from trading accrue to insiders, the associated proprietary costs are incurred by shareholders. There are at least two, nonmutually exclu- sive mechanisms or channels through which insiders reduce trading activities in the face of high proprietary costs. First, firms with high proprietary costs may impose and/or more strictly enforce insider trading restrictions to prevent costly information transfers. For instance, in conjunction with trading window restrictions, many firms employ event-specific blackout periods such as prior to announcements about new

products, clinical trials, and/or mergers and acquisi- tions. Second, insiders may voluntarily refrain from trading if their equity holdings cause them to sufficiently internalize the associated proprietary costs. In which case, we expect insiders voluntarily reduce their trading activities in response to proprietary costs to a greater extent. Similar to Hypothesis 2, we also expect that insi- ders require higher trading profits when their equity holdings are higher to be willing to purchase shares. Accordingly, we make the following hypotheses.

Hypothesis 3A. The likelihood of imposing or more strictly enforcing insider trading restrictions is positively associated with proprietary costs.

Hypothesis 3B. The negative association between proprie- tary costs and insider purchases is stronger for firms in which insiders have higher share ownership.

Hypothesis 3C. The positive association between proprie- tary costs and the profitability of insider purchases is stron- ger for firms in which insiders have higher share ownership.

3. Sample and Variable Measurement 3.1. Sample Selection We obtain data on insiders’ open market stock pur- chases from the Thomson Reuters Insider Filings data- base, which provides transactions by insiders subject to disclosure requirements under the Securities Exchange Act of 1934. We focus on top five executives (chief execu- tive officer (CEO), chief financial officer (CFO), chief operating officer (COO), president, and chairman of the board) as they are more likely to possess proprietary information (Peress 2010, Cheng et al. 2016) and thus, be more sensitive to the proprietary costs associated with information transfers. We obtain financial data from Compustat, stock price data from CRSP, analyst forecast data from I/B/E/S, institutional and executive owner- ship data from Thomson Reuters, and news coverage information from RavenPack News Analytics. Patent data are obtained from the Darden School of Business’ Global Corporate Patent Data set (Bena et al. 2017) and product similarity measures from the Hoberg-Phillips Data Library (https://hobergphillips.tuck.dartmouth. edu). Finally, we gather data on noncompete agreement enforceability scores from table 4 of Ertimur et al. (2018), product announcement data from S&P Capital IQ, and federal judge ideology from the Federal Judicial Center’s website (https://www.fjc.gov).

We use two samples in our tests: (i) a firm-year sam- ple for the purchase analyses and (ii) an insider purchase transactions sample for the profitability analyses. The sample period and size vary due to differences in data availability. Our primary analyses are based on samples of 71,255 firm-years (9,152 distinct firms) and 132,328 insider purchases (at 6,337 distinct firms) between 1997 and 2017.

3.2. Measuring Proprietary Costs As discussed in Section 2, a firm possesses proprietary information (e.g., expected sales growth, progress of R&D efforts, clinical trials results, FDA approval deci- sions, technological advancements, etc.) that it seeks to keep confidential. Proprietary costs relate to the deterio- ration of the firm’s competitive position, and hence, the firm’s value, if proprietary information is obtained by competitors (Healy and Palepu 2001). King et al. (1990) argue property rights associated with innovations and intangible assets are not perfectly enforceable and are a primary source of proprietary costs when competitors expropriate some of their value. Moreover, firms face competitive pressures from different sources, including competition from existing rivals, competition from potential rivals, technological competition, and direct product market competition (Bloom et al. 2013, Hoberg and Phillips 2016, Cao et al. 2018). Thus, a firm’s proprie- tary costs depend primarily on the nature of the firm’s innovative and operational activities (e.g., R&D, invest- ments, production, distribution, products, and services) and the characteristics of the firm’s competitive environ- ment (Healy and Palepu 2001).

Because of the multidimensional aspect of proprietary costs, we use several time-varying, firm-specific mea- sures from prior literature to estimate these costs.10 The first measure is R&D intensity. R&D activities stimulate product innovation and technological change. Thus, a firm’s R&D expenditures represent how actively the firm engages in innovative activities, which arguably carry significant amounts of proprietary information (Hall and Ziedonis 2001). Consequently, firms with higher R&D expenditures tend to face higher proprie- tary costs (Wang 2007, Ellis et al. 2012, Albring et al. 2016). For example, Ellis et al. (2012) find that firms with higher R&D expenditures are less likely to disclose major customers. Their results are consistent with higher R&D firms having higher proprietary costs. We measure R&D intensity, R&D, as the decile ranking of R&D expenditures divided by total sales. Observations with missing or zero R&D expenditures represent 61% of all observations. Accordingly, we use decile rankings, and assign a zero rank to observations with missing or zero R&D expenditures.11

Our second measure is the annual number of patent applications. Firms with more patent filings likely pos- sess higher degrees of confidentiality and thus face higher proprietary costs. This measure is widely used in the literature to capture innovation (Aghion et al. 2005, He and Tian 2013). We use patent filings rather than pat- ent grants as the former better captures the creation of patents and thus the time of innovation (Griliches et al. 1987, He and Tian 2013, Chang et al. 2015).12 Due to the skewness of the data, we define PatentApps as the natu- ral logarithm of one plus the number of patent applica- tions filed.13

https://hobergphillips.tuck.dartmouth.edu

https://www.fjc.gov

Our third measure (ProdSimilarity) is a firm-level mea- sure of product similarity that captures the intensity of competition from existing product market rivals. The intuition for this measure is that the greater the similar- ity between a firm’s products and products from other firms, the more substitutable the products are, and hence, the greater the competitive pressures the firm faces. Thus, high product similarity reflects intense product market competition. Prior studies (Gow et al. 2021, Ryou et al. 2022) use product similarity to measure product market competition. We expect the proprietary costs of insider purchases increase with the intensity of product market competition and therefore product simi- larity. ProdSimilarity is a text-based measure of product similarity using product descriptions in 10-K filings (Hoberg and Phillips 2010, 2016). The measure is esti- mated using firm-by-firm pairwise cosine similarity scores to quantify product similarity between two firms. ProdSimilarity for a firm is the sum of the pair- wise similarities between the firm and all other Com- pustat firms.

Because each measure captures different dimensions of proprietary costs and is not equally applicable to all firms, we construct a composite measure, PropCost, simi- lar to Albring et al. (2016). Moreover, commonly used measures of proprietary costs, including the ones used in this study, are relatively imprecise and suffer from measurement error (Beyer et al. 2010). Thus, another advantage of using a composite measure is that it reduces the effects of idiosyncratic measurement errors on our results. We standardize each measure (R&D, PatentApps, and ProdSimilarity) to have zero mean and unit variance and then sum these standardized values to obtain PropCost.14

Furthermore, we consider two other firm-level mea- sures of proprietary costs. First, TechSimilarity captures technological similarity. The intuition for this measure is consistent with ProdSimilarity: The greater the similarity between a firm’s technologies and those of other firms, the more substitutable the technologies are, and hence, the greater the competitive pressures the firm faces. TechSimilarity is based on firm-by-firm pairwise techno- logical similarity scores that are estimated using the cosine similarity of 127 patent technology classes from the Cooperative Patent Classification (Jaffe 1986, Bloom et al. 2013). TechSimilarity for a firm is the natural loga- rithm of the sum of pairwise technological similarities between the firm and all other Compustat firms for the sample period. Second, CompWords is a firm-level mea- sure of competition based on the number of competition- related words (e.g., “competition,” “competitor,” … ) scaled by the total number of words in a 10-K filing (Li et al. 2013). Tables OA1 and OA2 in the Online Appen- dix present results of our main specifications using these two measures along with the three individual measures in PropCost.15

3.3. Insider Trading Variables Insider purchases, Purchases, is the total number of shares purchased by insiders during the year, scaled by the total number of shares outstanding at the beginning of the year.16 We use this measure because it captures the intensity of trading by insiders and thus likely reflects the private information that the trades are based on. This measure is similar to those used by prior research (Beneish and Vargus 2002, Brochet 2010). Be- cause the range of Purchases is a small positive interval around zero, we multiply Purchases by 1,000.

We use three measures of trading profits estimated over the 360 calendar days following the purchase date.17 Daily alpha, Alpha, is the intercept from the Car- hart (1997) four-factor model (Jagolinzer et al. 2011, Dai et al. 2015). BHAR is the buy-and-hold abnormal stock returns estimated using the four-factor model (Ravina and Sapienza 2010, Dai et al. 2016). Raw is buy-and-hold raw stock returns.

3.4. Summary Statistics Descriptive statistics and correlation coefficients are pro- vided in Tables 1 and 2, respectively. Insider purchases are relatively infrequent and occur in 27.1% of firm-year observations. The mean (median) value of PropCost is 0.092 (�0.724). PropCost exhibits a high degree of varia- tion. The standard deviation is 1.859 and the interquar- tile range varies between �1.517 and 1.408. Inside ownership is relatively small: the mean (median) owner- ship is 1.9% (0.3%).

When classifying firm-year observations by PropCost, the mean value of Purchases is 0.469 and 0.693, respec- tively, for high and low PropCost portfolios (Table 1, Panel B). This provides prima facie evidence of a nega- tive association between proprietary costs and insider purchases, consistent with Hypothesis 1. Similarly, Prop- Cost is negatively correlated with Purchases. However, the magnitude of the correlation is modest (�0.05). Inter- estingly, while InsideOwn is negatively correlated with PropCost (�0.14), it is positively correlated with Purchases (0.12). The correlations between PropCost and the control variables are relatively modest, and none are larger than 0.15 in magnitude.

In untabulated descriptive statistics, the individual proprietary cost measures in PropCost, R&D, PatentApps, and ProdSimilarity, exhibit substantial variation with high standard deviations. The correlations between the individual measures are varied, as they reflect different dimensions of proprietary costs. R&D and PatentApps are positively correlated (0.45), as expected. ProdSimilar- ity is negatively correlated with both R&D and Paten- tApps, although the magnitudes are modest (both 0.14).18 Overall, the correlation results highlight the mul- tidimensional aspects of competition and indicate that PropCost captures proprietary costs for a wide variety of firms.

4. Analyses and Results 4.1. Proprietary Costs and Insider Purchases 4.1.1. Proprietary Costs and Insider Purchase Activ- ity. To analyze the association between proprietary costs and insider purchases, we estimate the following Tobit regression for firm i in year t with standard errors clus- tered by firm and year:

Purchasesi, t � α + β1PropCosti, t�1 + β2Sizei, t�1

+ β3BMi, t�1 + β4PriorReti, t�1

+ β5Coveragei, t�1 + β6InstOwni, t�1

+ β7Turnoveri, t�1 + β8Litigationi, t�1

+ β9Volatilityi, t�1 + β10InsideOwni, t�1

+ β11Spreadi, t�1 + X γj Industryj

+ X δtYeart + εi, t: (1)

Hypothesis 1 predicts that the coefficient on PropCost will be negative. We use a Tobit model because the majority of firm-year observations have values of Pur- chases equal to zero, which corresponds to the corner solution (type I) Tobit model (Woolridge 2002). The appendix provides all variable definitions.

We include several firm characteristics that are associated with insider trading decisions (Clinch 1991, Ke et al. 2003, Frankel and Li 2004, Piotroski and Roul- stone 2005, Cheng and Lo 2006, Huddart et al. 2007, Ravina and Sapienza 2010, Gao et al. 2014, Billings and Cedergren 2015, Hillegeist and Weng 2021). We include firm size (Size), book-to-market ratio (BM), lagged stock returns (PriorRet), analyst coverage (Cov- erage), institutional ownership (InstOwn), share turn- over (Turnover), ex ante litigation risk (Litigation), stock return volatility (Volatility), and the percentage

Table 1. Descriptive Statistics

Panel A: Full sample

Q1 Mean Median Q3 Standard deviation

Proprietary cost measure PropCost �1.517 0.092 �0.724 1.408 1.859

Insider purchases Purchases 0.000 0.581 0.000 0.018 2.893

Firm characteristics Size 4.401 5.959 5.917 7.407 2.099 BM 0.295 0.635 0.520 0.828 0.736 PriorRet �0.197 0.155 0.071 0.349 0.660 Coverage 0.000 1.368 1.609 2.485 1.223 InstOwn 0.148 0.471 0.483 0.771 0.331 Turnover �0.634 0.056 0.160 0.823 1.074 Litigation 0.080 0.307 0.192 0.459 0.291 Volatility 0.019 0.034 0.028 0.042 0.021 InsideOwn 0.001 0.019 0.003 0.013 0.047 Spread 0.024 0.043 0.036 0.054 0.025

Panel B: High and low PropCost

Variable

High PropCost Low PropCost p value for test of mean differences

p value for test of median differencesMean Median Mean Median

PropCost 1.590 1.409 �1.404 �1.517 <0.01 <0.01 Purchases 0.469 0.000 0.693 0.000 <0.01 - Size 5.993 5.813 5.924 6.017 <0.01 <0.01 BM 0.570 0.468 0.699 0.572 <0.01 <0.01 PriorRet 0.161 0.062 0.150 0.077 0.02 <0.01 Coverage 1.423 1.609 1.313 1.386 <0.01 <0.01 InstOwn 0.485 0.502 0.457 0.462 <0.01 <0.01 Turnover 0.110 0.218 0.001 0.099 <0.01 <0.01 Litigation 0.332 0.214 0.283 0.173 <0.01 <0.01 Volatility 0.035 0.029 0.032 0.027 <0.01 <0.01 InsideOwn 0.014 0.003 0.024 0.004 <0.01 <0.01 Spread 0.044 0.038 0.041 0.034 <0.01 <0.01

Notes. Panel A presents selected descriptive statistics. Panel B presents descriptive statistics for insider-year observations classified as high (low) PropCost, where high (low) PropCost are observations above (below) the industry-year median of PropCost. The sample covers the period 1997 to 2017. PropCost is a composite measure of proprietary costs, defined as the sum of R&D, PatentApps, and ProdSimilarity, where each measure is standardized to have zero mean and unit variance. See the Online Appendix for additional variable definitions. All continuous variables are winsorized at 1% and 99%.

of shares held by top five insiders (InsideOwn). We include the annual average of daily bid-ask spreads (Spread) because information asymmetry is positively associated with insider trading (Aboody and Lev 2000, Ellul and Panayides 2018). We include industry and year fixed effects to control for systematic varia- tion in insider trading across industries and time.

Table 3 presents the results from estimating Equation (1). The results in column 1 show the PropCost coefficient is negative and highly significant (t statistic � �3.38). In robustness tests, we re-estimate Equation (1) using three alternative measures of insider purchases: (i) the dollar value of purchases scaled by beginning-of-year market capitalization (similar to Huddart and Ke 2007), (ii) the number of purchase transactions scaled by the number of active insiders, where an active insider is one who has at least one insider stock transaction during our sample period (Ke et al. 2003, Peress 2010), and (iii) the likeli- hood of at least one insider purchase occurring (Massa et al. 2015). The untabulated results are qualitatively similar to those in Table 3 and our inferences remain unchanged. Table OA1 in the Online Appendix reports the results from estimating Equation (1) after alterna- tively replacing PropCost with each of the individual pro- prietary cost measures. All the individual proprietary cost coefficients are negative and significant (t statistics range from �1.83 to �2.95).19 Although each individual measure captures different aspects of proprietary costs, these results are consistent with insiders reducing their insider purchases when proprietary costs are higher. Collectively, our results support Hypothesis 1.

In economic terms, a one standard deviation increase in PropCost is associated with a 0.270 share decrease in the predicted value of Purchases. This magnitude repre- sents a 46.47% decrease relative to the sample mean pur- chases. This form of marginal effect describes how the unobserved latent trading incentives change with respect to changes in proprietary costs. An alternative type of

marginal effect is with respect to how the expected value of the observed purchases changes as proprietary costs vary. This estimated marginal effect is �0.033, which represents a 10.56% decrease relative to the mean shares purchased by insiders. Thus, for both types of marginal effects, the results are economically significant.

The coefficients on the control variables are generally consistent with prior literature. Insider purchases are neg- atively associated with firm size, lagged stock returns, and institutional ownership, and are positively associated with book-to-market ratio and bid-ask spreads. Inside ownership is not significantly associated with insider purchases. This result is inconsistent with a potential alternative explanation whereby high ownership levels deter insiders from making additional purchases and insiders at firms with high proprietary costs having higher ownership levels. Finally, litigation risk is posi- tively associated with insider purchases, which is incon- sistent with litigation risk deterring insider purchases.

Opportunistic vs. Routine Trading. Cohen et al. (2012) classify insider trading as “routine” (i.e., less likely to be based on private information) or “op- portunistic” (i.e., more likely to be based on private in- formation). They find opportunistic trades are more informative in terms of predicting future firm stock returns, news, and events compared with routine trades. If concerns about proprietary costs influence insiders’ trading decisions, then we expect these con- cerns affect opportunistic trades more than routine trades, as they are likely to attract greater attention from outsiders, including rivals.

To test this prediction, we classify insider purchases as opportunistic or routine (Cohen et al. 2012). Given insiders’ highly undiversified portfolios, insider pur- chases are often thought to be primarily motivated by private information, and thus, rarely routine. However, routine purchases often occur after an insider receives a

Table 2. Correlation Coefficients

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

(1) PropCost 1.00 20.02 0.09 20.16 20.01 0.14 0.08 0.13 0.14 0.04 20.16 0.05 (2) Purchases 20.05 1.00 20.22 0.10 20.14 20.17 20.18 20.11 20.05 0.12 0.09 0.11 (3) Size 0.13 20.18 1.00 20.37 0.01 0.77 0.54 0.46 0.21 20.52 20.49 20.47 (4) BM 20.12 0.06 20.29 1.00 0.11 20.29 20.19 20.28 20.15 0.06 0.11 0.02 (5) PriorRet 0.01 20.02 20.10 0.11 1.00 0.06 0.09 20.07 20.03 20.06 20.02 20.06 (6) Coverage 0.15 20.13 0.76 20.20 0.01 1.00 0.52 0.51 0.28 20.29 20.39 20.25 (7) InstOwn 0.09 20.15 0.55 20.15 0.04 0.54 1.00 0.47 0.19 20.21 20.28 20.17 (8) Turnover 0.14 20.07 0.43 20.18 20.05 0.50 0.48 1.00 0.47 0.16 20.19 0.22 (9) Litigation 0.15 0.00 0.08 20.06 0.03 0.16 0.08 0.39 1.00 0.38 20.10 0.41 (10) Volatility 0.05 0.13 20.52 0.12 0.12 20.30 20.28 0.12 0.39 1.00 0.28 0.95 (11) InsideOwn 20.14 0.12 20.29 0.06 0.02 20.24 20.25 20.19 20.01 0.20 1.00 0.26 (12) Spread 0.07 0.12 20.49 0.09 0.12 20.27 20.25 0.16 0.43 0.93 0.16 1.00

Notes. This table presents Pearson (Spearman) correlation coefficients below (above) the diagonal. The sample covers the period 1997 to 2017. PropCost is a composite measure of proprietary costs, defined as the sum of R&D, PatentApps, and ProdSimilarity, where each measure is standardized to have zero mean and unit variance. Bold correlation coefficients are statistically significant at the p < 0.10 level. See the Online Appendix for additional variable definitions. All continuous variables are winsorized at 1% and 99%.

bonus, which is generally paid out in the same calendar month every year (Cohen et al. 2012). If an insider trades in the same calendar month for three consecutive years, then all subsequent trades by this insider are classified as routine trades. If an insider trades in three consecutive years and the trades do not meet the routine classifica- tion criteria, then all subsequent trades are classified as opportunistic unless the insider subsequently makes trades that meet the routine classification criteria. Based on this classification scheme, 35,348 purchase trans- actions (27%) are classified as routine, 17,419 (13%) as opportunistic, and 79,561 (60%) are unclassified. This classification rate is similar to that in Cohen et al. (2012). Clearly, this scheme produces classification

errors. These errors reduce our ability to reject the null hypothesis of no differences in the associations between proprietary costs and routine and opportu- nistic purchases.

Columns 2 and 3 present the results from estimating Equation (1) where Opportunistic Purchases and Routine Purchases are the respective dependent variables. The results indicate proprietary costs are negatively associated with both opportunistic and routine purchases (t statistics � �3.31 and �2.13, respectively). The absolute magnitude of the PropCost coefficient is twice as large for opportunis- tic purchases compared with routine purchases (�0.044 versus �0.022).20 A Wald chi-square test indicates the PropCost coefficients are significantly different (p � 0.02).

Table 3. Proprietary Costs and Insider Purchases

Variable

Dependent variable

Purchases Opportunistic Purchases Routine Purchases (1) (2) (3)

PropCost �0.145*** �0.044*** �0.022** (�3.38) (�3.31) (�2.13)

Size �0.666*** �0.106*** �0.049*** (�10.98) (�5.49) (�3.94)

BM 0.191** 0.017 �0.017 (2.27) (0.85) (�0.95)

PriorRet �0.808*** �0.124*** �0.099*** (�4.29) (�2.76) (�4.04)

Coverage 0.078 0.042** 0.008 (1.46) (2.16) (0.47)

InstOwn �1.547*** �0.273*** �0.171** (�4.62) (�2.89) (�2.44)

Turnover 0.092 0.052** �0.051*** (1.35) (2.56) (�2.73)

Litigation 0.436*** 0.122** 0.050 (2.85) (2.07) (1.35)

Volatility �6.969 �3.371** �3.345* (�1.26) (�2.41) (�1.91)

InsideOwn 1.819 �0.507* 0.420 (1.50) (�1.66) (1.35)

Spread 16.938*** 2.581* 1.593 (3.42) (1.80) (1.21)

p value for the test of equal (2) vs. (3) coefficients on PropCost 0.02 Industry fixed effects Yes Yes Yes Year fixed effects Yes Yes Yes Observations 71,255 57,750 57,750 Pseudo-R2 (%) 3.63 7.02 6.23

Notes. This table presents the results from Tobit regressions of proprietary costs on insider purchases. Standard errors are clustered by firm and year. t statistics are reported in parentheses below the coefficients. The sample covers the period 1997 to 2017. In column 1, the dependent variable is Purchases, the number of shares purchased by insiders during the year scaled by total number of shares outstanding at the beginning of the year. In column 2 (column 3), the dependent variable is Opportunistic Purchases (Routine Purchases), the number of opportunistic (routine) purchases during the year scaled by total number of shares outstanding at the beginning of the year. Insider purchases are classified as opportunistic or routine based on insiders’ trading patterns in the three preceding years. If an insider trades in the same calendar month for three consecutive years, then all subsequent trades by this insider are classified as routine. If an insider trades in three consecutive years and the trades do not meet the routine classification criteria, then all subsequent trades are classified as opportunistic. An insider may have trades classified as opportunistic and, subsequently, trades classified as routine if these subsequent trades meet the routine classification criteria. PropCost is a composite measure of proprietary costs, defined as the sum of R&D, PatentApps, and ProdSimilarity, where each measure is standardized to have zero mean and unit variance. Industry and year fixed effects are included but not reported for brevity. Industry grouping is defined based on the Fama and French (1997) 48-industry classification. See the Online Appendix for additional variable definitions. All continuous variables are winsorized at 1% and 99%.

***, **, and *Statistical significance at p < 0.01, p < 0.05, and p < 0.10 (two-tailed), respectively.

Overall, the results in Table 3 suggest insiders reduce their purchases when proprietary costs are higher. The reductions are concentrated among trades that are most likely to be based on private information, and hence, where proprietary costs are higher. Together, these results support Hypothesis 1 and are consistent with insiders limiting their purchases to avoid the risk of revealing proprietary information.

4.1.2. Proprietary Costs and Insider Trading Profits. In this section, we examine whether insiders’ trading prof- its over the 360 calendar days subsequent to the transac- tion date are higher when proprietary costs are higher. We estimate the following ordinary least squares (OLS) regression at the insider-purchase date level for insider k at firm i on transaction date d and year t, with standard errors clustered by firm and year:

Profitk , i, d, t � α + β1PropCosti, t�1 + β2Sizei, t�1

+ β3BMi, t�1 + β4PriorReti, [d�380;d�20]

+ β5Coveragei, t�1 + β6InstOwni, t�1

+ β7Turnoveri, [d�380;d�20]

+ β8Volatilityi, [d�380;d�20]

+ β9TradeSizek , i, d, t + β10Spreadk, i, d, t

+ X γjIndustryj +

X δtYeart + εk, i, d, t,

(2)

where Profit is one of our three measures of insider trad- ing profits, Alpha, BHAR, and Raw.

Following Brochet (2010), Ravina and Sapienza (2010), Skaife et al. (2013), and Gao et al. (2014), we include sev- eral firm- and transaction-level control variables in Equation (2). Firm-level controls include firm size (Size), book-to-market ratio (BM), analyst following (Coverage), institutional ownership (InstOwn), and the average bid- ask spread (Spread). Transaction-level controls include lagged stock returns (PriorRet), share turnover (Turn- over), stock return volatility (Volatility), and trade size (TradeSize). We include industry and year fixed effects.

The results from estimating Equation (2) are reported in Panel A of Table 4. The results indicate insiders at firms with higher proprietary costs earn significantly higher profits from their purchase transactions (t statis- tics range from 2.76 to 2.99). The economic magnitudes are significant as well. When daily alpha is used to mea- sure profitability, a one standard deviation increase in proprietary costs is associated with an increase in daily abnormal stock returns of 1.326 basis points, which corre- sponds to a 3.34% increase in annual returns. The mar- ginal effects are stronger for the other two profitability measures: 4.96% (4.79%) higher annual abnormal returns (raw returns). Overall, these findings suggest that when insiders at firms with higher proprietary costs purchase

shares, their trades are more likely to be based on (more valuable) private information. These findings provide empirical support for Hypothesis 2 and suggest proprie- tary costs increase the “hurdle rate” insiders use to evalu- ate their purchase opportunities.

Cohen et al. (2012) suggest that only opportunistic trades yield abnormal profits. Accordingly, we estimate Equation (2) separately for opportunistic and routine purchases and present the results in Panel B. The results for each profitability measure show the PropCost coeffi- cient is positive and significant (t statistics range from 1.77 to 2.14) for opportunistic purchases. The magnitudes of the coefficients are also 63%–93% larger than their Panel A counterparts. In contrast, only two of the Prop- Cost coefficients are significant (both at the 10% level) for routine purchases. Furthermore, Wald chi-square tests for all three profitability measures reject the null hypothe- sis that the PropCost coefficients are equal for opportunis- tic and routine purchases (p values range from <0.01 to 0.03). Thus, our results indicate opportunistic trades made at firms with higher proprietary costs earn signifi- cantly higher profits compared with routine trades at such firms. These results further support Hypothesis 2.

4.2. Alternative Explanations 4.2.1. Litigation and Regulatory Risk. An alternative explanation for the negative association between propri- etary costs and insider purchases is that proprietary costs are correlated with litigation and regulatory risk, and these risks are the ultimate factor driving the nega- tive association we document. This alternative explana- tion does not seem plausible in our setting primarily because these risks are quite low when insiders purchase their own stock before good news. For example, private class action suits on behalf of shareholders occur when shareholders suffer damages sufficiently large to make lawsuits worthwhile. However, shareholders do not suf- fer obvious damages when insiders purchase shares as stock prices do not decrease.21

The results in Table 3 are inconsistent with this alternative explanation as the coefficients on Litigation, the litigation risk measure based on predicted litigation probability from Kim and Skinner (2012), are significantly positive in columns 1 and 2 and insignificant in column 3. To provide additional evidence, we use an alternative measure of litigation risk, LiberalCourt, which is based on federal judge ideology (Huang et al. 2024). PropCost and LiberalCourt are positively correlated, but the magnitude (0.12) is relatively modest. The results when LiberalCourt replaces Litigation in Equation (1) are tabulated in the Online Appendix (Table OA3). The LiberalCourt coeffi- cient is insignificant, whereas the coefficient on PropCost is negative and significant at the 1% level.

One could argue that the influence of litigation risk on the association between proprietary costs and insider purchases varies nonlinearly. To address this concern,

Table 4. Proprietary Costs and Insider Purchase Profits

Panel A: All purchases

Variable

Dependent variable

Alpha BHAR Raw (1) (2) (3)

PropCost 0.008*** 0.030*** 0.029*** (2.76) (2.93) (2.99)

Size �0.022*** �0.055*** �0.042*** (�4.14) (�3.42) (�2.76)

BM 0.007** 0.051*** 0.042** (2.13) (2.96) (2.44)

PriorRet[-380,-20] 0.021*** 0.060*** 0.067*** (6.57) (4.22) (4.52)

Coverage 0.046*** 0.209*** 0.244*** (2.94) (4.27) (4.63)

InstOwn �0.026*** �0.048** �0.046* (�4.94) (�2.30) (�1.78)

Turnover[-380,-20] �0.025*** �0.032* �0.017 (�5.81) (�1.86) (�0.72)

Volatility[-380,-20] 2.011** 2.216* 2.392 (2.50) (1.75) (1.12)

TradeSize 0.001 0.001 �0.001 (1.56) (1.10) (�0.70)

Spread 0.621 0.580 0.464 (1.40) (0.65) (0.26)

Industry fixed effects Yes Yes Yes Year fixed effects Yes Yes Yes Observations 73,734 75,866 75,702 Adjusted R2 (%) 17.48 5.42 11.97

Panel B: Opportunistic and routine purchases

Variable

Dependent variable

Alpha BHAR Raw

Opportunistic purchases

Routine purchases

Opportunistic purchases

Routine purchases

Opportunistic purchases

Routine purchases

(1) (2) (3) (4) (5) (6)

PropCost 0.013* 0.006 0.058** 0.029* 0.051* 0.032* (1.77) (1.05) (2.14) (1.92) (1.95) (1.78)

p value for the test of equal coefficients on PropCost

(1) vs. (2) (3) vs. (4) (5) vs. (6) <0.01 <0.01 0.03

Control variables Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Observations 10,271 15,900 10,549 16,293 10,528 16,261 Adjusted R2 (%) 19.67 22.04 8.43 10.39 13.69 19.19

Notes. This table presents the results from OLS regressions of proprietary costs on insider trading profits conducted at the insider-purchase date level. Standard errors are clustered by firm and year. t statistics are reported in parentheses below the coefficients. The sample covers the period 1997 to 2017. Panel A reports insider trading profits for all insider purchases. Panel B reports insider trading profits for opportunistic purchases in columns 1, 3, and 5 and routine purchases in columns 2, 4, and 6. Insider purchases are classified as opportunistic or routine based on insiders’ trading patterns in the three preceding years. If an insider trades in the same calendar month for three consecutive years, then all subsequent trades by this insider are classified as routine. If an insider trades in three consecutive years and the trades do not meet the routine classification criteria, then all subsequent trades are classified as opportunistic. An insider may have trades classified as opportunistic and, subsequently, trades classified as routine if these subsequent trades meet the routine classification criteria. The dependent variable is insider trading profits estimated over the 360 calendar days following the transaction date using daily alpha (Alpha) the intercept from the Carhart (1997) four-factor model, buy-and-hold abnormal stock returns (BHAR) using the Carhart (1997) four-factor model, and buy-and-hold raw stock returns (Raw). PropCost is a composite measure of proprietary costs, defined as the sum of R&D, PatentApps, and ProdSimilarity, where each measure is standardized to have zero mean and unit variance. PriorRet, Turnover, and Volatility are measured over the period [�380, �20] before each trade, where day 0 is the transaction date. Industry and year fixed effects (and control variables in Panel B) are included but not reported for brevity. Industry grouping is defined based on the Fama and French (1997) 48-industry classification. See the Online Appendix for additional variable definitions. All continuous variables are winsorized at 1% and 99%.

***, **, and *Statistical significance at p < 0.01, p < 0.05, and p < 0.10 (two-tailed), respectively.

we divide firm-year observations into litigation risk ter- ciles alternatively using Litigation or LiberalCourt based on industry-year cutoff points. We estimate Equation (1) within each tercile. The results based on Litigation (Liber- alCourt) are presented in the Online Appendix (Table OA4, Panel A (Panel B)). The results show that the asso- ciation between proprietary costs and insider purchases is negative and significant within each tercile of litigation risk (t statistics range from �2.07 to �2.52 in Panel A and from �2.19 to �2.39 in Panel B). Moreover, the dif- ferences in the PropCost coefficients between the lowest and highest terciles are not significant. Overall, these results provide further evidence that our results are not driven by litigation risk.

Additionally, we examine whether proprietary costs are associated with the likelihood of class action litiga- tion. Using data from the Securities Class Action Filings data from Stanford Law School to identify class action litigation, we create an indicator variable, ClassAction, that equals one if a firm has a class action lawsuit in year t, and zero otherwise. We then estimate a logit version of Equation (1) after replacing the dependent variable with ClassAction. The untabulated results indicate the Prop- Cost coefficient is insignificant (t statistic � 1.44). Thus, we find no evidence that proprietary costs are associated with higher litigation risk, and hence, these results do not support the litigation-based alternative hypothesis.

Although insiders may face a low risk of litigation from their insider purchases (regardless of proprietary costs), they could be exposed to regulatory risk in terms of SEC enforcement actions. Thus, if proprietary costs are positively associated with regulatory risk, then this association could also explain our results. We explore this possibility by performing a set of analyses similar to those performed for litigation risk. We use data from Davidson and Pirinsky (2022) on SEC enforcement activ- ity related to insider trading. First, we include Enforce12 (Enforce3), in Equation (1), where Enforce12 (Enforce3) is the natural logarithm of the number of insider trading enforcement actions during the twelve (three) months before the beginning of year t. The results in Table OA5 in the Online Appendix show that, although neither of the two enforcement variables are significant, both of the PropCost coefficients are negative and significant at the 1% level. Second, we create terciles of regulatory risk based on enforcement activities during the previous 12 (3) months before the beginning of year t. We estimate Equation (1) within each tercile. The results in Table OA6 in the Online Appendix show that the PropCost coefficients are negative and significant at the 5% level for all terciles of Enforce12 (Panel A) and at the 1% level for the lowest and highest terciles of Enforce3 (Panel B). Furthermore, there are no significant differences in the PropCost coefficients between the lowest and the highest tercile for either enforcement measure. Third, we create an indicator variable, Investigated, that equals one if the

firm is investigated by the SEC for a Section 16 Insider Trading violation during or after year t, and zero other- wise. We then estimate a logit version of Equation (1) after replacing the dependent variable with Investigated. The untabulated results indicate the PropCost coefficient is insignificant (t statistic � �1.12).22 Thus, we find no evidence that proprietary costs are associated with higher regulatory risk.23

Finally, Huddart et al. (2007) argue that insiders face low levels of legal, regulatory, and reputational costs (collectively referred to by them as “jeopardy” costs) when they trade shortly after a quarterly earnings an- nouncement (QEA). We re-estimate Equation (1) where the dependent variable equals the number of shares pur- chased during the safe period ([QEA+ 3, QEA+ 22]) scaled by the number of shares outstanding at the begin- ning of the year. The untabulated results show that the PropCost coefficient is negative and significant. Thus, we continue to find that proprietary costs are negatively associated with insider purchases even when insider trades are completed during periods with low jeopardy costs.

4.2.2. Information Asymmetry. Another alternative exp- lanation for our results is that they are driven by the rela- tion between information asymmetry and insider trad- ing because two of our empirical proxies for proprietary costs (R&D and PatentApps) are positively correlated with information asymmetry.24 Prior theoretical litera- ture, including the seminal studies of Grossman and Sti- glitz (1980) and Kyle (1985), find that the expected trading profits of privately informed investors (which includes firm insiders) are increasing when information asymmetry is higher.25 The empirical literature provides supporting evidence. Aboody and Lev (2000) and Hud- dart and Ke (2007) use R&D-based indicator variables to distinguish between high and low information asymme- try firms, whereas Ellul and Panayides (2018) and Wu (2021) use exogenous reductions in analyst coverage to identify increases in asymmetry. In each study, higher information asymmetry is positively associated with the profitability of insider trades. These results suggest that insiders at high R&D firms, which have high levels of information asymmetry, exploit the private information they have regarding their firm’s R&D activities and, given the importance of these activities for the firm’s future prospects, reap higher trading profits from their insider trades.26 As PropCost is positively correlated with measures of information asymmetry, our insider profit- ability results could be driven by information asymme- try instead of proprietary costs.

When information asymmetry between insiders and investors is high, insiders have more profitable opportu- nities to trade on their private information. Consistent with this observation, prior studies predict and find that insiders at firms with higher information asymmetry

trade more. For example, Huddart and Ke (2007) show that in an extension of the model in Kyle (1985) where the insider’s private information is imperfect, the quan- tity an insider trades increases with the precision of their private information (i.e., when information asymmetry is higher).27 The underlying reason is that more precise private information is positively associated with the expected price adjustment, and hence, the insider trades more aggressively. Consistent with this prediction, they find that trading activity (purchases and sales combined) is significantly higher among insiders at R&D firms compared with non-R&D firms (their table 6). Aboody and Lev (2000) report similar (untabulated) univariate results. In addition, Wu (2021) finds that following an exogenous decrease in analyst coverage, there are signif- icant increases in both insider purchase activity and the probability of insider purchases (his tables 8 and 9, respectively).

In summary, the literature on insider trading and information asymmetry predicts and finds that higher information asymmetry is positively associated with the profitability of insider trades and is positively associated with insider trading activity. These results are consistent with insiders in high information asymmetry firms exploiting their private information by engaging in more insider trading activity and receiving higher trad- ing profits. In contrast, we hypothesize and provide empirical evidence that while higher proprietary costs are positively associated with insider trading profits, they are negatively associated with insider purchases. These results suggest that insiders in high proprietary firms refrain from purchasing their own stock to avoid revealing proprietary information but earn higher trad- ing profits when they trade. This prediction regarding how proprietary costs are related to insider trading activity is in the opposite direction than the prediction on how information asymmetry is related to insider trading activity. As our results are inconsistent with one of the two predictions of the information asymmetry explanation, we do not think that this alternative expla- nation is driving our results.28

4.2.3. Earnings Volatility. In his model of the relation between product market competition and insider trad- ing, Peress (2010) analytically shows that higher product market competition increases earnings volatility, which in turn reduces insider trading. He finds empirically that insider purchases are negatively associated with product market competition. Thus, results in Peress (2010) sug- gest an alternative explanation for the negative associa- tion between proprietary costs and insider purchases: Insiders reduce their purchases in highly competitive product markets due to higher uncertainty caused by higher earnings volatility.

The results in Table 3 provide mixed support for this alternative explanation as Volatility is not significantly

associated with insider purchases in column 1 but is neg- atively associated in columns 2 and 3. To explore this issue further, we construct the variable PCM, the industry-adjusted price-cost margin, which is an inverse measure of competition. Peress (2010) finds that PCM is positively associated with insider trading (purchases and sales combined), which indicates that higher levels of competition result in less insider trading. The correla- tion between PropCost and PCM is small and negative (�0.07), suggesting the two variables capture different underlying constructs.

We include PCM as an additional variable in Equation (1). The results are tabulated in the Online Appendix (Table OA7). The results show that the PropCost coeffi- cient is significantly negative (t statistic � �3.51). In con- trast, the PCM coefficient is small in magnitude and statistically insignificant (t statistic � �0.73). Thus, the results support our interpretation that insiders reduce their purchases when proprietary costs are higher as opposed to the volatility-based explanation in Peress (2010).29

4.3. Event-Driven Changes in Proprietary Costs The evidence presented in Table 3 is consistent with pro- prietary costs reducing insider purchases. However, it is possible our proprietary cost measures are correlated with omitted firm characteristics that are also associated with insider purchase decisions. To help address this concern, we examine two distinct events with elevated proprietary costs, including one which is uniquely exog- enous to firms. These analyses enable us to exploit event-driven, and even exogenous, variation in proprie- tary costs.

4.3.1. Insider Purchases and Noncompete Agreement Enforceability. A noncompete agreement (NCA) is a contract between a firm and its employee that prevents the employee from joining (or forming) a rival com- pany. NCAs are used frequently as 80% of S&P 1500 firms have NCAs with their top executives (Garmaise 2011). NCAs represent an important mechanism that binds employees to a firm and helps mitigate the reve- lation of proprietary information after employee depar- tures (Garmaise 2011). There is substantial variation in NCA enforceability across states ranging from low enforceability states (e.g., California) where NCAs are essentially unenforceable to high enforceability states (e.g., Florida) where firms are likely to prevail over for- mer employees in court (Garmaise 2011, Ertimur et al. 2018). Garmaise (2011) finds an increase in NCA enforceability reduces the frequency that executives move to within-industry rivals and increases average tenure length (Marx et al. 2009, Ertimur et al. 2018). Thus, NCA enforceability has powerful effects on employee mobility.

When a rival hires a firm’s employee, it can access proprietary information that often extends beyond tech- nical knowledge, including operational and client data, future strategic plans, and details of the firm’s profit areas (Aobdia 2018). Accordingly, Aobdia (2015, 2018) argue that higher NCA enforceability reduces the amount of proprietary information rivals can obtain through hiring the firm’s employees. Without access to employees, rivals rely more on the firm’s public disclo- sures to obtain proprietary information. Therefore, the proprietary costs associated with disclosures (and insider purchases) increase with NCA enforceability. Aobdia (2018) provides supporting evidence and finds firms reduce their disclosure activities when NCA enforceability is high. Consistent with this idea, Aobdia (2015) finds firms headquartered in high NCA en- forceability states are less likely to share auditors, which reflects heightened information-spillover concerns. Changes in NCA enforceability are exogenous to firms because NCA enforceability is determined by each indi- vidual state. Thus, by providing exogenous variation in proprietary costs, changes in NCA enforceability allevi- ate potential endogeneity and measurement issues pre- sent with other proprietary cost measures (Beyer et al. 2010, Aobdia 2018). Accordingly, we expect that increases (decreases) in NCA enforceability result in decreases (increases) in insider purchases because the risk that insider trades reveal proprietary information increases (decreases) with NCA enforceability.30

We estimate the following Tobit regression for firm i in year t with standard errors clustered by state and year:

Purchasesi, t � α + β1Enforce∆i, t + β2Sizei, t�1 + β3BMi, t�1

+ β4PriorReti, t�1 + β5Coveragei, t�1

+ β6InstOwni, t�1 + β7Turnoveri, t�1

+ β8Litigationi, t�1 + β9Volatilityi, t�1

+ β10InsideOwni, t�1 + β11Spreadi, t�1

+ X γjIndustryj +

X δtYeart

+ X λs States + εi, t:

(3)

Enforce∆ is an indicator variable equal to 1 (�1) if there is an increase (decrease) in the firm’s headquarters state’s NCA enforceability score in the current or any prior year and 0 otherwise (Garmaise 2011). Hypothesis 1 predicts that the coefficient on Enforce∆ will be negative, which suggests insider purchases are negatively associated with state-level changes in NCA enforceability. We obtain state-level NCA enforceability scores from table 4 of Ertimur et al. (2018). The index varies from zero (extremely low enforceability) to nine (extremely high enforceability) and is available from

1986 through 2013.31 We include industry, year, and state fixed effects to control for other industry-, year-, and state-specific factors correlated with Enforce∆ that could also be associated with insider purchases. Impor- tantly, including year and state fixed effects causes Equation (3) to be equivalent to a difference-in- differences (DnD) research design because Enforce∆ is a state level variable (Bertrand et al. 2004, Defond and Lennox 2017).32 Therefore, Enforce∆ is only identified by state-level increases or decreases in enforceability. Three states have changes in Enforce∆ of two (or more), whereas 10 states have one or more changes equal to one.

Next, we estimate Equation (3) after replacing Enforce∆ with BigEnforce∆, an indicator variable equal to 1 (�1 ) if there is a large increase (decrease) in the firm’s headquarters state’s enforceability score of two or more in the current or any prior year and 0 otherwise. The three states with large changes are (i) Texas (with a change of �2 in 1994 due to a Texas Supreme Court deci- sion); (ii) Florida (with a change of +2 in 1996 due to a new state law); and (iii) Louisiana (with a change of �4 in 2001 due to a Louisiana Supreme Court decision and a change of +4 in 2003 due to a new state law).

Table 5, columns 1 and 2, present the results from esti- mating Equation (3) with Enforce∆ and BigEnforce∆, respectively, as the independent variables. As expected, the coefficient on Enforce∆ is negative and significant (t statistic��2.46). Similarly, the coefficient on BigEnforce∆ is negative and significant (t statistic � �2.95). The magni- tudes of the Enforce∆ and BigEnforce∆ coefficients in col- umns 1 and 2 indicate insiders purchase 12.42% and 19.39% fewer shares, respectively, in states with increased NCA enforceability compared with states with no increase. These findings suggest insiders significantly change their purchasing behavior after their firms’ headquarters states change the level of NCA enforceability.

Following Aobdia (2018), we estimate a specification that focuses on the change in enforcement in Texas because of the large number of firms headquartered in this state coupled with the large decrease in NCA enforceability.33 We replace Enforce∆ in Equation (3) with Texas × Post.34 Texas is an indicator variable equal to one if the firm is headquartered in Texas, and zero otherwise. Post is an indicator variable equal to �1 for years 1995 or 1996, the two years after the 1994 Texas Supreme Court decision that decreased the NCA enforceability score by 2, and 0 for years 1992–1994.35

Based on Hypothesis 1 and the results in columns 1 and 2, we expect the coefficient on Texas × Post will be nega- tive, which is consistent with insiders increasing their purchasing activity following the large decrease in NCA enforcement, relative to the concurrent changes by insi- ders whose firms are headquartered in other states. The results presented in column 3 show that the Texas × Post coefficient is negative and significant (t statistic ��2.19),

Table 5. Insider Purchases and Noncompete Agreement Enforceability

Dependent variable: Purchases

Variable

All High

enforceability Low

enforceability

(1) (2) (3) (4) (5) (6)

Enforce∆ �0.361** �0.047 (�2.46) (�0.19)

BigEnforce∆ �0.557*** (�2.95)

Texas × Post �0.634** (�2.19)

LocalRival 0.774** (2.30)

Enforce∆ × LocalRival �1.527** (�2.11)

PropCost �0.120** �0.009 (�2.02) (�0.10)

Size �0.737*** �0.732*** �0.874*** �0.677*** �0.595*** �0.958*** (�10.20) (�10.44) (�7.25) (�8.06) (�6.47) (�7.18)

BM 0.270*** 0.267*** 0.349 0.235** 0.113 0.519** (2.68) (2.76) (1.42) (2.11) (0.85) (2.20)

PriorRet �0.729*** �0.736*** �0.308*** �0.757*** �0.778*** �0.732*** (�4.91) (�4.96) (�14.99) (�3.96) (�4.00) (�4.27)

Coverage 0.040 0.046 �0.017 0.069 0.117 0.233* (0.72) (0.88) (�0.16) (1.12) (1.34) (1.90)

InstOwn �1.990*** �1.918*** �3.131*** �1.778*** �1.640*** �1.810*** (�6.87) (�6.45) (�5.43) (�4.90) (�4.00) (�4.17)

Turnover 0.175*** 0.175*** 0.158 0.128 0.058 0.089 (2.86) (2.89) (1.25) (1.61) (0.64) (0.81)

Litigation 0.466** 0.450** �0.274 0.504*** 0.502* 0.482 (2.57) (2.56) (�1.25) (3.55) (1.95) (1.40)

Volatility 8.629** 9.048*** 3.508 �4.563 �0.843 �13.278 (2.51) (2.67) (0.32) (�0.93) (�0.12) (�1.48)

InsideOwn 1.838** 1.765** 0.340 2.158 5.100*** �0.328 (2.21) (2.17) (0.48) (1.42) (2.81) (�0.19)

Spread �1.252 �1.096 �1.614 16.541*** 14.644** 14.180*** (�0.47) (�0.41) (�0.27) (3.48) (1.99) (2.92)

p value for the test of equal coefficients on PropCost

(5) vs. (6) 0.11

Industry fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes State fixed effects Yes Yes Yes Yes Yes Yes Observations 82,099 85,017 15,880 54,388 28,586 12,239 Pseudo-R2 (%) 3.25 3.23 2.62 3.84 3.83 4.19

Notes. This table presents the results from Tobit regressions of insider purchases following noncompete agreement enforcements. Standard errors are clustered by state and year. t statistics are reported in parentheses below the coefficients. The sample covers the period 1986 to 2013 in columns 1 and 2, 1992 to 1996 in column 3, 1998 to 2013 in column 4, and 1997 to 2013 in columns 5 and 6. The dependent variable is Purchases, the number of shares traded purchased by insiders during the year scaled by total number of shares outstanding at the beginning of the year. Enforce∆ is an indicator variable equal to 1 (�1) if there is an increase (decrease) in the firm’s headquarters state’s enforceability score in the current or any prior year, and 0 otherwise. We exclude firms headquartered in states where the enforceability score increases and then decreases in a later year, or vice versa. BigEnforce∆ is an indicator variable equal to 1 (�1) if there is a large increase (decrease) in the firm’s headquarters state’s enforceability score of two or more in the current or any prior year and 0 otherwise. Texas is an indicator variable equal to one if the firm is headquartered in Texas, and zero otherwise. Post is an indicator variable equal to �1 for years 1995 or 1996, the two years after the 1994 Texas Supreme Court decision that decreased the NCA score enforceability by two, and 0 for years 1992 to 1994. LocalRival is the decile ranking of the total number of establishments in a given three-digit NAICS industry present in the firm state minus one, divided by the total number of establishments with the same NAICS in the United States minus one. Columns 5 and 6 report results for observations classified as high and low enforceability, respectively, where high (low) enforceability states have enforceability scores in the highest (lowest) tercile each year. PropCost is a composite measure of proprietary costs, defined as the sum of R&D, PatentApps, and ProdSimilarity, where each measure is standardized to have zero mean and unit variance. Industry, year, and state fixed effects are included but not reported for brevity. Industry grouping is defined based on the three-digit NAICS industry classification. See the Online Appendix for additional variable definitions. All continuous variables are winsorized at 1% and 99%.

***, **, and *Statistical significance at p < 0.01, p < 0.05, and p < 0.10 (two-tailed), respectively.

as expected. In summary, the results in columns 1–3 pro- vide strong evidence that state-level changes in NCA enforceability cause subsequent changes in the purchase activities of insiders whose firms are headquartered in those states. These results are consistent with insiders changing their purchase behavior due to changes in the proprietary costs associated with their purchases.

Next, we examine the effects of local competition on our results. These analyses are germane because NCAs are rarely enforceable beyond the state in which they are issued (Garmaise 2011, Marx 2011). Thus, these tests are more powerful because they capture the influence of NCAs on insiders’ purchase decisions conditional on the presence of local rivals. Employee mobility between local competitors will be high when local markets are highly competitive and NCA enforceability is weak (Gardner 2005, Boschma et al. 2009). Under such condi- tions, the proprietary costs of disclosure are lower because rivals have greater access to the underlying pro- prietary information. Consistent with lower proprietary costs under these conditions, Breschi and Lissoni (2009) find disclosure activities are positively associated with the presence of local rivals. In contrast, local competitors are less able to learn proprietary information by hiring a firm’s employees when NCA enforceability is high. This lack of mobility, in turn, increases the proprietary costs of disclosures. Consistent with this prediction, Aobdia (2018) finds NCA enforceability moderates the positive effect of local rivals on disclosure activities.

Accordingly, we examine whether the negative asso- ciation between NCA enforceability and insider pur- chases is affected by the presence of local rivals. We include two additional variables in Equation (3): (i) Local- Rival, the decile ranking of the total number of same- industry establishments in a given state minus one, divided by the total number of same-industry establish- ments in the United States, minus one and (ii) the inter- action term Enforce∆ × LocalRival. In line with Aobdia (2018), we identify industry rivals using the North American Industry Classification System (NAICS) at the three-digit level to ensure an adequate number of local rivals per firm. Because LocalRival is defined using three- digit NAICS industries, we include industry fixed effects based on three-digit NAICS industry groupings for this analysis.

The results are presented in column 4. As expected, the interaction variable, Enforce∆ × LocalRival is signifi- cantly negative (t statistic � �2.11). In addition, the Enforce∆ coefficient is insignificant (t statistic � �0.19). These results indicate that the effects of NCAs on insider purchases are driven by the degree of in-state competi- tion, which is consistent with NCAs being enforced at the state level (Garmaise 2011, Marx 2011, Aobdia 2018).

Finally, Enforce∆ captures state-level differences in proprietary costs (Garmaise 2011, Aobdia 2015, Ertimur et al. 2018). As such, it ignores firm-specific variation in

proprietary costs that are not related to NCA enforce- ability. To capture both sources of variation, we divide the sample based on whether firm-year observations are headquartered in high or low enforceability states. States are classified as high (low) enforceability if they have enforceability scores that are in the highest (lowest) ter- cile in each sample year, which includes scores from five to nine (zero to two).36 We estimate Equation (1) sepa- rately on each sample to examine insiders’ purchase decisions in settings where firm-specific proprietary costs are expected to be higher or lower. We expect that firm-level proprietary costs influence insiders’ trading decisions more in states where NCA enforceability is high because rivals have fewer alternative ways of obtaining proprietary information.

The results are presented in columns 5 and 6. As expected, the PropCost coefficient is negative and signifi- cant (t statistic � �2.02) in the high enforceability sam- ple. In contrast, the PropCost coefficient in the low enforceability sample is insignificant (t statistic � �0.10). The absolute magnitude of the PropCost coefficient is much larger for the high enforceability sample com- pared with the low enforceability sample (�0.120 versus �0.009), yet a Wald chi-square test indicates the coeffi- cients are not significantly different (p � 0.11). These results are consistent with more proprietary information conveyed by insider trades in high NCA enforceability states, most likely because rivals are less able to obtain the underlying proprietary information through other means, such as hiring the firm’s employees.

Together, the results in Table 5 provide further sup- port for Hypothesis 1. Using exogenous changes in pro- prietary costs, our findings indicate that insiders reduce their purchase activities when the costs of revealing pro- prietary information are higher. As such, they help reduce concerns that our prior results could be biased due to endogeneity concerns.

4.3.2. Insider Purchases Before Product Launches. In this section, we examine how insider purchases are related to new product announcements. As discussed in Section 2, we expect proprietary costs are especially high prior to new product launches due to the potential reac- tions by competitors imposing substantial proprietary costs. Consistent with proprietary cost concerns, insider trading policies frequently prohibit insiders from trad- ing based on private information regarding products under development and often impose blackout periods before new product announcements (Bettis et al. 2000). Therefore, we expect insiders are less likely to purchase shares during product development periods because proprietary costs are especially high then.

We obtain product related announcements for the 2002 to 2017 period from S&P Capital IQ. This database starts in 2002 and includes mostly unscheduled corporate infor- mation events from national newswires and newspapers,

including Reuters, Dow Jones, Comtex, Bloomberg Busi- ness News, and CNN. To ensure the announcements refer to new products, we require the press releases to include either “introduce” and “new” or “launch” and “new” in the headlines. To account for possible selection biases in Capital IQ’s coverage decisions, we restrict the sample to firms with at least one product announcement during the sample period. This procedure initially results in 10,393 product announcements, which is reduced to 6,754 announcements after matching to Compustat and excluding observations with missing data. To cleanly identify proprietary periods before product launches, we exclude observations that have overlapping proprietary and nonproprietary periods due to multiple product launches. This process results in a final sample of 3,551 new product announcements made by 3,304 distinct firms.

To test whether insiders reduce purchase activities during preannouncement proprietary periods, we esti- mate the following Tobit regression for firm i with prod- uct announcement date d in year t with standard errors clustered by firm and year:

Purchasesi, [d�(365∗k);d�1], t

� α + β1PropPeriodi, [d�(365∗k);d�1] + β2Sizei, t�1

+ β3BMi, t�1 + β4PriorReti, t�1 + β5Coveragei, t�1

+ β6InstOwni, t�1 + β7Turnoveri, t�1 + β8Litigationi, t�1

+ β9Volatilityi, t�1 + β10InsideOwni, t�1 + β11Spreadi, t�1

+ X γjIndustryj +

X δtYeart + εi, [d�(365∗k);d�1], t: (4)

PropPeriod is an indicator variable equal to one if the observation is in a proprietary period and zero other- wise. A proprietary period is the k-year long calendar- year period (k� 1, 2) immediately prior to product announcement date d.

Columns 1 and 2 of Table 6 present the results from estimating Equation (4) with two- and one- year proprie- tary periods, respectively. The coefficients on PropPeriod are negative and significant (t statistics � �2.47 and �2.11, respectively). The magnitudes of the PropPeriod coefficients indicate insiders purchase 21.80% and 18.03% fewer shares during proprietary periods com- pared with nonproprietary periods, respectively. These results are consistent with insiders reducing purchases prior to new product announcements to reduce the risk of proprietary information transfers.37

As an alternative specification, we divide the sample based on whether the observations are during proprie- tary or nonproprietary periods. We estimate Equation (1) separately on each sample. This specification allows us to distinguish between the effects of proprietary costs during periods when insiders are expected to be more or

less cognizant of potential information transfers. The results in columns 3 and 4 show that during high propri- etary cost periods, the PropCost coefficients are negative and significant (t statistics � �2.96 and �4.24, respec- tively). In contrast, the PropCost coefficient in column 5 is insignificant (t statistic � �0.40) during nonproprietary periods when concerns about information transfers are lower.

Overall, the results in Table 6 are consistent with insi- ders considering potential proprietary costs when mak- ing their purchase decisions during periods before product launch announcements.38 These findings pro- vide additional evidence supporting Hypothesis 1.

4.4. Proprietary Costs and Insider Purchases by Informativeness of Insider Trading

Hypothesis 1 implies that the negative association between proprietary costs and insider purchases is stronger when insider trades are more informative to rivals. We use four proxies for how informative insider purchases are likely to be to rivals. For each proxy, we divide the sample into high- and low-cost samples. We then separately estimate Equation (1) for each sample and report the results in Table 7.

First, top five executives are the most likely insiders to possess proprietary information (Peress 2010, Cheng et al. 2016), and hence, their trades are likely to be more informative than trades by other insiders. Accordingly, the association between proprietary costs and insider purchases should be relatively stronger for top five executives compared with other insiders. Thus, we include non–top five insider purchase transactions solely for this test. We classify each insider purchase as com- pleted by a top five insider (CEO, CFO, COO, president, and chairman of the board) or a non–top five insider (other officers and directors). The results are reported in columns 1 and 2, respectively. As expected, the Prop- Cost coefficient is negative and significant for top five purchases (t statistic � �3.38) but is insignificant for non-top five purchases (t statistic � 0.58). Moreover, the coefficient magnitudes (�0.145 and 0.020, respec- tively) are significantly different (p < 0.01). Thus, the results are consistent with top-level insiders being more sensitive to proprietary costs and reducing their purchases accordingly.

Second, we expect firm complexity is negatively associated with the amount of private information con- veyed by insiders’ purchase decisions (Frankel et al. 2006, Cohen and Lou 2012). For example, Frankel et al. (2006) find analyst reports are less informative for firms with multiple segments because of high information processing costs. Similarly, we expect the difficulty riv- als have in interpreting and drawing specific inferences about the firm’s prospects and its products from insider purchases is increasing with firm complexity.

Thus, the ability of rivals to act on insider trading sig- nals (and hence impose proprietary costs) is decreasing with complexity.

We classify firm-year observations into low and high complexity samples, where the low (high) complexity firms have single (multiple) business or geographic seg- ments for the year. The results are reported in columns 3 and 4. Both PropCost coefficients are significantly nega- tive (t statistics � �2.77 and �2.29, respectively). How- ever, the magnitude of the low complexity coefficient is more than twice as large as the high complexity coeffi- cient (�0.238 and �0.105, respectively). Moreover, the magnitudes are significantly different (p � 0.01). Thus, the results suggest insiders at less complex firms are

relatively more sensitive to proprietary costs when mak- ing their purchase decisions.

Third, we expect the ability of rivals to interpret and act on insider trading signals decreases with firm size. In addition to small firms generally being less complex, small firms are more vulnerable to potential actions by rivals (Bernard 2016). Hence, we expect the negative association between proprietary costs and insider pur- chases is relatively stronger for smaller firms. We classify firm-year observations into small and large samples, where small (large) firms have below (above) median industry market capitalization at the beginning of the year. The results in columns 5 and 6 are consistent with our prediction. Although both PropCost coefficients are

Table 6. Insider Purchases Before Product Launches

Variable

Dependent variable: Purchases

All Proprietary periods Nonproprietary periods

2 years 1 year 2 years 1 year (1) (2) (3) (4) (5)

PropPeriod �0.161** �0.136** (�2.47) (�2.11)

PropCost �0.013*** �0.023*** �0.010 (�2.96) (�4.24) (�0.40)

Size �0.181*** �0.195*** �0.116*** �0.141*** �0.210*** (�4.04) (�4.38) (�41.91) (�53.26) (�4.18)

BM 0.086*** 0.071* 0.215*** 0.221*** 0.053 (2.68) (1.92) (18.46) (18.25) (1.40)

PriorRet 0.017 0.020 0.079*** 0.182*** 0.002 (0.22) (0.26) (11.99) (38.59) (0.03)

Coverage �0.024 �0.022 �0.047*** �0.086*** �0.012 (�0.66) (�0.61) (�6.21) (�11.14) (�0.32)

InstOwn �0.279** �0.254** �0.584*** �0.681*** �0.203* (�2.51) (�2.37) (�21.12) (�24.72) (�1.88)

Turnover �0.009 0.009 �0.044*** 0.038*** 0.023 (�0.22) (0.21) (�4.94) (4.34) (0.44)

Litigation 0.095 0.026 0.343*** 0.082*** �0.004 (1.06) (0.25) (11.23) (2.58) (�0.04)

Volatility �2.641 �1.787 �2.848*** 1.873*** �1.783 (�0.47) (�0.29) (�5.31) (3.50) (�0.32)

InsideOwn �0.483 �0.104 �2.258*** �0.785*** �0.077 (�0.38) (�0.08) (�11.02) (�4.20) (�0.06)

Spread 9.962*** 9.366** 5.912*** �1.117** 10.149*** (2.71) (2.27) (13.62) (�2.55) (2.96)

Industry fixed effects Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Observations 13,410 11,937 3,598 2,135 9,715 Pseudo-R2 (%) 8.47 8.46 9.53 9.53 8.59

Notes. This table presents the results from Tobit regressions of proprietary costs on insider purchases before product launches in columns 1 and 2 and during (outside of) proprietary periods before product launches in columns 3 and 4 (column 5). Standard errors are clustered by firm and year. t statistics are reported in parentheses below the coefficients. The sample covers the period 2002 to 2017. The dependent variable is Purchases, the number of shares purchased by insiders during the year scaled by total number of shares outstanding at the beginning of the year. PropPeriod is an indicator variable equal to one if the observation is in a proprietary period, zero otherwise. In columns 1 and 3 (columns 2 and 4), a proprietary period is the two (one) calendar-year period immediately prior to a product announcement date with no other product announcement during the proprietary period. In column 5, a nonproprietary period is any period outside of a proprietary period. PropCost is a composite measure of proprietary costs, defined as the sum of R&D, PatentApps, and ProdSimilarity, where each measure is standardized to have zero mean and unit variance. Industry and year fixed effects are included but not reported for brevity. Industry grouping is defined based on the Fama and French (1997) 48-industry classification. See the Online Appendix for additional variable definitions. All continuous variables are winsorized at 1% and 99%.

***, **, and *Statistical significance at p < 0.01, p < 0.05, and p < 0.10 (two-tailed), respectively.

significantly negative (t statistics � �3.05 and �2.03, respectively), the coefficient magnitudes (�0.196 and �0.068, respectively) are significantly different (p< 0.01). Thus, these results indicate that insiders at smaller firms are relatively more sensitive to proprietary costs.

Finally, we investigate whether the negative associa- tion between proprietary costs and insider purchases varies with media coverage. Dai et al. (2015) show the media play an important role in disseminating insider trading information. Hence, firms facing high proprie- tary costs are likely to reduce their purchases more when media coverage of insider trading is higher. Thus, we expect the negative association between proprietary costs and insider purchases is increasing with media coverage.

We classify firm-year observations into high and low media coverage samples, where high (low) media cover- age firms have above (below) median industry news coverage of insider trading in the same size decile in the prior year. The results are presented in columns 7 and 8 and are consistent with our prediction. The PropCost coefficient is negative and significant for high media coverage firms (t statistic � �2.31) but is insignificant for low media coverage firms (t statistic ��0.85). The coeffi- cient magnitudes (�0.104 and �0.028, respectively) are significantly different (p � 0.07). Thus, the results are consistent with insiders at high media coverage firms being relatively more sensitive to proprietary costs.39

In summary, the results in Table 7 show the negative association between proprietary costs and insider pur- chases is more pronounced when insider purchases are more likely to be informative to rivals. Collectively, these results provide additional support for Hypothesis 1.

4.5. Why Do Insiders Reduce Trades in the Presence of Proprietary Costs?

The evidence discussed previously strongly supports Hypotheses 1 and 2, and collectively indicates that insider purchases are negatively associated with propri- etary costs. A remaining question is: Why do insiders internalize proprietary costs borne by shareholders? Here, we explore two potential (nonmutually exclusive) mechanisms through which proprietary costs lead to reduced insider purchases.

4.5.1. Proprietary Costs and Insider Trading Restric- tions. When firms face high proprietary costs, they can directly limit the private information flow from insider trading by imposing and/or more strictly enforcing ITPs. There is substantial heterogeneity among the speci- fic provisions in ITPs and how strictly firms enforce them (Bettis et al. 2000, Jagolinzer et al. 2011).40 Bettis et al. (2000) find the most common ITP provision is a trading window policy based on QEA dates where insi- ders are prohibited from trading during blackout

Table 7. Proprietary Costs and Insider Purchases by Informativeness of Insider Trading

Dependent variable: Purchases

Variable Top 5 Non-top 5

Low complexity

High complexity Small firms Large firms

High media coverage

Low media coverage

(1) (2) (3) (4) (5) (6) (7) (8)

PropCost �0.145*** 0.020 �0.238*** �0.105** �0.196*** �0.068** �0.104** �0.028 (�3.38) (0.58) (�2.77) (�2.29) (�3.05) (�2.03) (�2.31) (�0.85)

p value for the test of equal coefficients on PropCost

(1) vs. (2) (3) vs. (4) (5) vs. (6) (7) vs. (8) <0.01 0.01 <0.01 0.07

Control variables Yes Yes Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Observations 71,255 71,255 23,623 47,632 35,145 36,110 18,165 28,923 Pseudo-R2 (%) 3.63 1.68 2.77 4.14 2.02 4.84 5.89 3.51

Notes. This table presents the results from Tobit regressions of proprietary costs on insider purchases. Standard errors are clustered by firm and year. t statistics are reported in parentheses below the coefficients. The sample covers the period 1997 to 2017. The dependent variable is Purchases, the number of shares purchased by insiders during the year scaled by total number of shares outstanding at the beginning of the year. PropCost is a composite measure of proprietary costs, defined as the sum of R&D, PatentApps, and ProdSimilarity, where each measure is standardized to have zero mean and unit variance. Columns 1 and 2 report results for insiders classified as top five and non-top five officers and directors, respectively. Columns 3 and 4 report results for firm-year observations classified as low and high complexity, respectively, where low (high) complexity firms have single (multiple) business or geographic segments for the year. Columns 5 and 6 report results for firms classified as small and large firms, respectively, where small (large) firms have below (above) median industry market value of equity at the beginning of the year. Columns 7 and 8 report results for firms classified as high and low media coverage, respectively, where high (low) media coverage firms have above (below) median industry news coverage of insider trading in the same size decile in the prior year. Control variables and industry and year fixed effects are included but not reported for brevity. Industry grouping is defined based on the Fama and French (1997) 48-industry classification. See the Online Appendix for additional variable definitions. All continuous variables are winsorized at 1% and 99%.

***, **, and *Statistical significance at p < 0.01, p < 0.05, and p < 0.10 (two-tailed), respectively.

periods leading up to and including QEAs and are only allowed to trade during safe periods following QEAs.

Although some firms publicly disclose their ITPs, the vast majority of firms keep theirs private. Thus, we can- not directly observe whether a firm has an ITP, how restrictive it is, or how strictly the firm enforces it. Instead, we follow Roulstone (2003) and use a trading window measure based on the distribution of trades rel- ative to QEA dates to indirectly infer the existence of insider trading restrictions. Although this proxy is noisy, it is positively correlated with the existence and restric- tiveness of ITPs (Jagolinzer et al. 2011).

To examine whether firms with higher proprietary costs impose and/or more strictly enforce insider trad- ing restrictions, we estimate the following model for firm i in year t with standard errors clustered by firm and year:

Safei, t � α + β1PropCosti, t�1 + β2Sizei, t�1 + β3BMi, t�1

+ β4PriorReti, t�1 + β5Coveragei, t�1

+ β6InstOwni, t�1 + β7Turnoveri, t�1

+ β8Litigationi, t�1 + β9Volatilityi, t�1

+ β10InsideOwni, t�1+ β11Spreadi, t�1

+ X γjIndustryj +

X δtYeart + εi, t: (5)

Safe is the percentage of insider shares traded between 3 and 32 trading days (inclusive) following a QEA during a three-year rolling window (Bettis et al. 2000, Jagolinzer et al. 2011). Untabulated results are essentially identical when we identify shares traded between 3 and 12 or 3 and 22 trading days following a QEA. This safe trade measurement period is consistent with Jagolinzer et al. (2011), who find the mean length of blackout periods is 45.81 calendar days prior to QEAs. The mean (median) value of Safe is 0.713 (0.785) over the sample period.

We also estimate a logistic regression where we replace Safe with Restriction, an indicator variable equal to one if the firm has an inferred ITP and zero otherwise. In line with Roulstone (2003), an ITP is inferred in year t if the percentage of safe trades is greater than or equal to 75% based on a three-year rolling window period (years t � 2 to t). The 75% cutoff is based on the survey findings in Bettis et al. (2000) and Jagolinzer et al. (2011).41

The OLS and logistic results from estimating Equation (5) are presented in Panel A of Table 8, columns 1 and 2, respectively. Both PropCost coefficients are significantly positive (t statistics � 1.95 and 1.99, respectively). These results indicate that when proprietary costs are higher, insiders purchase a larger fraction of their shares during safe periods. These results are consistent with Hypothe- sis 3A.

To provide additional evidence, we revisit our two settings with event-driven variation in proprietary costs: changes in NCA enforceability and product launches.

We estimate Equations (3) and (4) where Safe is the dependent variable. The untabulated results are consis- tent with our expectations. First, the Enforce∆ coefficient in Equation (3) is positive and significant (t statistic � 2.05). Thus, when firms’ headquarters states increase the level of NCA enforceability, insider purchases are more concentrated during safe periods. Second, the two- and one-year PropPeriod coefficients in Equation (4) are posi- tive and significant (t statistics � 4.48 and 5.55, respec- tively). These results indicate that when proprietary costs are higher before product announcements, insiders purchase significantly more of their shares during safe periods. Thus, the results in Tables 5 and 6 may be due, at least in part, to firms being more likely to impose or strictly enforce insider trading policies during periods when proprietary costs are expected to be higher. As such, they are consistent with the results in Table 8, Panel A, and further support Hypothesis 3A.

4.5.2. Executives’ Incentive Alignment. When insiders’ incentives are more aligned with shareholders, they are more likely to internalize the potential erosion of their firms’ competitive advantage caused by their trades. Thus, when incentive alignment is higher, insiders are more likely to “voluntarily” reduce their trading activi- ties even in the absence of explicit insider trading poli- cies. Accordingly, Hypothesis 3B predicts the negative association between proprietary costs and insider pur- chases is stronger when insiders have greater firm own- ership. To examine this hypothesis, we classify insider- year observations into high and low ownership samples, where high (low) ownership observations have shares owned by each insider scaled by total shares outstand- ing that are above (below) the industry-year median. We estimate Equation (1) for each sample separately.

The results are presented in Panel B of Table 8. For both the high and low ownership samples, the PropCost coefficient is negative and significant (t statistics � �2.68 and �3.22, respectively). In addition, the magnitude of the PropCost coefficient in the high ownership sample is over twice as large as in the low ownership sample (�0.054 and �0.026, respectively). The difference is sta- tistically significant (p � 0.01). These results indicate that high-ownership insiders are more sensitive to proprie- tary costs when making their purchase decisions com- pared with low-ownership insiders. Thus, these results provide empirical support for Hypothesis 3B.

If the costs of trading are incrementally higher for high-ownership insiders, they will require higher expected profits to be willing to trade. Hence, Hypothe- sis 3C predicts that the profitability of insider purchases is higher when insider ownership is higher. To examine this hypothesis, we estimate Equation (2) on the same high and low ownership samples as in Panel B. The results are reported in Panel C. They show that the Prop- Cost coefficients are significantly positive for each

Table 8. Proprietary Costs and Mechanisms Underlying Insider Purchases

Panel A: Insider trading restrictions

Variable

OLS Logit Safe Pr(Restriction � 1) (1) (2)

PropCost 0.003* 0.028** (1.95) (1.99)

Control variables Yes Yes Industry fixed effects Yes Yes Year fixed effects Yes Yes Observations 69,183 69,183 Observations Restriction � 1 37,643 Adjusted/pseudo-R2 (%) 5.17 2.07

Panel B: Incentive alignment: Ownership and insider purchases

Dependent variable: Purchases

Variable High ownership Low ownership

(1) (2)

PropCost �0.054*** �0.026*** (�2.68) (�3.22)

p value for the test of equal coefficients on PropCost (1) vs. (2) 0.01

Control variables Yes Yes Industry fixed effects Yes Yes Year fixed effects Yes Yes Observations 114,903 115,381 Pseudo-R2 (%) 6.16 10.56

Panel C: Incentive alignment: Ownership and insider purchase profits

Dependent variable

Variable

Alpha BHAR Raw

High ownership

Low ownership

High ownership

Low ownership

High ownership

Low ownership

(1) (2) (3) (4) (5) (6)

PropCost 0.011** 0.006* 0.042** 0.028** 0.040*** 0.029** (2.46) (1.84) (3.09) (2.15) (3.30) (2.26)

p value for the test of equal coefficients on PropCost

(1) vs. (2) (3) vs. (4) (5) vs. (6) <0.01 0.02 0.05

Control variables Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Observations 31,203 27,397 31,983 28,134 31,903 28,103 Adjusted R2 (%) 20.80 17.82 7.60 7.12 15.11 14.84

Notes. Panel A of this table presents the results from an OLS regression of proprietary costs on the percentage of insider shares traded during safe periods (column 1) and a logistic regression of proprietary costs on the likelihood of an inferred blackout policy (column 2). In column 1, the dependent variable is Safe, the percentage of insider shares traded between three and 32 trading days (inclusive) following quarterly earnings announcements, measured using three-year rolling windows. In column 2, the dependent variable is Restriction, an indicator variable equal to one if the firm has an insider trading policy based on trading windows, inferred if the percentage of safe trades is greater than or equal to 75% based on a three-year rolling window period, and zero otherwise. Panel B of this table presents the results from Tobit regressions of insider purchases on proprietary costs. The dependent variable is Purchases, the number of shares purchased by insiders during the year scaled by total number of shares outstanding at the beginning of the year. Panel C presents the results from OLS regressions of insider trading profits on proprietary costs conducted at the insider-purchase date level. Insider trading profits are estimated over the 360 calendar days following the transaction date using daily alpha (Alpha) the intercept from the Carhart (1997) four-factor model, buy-and-hold abnormal stock returns (BHAR) using the Carhart (1997) four-factor model, and buy-and-hold raw stock returns (Raw). Column 1 (2) of Panel B and columns 1, 3, and 5 (2, 4, and 6) of Panel C report results for insider- year observations classified as high (low) ownership, where high (low) ownership are observations above (below) the industry-year median of shares owned by an insider scaled by total shares outstanding. In all panels, standard errors are clustered by firm and year. t statistics are reported in parentheses below the coefficients. The sample covers the period 1997 to 2017. PropCost is a composite measure of proprietary costs, defined as the sum of R&D, PatentApps, and ProdSimilarity, where each measure is standardized to have zero mean and unit variance. Control variables and industry and year fixed effects are included but not reported for brevity. Industry grouping is defined based on the Fama and French (1997) 48-industry classification. See Online Appendix for additional variable definitions. All continuous variables are winsorized at 1% and 99%.

***, **, and *Statistical significance at p < 0.01, p < 0.05, and p < 0.10 (two-tailed), respectively.

ownership sample and each measure of trade profitabil- ity (t statistics range between 1.84 and 3.30). Moreover, for each measure of trade profitability, the PropCost coef- ficient in the high ownership sample is significantly larger than the corresponding PropCost coefficient in the low ownership sample (p values range between < 0.01 and 0.05). These results provide support for Hypothesis 3C and additional evidence that proprie- tary costs influence insider trading.42

4.6. Investment Sensitivity to Rival Firms’ Insider Purchases and Proprietary Costs

We expect that one mechanism through which insider purchases can impose proprietary costs is by providing rival firms with an opportunity to make improved infer- ences about the prospects of the insiders’ firms and thus, make better business decisions. We examine whether firms consider the information conveyed by insider pur- chases at rival firms when making investment decisions and what role proprietary costs play. Prior literature indi- cates that rivals can learn investment-related information directly from rivals’ disclosures about earnings and resta- tements (Durnev and Mangen 2009, Badertscher et al. 2013) and indirectly from their rivals’ stock prices, where the amount of information they can learn is influenced by their disclosures (Foucault and Fresard 2014, Edmans et al. 2017). Managerial learning from stock prices is empirically captured by the sensitivity of investment to Tobin’s q, which is referred to as investment-price sensi- tivity (IPS) (Chen et al. 2007, Bond et al. 2012, Foucault and Fresard 2014). To capture managerial learning from both the focal firm’s and its rivals’ stock prices, and inves- tigate the role of insider purchases, we estimate versions of the following regression model for firm i in year t with firm and year fixed effects, as well as standard errors clus- tered by firm and year:

Investmenti, t � α + β1Rival_Qi, t�1 + β2Rival_CFOi, t�1

+ β3Rival_Assetsi, t�1

+ β4Rival_Purchasesi, t�1

+ β5Rival_Qi, t�1 ∗ Rival_Purchasesi, t�1

+ β6Qi, t�1 + β7CFOi, t�1 + β8Assetsi, t�1

+ β9Purchasesi, t�1 + β10Qi, t�1

∗ Purchasesi, t�1 + X γkFirmk +

X δtYeart

+ εi, t: (6)

Investment is capital expenditures divided by the beginning-of-year property, plant, and equipment. Q is Tobin’s q, measured as the ratio of the market value of assets to the book value of assets. CFO is cash flow from operations (CFO), measured as income before extraordi- nary items plus depreciation divided by beginning-of- year total assets. Assets is the natural logarithm of total

assets. Other variables are as defined previously. Rival firms consist of all firms in the same industry (excluding the focal firm), where text-based product market indus- try classifications are from Hoberg and Phillips (2016). Each rival firm variable (Rival_X) is the average value of variable X across all rival firms. Finally, to help with the interpretation of the coefficients, each independent vari- able in Equation (6) is demeaned and scaled by its stan- dard deviation.

The results from estimating Equation (6) are presented in Table 9. Column 1 presents the results from a baseline version where insider purchase variables are not included. Consistent with prior literature (Foucault and Fresard 2014, Edmans et al. 2017), investments are posi- tively associated with the stock prices of both the focal firm and rival firms, as the Q and Rival_Q coefficients are both positive and significant (t statistics � 14.89 and 2.12, respectively). Column 2 presents the results where the insider purchase variables and their interactions with Tobin’s q are included. The Rival_Purchases coeffi- cient is not significantly associated with investments, whereas the Rival_Q × Rival_Purchases coefficient is posi- tive and significant (t statistic � 2.33). This result sug- gests that focal firms learn more investment-related information from their rivals’ stock prices (i.e., rival IPS increases) when their rivals’ insiders purchase more shares. This result indicates that firms consider informa- tion about rivals’ insider purchases when they make their investment decisions.43 As such, this result sup- ports the plausibility of our hypothesis that insiders’ purchases convey valuable information to rival firms and therefore incur proprietary costs.

Next, we examine whether the association between firms’ investment decisions and rivals’ insider purchases and stock prices vary with rivals’ proprietary costs. We divide the sample based on the level of proprietary costs at rival firms. Firm-year observations are classified as high (low) Rival_PropCost if they are above (below) the median for the year of the average value of PropCost across all rival firms. The results from estimating Equa- tion (6) on the high (low) proprietary cost sample are reported in column 3 (4). The results show that the Riv- al_Q × Rival_Purchases coefficient is significantly positive (t statistic � 2.75) in the high Rival_PropCost sample. In contrast, the Rival_Q × Rival_Purchases coefficient in the low Rival_PropCost sample is insignificant (t statistic � 1.02). The absolute magnitude of the Rival_Q × Rival_ Purchases coefficient is much larger for the high Rival_ PropCost sample compared with the low Rival_PropCost sample (0.022 versus 0.004), and a Wald chi-square test indicates the coefficients are significantly different (p < 0.01). The results indicate that learning from rivals’ stock prices is more pronounced when insiders purchase more shares at high proprietary cost rival firms. These results provide additional support for our proprietary cost hypothesis.

Overall, the evidence in Table 9 suggests that firms’ investment decisions are influenced by insider pur- chases at their rivals and that the degree of influence is increasing with rivals’ proprietary costs. These results support the basic premise of our paper: insider pur- chases are informative signals to competitors and the strength of these signals is increasing with proprietary costs. In addition, these findings are important and novel because they suggest that insider purchases have real effects on rivals’ investment decisions. Moreover, they further reduce the plausibility of the alternative hypotheses discussed in Section 4.2 because we do not expect investment decisions will be associated with insider purchases at rivals unless those purchases are informative (i.e., convey proprietary information).

5. Conclusion In this paper, we argue that insider trading can reveal proprietary information to rivals (directly or indirectly) and analyze whether insiders’ trading decisions reflect these proprietary costs. Insiders at high proprietary cost firms have incentives to avoid revealing proprietary information through their trading activities, and thus, we hypothesize that proprietary costs are negatively associated with insider purchases. Using a composite measure of proprietary costs that captures multiple dimensions of competition, and hence, proprietary costs, we find a negative association between proprietary costs and insider purchases. In addition, we hypothesize and find supporting evidence that insiders at high pro- prietary cost firms earn significantly higher abnormal

Table 9. Proprietary Costs and Rival Companies Investment

Variable

Dependent variable: Investment

All High Rival_PropCost Low Rival_PropCost

(1) (2) (3) (4)

Rival_Q 0.0216** 0.0164** 0.0363*** 0.0067** (2.12) (2.66) (3.50) (2.31)

Rival_CFO �0.0249*** �0.0147*** �0.0167*** �0.0032 (�4.92) (�3.91) (�3.19) (�0.77)

Rival_Assets �0.007 �0.007 �0.006 �0.003 (�1.23) (�1.38) (�0.65) (�0.84)

Rival_Purchases 0.005 �0.001 0.002 (1.59) (�0.19) (0.64)

Rival_Q × Rival_Purchases 0.013** 0.022** 0.004 (2.33) (2.75) (1.02)

Q 0.144*** 0.116*** 0.115*** 0.111*** (14.89) (21.55) (15.13) (12.10)

CFO �0.004 0.047*** 0.028*** 0.099*** (�0.49) (7.02) (3.98) (7.97)

Assets �0.046*** �0.068*** �0.021 �0.122*** (�2.92) (�4.81) (�1.11) (�5.56)

Purchases �0.001 �0.002 0.001 (�0.40) (�0.46) (0.31)

Q × Purchases 0.003 0.005 0.010 (0.88) (1.02) (1.15)

p value for test of equal coefficients on Rival_Q × Rival_Purchases (3) vs. (4) <0.01

Firm fixed effects Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Observations 88,229 67,034 32,429 33,215 Adjusted R2 (%) 33.31 31.34 30.64 33.13

Notes. This table presents the results from OLS regressions of rival firms’ insider purchases and proprietary costs on investment, where each variable is demeaned and scaled by its standard deviation. Column 3 (4) reports results for firm-year observations classified as high (low) Rival_PropCost, where high (low) Rival_PropCost are observations above (below) the median for the year of the average value of PropCost across all rival firms, which consist of all firms (excluding the focal firm) in the same industry. Standard errors are clustered by firm and year. t statistics are reported in parentheses below the coefficients. The sample covers the period 1997 to 2017. The dependent variable is Investment, capital expenditures scaled by property, plant, and equipment at the beginning of the year. PropCost is a composite measure of proprietary costs, defined as the sum of R&D, PatentApps, and ProdSimilarity, where each measure is standardized to have zero mean and unit variance. Firm and year fixed effects are included but not reported for brevity. Industry grouping is defined based on the Fama and French (1997) 48-industry classification. See the Online Appendix for additional variable definitions. All continuous variables are winsorized at 1% and 99%.

***, **, and *Statistical significance at p < 0.01, p < 0.05, and p < 0.10 (two-tailed), respectively.

trading profits. The results of cross-sectional analyses show the negative association between proprietary costs and insider purchases is stronger when they are more likely to be informative or useful to rivals. Moreover, we examine two settings with event-driven, and even ex- ogenous, variation in proprietary costs. First, we investi- gate variation in the enforceability of employee noncompete agreements. Consistent with our expecta- tions, insiders reduce their purchases following an exog- enous increase in enforceability, which represents an increase in proprietary costs. Second, we examine insider purchases prior to new product launches, when proprietary costs are particularly high. We find insiders significantly reduce their purchases during the prean- nouncement period, consistent with them minimizing the risk of information transfers.

We investigate two nonmutually exclusive mecha- nisms by which insiders internalize proprietary costs, which are borne by shareholders. First, firms with higher proprietary costs are more likely to impose and/or more strictly enforce insider trading restrictions. Second, we find insiders’ purchase decisions are more sensitive to proprietary costs when their wealth is more closely linked to that of shareholders.

Finally, we shed light on the premise that insider pur- chases are informative to competitors and the strength of these signals is increasing with proprietary costs. We

find that firms’ investment decisions are associated with the purchase decisions of their rivals’ insiders, and the strength of the association is increasing with rivals’ pro- prietary costs. These novel findings document real effects of insider trading on rival firms.

Overall, we provide evidence that product market considerations are an important, but previously undocu- mented factor associated with insider trading decisions and profitability. These findings are likely of interest to regulators and corporate boards in setting insider trad- ing policies, and help investors interpret insider trading signals.

Acknowledgments The authors thank the associate editor, two anonymous refer- ees, DuckKi Cho, Patricia Dechow, Shawn Huang, Wei Huang, Artur Hugon, Erin Jordan, Adelphine Kogut, Yinghua Li, An-Ping Lin, Michal Matejka, Liwei Weng, Brian Wenzel, Yifang Xie, Ira Yeung, Wei Zhang, and workshop participants at Arizona State University, Chinese University of Hong Kong, City University of Hong Kong, Hong Kong Polytechnic University, National University of Singapore, Peking Univer- sity, Singapore Management University, Temple University, University of Cambridge, University of Illinois at Chicago, the AAA Annual Meeting, the Corporate Governance and Execu- tive Compensation Research Series, and the Hawaii Account- ing Research Conference for helpful feedback and suggestions.

Appendix. Variable Definitions

Variable Definition

Proprietary cost measures PropCost Sum of R&D, PatentApps, and ProdSimilarity, where each individual measure is standardized to have

zero mean and unit variance. R&D Decile ranking of R&D expenditures scaled by total sales (ranking determined every year and scaled

between zero and one). PatentApps Natural logarithm of one plus the number of patent applications filed to the U.S. Patent and Trademark

Office (USPTO) during the fiscal year. ProdSimilarity Sum of the pairwise product similarities between the firm and all other firms in Compustat in the fiscal

year. A firm-by-firm pairwise product similarity score is estimated using the cosine similarity of firms’ product spaces based on business descriptions from 10-K filings (see Hoberg and Phillips 2016).

Enforce∆ Indicator variable equal to 1 (�1) if there is an increase (decrease) in the noncompete agreement enforceability score of the firm headquarters state in the current or any prior year, and 0 otherwise, using historical headquarters locations. Noncompete enforceability scores for the year in the state where the firm’s headquarters are located were originally developed by Garmaise (2011) for the period from 1992 to 2004 and extended by Ertimur et al. (2018) for the period from 1980 to 2013.

BigEnforce∆ Indicator variable equal to 1 (�1) if there is a large increase (decrease) in the firm’s headquarters state’s noncompete agreement enforceability score of two or more in the current or any prior year, and 0 otherwise, using historical headquarters locations.

Texas Indicator variable equal to one if the firm is headquartered in Texas, and zero otherwise. Post Indicator variable equal to �1 for years 1995 or 1996, the two years after the 1994 Texas Supreme Court

decision that decreased the NCA score enforceability by two, and 0 for years 1992 to 1994. LocalRival Decile ranking of the total number of establishments in a given three-digit NAICS industry present in

the state minus one, divided by the total number of establishments with the same NAICS in the U.S. minus one (ranking determined every year and scaled between zero and one).

PropPeriod Indicator variable equal to one if the observation is during a proprietary period, and zero otherwise. A proprietary period is the one- or two-calendar-year period immediately prior to a product announcement date with no other product announcement during the proprietary period.

Appendix. (Continued)

Variable Definition

Insider trading measures Purchases Total number of shares purchased by insiders during the fiscal year, scaled by the total number of

shares outstanding at the beginning of the year, multiplied by 1,000. Opportunistic Purchases Total number of opportunistic shares purchased during the fiscal year, scaled by the total number of

shares outstanding at the beginning of the year, multiplied by 1,000. If an insider trades in three consecutive years and the trades do not meet the routine classification criteria, then all subsequent trades are classified as opportunistic. An insider may have trades classified as opportunistic and, subsequently, trades classified as routine if these subsequent trades meet the routine classification criteria.

Routine Purchases Total number of routine shares purchased during the fiscal year, scaled by the total number of shares outstanding at the beginning of the year, multiplied by 1,000. If an insider trades in the same calendar month for three consecutive years, then all subsequent trades by this insider are classified as routine.

Alpha Daily alpha, the intercept from the Carhart (1997) four-factor model, estimated over 360 calendar days following the transaction date, multiplied by 100.

BHAR Buy-and-hold abnormal stock returns estimated using the Carhart (1997) four-factor model, estimated over 360 calendar days following the transaction date.

Raw Buy-and-hold raw returns over 360 calendar days following the transaction date. Safe The percentage of insider shares traded classified as safe, using three-year rolling windows, where safe

shares are from insider trades made between three and 32 trading days (inclusive) following quarterly earnings announcements.

Restriction Indicator variable equal to one if the firm has an inferred insider trading policy (ITP) based on trading windows, and zero otherwise. An ITP is inferred if the percentage of safe trades is greater than or equal to 75% based on a three-year rolling window period.

Firm-level and transaction-level characteristics Assets Natural logarithm of total assets. BM Ratio of book value to market value of equity. CFO Income before extraordinary items plus depreciation scaled by total assets at the beginning of the year. Coverage Natural logarithm of one plus the number of analysts issuing earnings forecasts. InsideOwn The number of shares held by a top five executive (CEO, CFO, COO, President, and Chairman of the

Board), scaled by total shares outstanding. Number of shares held is compiled from insider transactions and holdings information on Forms 3 through 5.

InstOwn Number of shares held by institutional investors scaled by total shares outstanding. Investment Capital expenditures scaled by property, plant, and equipment at the beginning of the year. LiberalCourt Litigation risk measure based on federal judge ideology from Huang et al. (2024), measured using the

probability that a three-judge panel randomly selected from a circuit court has at least two judges appointed by Democratic presidents, that is, [C(x, 3) + C(x, 2) × C(y � x, 1)]/C(y, 3), where C(n, r) is a binomial coefficient indicating the number of possible combinations of r objects from a set of n distinct objects, x is the number of Democratic appointees in the circuit, and y is the total number of judges in the circuit. Circuit court judges’ appointing presidents are obtained from the Federal Judicial Center’s website.

Litigation Litigation measure based on the predicted litigation probability using Model (3) of Kim and Skinner (2012), which regresses the indicator variable of litigation on industry membership and measures of firm characteristics (including size, growth, stock performance, skewness, volatility, and liquidity).

PriorRet Buy-and-hold raw stock returns. Q Ratio of the market value of assets to the book value of assets, where market value of assets is defined

as total assets plus market value of equity minus book value of equity. Rival_X Average value of variable X across all rival firms, which consist of all firms (excluding the focal firm) in

the same TNIC industry and where text-based product market industry classifications are from Hoberg and Phillips (2016).

Size Natural logarithm of the market value of the equity. Spread Average of daily bid-ask spreads over the year, where bid-ask spread is bid price minus ask price scaled

by the average of the bid and ask prices. TradeSize Dollar value of the daily purchases by an insider scaled by the natural logarithm of the market value of

equity of the firm. Turnover Natural logarithm of the ratio of the number of shares traded scaled by number of shares outstanding at

the beginning of the year. Volatility Variance of daily stock returns.

Endnotes 1 Most firms will find it impossible or excessively costly to con- stantly monitor all existing and potential rivals for developments that could erode their competitive positions. One low-cost approach to streamline this process is for firms to monitor the insider pur- chases made by executives at rival firms. Observing insider pur- chases at a rival firm may spur the focal firm to redirect its competitive intelligence efforts. 2 Prior literature identifies three key drivers of the intensity of inter- firm rivalry: awareness, motivation, and capability (Smith et al. 2001). We expect that the proprietary costs related to insider trading are most strongly related to the awareness driver, where greater awareness results in more intense reactions by competitors. 3 Section 4.2 presents the results of supplementary analyses provid- ing evidence that our results are unlikely to be driven by regulatory or litigation costs or by information asymmetry. 4 ITPs frequently state that proprietary or confidential information is the sole property of the firm and that insiders are prohibited from using or personally benefiting from this information in any way. Thus, these IPTs implicitly prohibit insiders from trading on propri- etary information. Among these, some explicitly prohibit trading based on proprietary information. For example, FNX Mining’s ITP states: “All those with access to material confidential information are prohibited from using such information in trading in the Cor- poration’s securities,” and Commercial Metals Company’s ITP states: “No employee shall release, divulge, or expose any data, pro- prietary information or information concerning anticipated com- pany activities prior to the company’s public release of such information. Further, no employee shall use such data or informa- tion to the personal advantage of the employee including trading or causing to be traded Commercial Metals Company stock before the information is publicly released.” 5 Consider the following excerpt from NDS Limited’s ITP: “The Company’s employees are responsible for protecting the Com- pany’s confidential and proprietary information. No employee shall disclose confidential or proprietary information to a third-party without proper authorization or use such information for his or her own personal benefit [ … ] Proprietary information includes, with- out limitation, information relating to trade secrets, patents, research studies and results, manufacturing techniques and market- ing strategies. [ … ] Proprietary information is a Company asset. [ … ] Improper disclosure or use could [ … ] substantially weaken the Company’s competitive position.” 6 In this respect, we view insider trades in the same vein as manage- ment forecasts of quarterly earnings. Although firms act as if fore- casts incur proprietary costs (Ali et al. 2014, Huang et al. 2017), it is unclear how a competitor would directly glean proprietary informa- tion from the forecast itself or use it to gain competitive advantage. We expect the voluntary forecast triggers further investigations by competitors, which then result in specific actions that impose propri- etary costs on the disclosing firm. Furthermore, we are unaware of any empirical evidence that firms are negatively affected by issuing voluntary disclosures that convey proprietary information. Despite this lack of evidence, it is widely accepted that proprietary costs exist and are an important determinant of voluntary disclosure decisions. 7 As former General Electric CEO Jack Welch observed, success in most competitive rivalries “is a result of being able to respond rap- idly to real changes as they occur” (Welch and Byrne 2003). 8 Bernard et al. (2020) develop a novel pairwise measure of firms’ acquisition of rivals’ public disclosures based on unique IP addresses. Their results suggest that rivals’ public information, including insider trading information, is valuable as it helps facili- tate investment, product acquisition, and differentiation decisions. We thank the authors for generously providing us with these data.

9 Insiders have strong career- and wealth-based incentives to increase shareholder wealth. Thus, we expect them to, at least par- tially, internalize the proprietary costs associated with their trades by also considering the benefits and costs to shareholders when making their trading decisions. 10 Industry concentration has been used in prior studies as a measure of product market competition. However, certain studies argue high industry concentration reflects industries with fewer firms, and thus lower competition among existing rivals (Li 2010), whereas other stud- ies argue high industry concentration reflects more intense competition among existing rivals (Raith 2003, Ali et al. 2014). Given these mixed arguments and empirical findings, Beyer et al. (2010) conclude “that there is no clear empirical evidence to date on how proprietary costs, as proxied by the level of competition in an industry, are related to vol- untary disclosures” (p. 306). Given this ambiguity, we do not include industry concentration as one of our proprietary cost measures. 11 Our results are qualitatively similar and our inferences remain unchanged if we (i) redefine R&D after excluding missing and zero R&D expenditures; (ii) scale R&D expenditures by total expenses, assets, or average assets; (iii) include an indicator variable for miss- ing R&D observations (Koh and Reeb 2015); or (iv) replacing miss- ing R&D observations with the industry-mean value and include a missing R&D indicator variable (Koh and Reeb 2015). 12 Typically, a patent application is not publicly disclosed until 18 months following the application. Thus, proprietary costs are espe- cially high during this pre-disclosure period that precedes the pat- ent decision. Once a patent has been disclosed, proprietary costs should be low. Consistent with this dynamic, insider purchases are negatively associated with the number of patent applications, but are not significantly associated with number of disclosed patent applica- tions (results untabulated). These results indicate patent applica- tions are a measure of proprietary costs. 13 Filing a patent application represents the voluntary disclosure of a successful R&D outcome (Glaeser 2018). If the factors determining the patent application choice are endogenously associated with both insiders’ purchase decisions and our proprietary cost mea- sures, then our results could be biased. As discussed later, the results are qualitatively similar and our inferences are unchanged if PatentApps is excluded from the composite measure of proprietary costs. 14 Our main results are qualitatively similar and our inferences remain unchanged if PropCost is constructed using principal component analy- sis or factor analysis on R&D, PatentApps, and ProdSimilarity. 15 We include R&D, PatentApps, and ProdSimilarity in PropCost to preserve sample size. Including TechSimilarity (TechSimilarity and CompWords) in our composite measure reduces the sample size by 49% (56%). Our main results are qualitatively similar and our infer- ences remain unchanged if the composite measure (i) includes any combination of three, four, or five of the individual measures or (ii) is constructed using principal component analysis or factor analysis for any combination of three, four, or five of the individual mea- sures (with one exception when combining PatentApps, ProdSimilar- ity, and TechSimilarity, primarily due to the resulting smaller sample size). 16 As discussed in Section 4.1.1, our results are robust to using three alternative measures of insider purchases. 17 Our twelve-month time horizon is reasonable because insiders have to hold their shares for at least 180 days. Section 16(b) of the Securities and Exchange Act of 1934 requires insiders to disgorge “short-swing profits” made within 180 days. Results are qualita- tively similar when we estimate trading profits using a six-month holding period. 18 The negative correlations are not surprising as R&D expenditures and patent applications summarize activities intended to introduce

product differentiation and create barriers to entry. For example, Hoberg and Phillips (2016) find that when firms spend more on R&D, they experience significant reductions in ex post product similarity. 19 In contrast to our results, Huddart and Ke (2007) find that trading activity (both purchases and sales) is significantly higher among insiders at firms with positive R&D expense during the prior year compared with insiders at non-R&D firms (their table 6). Aboody and Lev (2000) discuss similar (untabulated) univariate results. A key distinction is that Huddart and Ke (2007) and Aboody and Lev (2000) combine insider purchases and sales together. To reconcile these results with ours, we examine the association between R&D activity and insider trading using alternatively (i) all trades (pur- chases and sales), (ii) purchases, or (iii) sales. We also use two mea- sures of R&D activity: RND, an indicator variable equal to one if the prior year’s R&D expenditures are positive and zero otherwise (Huddart and Ke 2007), and R&D, our measure of R&D intensity. In untabulated results, we find a significantly positive association between R&D activity and either all insider trades (purchases and sales) or insider sales, consistent with the results in Huddart and Ke (2007). In contrast, we find a significantly negative association between R&D activity and insider purchases, consistent with the results in our study. 20 The absolute values of all of the coefficients in columns 2 and 3 are much smaller than their full-sample counterparts in column 1, including those for PropCost (�0.044 and �0.022, respectively, ver- sus �0.145). These differences are primarily due to the differences in the magnitude of the dependent variables. 21 In contrast, insiders face substantial litigation risk when they sell before bad news. For example, Johnson et al. (2007) find litigation risk increases after abnormal insider sales, especially after the Pri- vate Securities Litigation Reform Act of 1995. Billings and Cederg- ren (2015) find the probability of being sued significantly increases when insiders engage in sales prior to the announcement of nega- tive earnings news and fail to provide prior warnings. 22 Our results are consistent with prior evidence that insiders face minimal regulatory risk when they purchase their own firm’s shares (Cohen et al. 2012, Del Guercio et al. 2017). Thus, despite its prima facie illegality, insiders face minimal regulatory risk when they exploit their private information by purchasing their own stock. 23 As an alternative measure of SEC enforcement activity, we follow Kedia and Rajgopal (2011) and define DistToSEC as the distance from a firm’s headquarters to the SEC’s headquarters, regional headquarters or district office (for observations starting in 2007 and onward), whichever one is closer. The correlation between PropCost and DistToSEC is �0.033. We also estimate Equation (1) and include DistToSEC as an additional explanatory variable. The untabulated results show that the coefficient on PropCost remains negative and significant (coefficient � �0.147; t statistic � �3.40). These findings provide additional evidence that proprietary costs are not associ- ated with SEC enforcement activity and the alternative explanation that our results are driven by proprietary costs being corelated with regulatory risk is unlikely to be plausible. 24 Equations (1) and (2) include the average bid-ask spread, Spread, to control for information asymmetry. Moreover, we include several other control variables that are correlated with information asym- metry: Size, Coverage, InstOwn, and Volatility. Thus, the PropCost coefficient captures variation that is incremental to these informa- tion asymmetry-related control variables, which reduces the plausi- bility of this alternative explanation. 25 Information asymmetry increases with the precision of the insi- der’s private information relative to the underlying level of uncer- tainty about firm value that outside investors have.

26 Consistent with mechanism, the evidence in Aboody and Lev (2000) indicates that insiders trade based on their private informa- tion about changes in future R&D expenditures, which are associ- ated with future returns. These trades yield higher trading profits. 27 In price-taking models of informed trading, such as Grossman and Stiglitz (1980), the quantity of trade increases with information asymmetry (Huddart and Ke 2007). 28 As discussed in Endnote 39, certain of the cross-sectional analy- ses in Table 7 provide further evidence that our results are not driven by information asymmetry, and thus, provide further insights on how proprietary costs and information asymmetry are associated with insider purchases in distinct ways. 29 In untabulated analyses, we include PCM in Equation (2). The results show that for all three profit measures, the PropCost coeffi- cients are positive and significant (t statistics range between 2.99 and 3.11). 30 Garmaise (2011) finds some evidence that increases in NCA enforceability are associated with decreases in R&D, explaining that higher NCA enforceability may discourage firms from investing in their own human capital, which in turns leads to less R&D invest- ment. Ceteris paribus, a decrease in R&D decreases proprietary costs. Thus, in our setting, a possible negative association between Enforce∆ and R&D would make it less likely to find significant neg- ative coefficients on Enforce∆ or Enforce∆ × LocalRival, which repre- sent an increase in proprietary costs. 31 To avoid potential biases from using current firm headquarters locations, we use historical headquarters locations (Aobdia 2018). Moreover, we exclude firms headquartered in states where the enforceability score first increases and then later decreases, or vice versa. Untabulated analyses show that including these observations does not qualitatively change the signs, magnitudes or significance levels of our results, and hence, our inferences remain unchanged. Finally, our results are qualitatively similar, and our inferences remain unchanged when we replace Enforce∆ with the continuous enforceability score. 32 Our Tobit regression in Equation (3) differs slightly from the DnD specification in Aobdia (2018), that is, the OLS regression in his equation (3). Specifically, we include state fixed effects, whereas he includes firm fixed effects. We are not able to include firm fixed effects because our Tobit regressions do not converge and/or gener- ate standard errors when large numbers of fixed effects are included in the model. For a similar reason, we are not able to include state × year fixed effects in our model with local competi- tion (discussed later), which are included in Aobdia (2018). Further- more, in a nonlinear DnD model, the treatment effect is based on the conditional expectation of the observed outcome minus the con- ditional expectation of the potential outcome without treatment, obscuring the interpretation of the coefficient estimates. However, the sign of the treatment effect is the same as the sign of the interac- tion coefficient in a linear DnD model (Puhani 2012). Thus, we focus primarily on the sign of the coefficient estimates when estimating Equation (3) to examine the direction of the marginal effects of changes in NCA enforcement. 33 We restrict the sample to 1992–1996 to focus on the period sur- rounding the change in enforceability, following Aobdia (2018). 34 Because of the inclusion of state and year fixed effects, Texas and Post are excluded from the regression. 35 Post is equal to �1 (0) in 1995–1996 (1992–1994) to reflect the decrease (no change) in NCA enforceability, consistent with Enforce∆ and BigEnforce∆ equal to 1 (�1) for increases (decreases) and large increases (decreases), respectively, in NCA enforceability. 36 Recall that the enforceability scores are integers that vary between zero and nine at the state level. Given that the number of firms in each state vary, the number of observations in the High

Enforceability tercile (28,586) is different from the number of obser- vations in the Low Enforceability tercile (12,239). 37 Our results are qualitatively similar and our inferences remain unchanged if we redefine the proprietary period as the nine-month, six-month, or three-month period immediately prior to a product announcement date, with the PropPeriod coefficients being negative and significant in each case (coefficients � �0.177, �0.362, and �0.693; t statistics � �2.93, �3.95, and �3.79, respectively). 38 Our results are consistent with an alternative explanation whereby SEC scrutiny of insider trades and/or litigation risk is higher during these high proprietary cost periods, which in turn would deter insiders from trading during them. To address this possibility, we analyze whether the probability of an enforcement action (class action securities litigation) is higher during the high proprietary cost periods preceding product launches. The results in the Online Appendix, Table OA8 (Table OA9), show that there is no significant increase in the probability of a SEC enforcement action (class action securities litigation) during these periods. 39 In addition, the results in columns 3 and 4 (7 and 8) are inconsis- tent with the information asymmetry alternative explanation (dis- cussed in Section 4.2.2) whereby insiders trade more when information asymmetry is higher. The reason is that information asymmetry is higher for multisegment firms (Cohen and Lou 2012) and firms with low media coverage (Bushee et al. 2010). These cross-sectional results provide further evidence that our results are distinct from those in the prior literature that examines the relation between information asymmetry and insider trading. 40 For example, Jagolinzer et al. (2011) find that 80% of ITPs require insider trades to be pre-approved by the general counsel. Thus, we expect our identification of trading window restrictions also proxies for more restrictive insider trading policies in general. However, we are not able to directly assess these other types of restrictions. To the extent our trading-window measures do not generally reflect the actual trading restrictions that firms impose on insiders, then our results need to be interpreted accordingly. 41 Jagolinzer et al. (2011) find this method correctly classifies 81.13% (90.02%) of trades occurring inside blackout (safe) periods. They no longer find significant results when using inferred windows instead of actual windows. They conclude researchers “should exercise cau- tion when interpreting null results based on inferred windows.” As we find significant results for our restriction tests, their caution is not applicable in our setting. 42 Under the information asymmetry alternative explanation (see Sec- tion 4.2.2), one would not expect insider ownership to moderate the positive association between PropCost and insider trading profits. 43 To our knowledge, this is the first study to document real effects of voluntarily disclosed information by rival firms.

References Aboody D, Lev B (2000) Information asymmetry, R&D, and insider

gains. J. Finance 55(6):2747–2766. Aghion P, Bloom N, Blundell R, Griffith R, Howitt P (2005) Compe-

tition and innovation: An inverted-U relationship. Quart. J. Econom. 120(2):701–728.

Agrawal A, Jaffe JF (1995) Does Section 16b deter insider trading by target managers? J. Financial Econom. 39(2–3):295–319.

Albring S, Banyi M, Dhaliwal D, Pereira R (2016) Does the firm information environment influence financing decisions? A test using disclosure regulation. Management Sci. 62(2):456–478.

Ali A, Klasa S, Yeung E (2014) Industry concentration and corporate disclosure policy. J. Accounting Econom. 58(2–3):240–264.

Aobdia D (2015) Proprietary information spillovers and supplier choice: Evidence from auditors. Rev. Accounting Stud. 20(4): 1504–1539.

Aobdia D (2018) Employee mobility, noncompete agreements, product-market competition, and company disclosure. Rev. Accounting Stud. 23(1):296–346.

Arif S, Kepler JD, Schroeder J, Taylor D (2022) Audit process, pri- vate information, and insider trading. Rev. Accounting Stud. 27(3):1125–1156.

Badertscher BA, Hribar SP, Jenkins NT (2011) Informed trading and the market reaction to accounting restatements. Accounting Rev. 86(5):1519–1547.

Badertscher B, Shroff N, White HD (2013) Externalities of public firm presence: Evidence from private firms’ investment deci- sions. J. Financial Econom. 109(3):682–706.

Bena J, Ferreira MA, Matos P, Pires P (2017) Are foreign investors locusts? The long-term effects of foreign institutional owner- ship. J. Financial Econom. 126(1):122–146.

Beneish MD, Vargus ME (2002) Insider trading, earnings quality, and accrual mispricing. Accounting Rev. 77(4):755–791.

Bernard D (2016) Is the risk of product market predation a cost of disclosure? J. Accounting Econom. 62(2–3):305–325.

Bernard D, Blackburne T, Thornock J (2020) Information flows among rivals and corporate investment. J. Financial Econom. 136(3):760–779.

Bertrand M, Duflo E, Mullainathan S (2004) How much should we trust differences-in-differences estimates? Quart. J. Econom. 119(1):249–275.

Bettis JC, Coles JL, Lemmon ML (2000) Corporate policies restricting trading by insiders. J. Financial Econom. 57(2):191–220.

Beyer A, Cohen DA, Lys TZ, Walther BR (2010) The financial reporting environment: Review of the recent literature. J. Accounting Econom. 50(2–3):296–343.

Bhattacharya U (2014) Insider trading controversies: A literature review. Annu. Rev. Financial Econom. 6:385–403.

Billings MB, Cedergren MC (2015) Strategic silence, insider selling and litigation risk. J. Accounting Econom. 59(2–3):119–142.

Blackburne T, Kepler JD, Quinn PJ, Taylor D (2021) Undisclosed SEC investigations. Management Sci. 67(6):3403–3418.

Bloom N, Schankerman M, Van Reenen J (2013) Identifying technol- ogy spillovers and product market rivalry. Econometrica 81(4): 1347–1393.

Bond P, Edmans A, Goldstein I (2012) The real effects of financial markets. Annu. Rev. Financial Econom. 4(1):339–360.

Boschma R, Eriksson R, Lindgren U (2009) How does labour mobil- ity affect the performance of plants? The importance of related- ness and geographical proximity. J. Econom. Geography 9(2): 169–190.

Botosan CA, Stanford M (2005) Managers’ motives to withhold seg- ment disclosures and the effect of SFAS no. 131 on analysts’ information environment. Accounting Rev. 80(3):751–771.

Breschi S, Lissoni F (2009) Mobility of skilled workers and co-invention networks: An anatomy of localized knowledge flows. J. Econom. Geography 9(4):439–468.

Brochet F (2010) Information content of insider trades before and after the Sarbanes-Oxley Act. Accounting Rev. 85(2):419–446.

Brown S, Hillegeist SA, Lo K (2004) Conference calls and informa- tion asymmetry. J. Accounting Econom. 37(3):343–366.

Bushee BJ, Core JE, Guay W, Hamm SJW (2010) The role of the business press as an information intermediary. J. Accounting Res. 48(1):1–19.

Cao SS, Ma G, Tucker JW, Wan C (2018) Technological peer pres- sure and product disclosure. Accounting Rev. 93(6):95–126.

Carhart MM (1997) On persistence in mutual fund performance. J. Finance 52(1):57–82.

Chang X, Fu K, Low A, Zhang W (2015) Non-executive employee stock options and corporate innovation. J. Financial Econom. 115(1):168–188.

Chen Q, Goldstein I, Jiang W (2007) Price informativeness and investment sensitivity to stock price. Rev. Financial Stud. 20(3): 619–650.

Chen H, Cohen L, Gurun U, Lou D, Malloy C (2020) IQ from IP: Simplifying search in portfolio choice. J. Financial Econom. 138(1):118–137.

Cheng Q, Lo K (2006) Insider trading and voluntary disclosures. J. Accounting Res. 44(5):815–848.

Cheng CSA, Huang HH, Li Y (2016) Does shareholder litigation deter insider trading? J. Law Finance Accounting 1(2):275–318.

Clinch G (1991) Employee compensation and firms’ research and development activity. J. Accounting Res. 29(1):59–78.

Cohen L, Lou D (2012) Complicated firms. J. Financial Econom. 104(2):383–400.

Cohen L, Malloy C, Pomorski L (2012) Decoding inside information. J. Finance 67(3):1009–1043.

Cziraki P, Lyandres E, Michaely R (2021) What do insiders know? Evidence from insider trading around share repurchases and SEOs. J. Corporate Finance 66:1–24.

Dai L, Parwada JT, Zhang B (2015) The governance effect of the media’s news dissemination role: Evidence from insider trad- ing. J. Accounting Res. 53(2):331–366.

Dai L, Fu R, Kang JK, Lee I (2016) Corporate governance and the profitability of insider trading. J. Corporate Finance 40:235–253.

Damodaran A, Liu CH (1993) Insider trading as a signal of private information. Rev. Financial Stud. 6(1):79–119.

Davidson RH, Pirinsky C (2022) The deterrent effect of insider trad- ing enforcement actions. Accounting Rev. 97(3):227–247.

Debruyne M, Moenaert R, Griffinc A, Hart S, Huitink EJ, Robbenf H (2002) The impact of new product launch strategies on com- petitive reaction in industrial markets. J. Production Innovation Management 19:159–170.

Defond ML, Lennox CS (2017) Do PCAOB inspections improve the quality of internal control audits? J. Accounting Res. 55(3):591–627.

Del Guercio D, Odders-White ER, Ready MJ (2017) The deterrent effect of the securities and exchange commission’s enforcement intensity on illegal insider trading: Evidence from run-up before news events. J. Law Econom. 60(2):269–307.

Durnev A, Mangen C (2009) Corporate investments: Learning from restatements. J. Accounting Res. 47(3):679–720.

Dye RA (1986) Proprietary and nonproprietary disclosures. J. Bus. 59(1):331–366.

Edmans A, Jayaraman S, Schneemeier J (2017) The source of infor- mation in prices and investment-price sensitivity. J. Financial Econom. 126(1):74–96.

Edwards J (2014) Mastering Strategic Management, 1st Canadian ed. (BC Campus, Victoria, BC).

Ellis JA, Fee CE, Thomas SE (2012) Proprietary costs and the disclo- sure of information about customers. J. Accounting Res. 50(3): 685–727.

Ellul A, Panayides M (2018) Do financial analysts restrain insiders’ informational advantage? J. Financial Quant. Anal. 53(1):203–241.

Ertimur Y, Rawson C, Rogers JL, Zechman SLC (2018) Bridging the gap: Evidence from externally hired CEOs. J. Accounting Res. 56(2):521–579.

Fama EF, French KR (1997) Industry costs of equity. J. Financial Econom. 43:153–193.

Foucault T, Fresard L (2014) Learning from peers’ stock prices and corporate investment. J. Financial Econom. 111(3):554–557.

Frankel R, Li X (2004) Characteristics of a firm’s information envi- ronment and the information asymmetry between insiders and outsiders. J. Accounting Econom. 37(2):229–259.

Frankel R, Kothari SP, Weber J (2006) Determinants of the informa- tiveness of analyst research. J. Accounting Econom. 41(1–2):29–54.

Gao F, Lisic LL, Zhang IX (2014) Commitment to social good and insider trading. J. Accounting Econom. 57(2–3):149–175.

Gardner TM (2005) Interfirm competition for human resources: Evi- dence from the software industry. Acad. Management J. 48(2): 237–256.

Garmaise MJ (2011) Ties that truly bind: Noncompetition agree- ments, executive compensation, and firm investment. J. Law Econom. Organ. 27(2):376–425.

Glaeser S (2018) The effects of proprietary information on corporate disclosure and transparency: Evidence from trade secrets. J. Accounting Econom. 66(1):163–193.

Gosnell T, Keown AJ, Pinkerton JM (1992) Bankruptcy and insider trading: Differences between exchange-listed and OTC firms. J. Finance 47(1):349–362.

Gow ID, Larcker DF, Zakolyukina AA (2021) Non-answers during conference calls. J. Accounting Res. 59(4):1349–1384.

Griliches Z, Pakes A, Hall B (1987) The value of patents as indica- tors of inventive activity. Dasgupta P, Stoneman P, eds. Economic Policy and Technological Performance (Cambridge University Press, Cambridge, UK), 97–124.

Grossman S, Stiglitz J (1980) On the impossibility of informationally efficient markets. Amer. Econom. Rev. 70(3):393–408.

Hall BH, Ziedonis RH (2001) The patent paradox revisited: An empirical study of patenting in the U.S. semiconductor indus- try, 1979–1995. RAND J. Econom. 32(1):101–128.

He J, Tian X (2013) The dark side of analyst coverage: The case of innovation. J. Financial Econom. 109(3):856–878.

Healy P, Palepu K (2001) Information asymmetry, corporate disclo- sure, and the capital markets: A review of the empirical disclo- sure literature. J. Accounting Econom. 31(1–3):405–440.

Hillegeist SA, Weng L (2021) Quasi-indexer ownership and insider trading: Evidence from Russell index reconstitutions. Contempo- rary Accounting Res. 38(3):2192–2223.

Hoberg G, Phillips G (2010) Product market synergies and competi- tion in mergers and acquisitions: A text-based analysis. Rev. Financial Stud. 23(10):3773–3811.

Hoberg G, Phillips G (2016) Text-based network industries and endogenous product differentiation. J. Political Econom. 124(5): 1423–1465.

Huang A, Hui KW, Zheng Y (2024) Judge ideology and opportunis- tic insider trading. J. Financial Quant. Anal. Forthcoming.

Huang Y, Jennings R, Yu Y (2017) Product market competition and managerial disclosure of earnings forecasts: Evidence from import tariff rate reductions. Accounting Rev. 92(3):185–207.

Huddart SJ, Ke B (2007) Information asymmetry and cross-sectional variation in insider trading. Contemporary Accounting Res. 24(1): 195–232.

Huddart S, Ke B, Shi C (2007) Jeopardy, non-public information, and insider trading around SEC 10-K and 10-Q filings. J. Accounting Econom. 43(1):3–36.

Jaffe AB (1986) Technological opportunity and spillovers of R&D: Evidence from firms’ patents, profits, and market values. Amer. Econom. Rev. 76(5):984–1002.

Jagolinzer A, Larcker DF, Taylor DJ (2011) Corporate governance and the information content of insider trades. J. Accounting Res. 49(5):1249–1274.

Johnson MF, Nelson KK, Pritchard A (2007) Do the merits matter more? The impact of the private securities litigation reform act. J. Law Econom. Organ. 23(3):627–652.

Ke B, Huddart S, Petroni K (2003) What insiders know about future earnings and how they use it: Evidence from insider trades. J. Accounting Econom. 35(3):315–346.

Kedia S, Rajgopal S (2011) Do the SEC’s enforcement preferences affect corporate misconduct? J. Accounting Econom. 51(3):259–278.

Kim I, Skinner DJ (2012) Measuring securities litigation risk. J. Accounting Econom. 53(1–2):290–310.

King R, Pownall G, Waymire G (1990) Expectations adjustment via timely management forecasts: Review, synthesis, and sugges- tions for future research. J. Accounting Literature 9:113–144.

Koh PS, Reeb DM (2015) Missing R&D. J. Accounting Econom. 60(1):73–94.

Kyle A (1985) Continuous auctions and insider trading. Econometrica 53:1315–1335.

Lee H, Smith KG, Grimm CM, Schomburg A (2000) Timing, order and durability of new product advantages with imitation. Stra- tegic Management J. 21(1):23–30.

Li X (2010) The impacts of product market competition on the quan- tity and quality of voluntary disclosures. Rev. Accounting Stud. 15(3):663–711.

Li F, Lundholm R, Minnis M (2013) A measure of competition based on 10-K filings: A measure of competition based on 10-k filings. J. Accounting Res. 51(2):399–436.

Marion BW (1998) Competition in grocery retailing: The impact of a new strategic group on BLS price increases. Rev. Industry Organ. 13:381–399.

Marx M (2011) The firm strikes back: Noncompete agreements and the mobility of technical professionals. Amer. Sociol. Rev. 76(5):695–712.

Marx M, Strumsky D, Fleming L (2009) Mobility, skills, and the Michi- gan non-compete experiment. Management Sci. 55(6):875–889.

Massa M, Qian W, Xu W, Zhang H (2015) Competition of the informed: Does the presence of short sellers affect insider sell- ing? J. Financial Econom. 118(2):268–288.

Meulbroek LK (1992) An empirical analysis of illegal insider trad- ing. J. Finance 47(5):1661–1699.

Peress J (2010) Product market competition, insider trading, and stock market efficiency. J. Finance 65(1):1–43.

Piotroski J, Roulstone D (2004) The influence of analysts, institu- tional investors, and insiders on the incorporation of market, industry, and firm-specific information in stock prices. Account- ing Rev. 79(4):1119–1151.

Piotroski J, Roulstone D (2005) Do insider trades reflect both con- trarian beliefs and superior knowledge about future cash flow realizations? J. Accounting Econom. 39(1):55–81.

Plumlee MA, Xie Y, Yan M, Yu JJ (2015) Bank loan spread and pri- vate information: Pending approval patents. Rev. Accounting Stud. 20(2):593–638.

Puhani PA (2012) The treatment effect, the cross difference, and the interaction term in nonlinear “difference-in-differences” mod- els. Econom. Lett. 115(1):85–87.

Raith M (2003) Competition, risk, and managerial incentives. Amer. Econom. Rev. 93(4):1425–1436.

Ravina E, Sapienza P (2010) What do independent directors know? Evidence from their trading. Rev. Financial Stud. 23(3):962–1003.

Roulstone DT (2003) The relation between insider-trading restric- tions and executive compensation. J. Accounting Res. 41(3): 525–551.

Ryou JW, Tsang A, Wang KT (2022) Product market competition and voluntary corporate social responsibility disclosures. Con- temporary Accounting Res. 39(2):1215–1259.

Seyhun HN (1986) Insiders’ profits, costs of trading, and market effi- ciency. J. Financial Econom. 16(2):189–212.

Skaife HA, Veenman D, Wangerin D (2013) Internal control over financial reporting and managerial rent extraction: Evidence from the profitability of insider trading. J. Accounting Econom. 55(1):91–110.

Smith KG, Ferrier WJ, Ndofor H (2001) Competitive dynamics research: Critique and future directions. Handbook of Stra- tegic Management (Blackwell Publishers, Ltd., London), 315–361.

Stocken PC (2000) Credibility of voluntary disclosure. RAND J. Econom. 31(2):359–374.

Verrecchia RE (1983) Discretionary disclosure. J. Accounting Econom. 5(3):179–194.

Verrecchia RE (2001) Essays on disclosure. J. Accounting Econom. 32(1–3):97–180.

Wang IY (2007) Private earnings guidance and its implications for disclosure regulation. Accounting Rev. 82(5):1299–1332.

Wang RD, Shaver JM (2016) The multifaceted nature of competitive response: Repositioning and new product launch as joint response to competition. Strategy Sci. 1(3):148–162.

Welch J, Byrne JA (2003) Jack: Straight from the Gut (Grand Central Publisher, New York).

Woolridge J (2002) Econometric Analysis of Cross Section and Panel Data (MIT Press, Cambridge, MA).

Wu W (2021) Information asymmetry and insider trading. Working paper, Texas A&M University, College Station, TX.

Copyright of Management Science is the property of INFORMS: Institute for Operations Research & the Management Sciences and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

Do Proprietary Costs Deter Insider Trading?

Introduction
Hypothesis Development
Sample and Variable Measurement
Analyses and Results
Conclusion