It's Not Who You Know—It's Who Knows You: Employee Social Capital and Firm Performance

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) It’s Not Who You Know—It’s Who Knows You: Employee Social Capital and Firm Performance DuckKi Cho, Lyungmae Choi, Michael Hertzel, Jessie Jiaxu Wang 2023-020 Please cite this paper as: Cho, DuckKi, Lyungmae Choi, Michael Hertzel, and Jessie Jiaxu Wang (2023). “It’s Not Who You Know—It’s Who Knows You: Employee Social Capital and Firm Performance,” FinanceandEconomicsDiscussionSeries2023-020. Washington: BoardofGovernorsofthe Federal Reserve System, https://doi.org/10.17016/FEDS.2023.020. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

It’s Not Who You Know—It’s Who Knows You: Employee Social Capital and Firm Performance† DuckKi Cho Lyungmae Choi Michael Hertzel Jessie Jiaxu Wang* Abstract We show that the social capital embedded in employees’ networks contributes to firm performance. Using novel, individual-level network data, we measure a firm’s social capital derived from employees’ connections with external stakeholders. Our directed network data allow for differentiating those connections that know the employee and those that the employee knows. Results show that firms with more employee social capital perform better; the positive effect stems primarily from employees being known by others. We provide causal evidence exploiting the enactment of a government regulation that imparted a negative shock to networking with specific sectors and provide evidence on the mechanisms. JEL codes: G30, G41, L14 Keywords: Social capital; Social networks; Labor and finance † We thank Kenneth Ahern, Ilona Babenko, Sreedhar Bharath, Alice Bonaime, Jillian Grennan, Jessica Jeffers, Theresa Kuchler, Jongsub Lee, Michael Lee, Denis Sosyura, Nicholas Wilson, and conference and seminar participants at the 2020 UA-ASU Junior Finance Conference, 2020 Northern Finance Association Conference, WAPFIN@STERN 2020, 2nd Finance Junior Conference at Indiana University, 2021 Midwest Finance Association Conference, 2021 Society of Labor Economists meeting, 2021 Future of Financial Information Conference, 2021 Financial Intermediation Research Society Conference, 2021 KIF-KAEA-KAFA joint symposium, 2021 China International Conference in Finance, 2022 Napa/Sonoma Conference, 2023 American Finance Association Conference, Arizona State University, Nanyang Technological University, and Peking University for helpful comments. We are grateful to Drama & Company for providing the data. DuckKi Cho acknowledges the PHBS Dean’s Research Fund for financial support. Lyungmae Choi acknowledges the General Research Fund of the Research Grants Council of Hong Kong (Project no. 21502018) for financial support. The views expressed in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or of anyone else associated with the Federal Reserve System. * Cho, duckki.cho@phbs.pku.edu.cn, Peking University, HSBC Business School; Choi, lyungmae.choi@cityu.edu.hk, City University of Hong Kong; Hertzel, Michael.Hertzel@asu.edu, W. P. Carey School of Business, Arizona State University; and Wang, jessiejiaxuw@gmail.com, Board of Governors of the Federal Reserve System and Arizona State University.

1. Introduction The role of physical capital, human capital, and intellectual capital in corporations is well studied. Yet, less attention has been paid to a firm’s social capital, consisting of the relationships that a firm and its employees have built with economically related agents outside the firm. Social capital is a broad concept that can be understood as the norms of reciprocity and trust within social networks (Putnam, 2000). The literature has shown that an individual’s social capital—such as the size of their Rolodex—provides them with benefits and access to resources (Bourdieu, 1986; Coleman, 1988; Lin, 2002; Glaeser et al., 2002).1 An individual’s social capital should also be useful at the firm level since employees, including both management and rank and file, interact directly with business partners, clients, and other stakeholders. Yet, due to the latent nature of social networks, how the social capital embodied in employees’ connections contributes to firm value and performance remains much unexplored.2 In this paper, we aim to establish a causal link between employee social capital and firm performance. We construct a novel firm-level measure of employee social capital using professional connections that a firm’s employees, across all job levels, have built with business contacts outside the firm.3 To provide causal evidence, we exploit the enactment of a government regulation as a plausibly exogenous shock. We also identify the types of employee connections that are valuable to firms and discover the economic benefits that firms obtain from these connections, thus contributing to a more granular understanding of social capital in corporations. To measure employee social capital, we leverage a unique cultural practice in Asia: the exchange of business cards when people make connections. We have full access to data from the professional networking app “Remember,” to which users upload business cards they have collected from others. Remember has a near-monopoly of business card management in Korea, with its users making up 18% of the total full-time employees of the country. The comprehensive data on card collections of every user allow us to identify the professional networks of individual employees and 1 A complementary approach measures social capital at the country or regional level using metrics such as the civic engagement of the population or civic norms and trust. These studies find that regions with more social capital have better economic outcomes (e.g., Knack and Keefer, 1997; La Porta et al., 1997; Guiso et al., 2004, 2008) and that firms in these regions suffer less from agency problems (Hasan et al., 2017; Hoi et al., 2019). 2 Limited by data availability on networks, the literature that uses the network approach focuses almost exclusively on benefits firms obtain from their well-connected executives and board members (e.g., Cai and Sevilir, 2012; Engelberg et al., 2012; Larcker et al., 2013). 3 Our construction of employee social capital distinguishes it from relationships within the firm (see, e.g., Jeffers and Lee, 2019) or norms and values that are shared within the firm, also referred to as corporate culture (see, e.g., Guiso et al., 2015; Popadak, 2016; Graham et al., 2018; Grennan, 2022; Graham et al., 2022; Gorton et al., 2022; Grennan and Li, 2022). 1

quantify the connections each employee has built with people outside their firm. We further map the connections of public firm employees to the financial variables of their employers to obtain a matched employer-employee dataset. Several aspects of our data are novel. First, our final sample consists of 2.4 million employees, with 12.4 million connections between them. The data’s broad coverage of employees across ranks allows us to quantify employee social capital at the firm level. Second, because in Asian culture business cards are typically exchanged in face-to-face meetings (it is not the norm to pass on cards on behalf of others), our data depict real-world professional connections more reliably than those from online platforms, such as LinkedIn, where people can connect even though they have never met. Third, while card exchanges are mutual, uploading cards to the app is not necessarily mutual as users are more likely to upload the cards of contacts that they want to remember (apropos the name of the app). We refer to the network as directed: each connection is directed from the employee who uploads the card to the employee whose card is uploaded. We introduce three connection measures at the individual employee level—In-degree (number of others who have uploaded the employee as a contact), Out-degree (number of contacts uploaded by the employee), and Total degree (sum of In-degree and Out-degree). In-degree counts the people who remember the employee by uploading the employee’s card on the app, which we refer to as “who knows you”; Out-degree counts the contacts the employee remembers by uploading their cards, which we refer to as “who you know.”4 This directed nature of our network data enables us to move beyond “who knows who” and analyze the extent to which social capital—as distinguished by “who knows you” versus “who you know”—matters for the firm. Based on the three employee-level degree measures (In-degree, Out-degree, Total degree) within a firm, we construct firm-level measures of employee social capital (ESC) for a comprehensive sample of Korean public firms in OSIRIS Industrials from 2014 to 2018. Our initial analysis shows that, without regard to the direction of connections, the average Total degree of a firm’s employees is positively associated with firm profitability and sales growth in the following year. To investigate whether the direction of connections matters, we then separately examine ESC in-degree and ESC out-degree. Results show that the positive relation with future performance arises 4 Although none are perfect descriptors, we use “who knows you” and “who remembers you” interchangeably throughout the paper to describe an employee’s In-degree, which captures the extent to which the employee is in the Rolodex (or among the list of business contacts) of others. Similar descriptors apply to Out-degree, which measures the size of the employee’s Rolodex. A reciprocal connection where both parties upload each other’s cards (“know each other”) counts toward both the In-degree and Out-degree for each party. 2

mainly from ESC in-degree, which captures the extent to which a firm’s employees are remembered by their external contacts.5 While the social capital literature argues that networks benefit individuals, our findings imply that the extent to which employees can mobilize these benefits for their firm depends on whether their business contacts remember them. In this sense, having a broad network of business contacts who know you appears more valuable to your firm than having a broad network of contacts whom you know.6 Finally, we leverage the data’s coverage of employees across job levels to study employee social capital beyond the executive team, an aspect less explored in the literature. Our results emphasize the unique value of social capital embodied in non-executive employees. Establishing a causal link between employee social capital and firm performance requires a careful account of the endogeneity of networks. Despite our extensive robustness tests, concerns remain, such as reverse causality whereby better firm performance leads to the formation of connections. To address the endogeneity of employee social capital and reinforce its causal effect on firm performance, we exploit the 2016 enactment of the Kim Young-ran Act (the Act) as a plausibly exogenous shock to professional networking in Korea. The Act makes it illegal for media professionals (such as journalists) and public sector employees (such as public servants, lawmakers, and teachers), and their spouses to accept gifts or meals exceeding a specified limit, regardless of whether they are in exchange for favors. The Act is a suitable identification tool because of the uncertainty in the legislative process and its aggressive enforcement. Evidence suggests that the Act caused significant precautions among businesses, creating a chilling effect on social events and meetings with contacts in the media and the public sector. By limiting employees’ ability to extract benefits from their existing connections to these affected sectors, the Act constituted a negative shock to a firm’s employee social capital. 5 We discuss robustness checks in Internet Appendix II concerning issues with omitted variable bias, measurement error, and selection bias. Employee social capital may proxy for other variables that relate to firm performance. For example, sales personnel who serve as customer touchpoints are, by nature, active in exchanging cards, such that the observed relation between employee connections and sales growth might simply reflect firms’ sales activities. Our results, however, are robust to excluding connections of a firm’s customer-facing employees or excluding the connections with external contacts in customer industries. Firms with well-connected employees might also have high employee technical skills or high employee satisfaction, both related to superior firm performance. Following the strategy in Cohen et al. (2010), we exclude firms that are popular employers among skilled employees and find the results continue to hold. Finally, we show the robustness of our results against potential measurement error and selection bias in constructing firm-level employee social capital caused by differential app usage among a firm’s employees. 6 Although appearing less useful to employers, “who you know” can be an asset for employees themselves. To the extent that employees uploading contacts from other firms—as measured by ESC out-degree—expands outside job opportunities, as shown by Gortmaker et al. (2020) using data from LinkedIn, the resources mobilized through these connections do not necessarily accrue to their employer. 3

We use a difference-in-differences framework surrounding the enactment of the Act. The treatment intensity is the fraction of a firm’s preexisting employee social capital derived from its employees’ connections with the media and the public sector. Since some firms have employees more connected to these two sectors, we can estimate differences in performance before and after the Act between firms with differential exposure. We find that firms with employees more connected to these two sectors experience a greater decline in performance after the Act relative to those less connected. For instance, a one standard deviation increase in treatment intensity yields an increase in Tobin’s q of 17.5% relative to the sample mean before the Act, but only by 4.4% after. The differential effect does not appear in pre-treatment years and persists over the years after. Our results are robust to matching treatment to control firms based on industry and observable firm characteristics and to excluding firms that are economically linked to the two sectors directly affected by the Act, such as customers and suppliers of the media and the public sector. Using an event study approach, we examine stock price reactions around the court ruling date of the Act. Consistent with the value of firms’ employee social capital being destroyed by the limits on social interactions imposed by the Act, we find a significantly negative cumulative abnormal return of −0.61% (p-value = 0.017) for firms with employees more connected to the media and the public sector over the [−3, 3] event window, and a differential cumulative abnormal return of −1.02% (p-value = 0.019) relative to firms that are less connected. To shed light on the mechanisms through which employee social capital contributes to firm value, we consider the benefits that firms get from their employees’ connections with the sectors affected by the Act—the media and the public sector. Motivated by the literature on media coverage and firm value, we predict that employees’ media connections foster reciprocity and information sharing with journalists, which in turn promotes media coverage of the firm. Indeed, we find that firms with more employee media connections have substantially more news articles and a greater fraction of news articles with a positive tone. Moreover, the positive effects diminish after the enactment of the Act, reinforcing our causal inference. We then turn to the benefits of employee connections with the public sector. Drawing on evidence that public officers allocate more procurement contracts to firms with a connected CEO, we expect that employees with public sector connections also help their firms secure procurement contracts. This is indeed what we find. For example, a one standard deviation increase in the fraction 4

of employee social capital accumulated from public sector connections leads to a 6.8% increase in the number of newly signed contracts before the Act and only a 3.4% increase after.7 Our study adds to the burgeoning literature on the role of social capital in corporations. Since relationships of a firm are difficult to observe and measure, existing metrics for firm social capital largely rely on corporate social responsibility efforts or norms and social interactions in local areas around corporate headquarters, such as voter turnout, census response rate, density of sports clubs, and friendship links on Facebook. This literature finds that firms that entered a financial crisis with more social capital perform better (Lins et al., 2017; Servaes and Tamayo, 2017) and that firms operating in areas with higher social capital have better access to finance (Hasan et al., 2017; Kuchler et al., 2022), suffer less from agency problems (Hoi et al., 2019), and have earnings news more rapidly incorporated into stock prices (Hirshleifer et al., 2021). Adding to this literature, we develop a novel measure of a firm’s social capital using the professional connections of its employees, and show that otherwise similar firms with more employee social capital perform better, thus shedding light on the drivers of firm productivity (Syverson, 2011). Our study also complements prior work that identifies the benefits of managerial networks, such as high announcement returns in mergers and acquisitions (Cai and Sevilir, 2012), better firm performance (Larcker et al., 2013; Cai and Szeidl, 2017; Dass et al., 2014), favorable lending terms (Engelberg et al., 2012; Haselmann et al., 2018; Karolyi, 2018), and survival during a financial crisis (Acemoglu et al., 2016).8 Adding to this literature, we present novel evidence that executives are not the only group that possesses beneficial connections for their firms; employee connections across all job ranks matter for firm outcomes. More importantly, by exploiting the directed feature of our data, we uniquely show that the value of employee social capital to a firm comes mainly from employees being remembered by their external contacts. Finally, our study leverages the Asian cultural practice of exchanging business cards, which provides a unique institutional setting for identifying interpersonal networks. Although our evidence draws from Korean firms, the effects of social ties on business outcomes have been documented in diverse business cultures, such as the US (Hochberg et al., 2007; Shue, 2013), China (Cai and Szeidl, 7 A possible underlying channel is that employees’ media and public sector connections facilitate favor exchanges with journalists and public officials (which may include bribery). Although bribery reflects a dark side from a societal perspective, it represents a favor exchange facilitated through employee networks that benefits the firm. We elaborate on this point in Section 4.5 8 Other studies point out potential downsides to the firm with well networked executives: connections could weaken effective monitoring of board members, increase the entrenchment of CEOs, and lead to rent-seeking coalitions (Hwang and Kim, 2009; Fracassi and Tate, 2012; Ishii and Xuan, 2014; Khanna et al., 2015; Gompers et al., 2016). 5

2017), Germany (Haselmann et al., 2018), the UK (Rossi et al., 2018), and the global setting (Houston et al., 2018), suggesting that the insights are general and broadly contribute to our understanding of social capital. This paper proceeds as follows. Section 2 describes the data and the construction of firmlevel employee social capital. Section 3 examines the relation between employee social capital and firm performance. In Section 4, we provide causal evidence using the enactment of the Kim Youngran Act as a quasi-natural experiment, and Section 5 concludes. 2. Data and summary statistics 2.1. Remember, a professional networking app We exploit a unique dataset extracted from a professional networking app, Remember, which was developed by the Korean mobile and web service provider Drama & Company. Since its launch in January 2014, Remember has become the most popular business card management app in Korea, with virtually no domestic competitors.9 As of December 2018, the app had around 2.5 million users, equivalent to 18.1% of the total number of full-time employees in Korea. To track their professional network, app users upload the business cards they collected in face-to-face meetings. Then typists hired by the app hand-type the scanned cards into the database, which renders the network data free of automatic recognition errors. Through the app, users can keep track of their professional networks, use search criteria to connect to calls, texts, emails, and addresses, and add updates about promotions or new job titles. Unlike online networking platforms (e.g., LinkedIn, Facebook, or Twitter), the network of a user is not visible to others. 2.2. Business card data and individual employee-level connections The cultural norms in Korea strongly support the notion that tracking business card exchanges is a useful way to identify employees’ professional networks. As in most other Asian countries, in Korea, exchanging business cards in face-to-face meetings is more than an exchange of personal details; it is a ritual for building professional connections. It is widely believed that, besides being an ice breaker, the exchange of business cards can help establish a positive first impression and boost professional credibility. Business cards are also a physical reminder that one 9 The Remember app won the Google Play Awards in 2015 and 2016 and received the Brand of the Year Korea for four consecutive years, from 2015 through 2018. The app is accessible at rememberapp.co.kr, and is available free of charge from Google Play and the App Store. Figure IA.3 in the Internet Appendix illustrates how the app appears in the App Store, the app’s user interface, and how to upload business cards. 6

has met the contact rather than simply googled them. In addition, exchanging cards helps the two parties bond and build trust by encouraging follow-up social events.10 Tracing the exchange of business cards using our dataset is thus a feasible and reasonable way to identify Koreans’ professional networks. From each card uploaded by each app-user by December 2018, we obtain detailed information about the business contact, including an individual identifier (defined by coded name and coded mobile number to comply with user privacy laws), email, firm name, job position, and timestamp of card upload. The unit of observation is the connection pair consisting of the app-user who uploads the card and the business contact whose card is uploaded. Since our goal is to count connections among employees, we exclude connections that involve individuals who do not have a firm name on their card, whose email domain is inconsistent with their firm, or whose firm does not have a Korea Investors Service (KIS) identifier (a corporate registration number for listed and unlisted firms). To focus on interfirm connections, we keep connections between employees from different firms (with different KIS identifiers).11 Internet Appendix I provides more details on our data and an illustrative example. In general, cards are mutually exchanged between two parties, but the uploading of cards is not necessarily mutual. For example, after Aaron and Bob meet and exchange cards, Aaron uploads Bob’s card, but Bob does not upload Aaron’s card. Following the network literature (Jackson, 2008; Newman, 2010), this feature implies our network data are directed. Specifically, in social networks, individuals (nodes) form connections (links) to other individuals; the nodes and links constitute the network. If the links have a specified direction and are not necessarily mutual, the network is directed. The literature visualizes directed networks by drawing links as arrows to indicate the direction. So there can be links pointing inward to and outward from each node. The number of links pointing inward to each node is the in-degree, and the number of links pointing outward is the outdegree. The total degree of a node is the sum of its in- and out-degree. Applying these concepts to our data, each connection is a link directed from the user who uploads the card to the contact whose card is uploaded. The example of Aaron uploading Bob’s card counts as an out-degree for Aaron, and an in-degree for Bob. Users are most likely to remember 10 As discussed extensively in the Economist (May 2015), “business cards are doubly useful. They can be a quick way of establishing connections, particularly in Asia, where they are something of an obsession . . . exchanging business cards still seems to be an excellent way to initiate a lasting relationship. The ritual swapping of paper rectangles may be old-fashioned but on it will go.” Also see “Why Business Cards Still Matter,” BBC, September 2016, www.bbc.com/worklife/article/20160914-how-a-small-yet-mighty-bit-of-paper-can-still-get-you-a-job. 11 Since internal networking platforms (e.g., intranet) are often available, it is less common for employees to exchange cards within a firm. Consistent with this cultural norm, intra-firm connections are relatively less frequent in our data. 7

those business contacts whose cards they uploaded—as suggested by the name of the app. To capture this feature, we define three degree measures at the employee-year level. In-degree is the number of employees of other firms who have uploaded the employee as a contact by a given year (“who knows you”). Out-degree is the number of external business contacts uploaded by the employee by a given year (“who you know”). Total degree is the sum of In-degree and Out-degree. A reciprocal relationship, which occurs when both parties upload each other’s cards, counts toward both the Indegree and Out-degree for each party, thereby increasing the Total degree of each party by two. Since our interest is in the performance of publicly listed firms, we keep the connections in which at least one of the two individuals is a public firm employee. This network consists of 12.4 million connections between 2.4 million employees; among them, 17.4% are app-users and 43.0% work for public firms. There are 126,987 firms with KIS identifiers; among them, 1,866 are public firms. To analyze the performance of Korean public firms, we use the OSIRIS Industrials database, which contains financial information on publicly listed industrial firms worldwide. Our data cover firms in a wide array of sectors (see Table IA.2 in the Internet Appendix). Panel A of Table 1 presents summary statistics of connections for the public firm employees in our sample. There are 119,423 app-user employees. An average app-user employee has been uploaded as a contact by 26 app-users outside the firm (In-degree) and has uploaded 57 contacts from other firms (Out-degree). The sum of the two degrees, Total degree, has a mean of 83. All degree measures have a median much lower than the mean, suggesting that the distributions are highly right skewed. There are 896,600 non-app-users working for public firms. Non-app-users enter the network when their cards are uploaded by app-users and thus, by definition, only have links pointing inward.12 On average, a non-app-user, whose In-degree (which also equals Total degree) is around five, is uploaded as a contact by five app-users outside the firm. Pooling the app-users and non-app-users together, an average public firm employee in the network is uploaded by seven others as a business contact and has a total degree of 14. [Table 1 about here] Our data have several advantages in identifying employees’ professional networks. First, the data’s broad coverage of employees (including management and rank and file) allows us to map employee-level connections to their firms to construct a matched employer-employee dataset. This 12 We discuss potential measurement error and selection bias caused by not observing the Out-degree of non-app-users in Internet Appendix III. 8

feature overcomes a limitation of the literature that has focused primarily on managerial networks. Second, because business cards are typically exchanged in a face-to-face meeting, our data depict real-world professional relationships more reliably than online networks such as LinkedIn. An uploaded card is a physical imprint that the two people indeed met rather than simply connected via an online invitation. Third, since the connections of an employee are not publicly visible, one’s Indegree and Out-degree are unlikely to strategically influence each other. Fourth, the directed nature of the data allows us to move beyond “who knows who” and analyze the extent to which social capital—as distinguished by “who knows you” versus “who you know”—matters for firm outcomes. 2.3. Firm-level employee social capital (ESC) To examine the extent to which employees’ professional connections contribute to the employer’s performance, we construct measures of firm-level employee social capital (ESC) based on the employee-level degree measures. Our strategy is to average across the employee-level degrees to obtain a proxy for the connectedness of the representative employee of each firm. We utilize the direction of connections to decompose firm-level employee social capital into ESC indegree and ESC out-degree. ESC in-degree is the average In-degree across a firm’s employees in the network; it quantifies the number of times a firm’s employees have been uploaded as business contacts. As noted earlier, non-app-users enter the network when their cards are uploaded by others and thus, only have In-degree. Accordingly, ESC out-degree is the average Out-degree across the app-user employees of a firm; it quantifies the number of external business contacts that a firm’s app-user employees have uploaded. Finally, ESC total degree is the average Total degree across a firm’s employees in the network.13 2.4. Sample construction and summary statistics To construct our sample, we start with Korean public firms from the annual OSIRIS Industrials database from 2014 through 2018. We match the 1,866 public firms in the network data with OSIRIS Industrials using firm names. We use three measures for firm performance: Tobin’s q is the market value of assets divided by the book value of assets; ROA (return on assets) is earnings before interest, tax, depreciation, and amortization (EBITDA) divided by the lagged total assets;14 13 To reduce measurement error when taking averages, we restrict our sample to firm-year observations with at least ten employees observed in the network. Our results are robust to using alternative thresholds for the minimum number of employees who appear in the network; see discussions in Internet Appendix III on potential measurement error. 14 Using EBIT instead of EBITDA to measure ROA does not change our results. 9

Sales Growth is the annual log growth rate of sales. The definitions of all variables are provided in Internet Appendix II. We drop firm-year observations with missing data for the main variables in the baseline regressions. To reduce the effects of outliers, we winsorize all potentially unbounded variables at the 1st and 99th percentiles of the distribution. The final sample consists of 5,340 firmyear observations and covers 1,553 unique firms. Panel B of Table 1 reports summary statistics for our firm-year sample. ESC in-degree has a mean of 3.7 and a median of 3.1; ESC total degree has a mean of 6.8 and a median of 5.3. These numbers show that employees of a firm, on average, have 6.8 connections with employees of other firms and that in 3.7 of those connections, they are uploaded as a business contact by others. In comparison, ESC out-degree has a mean of 31.0 and a median of 24.2 among users, suggesting that app-user employees of a firm, on average, upload 31.0 business contacts from other firms; ESC outdegree is larger in magnitude than ESC total degree because we observe a more complete picture of connections by app-user employees of a firm, as reported in Panel A of Table 1.15 The financial variables are comparable in magnitude to those of US firms during the same period; Korean firms have less skewed Tobin’s q, larger ROA, smaller Sales Growth, and lower Book Leverage. Summary statistics of firm-level ESC measures by sector are reported in Table IA.2. Our results are not driven by any particular sector. 3. Employee social capital and firm performance: baseline analysis This section provides baseline estimates of the relation between employee social capital and firm performance. In Section 3.1, we examine ESC total degree, without accounting for the direction of connections. In Section 3.2, we exploit the directed nature of our network data, considering both ESC in-degree and ESC out-degree. Section 3.3 evaluates employee social capital across executives and non-executive employees. 3.1. Employee social capital measured by total degree The social capital literature suggests that social ties are associated with valuable resources. For instance, Bourdieu (1986) considers social capital as “the actual or potential resources which are linked to possession of a durable network”; Putnam (2000) notes that social connections lead to reciprocity, trust, and better sharing of information; and Lin (2002) defines social capital as resources that can be accessed or mobilized through ties in the networks. Motivated by this literature, 15 The number of observations of ESC out-degree is slightly smaller than that of the other main variables; this is because some firm-year observations do not have app-user employees and thus are missing ESC out-degree. 10

we examine the relation between employee social capital and future firm performance by estimating the following specification: Y = 𝛽 +𝛽 ×ln(1+ESC )+𝛾′X +𝛼 +𝜀 , ( 1 ) i,t 0 1 i,t-1 i,t-1 j,t i,t where Y is one of the performance measures (Tobin’s q, ROA, or Sales Growth), ESC is the onei,t i,t-1 year lagged firm-level employee social capital, X is a set of one-year lagged time-varying firmi,t-1 specific control variables (R&D, book leverage, total assets, stock return volatility, firm age, and number of employees) commonly included in the literature (see, e.g., Anderson and Reeb, 2003), and 𝛼 is a full set of two-digit Standard Industrial Classification (SIC) industry-by-year fixed j,t effects. As our data have a short time span, much of the variation in firm-level ESC is in the cross section; hence, we include industry-by-year fixed effects to control for unobserved time-varying heterogeneity across industries in, for example, business performance, professional connectivity, or employee app usage. Since our ESC measures are right skewed, we take the log transformation to reduce the effects of outliers; our results are qualitatively robust to using ln(ESC) and also robust to not taking the log transformation. [Table 2 about here] The results when ESC takes the value of ESC total degree (average Total degree at year i,t-1 t-1 across employees of firm i who are in the network) are shown in columns (1)–(3) of Table 2 Panel A. The coefficient estimates on ln(1+ESC) are positive across all performance measures, and statistically significant for ROA and Sales Growth. The estimates imply that a one standard deviation increase in ESC from its mean is associated with an increase in ROA of 0.4 percentage points (=0.008×(ln(1+6.836+5.844)−ln(1+6.836))) and Sales Growth of 2.1 percentage points. The effects are significant, given the mean ROA of 4.3 percentage points and the mean Sales Growth of 4.1 percentage points over the sample period,16 suggesting a positive relation between firm performance and employee social capital based on employees’ total number of connections. 3.2. Does the direction of employee connections matter? In-degree versus out-degree We next exploit the directed nature of our data which allows us to separately account for the business contacts that remember the employee and the business contacts that the employee remembers. More specifically, by using our decomposition of employee social capital into ESC in- 16 Since ROA and Sales Growth have negative values in the distribution, we do not compute the percentage increase relative to the sample mean when evaluating the economic magnitudes. 11

degree, which measures “who knows you,” and ESC out-degree, which measures “who you know,” we consider whether the direction of connections matters. Our results of re-estimating equation (1) separately for ESC in-degree and ESC out-degree, reported in columns (4)–(9) of Panel A, provide strong evidence that the direction of connections plays a role in firm performance. All coefficient estimates on ESC in-degree, reported in columns (4)–(6), are positive and statistically significant at the 1% level. The estimated effects are economically meaningful: a firm with one standard deviation more ESC in-degree has a 9.4% higher Tobin’s q relative to the sample mean, a 0.9 percentage points higher ROA, and a 4.0 percentage points higher Sales Growth. By contrast, the coefficient estimates on ESC out-degree in columns (7)–(9) are insignificant or borderline significant. The estimated coefficients for ESC out-degree and economic significance are an order of magnitude smaller than those for ESC in-degree, which is also confirmed by the one-tailed tests (p-value < 1% for all three columns). For example, relative to the 9.4% increase in Tobin’s q for ESC in-degree noted above, the same increase in ESC out-degree from its mean is associated with only a 1.8% increase in Tobin’s q.17 To address concerns with omitted variables, measurement errors, and selection bias, we conduct a battery of robustness tests in Internet Appendix III. Our findings suggest that the positive relation between employee social capital and firm performance comes mainly from employees’ connections with external contacts who remember the firm’s employees. While social ties provide benefits, the extent to which employees can leverage these benefits for their employers depends on whether their business contacts remember them. Although our results show that out-degree connections are less useful to their employers, individuals may still derive personal benefits from these connections. For example, studies show that social networks are useful for individuals seeking outside job opportunities (e.g., Lin et al., 1981; Granovetter, 1973, 1995; Hacamo and Kleiner, 2021). If employees uploading contacts from other firms—as measured by ESC out-degree—reflects their desire and efforts to switch firms,18 the resources mobilized through these connections do not accrue to their employer. Overall, our baseline regressions show that firms with more employee social capital have significantly better performance in the next year; yet, compared with the rolodex that an employee possesses, being on others’ rolodex is a more robust indicator of employee social capital that can benefit the firm. 17 Our results are robust to controlling for the percentage of employees of a firm using the app. 18 This mechanism is consistent with the evidence in Gortmaker et al. (2020). They analyze micro-level data from LinkedIn and find that, after learning about their firms’ credit deterioration, workers start initiating connections on LinkedIn more frequently; this is followed by an increased likelihood of a job change afterward. 12

3.3. Does employee job level matter? Executives versus non-executive employees A key advantage of our data is the broad coverage of employees across ranks, which allows us to study the social capital embodied in employees beyond the executive team, an aspect scarcely examined in prior literature largely due to data limitations. While executives make major strategic decisions, non-executive employees, such as middle managers and rank-and-file employees, constitute most of a firm’s workforce and often closely interact with business partners and other key stakeholders. Understanding the social capital embodied in non-executive employees is important since decision-making and information processing within a firm are often decentralized by a hierarchical structure (Radner, 1992). Panel B of Table 2 presents results on the effects of employee social capital on firm performance across executives and non-executive employees.19 Results show that ESC in-degree is positively associated with firm performance measures for both executives and non-executives. For example, a one standard deviation increase in ESC in-degree of executives is associated with a 7.3% increase in Tobin’s q relative to the sample mean; and that of non-executive employees is associated with a 5.6% increase.20 Results are similar when we include ESC in-degree of executives and nonexecutives in the same regression (untabulated). While our findings echo existing studies on the value of executive networks based on undirected network data (e.g., Cai and Sevilir, 2012; Engelberg et al., 2012; Larcker et al., 2013), they also uniquely suggest that non-executive employees have beneficial connections that contribute to firm performance. 4. Causal evidence from the 2016 Kim Young-ran Act Although we conduct a battery of tests to mitigate concerns with omitted variable bias and measurement error (and to some extent reverse causality by using lagged ESC measures), the results of our analysis may still be subject to endogeneity concerns. To establish a causal relation between employee connections and firm performance, it is important to identify exogenous variation in employee social capital. In this section, we provide causal evidence by exploiting a quasi-natural experiment that imparted a negative shock to professional networking in Korea. 19 Job levels classified as executives include chairman, vice chairman, president, deputy president, executive vice president, and senior vice president; about 9.7% of the observed employees are executives. Non-executive employees include all other employees. 20 The number of observations varies slightly across regressions because a small number of firm-years do not have executives. Results are similar when we run the regressions on the same set of observations. 13

4.1. Exogenous shock to employee social capital: the 2016 Kim Young-ran Act We exploit the enactment of the Kim Young-ran Act (the Act) in September 2016 as an exogenous shock to social interactions with employees in specific sectors. Named after the former head of the Anticorruption and Civil Rights Commission, the Act makes it illegal for media professionals (such as journalists) and public sector employees (such as civil servants, lawmakers, and teachers), and their spouses to accept gifts of more than 50,000 Korean won (about 45 USD) or 100,000 won at events such as weddings and funerals; it also limits meal expenditures to 30,000 won per person.21 Violations of the Act are subject to severe penalties, including imprisonment.22 Although the Act was intended to prevent corruption, the gift and meal limits also resulted in fewer social events and meetings with contacts employed in the media and the public sector, thereby restricting firms’ ability to leverage their employee social capital with these sectors. As a culturally ingrained business practice in Korea, corporate employees would regularly treat clients, business partners, and public employees to dinners, drinks, and other entertainment as part of normal networking activity (Choi and Storr, 2019). Through engagement in these networking activities, professionals invest in their social capital, enhance trust, and share information. However, anecdotal evidence suggests that the Act has caused significant precautions among businesses in their interactions with the media and the public sector due to the severity of its penalties, its aggressive enforcement, as well as its somewhat abstract and vague provisions and the lack of precedents.23 For example, companies say “they are concerned about how to maintain business relationships they have built with government officials and the media over the years. The law’s definition of those related to work is ambiguous…as it excludes socializing as part of business formality.” This concern by firms is consistent with the observations that “reservation rates of restaurants in Seoul’s financial and legal districts and those near government complexes in Sejong and Daejeon, have rapidly dropped” and that Korean reporters were intentionally left off the invitation list in a launch event for Apple’s iPhone X. 21 The upper limits were adjusted in January 2018 to 100,000 won for non-cash gifts and to 50,000 won for cash gifts. 22 The Act imposes a punishment of imprisonment for up to three years, or a fine of up to 30 million Korean won on persons convicted of accepting money or goods valued at more than one million won from one person in one installment, regardless of whether such compensation was in exchange for favors or related to the recipient’s work. If the money or goods are worth less than one million won, a fine of up to five times the gift’s value is imposed. 23 See, for example, “Corporate Korea Braces for Change over Anti-Graft Law,” Korea Herald, September 27, 2016, www.koreaherald.com/view.php?ud=20160927000851; “Companies Still Need to be Cautious of Kim Young-ran Act,” Korea Herald, September 24, 2017, www.koreaherald.com/view.php?ud=20170922000818. 14

To provide more systematic evidence that the Act resulted in an exogenous shock to employee social capital with the media and the public sector, we examine changes in the formation of connections with these sectors around the Act. Specifically, we examine the fraction of a firm’s employee social capital (ESC in-degree) that is derived from connections with employees in the industries affected by the Act (ESC in-degreeAct), as identified using industry codes listed in Internet Appendix II.24 Our estimation results in Table IA.3 further show that the fraction dropped by 7.8% (= −0.266/3.414) after the enactment relative to the sample mean. Hence, the evidence is consistent with the Act discouraging the formation of new connections with personnel in the media and the public sector. Another aspect that makes the Act a useful identification tool is the uncertainty around whether the Act would be ruled constitutional. Right after bipartisan approval of the Act in 2015, the Korean Bar Association and the Korean Journalists Association filed a court petition questioning the law’s constitutionality on the grounds that it threatened freedom of speech. The Constitutional Court upheld the law on July 28, 2016, rejecting the petition. This series of unforeseen events supports our identifying assumption of orthogonality between the enactment and unobservables that affect firm performance. 4.2. Evidence for causality We assess the causal effect of employee social capital on firm performance using a difference-in-differences framework surrounding the enactment of the Kim Young-ran Act. Since some firms have more of their employee social capital derived from connections to the media and the public sector (thus have employee social capital more exposed to the Act) than others, we can estimate differences in performance between firms with differential exposure to the Act. The restrictions of the Act impair the ability of employees to access the resources embedded in their existing connections to the media and the public sector; hence, we hypothesize that firms with greater exposure experienced a bigger reduction in the value of their employee social capital. We test the predictions of our hypothesis by estimating the following regression model: 24 Our results in Section 3 show that the economic value of employee social capital to a firm comes mainly from its employees being remembered (uploaded) by others rather than the other way around. Hence, we focus on a firm’s ESC in-degree for this and the remaining tests. 15

Y = 𝛽 +𝛽 ×Act Exposure +𝛽 ×Act Exposure ×Post +𝛾′X +𝛼 +𝜀 , ( 2 ) i,t 0 1 i 2 i t i,t-1 j,t i,t where Y measures firm performance and Act Exposure, the treatment intensity, is calculated as the i,t i ratio ESC in-degree Act /ESC in-degree , where ESC in-degree Act is ESC in-degree in 2015 i,2015 i,2015 i,2015 that is due to connections to employees in industries subject to the Act.25 We measure the treatment intensity in 2015, before the enactment, to isolate it from the dynamic response of a firm’s employee social capital to the Act. The summary statistics of Act Exposure are shown in Panel B of Table IA.4 in the Internet Appendix. Post is a dummy variable for the years during and after the enactment (2016–2018). X is the same set of lagged control variables as in Table 2; 𝛼 is a full set of industryj,t by-year fixed effects. We are interested in 𝛽 , the coefficient of the interaction term, Act 2 Exposure×Post. If employee social capital indeed has a causal effect on firm performance, we expect firms with ESC more exposed to the Act to derive less value from their ESC after the Act than firms that are less exposed, i.e., we expect 𝛽 to be negative. 2 [Table 3 about here] Table 3 summarizes the results of estimating equation (2). The regression in column (1) excludes observations during the enactment year because the Act only became effective in the latter half of 2016. Consistent with our prediction, the estimate of 𝛽 is negative and significant at the 1% 2 level. Based on the positive and significant 𝛽 estimate, employee connections to the media and the 1 public sector contribute positively to a firm’s Tobin’s q before the Act; however, the negative 𝛽 2 estimate shows that the positive impact declines substantially after the Act. For instance, a one standard deviation increase in Act Exposure (0.038) leads to an increase in Tobin’s q by 17.5% (=0.038×6.578/1.432) relative to the sample mean before the Act, but only by 4.4% after. Our estimate is little changed when we control for Act Exposure measured by ESC out-degree in the regressions (see Panel A of Table IA.6 in the Internet Appendix); this robustness result reinforces our earlier finding on the value of “who knows you” to firms as opposed to “who you know.” Panel A of Table IA.6 also shows that the results are robust to alternative thresholds for the minimum number of employees or a minimum percentage of firm employees who appear in the network. Finally, we include observations in 2016 in column (2) of Table 3 and find little change in the magnitude and significance of our 𝛽 estimate. 2 25 We focus on Tobin’s q as our measure of firm performance in testing for causality since, as shown in Table IA.4 in the Internet Appendix, connections to industries affected by the Act have a significant and positive impact on firm performance, with the effect concentrated in Tobin’s q. 16

To test for the presence of pre-trends, in columns (3)–(4) we estimate an augmented version of equation (2) where we interact Act Exposure with an indicator variable for each year.26 The finding is visualized in Figure IA.4 in the Internet Appendix. Consistent with Act Exposure capturing an adverse shock to employee social capital, the decline in firm performance does not occur prior to the enactment. Starting from the enactment in 2016, the estimate becomes negative and remains negative and significant at the 1% level. Our results suggest no preexisting trend in firm performance before the enactment, reinforcing that the Act negatively affects firm performance by reducing employee social capital. To further assess the reliability of our identification strategy, we perform a placebo test. We randomly assign a Pseudo Exposure to each firm while maintaining the true distribution of Act Exposure and re-estimate column (1) in Table 3. By randomizing Act Exposure while holding all other variables fixed, we break the true link between employee social capital and firm performance, thereby imposing the null hypothesis on the data. We repeat this procedure 1,000 times and obtain the empirical distribution of the coefficient estimate on the interaction term. The true coefficient estimate (−4.930) falls well below the 1% threshold of this distribution, as reported in Table IA.5 in the Internet Appendix. This placebo test gives confidence that the negative estimate of 𝛽 is not a 2 statistical artifact. The exposure of a firm’s employee social capital to the Act is not randomly assigned. Firms with ESC more exposed to the Act tend to be larger in asset size and number of employees. It is likely they also had more frequent business interactions with the media and the public sector by 2015. We perform two robustness checks to address the issue of covariate balance. First, we use propensity score matching to generate a group of control firms similar to the treated firms and conduct the tests using this matched sample. We use a probit model to estimate the probability of being a treated firm (those with above-median Act Exposure in 2015). Then we match each treated firm to a control firm with replacement, using nearest neighbor matching with a maximum difference of 0.01. Panel A of Table 4 shows that the treated and control firms in the matched sample display indistinguishable differences. In Panel B, we estimate the same specifications as in Table 3 on the matched sample and find consistent results. Second, we use the full sample and interact firm-level 26 In column (3), we set 2015 as the baseline year and omit the 2015 interaction term (the outcome variable in year 2014 is dropped in our baseline analysis because we lag all control variables by one year). To highlight the insignificance of the pre-treatment interaction terms, in column (4) we extend our pre-treatment sample to include year 2014 and set 2014 as the baseline year, omitting the 2014 interaction term. 17

control variables with the Post dummy to control for any observable differences in characteristics related to the treatment that could lead to differences in performance around the enactment. We find the results continue to hold, as reported in Panel B of Table IA.6. [Table 4 about here] To alleviate concerns that adverse sectoral shocks to the industries directly affected by the Act (media and public sector) could spill over to treated firms through economic linkages rather than employee connections, we conduct subsample analyses in Panel C. Firms in the media and the public sector may be highly connected among themselves, thereby mechanically having a high Act Exposure; therefore, we drop firms that belong to the industries directly affected by the Act (26 firms) in column (1) and also drop firms that more broadly belong to the media and the publishing activities sectors (KSIC 58, 59) in column (2). In column (3), we further drop firms in the supplier and customer industries of the media and the public sector.27 To examine whether our results are driven by firms that have no employee connections to the affected industries, in column (4), we focus on the subsample with positive exposure of employee social capital to the Act. Across all these subsamples, the coefficient estimates on the interaction term remain negative and significant at the 1% level. These tests help rule out alternative explanations due to potential differences between the treated and control firms and economic spillovers. 4.3. Stock market reaction to the court ruling on the Kim Young-ran Act To reinforce a causal interpretation of our findings, we conduct an event study analysis of the stock market response to the Act. We focus on event days surrounding the date the court ruled that the Act was constitutional. After bipartisan approval, the Act faced a lengthy petition challenging its scope and constitutionality. The Korean Bar Association and the Korean Journalists Association argued that applying the law to journalists and private school teachers (and their spouses) infringed on freedom of the press and on the rights of private schools. However, the petition was eventually rejected at 2pm on July 28, 2016 when seven out of the nine Constitutional Court justices ruled that the Act was constitutional. We examine stock price reactions around the court ruling for firms differentially exposed to the Act. A negative market reaction for firms with ESC 27 We use the same method described in Internet Appendix III to identify the customer industries and a similar method to identify the supplier industries. Examples of supplier industries include manufacturers of newsprint, printing and reproduction of recorded media, infrastructure suppliers, and restaurants; examples of customer industries include the wholesale and retail sectors and sellers of motor vehicles and parts (with significant advertising expenses). 18

more exposed to the Act would buttress support for the causal effect of employee social capital on firm performance. [Table 5 about here] We divide firms into above-median and below-median subgroups based on Act Exposure (ESC in-degree Act /ESC in-degree ). We calculate average cumulative abnormal returns for i,2015 i,2015 each subgroup, both CAPM-adjusted and size-adjusted, for various windows around the court ruling date. As reported in Table 5, we find evidence of a negative market reaction to firms with ESC more exposed to the Act. For example, the average cumulative abnormal return over the [−3, 3] event window is −0.61% (p-value = 0.017) for firms with ESC more exposed to the Act and 0.41% for firms with ESC that is less exposed. The difference between the two groups is statistically significant with a p-value of 0.019.28 We also examine the cross-sectional pairwise correlation between Act Exposure and the cumulative abnormal returns and find that greater exposure to the Act is significantly associated with more negative stock price reactions. Meanwhile, we do not find significant market reactions when we construct Act Exposure using ESC out-degree. Taken together, the event study evidence supports the notion that employee social capital positively contributes to firm value. 4.4. Mechanisms: benefits of employee connections with the media and the public sector To shed light on the economic mechanisms through which employee social capital contributes to firm value, we proceed to identify benefits that a firm can extract from its employee connections to the sectors affected by the Act—the media and the public sector. We start by showing that the negative effect of the Act on the value of employee social capital demonstrated in Table 3 (where Act Exposure is measured using the sum of the connections to both affected sectors) is also observed separately for each of the affected sectors. Act ExposureMedia is the fraction of ESC in-degree in 2015 that is due to connections to media employees (ESC in-degree Media/ESC in-degree ); Act ExposurePublic is defined similarly. Panel 2015 2015 B of Table IA.4 presents summary statistics of these two variables. As shown in Panel A of Table 6, when we re-estimate equation (2) by setting the treatment intensity separately as 28 The observation that the return differentials are not significant for the [−1, 1] event window and are increasing with the length of the event windows suggests that firms’ social capital exposed to the Act might not be immediately known to the market as employee connections are latent. 19

Act ExposureMedia and Act ExposurePublic, we find results similar to what we find for the combined effect as captured by Act Exposure. Before the Act, employee connections to both the media and the public sector have a significant positive impact on firm Tobin’s q, and the impact declines for both sectors after the Act. [Table 6 about here] Given the positive value of employee social capital tied to each sector, we can now consider some specific benefits that firms can derive from their employee connections with these sectors. With respect to media connections, a large body of literature suggests that media coverage influences stock returns (Tetlock et al., 2008; Dougal et al., 2012; Gurun and Butler, 2012; Ahern and Sosyura, 2014). Gurun and Butler (2012) document that local media tend to display a “positive slant” toward local firms by using fewer negative words in news articles and that the positive slant strongly relates to firms’ equity value. Relatedly, Ahern and Sosyura (2014) find that firms actively manage media coverage to influence their stock prices. Like the positive slant when media covers local firms, media connections of a firm’s employees may lead to a positive slant in news coverage and a resulting positive effect on firm value. For instance, reporters who are well connected to a firm’s employees may have developed trust in those employees and therefore be more likely to report positive news about the firm. Media connections might also facilitate active media management by allowing firms to influence the timing and content of media coverage. We thus expect that all else equal, employee connections with the media foster more news coverage of the firm, and more news stories with a positive tone; moreover, if employee social capital is driving this relationship, we expect a decline in the positive impact of media connections after the Act. To test these predictions, we examine the effect of a firm’s employee social capital—derived from connections with the media—on media coverage of the firm before and after the Act; the results are reported in columns (1)–(2) in Panel B of Table 6. The dependent variable in column (1) is the log of the weighted number of news articles from RavenPack News Analytics covering a firm in a given year. To measure positive slant by media, we calculate the fraction of news articles covering a firm each year that are associated with a positive sentiment according to RavenPack’s sentiment series and use this measure as the dependent variable in column (2).29 29 We report results excluding observations in the enactment year of 2016 because the outcome variables reflect the cumulative outcomes throughout the year. Results are robust if we also include observations from 2016. 20

Consistent with the notion that media connections promote news coverage, we obtain a significant and positive coefficient on Act ExposureMedia. Moreover, consistent with the idea that reduced social interactions due to the Act undermine the benefits of media connections, the estimated coefficient for Act ExposureMedia×Post is significantly negative for both the number and the tone of news articles. For example, a one standard deviation increase in Act ExposureMedia increases the number of news articles by 13.0% (=0.029×4.495) and positive media coverage by 49.1% before the Act, but only increases news articles by 4.3% and positive media coverage by 14.8% after the Act. Taken together, these findings suggest that media connections lead to more favorable media coverage, enhancing firm performance. After the Act, the positive impact of media coverage declines substantially, consistent with the diminished contribution to Tobin’s q in Panel A as well as the event study results showing negative valuation effects. We now turn to investigating the benefits of employee social capital due to connections with the public sector. A nontrivial responsibility of public sector employees is public procurement, which accounts for 10–20% of GDP in developed countries (OECD, 2015). Schoenherr (2019) documents that Korean public officers who control the distribution of government contracts allocate significantly more procurement contracts to firms with connected CEOs. Similarly, we expect that firms with employees (including non-executive employees) who are better connected with the public sector may obtain more government contracts, thereby resulting in superior performance. To assess this prediction, we examine the effect of a firm’s employee connections with the public sector on public procurement contracting outcomes using data from the Korea online e- Procurement Service. Consistent with our prediction, findings in columns (3)–(5) in Panel B of Table 6 show that firms highly connected to public sector employees obtain more public procurement contracts, in terms of the number of newly signed contracts, their value in Korean won, and their value scaled by firm assets, respectively. The estimated effect is reduced by about half after the Act. For example, column (3) shows that a one standard deviation increase in Act ExposurePublic leads to a 6.8% increase in the number of newly signed contracts before the Act and only 3.4% after. We conduct a falsification test to ensure our results are not driven by unobserved firm characteristics that are correlated with exposure to the Act. Specifically, we swap the Act exposure variables and instead regress the media coverage outcomes on Act ExposurePublic and regress the procurement contracting outcomes on Act ExposureMedia. If our findings in Panel B indeed reflect a causal effect of media connections in promoting media coverage and of public sector connections in obtaining procurement contracts, we should not expect significant effects in this falsification test. 21

The results reported in Panel C of Table 6 confirm this prediction, thus supporting a causal interpretation of the mechanism in Panel B. In sum, Tables 3–6 provide causal evidence that a firm’s employee social capital tied to the media or the public sector contributes to its performance by promoting favorable media coverage of the firm or by enhancing its ability to obtain public procurement contracts. 4.5. Discussion Given the policy intention of the Act, a natural question is to what extent our results are due to the Act’s success in reducing the ability of firms to obtain resources (favorable news coverage and procurement contracts) by bribing their connections in the media and the public sector. Several points are worth discussing in this context. First, the social capital literature (e.g., Bourdieu, 1986) highlights favor exchanges and reciprocity as important channels through which social relations increase the ability of individuals to advance their economic interests. Despite the negative connotation (and potential negative welfare effects), the literature recognizes bribery for resources as an example of a favor exchange that is more easily achieved for individuals with greater social capital.30 For example, it is difficult to offer bribes to people who do not know or trust you. Hence, to the extent that results in Table 6 are driven by employees’ connections with journalists or public officials facilitating bribery for resources, this bribery channel is still consistent with the notion that employee social capital improves firm outcomes (although not necessarily social welfare). Second, our evidence suggests that a reduction in bribery is unlikely the only channel driving our results in Table 6. While bribery is not directly observable, a firm’s entertainment expenses are shown to include a significant bribe component (Cai et al., 2011; Kang et al., 2020). Using a firm’s entertainment expenses scaled by total assets as a proxy for bribery activities, we find that its correlation with Act Exposure is only 0.043, suggesting that firms with employees well connected with the media and the public sector do not seem to coincide with those that actively pay bribes. In addition, when we decompose Panel B of Table 6 into executives and non-executive employees in Panel D, we find that the connections by non-executive employees are also significantly valuable in bringing benefits to their firm. This result once again highlights our novel addition to existing evidence on the value of executives’ media connections and political connections. More importantly, 30 This “dark-side” view of social connections is consistent with the evidence of crony lending documented in Haselmann et al. (2018) and the distortive allocation of government resources to politically connected firms (Schoenherr, 2019). While these rent-seeking activities are not allocatively efficient, they do benefit the connected borrowers and firms. 22

to the extent that bribing for resources for their firm is mostly carried out by executives, bribery does not appear as the only driver of our findings. 5. Conclusion This paper provides novel evidence that a firm’s social capital derived from its employees’ professional connections is a valuable production factor contributing to firm performance. We use a comprehensive dataset from a professional networking app with broad coverage of individuallevel connections to measure firm-level employee social capital. Our analysis reveals that employee social capital is robustly and positively associated with firm performance. Our unique network data record the direction of connections, allowing us to separately account for those business contacts that remember the employee and those that the employee remembers. Our results show that the positive effect on firm performance manifests primarily when external stakeholders remember a firm’s employees. To establish a causal interpretation of our results, we exploit the enactment of the Kim Young-ran Act in 2016 which imparted a negative shock to networking with specific sectors. Our evidence suggests that firms with employee connections more exposed to the Act derive less value from their employee social capital after the Act than firms that are less exposed. The results support a causal role of employee social capital in boosting firm performance and creating firm value. This paper makes several contributions. First, our study uses a comprehensive measure of employee social capital and establishes its contribution to firm performance. We quantify employee social capital at the firm level by identifying interpersonal networks that cover employees at all job levels. Second, our employee social capital measures are directional. Our finding that being remembered by others is more productive than remembering others echoes a popular saying about professional networking: “It is not who you know—it is who knows you.” Third, our analysis of the connections with economically related industries provides novel insight into the economic mechanisms underlying the concomitant benefits of employee connections. One implication of our research is that social ties can be leveraged in business settings. Personal relationships and business contacts endow employees (and their firms) with resources, constituting an essential form of social capital that is convertible into firm value and performance. 23

References Acemoglu, Daron, Simon Johnson, Amir Kermani, James Kwak, and Todd Mitton, 2016, The value of connections in turbulent times: Evidence from the United States, Journal of Financial Economics 121, 368‒391. Ahern, Kenneth R., 2009, Sample selection and event study estimation, Journal of Empirical Finance 16, 466‒482. Ahern, Kenneth R., and Denis Sosyura, 2014, Who writes the news? Corporate press releases during merger negotiations, Journal of Finance 69, 241‒291. Anderson, Ronald C., and David M. Reeb, 2003, Founding-family ownership and firm performance: Evidence from the S&P 500, Journal of Finance 58, 1301‒1328. Bourdieu, Pierre, 1986, The forms of capital, In: John G. Richardson (Eds.), Handbook of Theory and Research for the Sociology of Education (Greenwood Press, New York, NY), 241‒258. Cai, Hongbin, Hanming Fang, and Lixin Colin Xu, 2011, Eat, drink, firms, government: An investigation of corruption from the entertainment and travel costs of Chinese firms, Journal of Law and Economics 54, 55‒78. Cai, Jing, and Adam Szeidl, 2017, Interfirm relationships and business performance, Quarterly Journal of Economics 133, 1229‒1282. Cai, Ye, and Merih Sevilir, 2012, Board connections and M&A transactions, Journal of Financial Economics 103, 327‒349. Choi, Seung Ginny, and Virgil Henry Storr, 2019, A culture of rent seeking, Public Choice 181, 101‒126. Coleman, James S., 1988, Social capital in the creation of human capital, American Journal of Sociology 94 (Supplement), S95‒S120. Dass, Nishant, Omesh Kini, Vikram Nanda, Bunyamin Onal, and Jun Wang, 2014, Board expertise: Do directors from related industries help bridge the information gap?, Review of Financial Studies 27, 1533‒1592. Dougal, Casey, Joseph Engelberg, Diego García, and Christopher A. Parsons, 2012, Journalists and the stock market, Review of Financial Studies 25, 639‒679. Engelberg, Joseph, Pengjie Gao, and Christopher A. Parsons, 2012, Friends with money, Journal of Financial Economics 103, 169‒188. Fracassi, Cesare, and Geoffrey Tate, 2012, External networking and internal firm governance, Journal of Finance 67, 153‒194. Glaeser, Edward L., David Laibson, and Bruce Sacerdote, 2002, An economic approach to social capital, Economic Journal 112, 437‒458. Gompers, Paul A., Vladimir Mukharlyamov, and Yuhai Xuan, 2016, The cost of friendship, Journal of Financial Economics 119, 626‒644. Gortmaker, Jeff, Jessica Jeffers, and Michael Junho Lee, 2020, Labor reactions to credit deterioration: Evidence from LinkedIn activity, Working paper, University of Chicago and Federal Reserve Bank of New York. Gorton, Gary B., Jillian Grennan, and Alexander K. Zentefis, 2022, Corporate culture, Annual Review of Financial Economics 14, 1‒27. Graham, John R., Jillian Grennan, Campbell R. Harvey, and Shivaram Rajgopal, 2018, Corporate culture: The interview evidence, Working paper, Duke University and Columbia Business School. Graham, John R., Jillian Grennan, Campbell R. Harvey, and Shivaram Rajgopal, 2022, Corporate culture: Evidence from the field, Journal of Financial Economics (forthcoming). Granovetter, Mark S., 1973, The strength of weak ties, American Journal of Sociology 78, 1360‒1380. Granovetter, Mark S., 1995, Getting a job: A study of contacts and careers (University of Chicago Press, Chicago, IL). Grennan, Jillian, 2022, Communicating culture consistently: Evidence from banks, Working paper, Duke University. Grennan, Jillian, and Kai Li, 2022, Corporate culture: A review and directions for future research, Working paper, Duke University and University of British Columbia. 24

Guiso, Luigi, Paola Sapienza, and Luigi Zingales, 2004, The role of social capital in financial development, American Economic Review 94, 526‒556. Guiso, Luigi, Paola Sapienza, and Luigi Zingales, 2008, Trusting the stock market, Journal of Finance 63, 2557‒2600. Guiso, Luigi, Paola Sapienza, and Luigi Zingales, 2015, The value of corporate culture, Journal of Financial Economics 117, 60‒76. Gurun, Umit G., and Alexander W. Butler, 2012, Don’t believe the hype: Local media slant, local advertising, and firm value, Journal of Finance 67, 561‒597. Hacamo, Isaac, and Kristoph Kleiner, 2021, Competing for talent: Firms, managers, and social networks, Review of Financial Studies (forthcoming). Hasan, Iftekhar, Chun Keung Hoi, Qiang Wu, and Hao Zhang, 2017, Social capital and debt contracting: Evidence from bank loans and public bonds, Journal of Financial and Quantitative Analysis 52, 1017‒1047. Haselmann, Rainer, David Schoenherr, and Vikrant Vig, 2018, Rent seeking in elite networks, Journal of Political Economy 126, 1638‒1690. Hirshleifer, David A., Lin Peng, and Qiguang Wang, 2021, Social networks and market reactions to earnings news, Working paper, University of California-Irvine, City University of New York, and Hong Kong Baptist University. Hochberg, Yael V., Alexander Ljungqvist, and Yang Lu, 2007, Whom you know matters: Venture capital networks and investment performance, Journal of Finance 62, 251‒301. Hoi, Chun Keung(Stan), Qiang Wu, and Hao Zhang, 2019, Does social capital mitigate agency problems? Evidence from chief executive officer (CEO) compensation, Journal of Financial Economics 133, 498‒519. Houston, Joel F., Jongsub Lee, and Felix Suntheim, 2018, Social networks in the global banking sector, Journal of Accounting and Economics 65, 237‒269. Hwang, Byoung-Hyoun, and Seoyoung Kim, 2009, It pays to have friends, Journal of Financial Economics 93, 138‒158. Ishii, Joy, and Yuhai Xuan, 2014, Acquirer-target social ties and merger outcomes, Journal of Financial Economics 112, 344‒363. Jackson, Matthew O., 2008, Social and economic networks (Princeton University Press, Princeton, NJ). Jeffers, Jessica, and Michael Junho Lee, 2019, Corporate culture as an implicit contract, Working paper, University of Chicago and Federal Reserve Bank of New York. Kang, Sungwon, Daehwan Kim, and Gyeonhyeong Kim, 2020, Corporate entertainment expenses and corruption in public procurement, Working paper, Korea Environment Institute, Konkuk University, and Seoul National University. Karolyi, Stephen Adam, 2018, Personal lending relationships, Journal of Finance 73, 5‒49. Khanna, Vikramaditya, E. Han Kim, and Yao Lu, 2015, CEO connectedness and corporate fraud, Journal of Finance 70, 1203‒1252. Knack, Stephen, and Philip Keefer, 1997, Does social capital have an economic payoff? A cross-country investigation, Quarterly Journal of Economics 112, 1251‒1288. Kuchler, Theresa, Yan Li, Lin Peng, Johannes Stroebel, and Dexin Zhou, 2022, Social proximity to capital: Implications for investors and firms, Review of Financial Studies 35, 2743‒2789. La Porta, Rafael, Josef Lakonishok, Andrei Shleifer, and Robert W. Vishny, 1997, Good news for value stocks: Further evidence on market efficiency, Journal of Finance 52, 859‒874. La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert W. Vishny, 1997, Trust in large organizations, American Economic Review 87, 333‒338. Larcker, David F., Eric C. So, and Charles C. Y. Wang, 2013, Boardroom centrality and firm performance, Journal of Accounting and Economics 55, 225‒250. Lin, Nan, 2002, Social capital: A theory of social structure and action (Cambridge University Press, Cambridge, UK). Lin, Nan, Walter Ensel, and John Vaughn, 1981, Social resources and strength of ties: Structural factors in occupational status attainment, American Sociological Review 46, 393‒405. 25

Lins, Karl V., Henri Servaes, and Ane Tamayo, 2017, Social capital, trust, and firm performance: The value of corporate social responsibility during the financial crisis, Journal of Finance 72, 1785‒1824. Newman, Mark E. J., 2010, Networks: An introduction (Oxford University Press, Oxford, UK). Organisation for Economic Co-operation and Development (OECD), 2015, Size of public procurement, In: Government at a Glance 2015 (OECD Publishing, Paris, France), 136‒137. Popadak, Jillian, 2016, Balancing governance and culture to create sustainable firm value, Brookings Institution Governance Studies 27, 1‒13. Putnam, Robert D., 2000, Bowling alone: The collapse and revival of American community (Simon & Schuster, New York, NY). Radner, Roy, 1992, Hierarchy: The economics of managing, Journal of Economic Literature 30, 1382‒1415. Rossi, Alberto G., David Blake, Allan Timmermann, Ian Tonks, and Russ Wermers, 2018, Network centrality and delegated investment performance, Journal of Financial Economics 128, 183‒206. Schoenherr, David, 2019, Political connections and allocative distortions, Journal of Finance 74, 543‒586. Servaes, Henri, and Ane Tamayo, 2017, The role of social capital in corporations: A review, Oxford Review of Economic Policy 33, 201‒220. Shue, Kelly, 2013, Executive networks and firm policies: Evidence from the random assignment of MBA peers, Review of Financial Studies 26, 1401‒1442. Syverson, Chad, 2011, What determines productivity?, Journal of Economic Literature 49, 326‒365. Tetlock, Paul C., Maytal Saar‐Tsechansky, and Sofus Macskassy, 2008, More than words: Quantifying language to measure firms’ fundamentals, Journal of Finance 63, 1437‒1467. 26

Table 1. Summary statistics: employee-level connections and firm-year sample This table provides summary statistics for our data. Panel A presents summary statistics of the employee-level connections as of December 2018, based on the 1,016,023 public firm employees of our sample. In-degree, which measures “who knows you,” is the number of employees of other firms who have uploaded the corresponding employee as a business contact as of December 2018. Out-degree, which measures “who you know,” is the number of business contacts of other firms uploaded by the focal app-user employee as of December 2018; given the nature of our data, Out-degree is only available for the 119,423 public firm employees who are app-users. Total degree is the sum of Indegree and Out-degree. Panel B presents summary statistics of the main variables for our firm-year sample. ESC indegree is the average In-degree across employees of firm i who are in the network in year t. ESC out-degree is the average Out-degree across app-user employees of firm i in year t. For reference, we also tabulate ESC out-degree computed as the average Out-degree across employees of firm i who are in the network in year t. ESC total degree is the average Total degree across employees of firm i who are in the network in year t. The sample period is 2014–2018. The definitions of all variables are provided in Internet Appendix II. Panel A. Employee-level connections as of December 2018 N Mean Median SD P25 P75 [App-users] In-degree 119,423 26.329 11 50.160 4 27 Out-degree 119,423 56.916 17 116.831 5 56 Total degree 119,423 83.244 30 161.819 11 84 [Non-app-users] In-degree = Total degree 896,600 4.820 2 9.826 1 5 [All public firm employees in the network (app-users + non-app-users)] In-degree 1,016,023 7.348 2 20.710 1 6 Total degree 1,016,023 14.038 2 61.652 1 7 Panel B. Firm-level employee social capital (ESC) measures and other main variables N Mean Median SD P25 P75 ESC in-degree 5,340 3.676 3.139 2.392 1.976 4.693 ESC out-degree 4,994 30.953 24.167 26.787 12.909 40.304 ESC out-degree (app-users + non-app-users) 5,340 3.210 2.031 4.190 0.740 4.057 ESC total degree 5,340 6.836 5.319 5.844 3.000 8.548 Tobin’s q 5,340 1.456 1.106 1.099 0.890 1.575 ROA 5,340 0.043 0.042 0.087 0.009 0.082 Sales Growth 5,340 0.041 0.037 0.324 -0.066 0.141 R&D 5,340 0.024 0.003 0.067 0.000 0.022 Book Leverage 5,340 0.101 0.062 0.115 0.001 0.165 ln(1+Assets) (in million Korean won) 5,340 12.248 12.013 1.343 11.341 12.950 Volatility 5,340 0.130 0.115 0.068 0.085 0.156 Firm Age 5,340 28.666 25 16.163 16 40 ln(1+Emp) 5,340 5.478 5.429 1.154 4.771 6.071 27

Table 2. Employee social capital and firm performance This table reports OLS regression estimates on the relation between employee social capital and future firm performance. We estimate the following specification: Y i,t = 𝛽 0 +𝛽 1 ×ln(1+ESC i,t-1 )+𝛾′ X i,t-1 +𝛼 j,t +𝜀 i,t , where Y is one of the performance measures (Tobin’s q, ROA, or Sales Growth), ESC is the one-year lagged firm-level employee social capital of firm 𝑖 in i,t i,t-1 year t-1; X is a set of lagged firm-specific control variables commonly included in the literature (Anderson and Reeb, 2003); 𝛼 is a full set of industry-by-year i,t-1 j,t fixed effects. Panel A reports the baseline estimates. Columns (1)–(3) report results when measuring employee social capital by ESC total degree, without accounting for the direction of connections; columns (4)–(9) report results when we measure employee social capital by ESC in-degree and ESC out-degree to differentiate the direction of connections. We perform one-tailed tests comparing the coefficient estimates of ESC in-degree and ESC out-degree and find the pvalues less than 0.01 for all three performance measures. Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. The sample period is 2015–2018 for output variables. The definitions of all variables are provided in Internet Appendix II. Panel A. Baseline estimates: ESC total degree, ESC in-degree, and ESC out-degree ESC total degree ESC in-degree (“who knows you”) ESC out-degree (“who you know”) Sales Sales Sales Dep. var. Tobin’s q ROA Tobin’s q ROA Tobin’s q ROA Growth Growth Growth (1) (2) (3) (4) (5) (6) (7) (8) (9) ln(1+ESC) 0.084 0.008** 0.038*** 0.330*** 0.021*** 0.098*** 0.042 0.004* 0.004 (0.053) (0.004) (0.012) (0.090) (0.007) (0.024) (0.030) (0.002) (0.007) R&D 4.634*** -0.182*** 0.420*** 4.536*** -0.187*** 0.397*** 4.565*** -0.176*** 0.398*** (0.576) (0.034) (0.125) (0.577) (0.034) (0.124) (0.573) (0.034) (0.125) Book Leverage 0.172 -0.138*** 0.076 0.160 -0.139*** 0.073 0.059 -0.134*** 0.091 (0.179) (0.016) (0.054) (0.178) (0.016) (0.053) (0.163) (0.016) (0.057) ln(1+Assets) -0.134*** 0.010*** -0.009 -0.142*** 0.009*** -0.011 -0.126*** 0.010*** -0.010 (0.022) (0.002) (0.008) (0.022) (0.002) (0.009) (0.022) (0.002) (0.009) Volatility 3.498*** -0.104*** 0.050 3.504*** -0.103*** 0.054 3.618*** -0.106*** 0.023 (0.388) (0.026) (0.080) (0.388) (0.026) (0.079) (0.409) (0.027) (0.083) Firm Age -0.005*** -0.000*** 0.000 -0.005*** -0.000*** 0.000 -0.005*** -0.000*** 0.000 (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) ln(1+Emp) 0.064*** 0.009*** -0.007 0.079*** 0.010*** -0.003 0.075*** 0.008*** -0.008 (0.023) (0.002) (0.006) (0.024) (0.002) (0.006) (0.024) (0.002) (0.006) Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Observations 5,340 5,340 5,340 5,340 5,340 5,340 4,994 4,994 4,994 Adjusted R2 0.248 0.148 0.035 0.252 0.150 0.038 0.252 0.142 0.035 28

Table 2. Employee social capital and firm performance (continued) In Panel B, firm-level employee social capital takes the lagged value of ESC in-degree averaged across executives (chairman, vice chairman, president, deputy president, executive vice president, and senior vice president) in columns (1)–(3) and averaged across non-executive employees (all other employees) in columns (4)–(6). We include the same set of lagged control variables as in Panel A. Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. The sample period is 2015–2018 for output variables. The definitions of all variables are provided in Internet Appendix II. Panel B. Executives versus non-executive employees Sales Sales Dep. var. Tobin’s q ROA Tobin’s q ROA Growth Growth (1) (2) (3) (4) (5) (6) ESC in-degree of ESC in-degree of executives non-executive employees ln(1+ ESC in-degree) 0.190*** 0.013*** 0.050*** 0.207** 0.032*** 0.090*** (0.056) (0.004) (0.013) (0.100) (0.008) (0.025) Controls Yes Yes Yes Yes Yes Yes Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Observations 5,321 5,321 5,321 5,340 5,340 5,340 Adjusted R2 0.251 0.151 0.036 0.249 0.154 0.037 29

Table 3. Causal evidence: the 2016 Kim Young-ran Act as an exogenous shock to employee social capital This table provides evidence on the causal effect of employee social capital on firm performance. We estimate the following difference-in-differences model surrounding the enactment of the Kim Young-ran Act: Y = 𝛽 +𝛽 ×Act Exposure +𝛽 ×Act Exposure ×Post +𝛾′X +𝛼 +𝜀 , i,t 0 1 i 2 i t i,t-1 j,t i,t where Y is Tobin’s q, Act Exposure =ESC in-degree Act /ESC in-degree , and ESC in-degree Act is ESC ini,t i i,2015 i,2015 i,2015 degree in 2015 that is due to connections to employees in industries subject to the Act. Post is an indicator variable t that equals one during and after the enactment year (2016–2018) and zero otherwise. d is an indicator variable for t year t. X is the same set of lagged controls as in Table 2; 𝛼 is a full set of industry-by-year fixed effects. Column i,t-1 j,t (1) reports results excluding the enactment year (2016); columns (2)–(4) report results including the year 2016. The sample period is 2015–2018 for output variables in columns (1)–(3) and is 2014–2018 for output variables in column (4). Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. The definitions of all variables are provided in Internet Appendix II. Dep. var. Tobin’s q (1) (2) (3) (4) Act Exposure 6.578*** 6.640*** 6.642*** 5.420*** (1.273) (1.272) (1.272) (1.050) Act Exposure × Post -4.930*** -4.726*** (1.132) (1.052) Act Exposure × d 1.169 2015 (0.793) Act Exposure × d -4.155*** -2.973*** 2016 (0.932) (0.849) Act Exposure × d -4.730*** -3.540*** 2017 (1.162) (1.006) Act Exposure × d -5.162*** -3.980*** 2018 (1.169) (0.983) R&D 5.431*** 5.066*** 5.065*** 4.969*** (0.689) (0.677) (0.678) (0.653) Book Leverage 0.183 0.233 0.232 0.227 (0.185) (0.182) (0.182) (0.177) ln(1+Assets) -0.139*** -0.146*** -0.146*** -0.139*** (0.025) (0.023) (0.023) (0.022) Volatility 3.403*** 3.400*** 3.396*** 3.238*** (0.449) (0.395) (0.395) (0.363) Firm Age -0.005*** -0.005*** -0.005*** -0.005*** (0.002) (0.001) (0.001) (0.001) ln(1+Emp) 0.076*** 0.067*** 0.067*** 0.068*** (0.024) (0.023) (0.023) (0.023) Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Including year 2016 No Yes Yes Yes Observations 3,778 5,101 5,101 6,048 Adjusted R2 0.242 0.245 0.245 0.243 30

Table 4. Causal evidence: robustness analyses Panel A uses a propensity score matched sample to estimate the specifications in Table 3. We use a probit regression to estimate the probability of being a treated firm (those with above-median Act Exposure in 2015) using the sample of 2015 with a set of industry fixed effects and the same set of control variables in 2015 as in Table 3. Each treated firm is matched to a control firm using nearest neighbor with replacement within each two-digit SIC industry, where the maximum absolute difference in propensity scores between the matched observations is 0.01. We first tabulate the means of the matched variables for the treated group (those with above-median Act Exposure) and the control group (those with below-median exposure) in the year 2015. We also report the mean differences between the two groups and their corresponding p-values based on heteroskedasticity-consistent standard errors. Panel B present the results estimating the specifications in Table 3 using the matched sample, and the same set of lagged control variables and fixed effects. In Panel C, we re-estimate the specification of column (1) in Table 3 using subsamples. Column (1) drops firms that belong to the industries directly affected by the Act (26 unique firms identified according to the industry codes in Internet Appendix II); column (2) additionally drops firms that belong more broadly to the media and the publishing activities sectors (KSIC 58, 59); column (3) further drops firms that belong to the supplier and customer industries of the media and the public sector using detailed Make-and-Use tables; column (4) focuses on a subsample with positive exposure of employee social capital to the Act. Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. The definitions of all variables are provided in Internet Appendix II. Panel A. Propensity score matched sample Above median Below median Above − Below p-value (Obs. = 635) (Obs. = 635) R&D 0.021 0.023 -0.002 0.587 Book Leverage 0.107 0.109 -0.002 0.679 ln(1+Assets) 12.347 12.304 0.043 0.574 Volatility 0.142 0.148 -0.006 0.189 Firm Age 29.191 30.710 -1.519 0.117 ln(1+Emp) 5.572 5.565 0.007 0.917 Panel B. Robustness tests based on the matched sample Dep. var. Tobin’s q (1) (2) (3) (4) Act Exposure 6.507*** 6.531*** 6.531*** 5.521*** (1.356) (1.353) (1.353) (1.177) Act Exposure × Post -4.651*** -4.409*** (1.232) (1.140) Act Exposure × d 0.964 2015 (0.878) Act Exposure × d -3.957*** -2.997*** 2016 (1.050) (1.002) Act Exposure × d -4.064*** -3.102*** 2017 (1.218) (1.099) Act Exposure × d -5.237*** -4.272*** 2018 (1.306) (1.150) Controls Yes Yes Yes Yes Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Including year 2016 No Yes Yes Yes Observations 3,541 4,811 4,811 5,721 Adjusted R2 0.266 0.265 0.265 0.264 31

Panel C. Subsamples Dep. var. Tobin’s q (1) (2) (3) (4) Act Exposure 8.010*** 8.350*** 8.190*** 6.362*** (1.419) (1.535) (2.232) (1.363) Act Exposure × Post -5.884*** -6.211*** -6.376*** -4.760*** (1.304) (1.407) (2.046) (1.196) Controls Yes Yes Yes Yes Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Including year 2016 No No No No Observations 3,708 3,464 2,686 3,344 Adjusted R2 0.247 0.251 0.222 0.234 32

Table 5. Stock market reaction to the court ruling on the Act This table reports the stock market reaction around July 28, 2016, when the Constitutional Court rejected the petition and ruled that the Kim Young-ran Act is constitutional. In the upper panel, we report the cumulative CAPM-adjusted abnormal returns in event windows [-1, 1], [-3, 3], and [-5, 5], where day 0 is the date of the announcement. Daily abnormal stock returns are computed based on the market model using the Korean equal-weighted market return as the market proxy. The estimation window is days [-200, -60] prior to the event date. In the lower panel, we report the cumulative size-adjusted abnormal returns in the same event windows. Following La Porta et al. (1997) and Ahern (2009), for each event window, we form a size-decile benchmark portfolio equally weighted using all stocks in that size decile, where size is measured as market capitalization as of one day prior to the start date of the event window. The daily size-adjusted abnormal returns are the difference between raw returns and the corresponding size-decile benchmark portfolios. In both panels, we report the average cumulative abnormal returns for firms with below-median exposure in column (1) and above-median exposure in column (2), where Act Exposure=ESC in-degree Act / 2015 ESC in-degree . Column (3) reports the mean difference between the above-median and the below-median 2015 subgroup; column (4) reports the cross-sectional pairwise correlation coefficient between Act Exposure and the cumulative abnormal returns. The p-values in square brackets are based on one-tailed tests for positive returns in column (1), for negative returns in columns (2)–(3), and for negative correlations in column (4), with the standard errors clustered at the industry (two-digit SIC) level. We exclude penny stocks with stock prices less than 1,000 Korean won (about 0.9 USD) as of June 28, 2016, one month prior to the court ruling. Act Exposure = ESC in-degree Act /ESC in-degree 2015 2015 Diff Correlation Below median Above median Above − Below coefficient (1) (2) (3) (4) [Cumulative CAPM-adjusted abnormal returns] [-1, 1] 0.07% -0.27% -0.34% -0.009 [0.325] [0.080] [0.083] [0.363] [-3, 3] 0.41% -0.61% -1.02% -0.076 [0.173] [0.017] [0.019] [0.020] [-5, 5] 0.62% -1.04% -1.66% -0.086 [0.131] [0.007] [0.008] [0.014] Observations 751 751 [Cumulative size-adjusted abnormal returns] [-1, 1] 0.16% -0.11% -0.27% -0.004 [0.182] [0.207] [0.098] [0.440] [-3, 3] 0.52% -0.43% -0.95% -0.065 [0.119] [0.041] [0.014] [0.035] [-5, 5] 0.65% -0.69% -1.33% -0.071 [0.128] [0.009] [0.013] [0.034] Observations 788 782 33

Table 6. Mechanisms: benefits of employee connections with the media and the public sector In Panel A, we estimate changes in the value of connections with the media and the public sector around the Act using: Y =𝛽 +𝛽 ×Act ExposureMedia (Public)+𝛽 ×Act ExposureMedia (Public)×Post +𝛾′X +𝛼 +𝜀 , i,t 0 1 i 2 i t i,t-1 j,t i,t where Y is Tobin’s q, Act ExposureMedia is ESC in-degree Media/ESC in-degree for columns (1)–(2) and i,t i i,2015 i,2015 Act ExposurePublic is ESC in-degree Public/ESC in-degree for columns (3)–(4); ESC in-degree Media (Public) is ESC i i,2015 i,2015 i,2015 in-degree in 2015 due to connections to the media (public) sector. Post is an indicator variable for during and after t the enactment year (2016–2018). X is the same set of lagged controls as in Table 2; 𝛼 is a full set of industry-byi,t-1 j,t year fixed effects. Columns (1) and (3) report results excluding the enactment year (2016), whereas columns (2) and (4) report results including 2016. Panel B reports results on the benefits of connections with the media and the public sector. Act Exposure is Act ExposureMedia for columns (1)–(2) and Act ExposurePublic for columns (3)–(5). Dependent variables in columns (1)–(2) are Media Coverage, the weighted count of news articles from RavenPack News Analytics covering a firm in a given year (the weight is the relevance score of each article provided by RavenPack; we only include articles with relevance scores greater than or equal to 75%), and Positive Media Coverage Ratio, the fraction of news articles with a positive sentiment (according to RavenPack’s BMQ sentiment series) covering a firm in a given year. Dependent variables in columns (3)–(5) are the natural logarithm of one plus the number of newly signed procurement contracts, the amount of newly signed procurement contracts in Korean won, and the amount of newly signed procurement contracts in Korean won scaled by the firm’s total assets. Panel C reports a falsification test where we repeat the analyses in Panel B but regress the media coverage outcomes on Act ExposurePublic for columns (1)–(2) and regress the procurement contracting outcomes on Act ExposureMedia for columns (3)–(5). Panel D repeats the analyses in Panel B when we differentiate the connections of executives (chairman, vice chairman, president, deputy president, executive vice president, and senior vice president) and non-executive employees (all other employees). Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. The definitions of all variables are provided in Internet Appendix II. Panel A. The value of connections with the media and the public sector: before and after the Act Act ExposureMedia Act ExposurePublic Dep. var. Tobin’s q (1) (2) (3) (4) Act ExposureMedia (Public) 8.016*** 8.070*** 6.181** 6.303*** (1.591) (1.588) (2.414) (2.407) Act ExposureMedia (Public)× Post -5.655*** -5.431*** -4.782** -4.735** (1.398) (1.290) (1.981) (1.899) R&D 5.455*** 5.092*** 5.449*** 5.085*** (0.697) (0.685) (0.686) (0.674) Book Leverage 0.183 0.233 0.185 0.235 (0.187) (0.185) (0.187) (0.183) ln(1+Assets) -0.141*** -0.148*** -0.124*** -0.132*** (0.025) (0.023) (0.025) (0.023) Volatility 3.377*** 3.376*** 3.445*** 3.443*** (0.451) (0.397) (0.447) (0.393) Firm Age -0.005*** -0.005*** -0.005*** -0.005*** (0.002) (0.001) (0.002) (0.001) ln(1+Emp) 0.080*** 0.070*** 0.068*** 0.059** (0.025) (0.024) (0.025) (0.024) Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Including year 2016 No Yes No Yes Observations 3,778 5,101 3,778 5,101 Adjusted R2 0.242 0.244 0.234 0.237 34

Panel B. The value of connections with the media and the public sector: economic benefits Act ExposureMedia Act ExposurePublic ln(1+Tot Amt of ln(1+Media Positive Media ln(1+# of Proc. ln(1+Tot Amt of Dep. var. Proc. Contracts Coverage) Coverage Ratio Contracts) Proc. Contracts) / Assets) (1) (2) (3) (4) (5) Act ExposureMedia (Public) 4.495*** 0.437** 3.756*** 19.837*** 0.091*** (1.564) (0.180) (1.111) (5.295) (0.027) Act ExposureMedia (Public) -2.991** -0.305* -1.878** -9.700** -0.040* × Post (1.445) (0.172) (0.839) (4.443) (0.022) Tobin’s q 0.116*** 0.013*** -0.003 -0.015 -0.000* (0.017) (0.004) (0.008) (0.041) (0.000) Book Leverage 0.131 -0.003 0.094 0.442 -0.003 (0.158) (0.027) (0.125) (0.538) (0.002) ROA -0.931*** -0.107*** -0.191* -1.668*** -0.005** (0.195) (0.027) (0.105) (0.521) (0.002) R&D 0.611** 0.020 -0.367** -1.883** -0.013*** (0.245) (0.040) (0.159) (0.772) (0.005) ln(1+Sales) 0.267*** 0.019*** 0.030*** 0.229*** -0.000 (0.025) (0.003) (0.011) (0.055) (0.000) Volatility -0.204 -0.017 0.143 1.049* 0.005 (0.181) (0.032) (0.104) (0.596) (0.003) Firm Age 0.009*** 0.001** 0.001 0.001 0.000 (0.001) (0.000) (0.001) (0.004) (0.000) ln(1+Emp) 0.069*** 0.009*** 0.107*** 0.576*** 0.002*** (0.024) (0.003) (0.014) (0.066) (0.000) Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Including year 2016 No No No No No Observations 3,775 3,775 3,775 3,775 3,775 Adjusted R2 0.343 0.164 0.241 0.264 0.194 Panel C. The value of connections with the media and the public sector: falsification test Act ExposurePublic Act ExposureMedia ln(1+Tot Amt of ln(1+Media Positive Media ln(1+# of Proc. ln(1+Tot Amt of Dep. var. Proc. Contracts Coverage) Coverage Ratio Contracts) Proc. Contracts) / Assets) (1) (2) (3) (4) (5) Act ExposurePublic (Media) 3.357* 0.320 -0.428 -2.443 -0.022** (1.889) (0.239) (0.495) (2.617) (0.011) Act ExposurePublic (Media) -2.868 -0.263 0.390 3.245 0.011 × Post (1.748) (0.236) (0.403) (2.483) (0.011) Controls Yes Yes Yes Yes Yes Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Including year 2016 No No No No No Observations 3,775 3,775 3,775 3,775 3,775 Adjusted R2 0.339 0.162 0.231 0.255 0.186 35

Panel D. The value of connections with the media and the public sector: executives versus non-executive employees Act ExposureMedia Act ExposurePublic ln(1+Tot Amt of ln(1+Media Positive Media ln(1+# of Proc. ln(1+Tot Amt of Dep. var. Proc. Contracts Coverage) Coverage Ratio Contracts) Proc. Contracts) / Assets) (1) (2) (3) (4) (5) [Executives] Act ExposureMedia(Public) 2.667*** 0.269*** 1.491*** 7.535*** 0.023** (0.744) (0.094) (0.438) (2.284) (0.010) Act ExposureMedia(Public) -2.013*** -0.182** -0.703** -2.660 -0.004 × Post (0.707) (0.080) (0.297) (1.745) (0.009) Observations 3,748 3,748 3,748 3,748 3,748 Adjusted R2 0.351 0.168 0.241 0.264 0.190 [Non-executive employees] Act ExposureMedia(Public) 5.317*** 0.502*** 3.015*** 16.979*** 0.081*** (1.707) (0.180) (1.117) (5.243) (0.026) Act ExposureMedia(Public) -3.654** -0.393** -1.498* -8.627** -0.039* × Post (1.548) (0.193) (0.822) (4.366) (0.021) Observations 3,770 3,770 3,770 3,770 3,770 Adjusted R2 0.343 0.164 0.236 0.260 0.191 Controls Yes Yes Yes Yes Yes Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Including year 2016 No No No No No 36

It’s Not Who You Know—It’s Who Knows You: Employee Social Capital and Firm Performance Internet Appendix 37

Internet Appendix I: Data on business card exchange network and an example This Appendix provides descriptive statistics for the business card exchange network data based on all business cards uploaded as of December 31, 2018. Number of connections 12,391,177 Number of employees 2,363,295 Number of employees who are app-users 411,039 Number of employees in public firms 1,016,023 Number of employees in public firms who are app-users 119,423 Number of firms with KIS identifiers 126,987 Number of public firms in OSIRIS Industrials 1,866 We use an example to illustrate the data structure of our business card exchange network and the method for constructing the measures of firm-level employee social capital. The example network is given by the following connection-level data, together with the network graph. Employee_ID_From Firm_ID_From Job_From Employee_ID_To Firm_ID_To Job_To A 1 Staff C 2 Staff A 1 Staff D 2 Vice president A 1 Staff E 3 Manager E 3 Manager A 1 Staff E 3 Manager B 1 Manager Employees A and E are app-users, and all other employees are non-app-users. Employee F does not appear in the network data. Each connection is a directed link from the app-user employee (Employee_ID_From) who uploads the card to the employee (Employee_ID_To) whose card is uploaded. For example, the first entry shows that employee A, a staff of firm 1, has uploaded a card of employee C, a staff of firm 2. This link counts toward the out-degree for A and the in-degree for C. Based on the connection-level data, we construct the measures of firm-level employee social capital (ESC). ESC in-degree is the average In-degree across the firm’s employees who are in the network. For example, the In-degree is one for both A and B, so firm 1’s ESC in-degree = 1. ESC out-degree is the average Outdegree across the firm’s app-user employees. Firm 1 has only one app-user employee, A, so its ESC out-degree equals the out-degree of employee A, which is three. Finally, ESC total degree is the average Total degree across the firm’s employees who are in the network. The total degree is four for employee A and one for employee B, so its ESC total degree = 2.5(=5/2). Firm 2 does not have ESC out-degree because we can only observe the out-degree of app-users. 1

Number of employees Number of app-user ESC in- ESC out- ESC total Firm_ID in the network employees in the network degree degree degree 1 2 1 1 3 2.5 2 2 0 1 - 1 3 1 1 1 2 3 2

Internet Appendix II: variable definitions Variable name Description Measures of employee social capital (ESC) ESC in-degree The average In-degree—the number of employees of other firms who have uploaded the employee as a business contact (“who knows you”) by the end of year t—across employees of firm i who are in the network in year t ESC out-degree The average Out-degree—the number of business contacts of other firms uploaded by the corresponding employee (“who you know”) by the end of year t—across app-user employees of firm i in year t ESC total degree The average Total degree—the sum of In-degree and Out-degree—across employees of firm i who are in the network in year t ESC: Excl. Sales ESC in which we exclude connections of a firm’s customer-facing employees who perform sales functions ESC: Excl. Customers ESC in which we exclude connections with individuals working in a firm’s customer industries ESC in-degree of non-app- The average In-degree—the number of employees of other firms who have uploaded user employees the employee as a business contact (“who knows you”) by the end of year t—across non-app-user employees of firm i who are in the network in year t ESC out-degree to app-users The average Out-degree to app-users—the number of app-user business contacts of other firms uploaded by the corresponding employee (“who you know”) by the end of year t —across app-user employees of firm i in year t ESC: Executives ESC based on the connections of executives (chairman, vice chairman, president, deputy president, executive vice president, and senior vice president) ESC: Non-exec emp ESC based on the connections of non-executive employees (all other employees) ESC in-degreeAct ESC in-degree using only the connections to employees in the industries subject to the Kim Young-ran Act according to the industry codes listed in the table below ESC in-degreeMedia ESC in-degree using only the connections to employees in the media (public) sector (ESC in-degreePublic) according to the industry codes listed in the table below ESC: Sum The sum of In-degree (or Out-degree) aggregated across employees of firm i who are in the network in year t Other variables Tobin’s q Market value of assets divided by book value of assets, in which market value of assets is the sum of market value of equity (common shares outstanding times fiscal-year closing price) and book value of assets minus book value of equity ROA Return on assets, calculated as EBITDA divided by the lagged total assets Sales Growth Log growth rate of sales R&D The ratio of R&D expenses to sales; the ratio is set equal to zero when R&D expenses are missing Book Leverage Total debt (sum of total long-term interest-bearing debt and current long-term debt) divided by total assets ln(1+Assets) Log of one plus total assets (in million Korean won) Volatility Stock return volatility of a firm during the past 24 months Firm Age Current year minus year of incorporation ln(1+Emp) Log of one plus total number of employees 3

Act Exposure ESC in-degree Act/ESC in-degree , that is, the fraction of ESC in-degree in 2015 2015 2015 that is due to connections to employees in industries subject to the Act (we use the industry codes listed in the table below to identify these connections) Post An indicator variable that takes the value of one during and after the enactment year (2016–2018) and zero otherwise d An indicator variable for year t t Act ExposureMedia (Public) ESC in-degree Media (Public)/ESC in-degree , that is, the fraction of ESC in-degree in 2015 2015 2015 that is due to connections to employees in the media (public) sector subject to the Act (we use the industry codes listed in the table below to identify these connections) ln(1+Media Coverage) Log of one plus the weighted count of news articles from RavenPack News Analytics covering a firm over a year in which the weight is the relevance score of each article which ranges from 0 to 100%. We only include news articles with relevance scores greater than or equal to 75%. Positive Media Coverage The ratio of positive media coverage to media coverage. Positive media coverage is Ratio the weighted count of news articles with BMQ sentiment scores of 100 from RavenPack News Analytics covering a firm over a year. The BMQ sentiment score represents the news sentiment of a given story according to the BMQ classifier, which specializes in short commentary and editorials. We only include news articles with relevance scores greater than or equal to 75%. ln(1+# of Proc. Contracts) Log of one plus the total number of newly signed procurement contracts of firm i in year t, from the Korea online e-Procurement Service, which is managed by the Public Procurement Service, Ministry of Economy and Finance ln(1+Tot Amt. of Proc. Log of one plus the total amount of newly signed procurement contracts of firm i in Contracts) year t, from the Korea online e-Procurement Service ln(1+Tot Amt. of Log of one plus the total amount of newly signed procurement contracts normalized Proc. Contracts / Assets) by total assets of firm i in year t, from the Korea online e-Procurement Service ln(1+Sales) Log of one plus sales List of industries subject to the Kim Young-ran Act KSIC code Sector Industry 5812 Media Publishing of newspapers, magazines, and periodicals 59114 Media Broadcasting program production 5912 Media Motion picture, video, and broadcasting program post-production activities 5913 Media Motion picture, video, and broadcasting program distribution activities 60 Media Broadcasting activities 63910 Media News agency activities 6411 Public Central bank 64991 Public Public fund management business 6513 Public Social security insurance 65303 Public Pension funding 6611 Public Administration of financial markets 66191 Public Securities issuance, management, deposit and settlement services 84 Public Public administration and defense; compulsory social security 85 Public Education 4

Internet Appendix III: robustness tests This Appendix presents a variety of robustness tests to address concerns with our results in Section 3.2 regarding the relation between employee social capital and firm performance. In particular, we discuss tests that allay concerns with omitted variable bias, measurement error, and selection bias. Omitted variables A concern is that omitted variables that are correlated with both employee social capital and firm performance may be driving our findings in Section 3.2. Although including industry-by-year fixed effects mitigates such concerns by controlling for unobservable industry-specific trends, we perform additional tests in Table IA.1 to further address this issue. One possibility is that the observed relation between ESC in-degree and sales growth might merely reflect a firm’s sales activities. Sales employees serve as customer touchpoints and are particularly active in exchanging business cards, such that firms with more sales employees may mechanically have greater sales as well as more employee connections. To alleviate this concern, we calculate ESC: Excl. Sales by excluding the connections of a firm’s customer-facing employees who perform sales functions. Specifically, we identify employees who perform sales functions by their job title and department information extracted from their business cards. Examples of job titles related to sales include sales representative, manufacturer’s representative, financial advisor, loan consultant; examples of departments involving sales include customer service, sales strategy, dealership, marketing communication, retail advisory, and marketing. Our method identifies 98,404 public firm employees as sales personnel. While connections with customer industries are clearly important to firms, to provide further evidence that our results are not a byproduct of sales activities, we also calculate ESC: Excl. Customers by excluding a firm’s employee connections with individuals working in its customer industries. To identify customer industries, we follow Frésard et al. (2020) and measure vertical relatedness using detailed Make-and-Use tables obtained from the Bank of Korea Economic Statistics System. Specifically, we use the 2014 Make-and-Use tables to construct a 328-by-328 industry flow matrix in which each cell indicates the dollar flows from an upstream industry to a downstream industry. We define industry j as a customer industry of industry i if the fraction of industry i’s total production used by industry j exceeds a threshold of 3%. As shown in Panel A of Table IA.1, the coefficients on ESC in-degree continue to be positive and statistically significant for both alternative measures, while those for ESC out-degree are not. Another possibility is that firms with well-connected employees might also have high employee technical skills or high employee satisfaction, and it is the employees’ skill or job satisfaction rather than their connections that drives superior firm performance. To alleviate this concern, we use a similar strategy as Cohen et al. (2010) and conduct subsample analyses. We first exclude firms that ranked at least once in the “top 20 most wanted employers by university students” during 2015–2018 according to the Job Korea Survey, such as Samsung Electronics and Hyundai Motor, because these firms tend to show high employee satisfaction and attract some of the most talented university graduates. We then drop financial firms (SIC codes 61, 62, 65, 67) and firms that are in the top three percentile of the asset size distribution, both of which are competitive in the market for talented employees. The 5

results, in Panel A of Table IA.1, show that ESC in-degree remains significantly related to firm performance, whereas the coefficient estimates of ESC out-degree largely remain insignificant, indicating that our results are not an artifact of a selected sample of employees with good technical skills or job satisfaction that drive firm performance. Finally, manufacturing is the dominant industry in Korea, accounting for 64% of public firms and 53% of employees. One might be concerned that the observed relation between ESC in-degree and firm performance is driven by supply chain relationships with prominent customers and suppliers. In untabulated analysis, we find that our results are robust when we exclude connections to customer and supplier industries or exclude manufacturing firms. Measurement error and selection bias Although our network data cover employees in a wide array of firms and industries, we do not observe the universe of employee connections. Thus, we examine the robustness of our results against potential measurement error and selection bias caused by (i) differential app usage among a firm’s employees, (ii) potential differences between app-users and non-app-users, and (iii) our aggregation approach to measuring firm-level employee social capital. First, the fact that our network data are based on the business card collections of app-users might introduce measurement error and selection bias. As discussed in Section 2 in the main text, ESC in-degree likely underestimates “who knows you” because it does not reflect external employees that remember the firm’s employees but do not use the app. To the extent that measurement error biases our estimates toward zero, partially observing employees’ Indegree biases against finding a significant effect of ESC in-degree. On the other hand, because we do not observe the Out-degree of non-app-user employees, ESC out-degree might also contain noise as it is measured on a smaller sample than ESC in-degree. To address this issue, we randomly assign Out-degree to non-app-users by drawing from the Outdegree distribution of app-users in the same firm with replacement; we then construct a bootstrapped ESC out-degree using the actual Out-degree of app-users and the bootstrapped Out-degree of non-app-users. Results based on the bootstrapped data show that the coefficient estimate of ESC out-degree is robustly small in magnitude and insignificant (see Figure IA.1), suggesting that the insignificance of ESC out-degree to firm performance is unlikely an outcome of measurement error. We repeat this procedure 500 times and find that none of the coefficient estimates based on the bootstrapped data are significant at the 5% level. Results are similar when we multiply the bootstrapped Out-degree of non-app-users with a scaler from 0.5 to 1.5 to account for potential differences between app-users and non-appusers. Second, app-users, by nature, are more likely to be tech-savvy and socially active than non-app-users. Since In-degree is observed for both app- and non-app-users, whereas Out-degree is observed only for app-users, a concern is that our decomposition of employee social capital by the direction of connections may pick up these or other differences between app- and non-app-users. To address this concern, in Panel B of Table IA.1, we examine ESC indegree of non-app-user employees to compare with our baseline estimates for ESC in-degree (measured for both appand non-app-user employees). If app-user employees drive our results, we should expect ESC in-degree of non-appuser employees not to be significant; however, the coefficient estimates on ESC in-degree continue to be positive and statistically significant. Similarly, we examine ESC out-degree to only those external contacts who are app-users to compare with our baseline estimates for ESC out-degree (to external contacts including app- and non-app-users), and still find similar results. Moreover, to directly compare the effects of ESC in-degree and ESC out-degree, we include 6

both measures in the same regression; and, since we observe a more complete picture of connections by app-users, we also run the same regressions when we construct both measures using only app-user employees of a firm. Our findings are robust in both cases. These tests suggest that our findings concerning the direction of connections are not an artifact of the asymmetry between app- and non-app-users. In Panel C of Table IA.1, we also perform a propensity score matching analysis to mitigate the potential effects of heterogeneous selection by matching each above-median ESC firm with a below-median firm on year, industry, and the controls in our baseline regression. Results confirm that firms with above-median ESC in-degree experience significantly better performance than their matched firms, whereas no significant difference is found for firms with different ESC out-degree. In addition, to evaluate whether the effects of ESC in-degree are evident for both firms with high performance and firms with low performance, we run quantile regressions and find that the estimated effect is equally strong among firms in different deciles of the performance distribution (shown in Figure IA.2). Third, errors could arise in measuring firm-level ESC since we average across the individual-level degree measures among the employees that are in the network. To reduce error when taking averages, we restrict our sample to observations with at least ten employees observed in the network. Panel D of Table IA.1 shows that our results are unchanged when we apply alternative thresholds for the minimum number or percentage of firm employees who are in the network. Relatedly, employees’ connections might collectively contribute to firm performance; hence, in lieu of averaging across employees, we calculate ESC: Sum as the sum of In-degree (or Out-degree) aggregated across the firm’s employees and find qualitatively similar results. These tests suggest that our results are robust to alternative sample selection and aggregation methods at the firm level. 7

Figure IA.1. Employee social capital and firm performance: measurement error in ESC out-degree To address the potential measurement error in constructing ESC out-degree because the Out-degree of non-app-users is unobservable, we randomly draw Out-degree for non-app-users from the distribution of app-users’ Out-degree in the same firm with replacement. We then reconstruct ESC out-degree using users’ actual Out-degree and non-appusers’ bootstrapped Out-degree and rerun the analyses in columns (7)–(9) of Panel A of Table 2. We repeat this procedure 500 times to generate a distribution of the estimated coefficients. This figure plots the kernel density of the coefficient distribution, with a vertical line indicating the actual coefficient estimates in columns (7)–(9) in Panel A of Table 2. 8

Figure IA.2. Employee social capital and firm performance: quantile regressions This figure plots quantile regression estimates on the relation between employee social capital and firm performance based on the specification in columns (4)–(6) of Panel A of Table 2. Firm-level employee social capital takes the lagged value of ESC in-degree (“Who Knows You”). In each panel, the solid red line represents the estimated coefficients on ln(1+ ESC in-degree) from quantile regressions, and the solid black line represents those from OLS estimates. The shaded area indicates the 95% confidence interval of quantile regression estimates, and the dotted line indicates that of OLS estimates. 9

Table IA.1. Employee social capital and firm performance: robustness analyses This table reports a battery of robustness tests for Panel A of Table 2. Panel A and Panel B present robustness checks for columns (4)–(9) in Panel A of Table 2. Panel A addresses omitted variables bias related to firm sales activities and employee technical skills/job satisfaction. We measure employee social capital by excluding connections of a firm’s customer-facing employees who perform sales functions (ESC: Excl. Sales) and by excluding connections with individuals working in a firm’s customer industries (ESC: Excl. Customers). We also repeat the analysis in columns (4)–(9) in Panel A of Table 2 using subsamples, which exclude, respectively, firms rated at least once in the “top 20 most wanted employers by university students” in 2015–2018, or financial firms (SIC codes 61, 62, 65, 67) and firms in the top three percentile of asset size distribution. Panel B addresses measurement error issues in ESC in-degree and ESC out-degree. In the upper panel, ESC is measured as ESC in-degree of non-app-user employees in columns (1)– (3) and ESC out-degree to app-users in columns (4)–(6). In the lower panel, we include both ESC in-degree and ESC out-degree in the same regression in columns (1)–(3). In columns (4)–(6), we focus on connections of app-users in measuring both ESC in-degree and ESC out-degree and require the firm-year observations to have at least ten appuser employees to reduce measurement errors. In all panels, we include the same set of lagged control variables and fixed effects as in Panel A. Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. The sample period is 2015–2018 for output variables. The definitions of all variables are provided in Internet Appendix II. Panel A. Omitted variables: sales activities and employee technical skills/job satisfaction ESC in-degree (“who knows you”) ESC out-degree (“who you know”) Sales Sales Dep. var. Tobin’s q ROA Tobin’s q ROA Growth Growth (1) (2) (3) (4) (5) (6) [Excluding connections of employees who perform sales functions] ln(1+ ESC: Excl. Sales) 0.389*** 0.020*** 0.093*** 0.050* 0.003 0.002 (0.084) (0.007) (0.024) (0.028) (0.002) (0.006) Observations 5,340 5,340 5,340 4,860 4,860 4,860 Adjusted R2 0.254 0.150 0.037 0.252 0.139 0.038 [Excluding connections with the customer industries] ln(1+ ESC: Excl. Customers) 0.309*** 0.014* 0.082*** 0.044 0.003 0.005 (0.083) (0.007) (0.025) (0.029) (0.002) (0.007) Observations 5,340 5,340 5,340 4,994 4,994 4,994 Adjusted R2 0.251 0.148 0.036 0.252 0.141 0.035 [Excluding top 20 most wanted employers by university students] ln(1+ESC) 0.329*** 0.021*** 0.083*** 0.043 0.004* 0.003 (0.090) (0.008) (0.021) (0.030) (0.002) (0.007) Observations 5,258 5,258 5,258 4,913 4,913 4,913 Adjusted R2 0.258 0.142 0.043 0.258 0.133 0.042 [Excluding financial sector and top 3% firms based on total assets] ln(1+ESC) 0.342*** 0.019** 0.081*** 0.044 0.004* 0.002 (0.093) (0.008) (0.022) (0.031) (0.002) (0.007) Observations 5,056 5,056 5,056 4,715 4,715 4,715 Adjusted R2 0.258 0.146 0.041 0.258 0.137 0.040 Controls Yes Yes Yes Yes Yes Yes Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year 10

Panel B. Measurement error in ESC in-degree and ESC out-degree Sales Sales Tobin’s q ROA Tobin’s q ROA Dep. var. Growth Growth (1) (2) (3) (4) (5) (6) [Differences in characteristics between app- and non-app-users] ESC in-degree of non-app-user employees ESC out-degree to app-users ln(1+ESC) 0.427*** 0.029*** 0.135*** 0.089* 0.005* 0.006 (0.110) (0.009) (0.029) (0.047) (0.003) (0.010) Observations 5,340 5,340 5,340 4,994 4,994 4,994 Adjusted R2 0.252 0.151 0.039 0.253 0.142 0.035 [ESC in-degree and ESC out-degree in the same regression] Based on app-users and non-app-users Based on app-users ln(1+ESC in-degree) 0.371*** 0.020** 0.118*** 0.416*** 0.023** 0.062** (0.103) (0.008) (0.028) (0.119) (0.010) (0.031) ln(1+ESC out-degree) -0.015 0.001 -0.014* -0.158 -0.004 -0.015 (0.032) (0.002) (0.007) (0.097) (0.008) (0.026) Observations 4,994 4,994 4,994 2,322 2,322 2,322 Adjusted R2 0.257 0.144 0.041 0.249 0.136 0.067 Controls Yes Yes Yes Yes Yes Yes Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year 11

Table IA.1. Employee social capital and firm performance: robustness analyses (continued) Panel C reports the results of a propensity score matching analysis. We match the above-median ESC firms with their below-median counterparts on year, industry (two-digit SIC), and the controls as in Table 2, using the nearestneighbor-matching algorithm with a caliper of 0.01, and with replacement. Standard errors in parentheses are bootstrapped based on five hundred replications with replacement. Panel D repeats the analysis in columns (4)–(9) of Panel A of Table 2 with alternative sample selection criteria where we restrict our sample to firm-year observations where at least 20 employees are observed in the network or at least 20% of the firm’s employees are observed in the network. We also present an alternative aggregation method of employee social capital: ESC: Sum is the sum of Indegree (or Out-degree) aggregated across employees of firm i in the network that year. We include an additional control, the number of employees of firm i in the network that year. In both panels, we include the same set of lagged control variables (unless specified) and fixed effects as in Table 2. Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. The sample period is 2015–2018 for output variables. The definitions of all variables are provided in Internet Appendix II. Panel C. Propensity score matching Tobin’s q ROA Sales Growth Number of matches (1) (2) (3) (4) Above median – Below median 0.203*** 0.014*** 0.065*** 2,456 (ESC in-degree) (0.047) (0.004) (0.016) Above median – Below median 0.025 0.005 -0.002 2,237 (ESC out-degree) (0.047) (0.004) (0.015) Panel D. Alternative sample selection criteria and measures of employee social capital ESC in-degree (“who knows you”) ESC out-degree (“who you know”) Sales Sales Dep. var. Tobin’s q ROA Tobin’s q ROA Growth Growth (1) (2) (3) (4) (5) (6) [At least 20 individuals] ln(1+ESC) 0.353*** 0.026*** 0.128*** 0.047 0.003 0.007 (0.097) (0.008) (0.025) (0.032) (0.002) (0.007) Observations 4,842 4,842 4,842 4,680 4,680 4,680 Adjusted R2 0.259 0.147 0.048 0.257 0.140 0.040 [At least 20% of employees] ln(1+ESC) 0.289*** 0.024*** 0.105*** 0.035 0.005* 0.007 (0.098) (0.008) (0.027) (0.040) (0.003) (0.008) Observations 4,209 4,209 4,209 4,014 4,014 4,014 Adjusted R2 0.263 0.170 0.043 0.267 0.154 0.039 [Sum of In-degree (Out-degree) across employees] ln(1+ESC: Sum) 0.251*** 0.016*** 0.067*** -0.004 0.002 0.007 (0.070) (0.006) (0.017) (0.022) (0.002) (0.005) Observations 5,340 5,340 5,340 4,994 4,994 4,994 Adjusted R2 0.254 0.150 0.037 0.253 0.142 0.036 Controls Yes Yes Yes Yes Yes Yes Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year 12

Internet Appendix IV: additional figures and tables Figure IA.3. Remember, the professional business card management app This figure displays screenshots of the Remember app’s user interface. Panel A shows the app available on App Store, Panel B presents the basic user interface, and Panel C illustrates how to scan and upload business cards using the app. Panel A. Remember on App Store Panel B. User interface Panel C. Uploading a card 13

Figure IA.4. Causal evidence: effect of employee social capital on firm performance year by year This figure plots the point estimates of 𝛽 in the following regression: t 2018 Y = 𝛽 +𝛽 ×Act Exposure +∑ 𝛽 ×Act Exposure ×d +𝛾′X +𝛼 +𝜀 , i,t 0 1 i t i t i,t-1 j,t i,t t=2015 where Y is Tobin’s q, Act Exposure =ESC in-degree Act /ESC in-degree , and ESC in-degree Act is ESC ini,t i i,2015 i,2015 i,2015 degree in 2015 that is due to connections to employees in industries subject to the Act. d is an indicator variable for t year t. We extend our pre-treatment sample to include the year 2014 and set 2014 as the baseline year, omitting the 2014 interaction term. The vertical bars correspond to the 95% confidence intervals based on standard errors clustered by firm. 14

Table IA.2. Descriptive statistics of the business card exchange network by sector This table presents descriptive statistics by sector (based on the KSIC codes) of the business card exchange network and the firm-level employee social capital measures as of December 2018. We report the number of public firm employees, the number of public firm employees who are app-users, the number of public firms in OSIRIS Industrials, and the average firm-level ESC measures: ESC in-degree, ESC out-degree, and ESC total degree. Business card Average firm-level exchange network employee social capital measures App-user Public ESC in- ESC out- ESC total Employee employee firms degree degree degree Agriculture, forestry and fishing 1,172 161 6 2.752 22.890 4.568 Mining and quarrying 32 5 3 18.929 73.000 34.571 Manufacturing 545,205 54,502 1,203 3.273 27.669 5.938 Electricity, gas, steam and air conditioning supply 17,698 1,892 11 3.145 25.507 5.670 Water supply; sewage, waste management, materials recovery 417 65 7 4.073 24.706 7.299 Construction 58,462 8,526 51 3.622 30.050 7.430 Wholesale and retail trade 74,745 8,441 148 3.663 29.820 6.694 Transportation and storage 23,843 2,924 26 3.619 37.821 7.231 Accommodation and food service activities 1,272 211 3 3.327 30.388 6.771 Information and communication 105,078 13,648 211 5.119 42.925 9.905 Financial and insurance activities 141,713 23,286 103 5.758 53.176 12.381 Real estate activities 347 100 2 9.217 92.867 21.470 Professional, scientific and technical activities 27,155 3,057 52 4.707 36.251 8.459 Business facilities management and business support services; rental and leasing activities 12,229 1,764 17 4.049 32.126 7.761 Education 2,289 279 10 4.323 32.527 7.758 Arts, sports, and recreation related services 2,467 317 12 3.315 19.571 5.168 Membership organizations, repair and other personal services 1,899 245 1 2.907 16.040 4.741 15

Table IA.3. Adverse impact of the 2016 Kim Young-ran Act on employee social capital We examine the adverse impact of the 2016 Kim Young-ran Act on social relations with the media and the public sector by estimating changes in the fraction of ESC subject to the Act around the enactment as follows: ESC in-degree Act i,t = 𝛽 +𝛽 ×Post +𝛾′X +𝛼 +𝜀 , ESC in-degree 0 1 t i,t-1 j i,t i,t ESC in-degree Act where i,t measures the fraction of a firm’s employee social capital that is derived from connections with ESC in-degree i,t employees in the industries affected by the Act. Post is an indicator variable that takes the value of one during and t after the enactment year (2016–2018) and zero otherwise. X is the same set of lagged control variables as in Table i,t-1 2; 𝛼 is a full set of two-digit SIC industry fixed effects. We no longer include year fixed effects in the regressions due j to the collinearity with the dummy variable Post. Since the Act became effective in the latter half of 2016, we report results excluding the enactment year of 2016 in column (1) and results including the year 2016 in column (2) for robustness; the sample period is 2015–2018 for output variables. Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. The definitions of all variables are provided in Internet Appendix II. Dep. var. ESC in-degreeAct/ ESC in-degree (%) (1) (2) Post -0.266*** -0.260*** (0.068) (0.062) R&D 0.496 0.549 (0.789) (0.831) Book Leverage -0.284 -0.114 (0.536) (0.538) ln(1+Assets) 0.498*** 0.492*** (0.111) (0.110) Volatility 1.609* 1.528* (0.891) (0.856) Firm Age 0.000 0.001 (0.005) (0.005) ln(1+Emp) -0.201* -0.178 (0.113) (0.112) Fixed effects Ind Ind Including year 2016 No Yes Observations 4,017 5,340 Adjusted R2 0.274 0.277 16

Table IA.4. Causal evidence: full measures of firm performance This table presents evidence that a firm’s employee social capital due to connections with industries affected by the Kim Young-ran Act has a positive impact on firm performance, with the effect concentrated in Tobin’s q, but not in ROA or Sales Growth. As in Table 3, we estimate the following difference-in-differences model surrounding the enactment of the Act: Y = 𝛽 +𝛽 ×Act Exposure +𝛽 ×Act Exposure ×Post +𝛾′X +𝛼 +𝜀 , i,t 0 1 i 2 i t i,t-1 j,t i,t where Y is Tobin’s q, ROA, and Sales Growth. Act Exposure =ESC in-degree Act /ESC in-degree , where i,t i i,2015 i,2015 ESC in-degree Act is ESC in-degree in 2015 that is due to connections to employees in industries subject to the Act. i,2015 Post is an indicator variable that takes the value of one during and after the enactment year (2016–2018) and zero t otherwise. X is the same set of lagged controls as in Table 2; 𝛼 is a full set of industry-by-year fixed effects. In i,t-1 j,t Panel A, columns (1)–(3) report results excluding the enactment year (2016), whereas columns (4)–(6) report results when we include the year 2016. The sample period is 2015–2018 for output variables. Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. In Panel B, we present summary statistics of the Act Exposure variables. The definitions of all variables are provided in Internet Appendix II. Panel A. Full measures of firm performance Sales Sales Dep. var. Tobin’s q ROA Tobin’s q ROA Growth Growth (1) (2) (3) (4) (5) (6) Act Exposure 6.578*** 0.152 0.178 6.640*** 0.156 0.185 (1.273) (0.099) (0.306) (1.272) (0.098) (0.308) Act Exposure × Post -4.930*** -0.173** -0.172 -4.726*** -0.148* -0.193 (1.132) (0.087) (0.338) (1.052) (0.080) (0.339) R&D 5.431*** -0.158*** 0.379*** 5.066*** -0.155*** 0.439*** (0.689) (0.040) (0.138) (0.677) (0.040) (0.134) Book Leverage 0.183 -0.132*** 0.075 0.233 -0.139*** 0.059 (0.185) (0.017) (0.057) (0.182) (0.016) (0.055) ln(1+Assets) -0.139*** 0.010*** -0.006 -0.146*** 0.009*** -0.007 (0.025) (0.002) (0.009) (0.023) (0.002) (0.009) Volatility 3.403*** -0.111*** 0.049 3.400*** -0.103*** 0.078 (0.449) (0.027) (0.093) (0.395) (0.026) (0.081) Firm Age -0.005*** -0.000*** -0.000 -0.005*** -0.000*** 0.000 (0.002) (0.000) (0.000) (0.001) (0.000) (0.000) ln(1+Emp) 0.076*** 0.010*** -0.007 0.067*** 0.010*** -0.007 (0.024) (0.002) (0.007) (0.023) (0.002) (0.006) Fixed effects Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Ind × Year Including year 2016 No No No Yes Yes Yes Observations 3,778 3,778 3,778 5,101 5,101 5,101 Adjusted R2 0.242 0.151 0.035 0.245 0.146 0.031 Panel B. Summary statistics of Act Exposure variables N Mean Median SD P25 P75 Act Exposure 3,778 0.036 0.026 0.038 0.012 0.049 Act ExposureMedia 3,778 0.019 0.008 0.029 0.000 0.024 Act ExposurePublic 3,778 0.017 0.013 0.019 0.005 0.024 17

Table IA.5. Causal evidence: randomization of the exposure to the Act This table reports the empirical distribution of the coefficient estimate on Pseudo Exposure × Post when re-estimating column (1) in Table 3 for 1,000 times using the bootstrapped sample. To obtain the bootstrapped sample, we randomly assign a false treatment intensity, Pseudo Exposure, to each firm by maintaining the true distribution of Act Exposure. We also plot the kernel density of the coefficient estimate distribution and draw a vertical line to indicate the actual coefficient of -4.930. Actual estimate Regression coefficient on Pseudo Exposure × Post Act Exposure × Post Mean p1 p5 p10 p25 p50 p75 p90 p95 p99 -4.930 0.045 -1.563 -1.081 -0.827 -0.389 0.062 0.476 0.858 1.069 1.687 18

Table IA.6. Causal evidence: additional robustness analyses This table presents robustness checks for the results in Table 3. Panel A considers alternative measures of Act Exposure and alternative sample selection criteria. In column (1), we additionally include Act Exposure out-degree and Act Exposure out-degree × Post to the estimation of equation (2). Here, Act Exposure out-degree = i ESC out-degree Act /ESC out-degree , and ESC out-degree Act is ESC out-degree in 2015 that is due to i,2015 i,2015 i,2015 connections to employees in industries subject to the Act. In columns (2) and (3), we repeat the analysis in column (1) of Table 3 with alternative sample selection criteria: we restrict our sample to firm-year observations where at least 20 employees are observed in the network in column (2), and to those where at least 20% of the firm’s employees are observed in the network in column (3). In panel B, we include the interaction terms between the firm-level control variables and the dummy variable Post to the estimation of equation (2). Column (1) reports results excluding the t enactment year of 2016; column (2) reports results including the year 2016. The sample period is 2015–2018 for output variables. Standard errors in parentheses are clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. The definitions of all variables are provided in Internet Appendix II. Panel A. Alternative measures of Act Exposure and alternative sample selection criteria Dep. var. Tobin’s q (1) (2) (3) Act Exposure 6.338*** 6.068*** 4.828*** (1.465) (1.266) (1.700) Act Exposure × Post -4.165*** -3.530*** -2.600* (1.315) (1.214) (1.543) Act Exposure out-degree 0.408 (0.772) Act Exposure out-degree × Post -0.782 (0.764) R&D 5.179*** 5.550*** 5.286*** (0.705) (0.693) (0.733) Book Leverage 0.026 0.124 0.054 (0.183) (0.188) (0.217) ln(1+Assets) -0.141*** -0.141*** -0.137*** (0.026) (0.025) (0.028) Volatility 3.585*** 3.420*** 3.690*** (0.491) (0.479) (0.509) Firm Age -0.005*** -0.005*** -0.004** (0.002) (0.002) (0.002) ln(1+Emp) 0.094*** 0.094*** 0.083*** (0.027) (0.027) (0.025) Fixed effects Ind × Year Ind × Year Ind × Year Including year 2016 No No No Observations 3,577 3,390 2,895 Adjusted R2 0.249 0.245 0.245 19

Panel B. Including the control variables interacted with the dummy variable Post Dep. var. Tobin’s q (1) (2) Act Exposure 7.380*** 7.380*** (1.319) (1.318) Act Exposure × Post -5.847*** -5.544*** (1.175) (1.100) R&D 1.997*** 1.997*** (0.712) (0.711) Book Leverage 0.564* 0.564* (0.314) (0.314) ln(1+Assets) -0.249*** -0.249*** (0.034) (0.034) Volatility 3.742*** 3.742*** (0.666) (0.666) Firm Age -0.010*** -0.010*** (0.002) (0.002) ln(1+Emp) 0.137*** 0.137*** (0.038) (0.038) R&D × Post 4.337*** 3.711*** (0.851) (0.805) Book Leverage × Post -0.481 -0.393 (0.359) (0.331) ln(1+Assets) × Post 0.141*** 0.123*** (0.033) (0.030) Volatility × Post -0.334 -0.352 (0.789) (0.729) Firm Age × Post 0.008*** 0.007*** (0.002) (0.002) ln(1+Emp) × Post -0.070* -0.081** (0.036) (0.034) Fixed effects Ind × Year Ind × Year Including year 2016 No Yes Observations 3,778 5,101 Adjusted R2 0.253 0.252 20