10:47 19/3/2019 RFS-OP-REVF180080.tex Page: 2021 20212061 Big Data as a Governance Mechanism Christina Zhu Wharton School, University of Pennsylvania This study empirically investigates two effects of alternative data availability: stock price informativeness and its disciplining effect on managers actions. Recent computing advancements have enabled technology companies to collect real-time, granular indicators of fundamentals to sell to investment professionals. These data include consumer transactions and satellite images. The introduction of these data increases price informativeness through decreased information acquisition costs, particularly in firms in which sophisticated investors have higher incentives to uncover information. I document two effects on managers. First, managers reduce their opportunistic trading. Second, investment efficiency increases, consistent with price informativeness improving managers incentives to invest and divest efficiently. (JEL G14, G12, G23, G34, O16, M12) Received June 1, 2017; editorial decision June 1, 2018 by Editor Wei Jiang. The Author has furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online. In this study, I examine whether the availability of alternative data improves price informativeness and helps discipline corporate managers. Price informativeness and the allocation of information in an economy are important because they have the potential to affect managers actions (e.g., Fishman and Hagerty 1989; Holmstrom and Tirole 1993; Brandenburger and Polak 1996). It can be empirically challenging to assess whether managers take different For their guidance, support, and many helpful insights, I am very grateful to the members of my dissertation committee: David Larcker (chair), Laurie Hodrick, Charles Lee, and Joseph Piotroski. I also thank Itay Goldstein, Wei Jiang, Andrew Karolyi (the Editors), Alexi Savov (discussant), Anne Beyer, Elizabeth Blankespoor, Lisa De Simone, Joseph Grundfest, Wesley Hartmann, Ron Kasznik, Alan Kwan, Rebecca Lester, Ivan Marinovic, and Peter Reiss; two anonymous reviewers; fellow PhD students; and seminar participants at Stanford University, Boston College, the University of Chicago, Columbia University, Duke University, Harvard University, London Business School, the University of Michigan, the University of North Carolina, Northwestern University, the University of Pennsylvania, the University of Toronto, Yale University, the Review of Financial Studies FinTech Workshops (Columbia, 2017 and Cornell Tech, 2018), and the 2017 CMU Accounting Mini-Conference for helpful suggestions. I thank two anonymous data providers, a marketing analytics company and a satellite image data provider, for providing proprietary data. I also thank Justin Zhen (Thinknum) and multiple anonymous investment professionals and industry experts for providing helpful institutional insight. This work was supported by Stanford University. All errors are my own. Supplementary data can be found on The Review of Financial Studies Web site. Send correspondence to Christina Zhu, The Wharton School, University of Pennsylvania, 1320 Steinberg Hall - Dietrich Hall, 3620 Locust Walk, Philadelphia, PA 19104; telephone: (215) 898-7783. E-mail: chrzhuwharton.upenn.edu. The Author(s) 2019. Published by Oxford University Press on behalf of The Society for Financial Studies. All rights reserved. For permissions, please e-mail: . doi:10.1093/rfs/hhy081 Downloaded from by Renmin University user on 05 December 2019 10:47 19/3/2019 RFS-OP-REVF180080.tex Page: 2022 20212061 The Review of Financial Studies / v 32 n 5 2019 actions when prices become more informative because of the endogenous nature of price informativeness and corporate disclosure. For example, a manager might choose to make less informative disclosures to benefit from personal trades in the firms stock. Similarly, he might disclose less when he chooses less efficient investments. To address the challenge of empirically studying this relation, I first test for an increase in price informativeness that results from technological innovations exogenous to the firms managers. I then evaluate the disciplining effects of this improved informativeness on managers opportunistic trading and real investment decisions. I test for a change in price informativeness using the growth in alternative data sets, some of which are referred to as “big data.” Alternative data are defined as data sets that are “not from a financial statement or report” (Quinlan and Associates 2017). In recent years, the proliferation of mobile devices, low-cost sensors, and other technologies has reduced data-gathering costs, leading to the birth of multiple start-ups that collect these alternative data. These data include point-of-sale transactions, satellite images, and clickstream data, and they are different from traditional sources of information (e.g., financial information from company filings, investor presentations, and analyst reports) in that they are granular, real-time data that are not derived from firm disclosures. The availability of these third-party data sets has reduced investors costs of acquiring information, such that investment professionals have begun to use these data in investment strategies (Bank of America Merrill Lynch 2016). Despite the increased use of these alternative data sets, there is limited or no research on their consequences for capital markets and managers actions. First, I examine the link between alternative data and improved price informativeness. Noisy rational expectations models predict that, when information acquisition costs decrease, the informational efficiency of stock price increases (Grossman and Stiglitz 1980; Diamond and Verrecchia 1981; Verrecchia 1982). In these models, prices do not perfectly convey the private signals of informed investors; consequently, a decrease in the cost of information acquisition improves price informativeness, defined as the correlation between price and fundamental information. However, it is less clear whether traders acquisition of the data can lead to an improvement in long-run price informativeness (McNichols and Trueman 1994). The data sets presumably contain short-horizon indicators of fundamentals (e.g., consumer transactions that have occurred but have not yet been announced by the firm). Therefore, it is an open question whether the availability of alternative data can improve the incorporation of long-horizon (i.e., 1-year-ahead) earnings into prices. To assess whether alternative data are indeed informative, I obtain access to two alternative data sources. The first data source contains consumer transactions from a marketing analytics platform built on a large panel of consumer-browsing data, which are passively collected from users that have installed antivirus software and sold to active portfolio managers. For example, 2022 Downloaded from by Renmin University user on 05 December 2019 10:47 19/3/2019 RFS-OP-REVF180080.tex Page: 2023 20212061 Big Data as a Governance Mechanism the data include unique checkout transactions completed at consumer-facing firms with an online presence (e.g., ). The second data source is a satellite image data partner that provides normalized car counts in parking lots of retailers. These car counts map to consumer transactions in stores and are relevant for firms with a retail store presence. These data cover 266 firms from 2014 to 2016. I show that aggregated signals from these data sets have predictive power for revenue and earnings that are not yet announced, and the data can predict announcement returns. 1 After verifying that these data sources contain incremental information content, I validate that investors use the data by showing that price reactions to earnings announcements are muted after alternative data from these data sources are available in June of 2014. Despite the prohibitive costs of these data sets (i.e., hundreds of thousands of dollars), the availability of alternative data results in a measurable increase in short- run price informativeness. Inferences are based on a difference-in-differences research design comparing the 266 firms covered by these alternative data sets to a group of matched firms that are economically similar but do not have much data coverage. Following these validation tests, I test for an increase in long-run price informativeness. The richness and granularity of the alternative data contain information that is typically not publicly disclosed by the manager, and this superior information can help investors incorporate fundamental information related to longer-term performance into prices. 2 I find evidence that, for firms affected by alternative data, current returns contain more information about future earnings. Cross-sectional tests find that the short-run and long-run effects are stronger in firms for which sophisticated investors have the highest incentives to uncover information (i.e., firms that sell discretionary consumer products and services, firms with higher market-to-book ratios, and more liquid firms). The results seem to be driven by sophisticated investors who presumably acquire these alternative data sets. After finding the improvement in long-run price informativeness after alternative data are available, I then focus on two potential effects on the manager. For the first effect, I investigate whether the investors use of alternative data reduces the managers opportunity to trade on his private information about future earnings. The literature on insider trading has found that managers exploit their superior information for personal trading gains (Piotroski and Roulstone 2005; Rogers 2008). When prices reflect future 1 Using both data sources, I find that these alternative data predict revenue and earnings that will be announced after the end of the quarter. They also predict announcement period returns when those revenue and earnings numbers are released. A long-short trading strategy earns 1.4% to 2.0% in abnormal returns in the 11-day -5, +5 window around the earnings announcement (see the Internet Appendix). 2 This assumption is consistent with Froot et al.s (2017) finding that managers do not disclose all of their private information. Similar to that used by Froot et al. (2017), the private information proxy in this paper is based on big data. However, Froot et al. (2017) do not test the impacts of these data on price informativeness or on managers actions, which are the focus of my paper. 2023 Downloaded from by Renmin University user on 05 December 2019 10:47 19/3/2019 RFS-OP-REVF180080.tex Page: 2024 20212061 The Review of Financial Studies / v 32 n 5 2019 earnings more quickly and completely, the manager has less of an opportunity to extract rents by trading on his private information. Consistent with this hypothesis, I find that insiders of firms affected by alternative data are less likely to purchase shares ahead of positive future earnings innovations. Furthermore, when insiders do trade, I find that the positive relation between insider trades and future earnings innovations is attenuated after alternative data are available. This evidence suggests that managers reduce the exploitation of their private information about future earnings through personal trades when prices reflect information from alternative data. The second managerial action effect I investigate is whether alternative data availability disciplines the manager to make better real investment decisions. Agency problems, which result from the separation of ownership and control, have been shown in several papers to induce empire building or overinvestment of free cash flow (Jensen 1986; Harford 1999; Bates 2005; Richardson 2006). Furthermore, concerns about reputation and reluctance to take action (i.e., the quiet life hypothesis) hinder the managers discontinuation of underperforming businesses (Kanodia et al. 1989; Boot 1992; Bertrand and Mullainathan 2003). When investment opportunities are declining, the optimal firm response is to curtail investment Wurgler (2000). Therefore, managers incentives to expand the size of the firm (i.e., empire building) instead of closing businesses (i.e., reputation and the quiet life) are misaligned with shareholders when investment opportunities are declining. In my setting, the documented increase in long-run price informativeness is consistent with alternative data providing information about future profitability, whether that profitability is related to assets in place or expected future investment opportunities. With respect to assets in place, alternative data might reveal granular information about which businesses should be closed. With respect to investment opportunities, alternative data might reveal superior information about which businesses to expand. I acknowledge that I cannot directly observe whether corporate managers are aware of alternative datas effect on sophisticated investor behavior and prices. Therefore, my tests are joint tests of this awareness and the effect on firm choices. Following Wurgler (2000), I define the level of investment efficiency as the responsiveness of the firm to investment opportunities (i.e., increasing investment when opportunities are expanding and, conversely, decreasing investment when opportunities are deteriorating). I test for changes in this responsiveness and find that the introduction of alternative data to the market is associated with a greater sensitivity of investment to deteriorating opportunities. Consistent with prior research, I do not find the symmetric effect when investment opportunities are expanding (Wurgler 2000; Bushman et al. 2011). In additional tests of investment efficiency, I find that the excess returns to announcements of discontinued operations are higher after alternative data availability. My study makes two main contributions. First, I contribute to the growing literature on the impact of technology on capital markets. Recent papers have 2024 Downloaded from by Renmin University user on 05 December 2019 10:47 19/3/2019 RFS-OP-REVF180080.tex Page: 2025 20212061 Big Data as a Governance Mechanism documented the capital market effects of multiple technological innovations, including algorithmic trading (Hendershott et al. 2011), high-frequency trading (a form of algorithmic trading studied in Brogaard et al. 2014), and robo- journalism (Blankespoor et al. 2018). My paper examines another technology- related impact on the capital markets: the impact of the use of alternative data in asset management. Asset managers invested an estimated $4 billion into alternative data in 2017 (Marenzi 2017), but to date there has been little research in this area. My study builds on the finding in prior research that alternative data sources can predict earnings and revenue (e.