Essays on reporting and information processing

(1)

Tilburg University

Essays on reporting and information processing de Kok, Ties DOI: 10.26116/center-lis-1904 Publication date: 2019 Document Version

Publisher's PDF, also known as Version of record

Link to publication in Tilburg University Research Portal

Citation for published version (APA):

de Kok, T. (2019). Essays on reporting and information processing. CentER, Center for Economic Research. https://doi.org/10.26116/center-lis-1904

General rights

Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain

• You may freely distribute the URL identifying the publication in the public portal Take down policy

(2)

Essays on reporting and information

processing

Proefschrift

ter verkrijging van de graad van doctor aan Tilburg University op gezag van de rector magnificus, prof. dr. E.H.L. Aarts, in het open-baar te verdedigen ten overstaan van een door het college voor promoties aangewezen commissie in de aula van de Universiteit op maandag 20 mei 2019 om 16.00 uur door

(3)

Promotiecommissie:

Promotores: prof. dr. P.P.M. Joos

prof. dr. J.F.M.G. Bouwens

Overige Leden: prof. dr. S. Hollander

(4)

(5)

(6)

Acknowledgments

In hindsight, I consider completing the PhD similar in spirit to climbing the Mount Everest. It is a challenge of persistence and self-discovery that is not possible to com-plete on your own. It requires people that support you and it gets a whole lot easier if there are people that are taking the challenge alongside you. I started the “climb” having no idea what I was embarking on, nor having a clear understanding as to why I wanted to conquer “the mountain” to begin with. However, unbeknownst to me at the time, the journey turned out to be the most important, with the final destination being just a bonus. More specifically, I have been incredibly fortunate to meet Yusiyu, my partner for life, at the beginning of this journey. Much of my development during the PhD I owe to her, and she made taking a leap of faith on the PhD one of the best decisions of my life.

I would also like to especially acknowledge the help, support, and friendship of Victor van Pelt. Victor, Yusiyu, and myself formed our PhD cohort, and I very much con-sider us akin to the “Three Musketeers”. I have benefited greatly, both personally and professionally, from having Victor as a friend, colleague, and co-author. He is an excel-lent researcher that is always willing to pitch in on research ideas and provide critical insights on how to solve research related problems. Furthermore, his assertive person-ality proved useful on more than one occasion, you can rely on Victor when something needs to be taken care of. I very much hope that we can continue our traditions of getting together for a beer in the future, wherever it may be!

(7)

Second of all, special thanks to my second supervisor Jan Bouwens. Jan and I only briefly overlapped at Tilburg but I still consider him an important mentor. If it wasn’t for him, I might not have been in the PhD to begin with given that he was the person I talked to, besides Willem Buijink, when I was considering the research master program. I also learned later that he was instrumental in allowing me to enter into the program, even though some of the formal criteria were not in my favor. Jan is also my first co-author and he opened my mind up to the world of “empirical management accounting”, but he did it in a way that did not restrict my interests in financial accounting topics. This is how it should be done! I greatly enjoy working with Jan and he is an incredibly nice person to be around. Jan has played an important supporting role throughout my PhD, and I am very grateful for that.

I also would like to express a thank you to my committee members: Stephan Hollander, Mark Clatworthy, and Ed deHaan. They have provided me with important feedback, and my papers have benefited from it greatly. I would like to especially thank Stephan for being a great colleague with whom I have always been able to discuss cool and interesting research ideas. We share a lot of common interests, and I will fondly re-member our interactions over Python in particular. Also, I would like to express a special thanks to Ed for being open minded enough to extend me an invitation to visit the University of Washington. I consider Ed an important third mentor besides Philip and Jan. His advice on my research projects and the US job market have been instrumental to my PhD. It is a privilege to join the University of Washington and be one of Ed his colleagues.

There are two remaining co-authors that I also would like to express a special thank you to, Christoph Sextroh and Arnt Verriest. Christoph joined Tilburg half-way through my PhD, and I consider myself lucky to call him my co-author. He is an exceptionally smart researcher and has a very kind and supportive personality. Besides working on research, I have fond memories of playing squash together and hanging out over beer and boardgames. Arnt and I never overlapped at Tilburg, but our paths crossed through our mutual co-authorship with Jan. I got to know Arnt as a great researcher, but more importantly as a kind and caring person. He is incredibly easy to hang out with and always has a way of making you feel comfortable around him, it truly is a pleasure to work with him.

(8)

master students and PhD students at Tilburg that I interacted with over the years; Martin, Ruishen, Jingwen, Tim, Ruidi, and Mathieu. You are all great people and you definitely made my PhD a lot more enjoyable. Besides the accounting PhDs I also have fond memories of playing video and board games with Clemens, Peter, Carlos, Ricardo, and Tung. I also would like to thank all the PhD students at the University of Washington for being so welcoming to me. In particular, Rosh was a great job market buddy and David was a great office mate. I would like to also especially thank John for his friendship and willingness to help me out during my visit. Even though he did not have to, he made sure that I had someone to hang out with from the very beginning, and without him my visit would have been a lot lonelier!

Thanks as well to the whole accounting department at Tilburg, they have provided me with a great environment to development myself in. There are some people that I would like to mention in particular. First of all, I would have not been here if it wasn’t for Willem Buijink spotting my potential in his bachelor course. You have opened up a new world for me Willem! I also would like to thank Bart Dierynck for his support during my PhD. A special thanks also to Bob van den Brand, most of my teaching was for his financial accounting course, and I always had a blast. Hetty Rutten has also set a very high bar with regards to administrative support, she is an instrumental part of the department and helped me a lot throughout my PhD. Lastly, I would like to acknowledge the help and training from Laurence van Lent, he is a very knowledgeable person and I learned a lot from him during the research master.

Finally, I owe everything to my parents, Jolanda de Kok-Hulzenga and Maarten de Kok. I consider their unconditional support and incredible upbringing instrumental to how I have developed as a person, both personally and professionally. In particular, their spirit of being super hardworking with a very down-to-earth mentality will always be a big inspiration to me. It is hard to describe how lucky I feel to have them as my parents. Furthermore, a special thanks to Anneloes, Joris, and Michiel. They are a fantastic group of siblings that are all achieving great things in their lives, and I could not be prouder of them! A last special thanks goes to two of my best friends, Johan and Tim. Playing video games with them online and getting together to hang out have always been important ways for me to decompress from the PhD. Thanks for sticking around, and I hope to welcome both of you to Seattle some day!

(9)

(10)

List of Figures

Chapter 2: 33

2.1 Estimation results for various period lengths. . . 33

2.2 Results of mediation model — Levels estimation . . . 34

2.3 Results of mediation model — Changes estimation . . . 34

2.4 Visualization of LDA model . . . 35

A.1 A small selection of market pressure Twitter messages. . . 43

B.1 Screen capture of M-Turk instructions . . . 45

B.2 Screen capture of M-Turk task . . . 46

Chapter 3: 87 3.1 Ex-ante trend for the main analysis. . . 87

3.2 Ex-ante trend for the selection analysis. . . 88

3.3 Ex-post trend for the selection analysis. . . 89

Chapter 4: 137 4.1 (a) Text-centric approach . . . 138

4.2 (b) User-centric approach . . . 138

4.3 Computational Pipeline to Estimate Variation in User Behavior . . . . 139

4.4 Illustration of the MTurk Instrument used to elicit User’s Behavior . . 140

4.5 Trend of IR Heterogeneity and Document/Text Features by Year . . . . 141

(14)

List of Tables

Chapter 2: 5

Table 1: Descriptive Statistics . . . 49

Table 2: Main Regressions . . . 51

Table 3: Reporting Quality Regressions . . . 53

Table 4: Frequency of Unverifiable Announcements . . . 54

Table 5: New Backers Regressions . . . 55

Table 6: Cross-Sectional Split Regressions . . . 56

Table 7: Market Pressure Heterogeneity (LDA) Regressions . . . 57

Chapter 3: 59 Table 1: Descriptive Statistics . . . 94

Table 2a: Descriptive Statistics . . . 96

Table 2b: Descriptive Statistics . . . 97

Table 3: Main Results . . . 98

Table 4: First Stage Heckman Selection Model . . . 99

Table 5: Second Stage Heckman Selection Model . . . 100

Table 6: Loan Officer Fixed-Effects . . . 101

Table 7: Loan outcome analysis . . . 102

Table 8: Screening analysis . . . 103

Table 9: Likelihood of acceptance analysis . . . 104

Table 10: Placebo Tests . . . 105

Chapter 4: 107 Table 1: Regressions of Participants’ Judgments on Marking Behavior . . . . 161

Table 2: Regressions of Reading Time on Financial Literacy and Text Com-plexity . . . 163

(15)

Table 4: Regressions of Marking Behavior on Financial Literacy and Text

Complexity . . . 165

Table 5: Top 15 Words and Bigrams in Sentences Marked by Medium or High Literacy Users Only . . . 166

Table 6: Predicted Marking Behavior of Financial Literacy Groups . . . 167

Table 7: Heterogeneity in Predicted Marking Behavior . . . 168

Table 8: Variable Descriptive Statistics for 10-K Filings . . . 169

Table 9: Analysis of Predicted Heterogeneity in Information Retrieval Using Post-Filing Date Market Model RMSE . . . 170

Table 10: Analysis of Predicted Heterogeneity in Information Retrieval Using Analyst Dispersion as Dependent Variable . . . 171

(16)

Chapter 1

(17)

(18)

In this dissertation, I present three empirical essays on a range of topics relating to reporting and information processing. All of these essays utilize state-of-the-art em-pirical techniques drawn from computer science along with new data sources to study fundamental accounting questions. More specifically, they cover four primary topics: internal and external reporting practices, narrative disclosures, recent advancements in reporting technologies, and the role of reporting in emerging markets.

Chapter 2 studies the reporting implications of recent technological advancements in funding structures and two-way communication channels. More specifically, this single author paper studies the role of reporting frequency in crowdfunding markets. Using data from one of the largest reward crowdfunding platforms, I provide the first empir-ical evidence on the relationship between reporting frequency and the level of market pressure in a scenario where the consumers of an entrepreneur also act as her funders. I develop a text-based measure for market pressure by classifying Twitter messages directed to the entrepreneurs with a machine learning algorithm trained using Amazon Mechanical Turk. The results show a negative association between reporting frequency and the level of market pressure. This result is driven primarily by the reporting part of the updates; a mediation analysis shows that only a small fraction of the relationship is driven by a consumption utility effect. Furthermore, this association is stronger when accompanied by higher quality reporting events, is not influenced by the frequency of unverifiable additional announcements, and is weaker during periods with a large presence of newly joined funders. These results highlight the fact that higher reporting frequencies can lead to reduced agency frictions in crowdfunding markets, even when these markets are characterized by strong myopic market preferences

(19)

decision process. Our findings indicate that this change enables better integration of soft information in credit decisions. These findings are robust to controlling for strate-gic loan-sorting behavior, manager fixed effects, and the likelihood of acceptance. We also document that this improved integration of soft information is driven by a change in behavior by the loan officers and that it results in better loan outcomes.

Chapter 4 studies the information retrieval process for narrative disclosures from a user perspective by combining innovative tracking techniques deployed on Amazon Mechanical Turk with state-of-the-art machine learning techniques. More specifically, we develop a comprehensive measure for variation in information retrieval based on observed user behavior that is also able to incorporate understudied text characteris-tics such as the semancharacteris-tics and content of a narrative. Using a tool that tracks reading and marking behavior in a controlled environment, we first document how users with varying degrees of financial literacy retrieve information from financial narratives. We find significant variation among financial literacy groups that cannot be solely ex-plained by text characteristics related to processing costs. Next, we use state-of-the-art machine-learning to predict variation in information retrieval for out-of-sample finan-cial narratives, and we show that these predictions are incrementally associated with the post-announcement return volatility. Overall, our results suggest that efforts by regulators and corporations to simplify text characteristics of corporate communica-tions might not resolve all differences in how users retrieve information from financial narratives.

(20)

Chapter 2 Reporting Frequency and Market

Pressure in Crowdfunding

(21)

(22)

2.1. Introduction

I study the relationship between reporting frequency and market pressure in crowd-funding markets. Reward crowdcrowd-funding is an alternative channel for entrepreneurs to bring their ideas to market by raising funds directly from their consumers rather than through traditional financial intermediaries such as venture capitalists. Entrepreneurs increasingly opt to use this funding channel because crowdfunding allows them to re-tain ownership and provides a cost-effective way to reduce demand uncerre-tainty before the investment decision (Strausz, 2017). Analysts estimate that by 2022 more than US $25 billion will have been raised worldwide through reward crowdfunding.1 _This

pop-ularity, however, is often considered puzzling from an economic perspective as the lack of regulation and monitoring mechanisms in these crowdfunding markets is expected to result in a moral hazard problem (Gutirrez Urtiaga and Saez Lacave, 2018). Disclosures by the entrepreneur provide a mechanism for alleviating this agency fric-tion. Recent studies examine the role of such disclosures and document that projects with higher upfront disclosure quality are able to raise more funding. (Ahlers,

Cum-ming, G¨unther, and Schweizer, 2015; Cascino, Correia, and Tamayo, 2018; Hornuf

and Schwienbacher, 2018; Madsen and McMullin, 2018). I extend these findings by studying how the frequency of disclosures interacts with funders’ monitoring behavior after they have provided their funding. In particular, I measure funders’ monitoring behavior as the level of market pressure that the incumbent funders impose on the entrepreneur. Drawing on various concepts in the literature, I define market pressure as communications from funders to the entrepreneur that are intended to steer the entrepreneur’s decisions in a specific direction.

It is not ex ante clear how the disclosure frequency affects the level of market pres-sure (Gigler, Kanodia, Sapra, and Venugopalan, 2014; Edmans, Heinle, and Huang, 2016). From the perspective of information asymmetry and agency frictions, a higher reporting frequency would be expected to reduce the need for market pressure. Empir-ical evidence shows that in a traditional capital market context, timelier information reduces the need for active stakeholders to exert market pressure owing to a reduc-tion in moral hazard (Demsetz and Lehn, 1985; Bushman, Chen, Engel, and Smith, 2004; Armstrong, Guay, and Weber, 2010). From the perspective of market myopia, however, the reporting frequency might also increase the level of market pressure. Ex-perimental evidence shows that exposing individuals to a higher evaluation frequency can adversely affect their decision horizon (Thaler, Tversky, Kahneman, and Schwartz, 1997; Gneezy and Potters, 1997; Van Der Heijden, Klein, M¨uller, and Potters, 2012).

(23)

The disclosure literature also suggests that higher reporting frequencies might attract funders with short-term preferences (e.g., Wagenhofer, 2014) and might result in more dissatisfied funders owing to increased exposure to the outcome variability (Guo, Finke, and Mulholland, 2015; Casas-arce, Louren¸co, and Mart´ınez-jerez, 2017).

To empirically study this relationship, I examine a type of reward crowdfunding called “Early Access”. Early-access crowdfunding differentiates itself from “traditional” re-ward crowdfunding by providing the funders with access to all development versions of the product. It is increasingly replacing platforms such as Kickstarter in the video game and other digital technology industries.2 _{The early-access model enables me to study}

the effect of reporting frequency on market pressure because the entrepreneurs pro-vide product updates at varying frequencies that reveal the progress in the product’s development and because the funder-entrepreneur interactions happen on publicly ob-servable communication channels. Furthermore, the funding period is open throughout the development period, which provides the entrepreneur with incentives to care about disclosures and market pressure, as the behavior of incumbent funders is a strong deter-minant for the investment decisions of future funders (Hildebrand, Puri, and Rocholl, 2017; Hornuf and Schwienbacher, 2018).

I collect publicly available data from the most popular early-access platform, Steam Early Access, using programmatic data-gathering techniques. The Steam platform is the largest digital distributor of games, with more than 125 million registered users and close to 15 million concurrent users during peak hours. My sample consists of 144 projects that became available on the Steam platform between May 2013 and February 2017 as part of the early-access program. An average project is, conservatively estimated, able to attract around US $1.5 million in funding from early-access funders. For my main analyses, this results in 916 observations, with each observation being a 10-week project period. This sample exhibits substantial variation in terms of the number of new product updates, as an average project has an update every 6 weeks but some projects update as frequently as once per week. Each product update reports on the progress made by the entrepreneur but also potentially provides funders with additional consumption utility for receiving new product features. To distinguish the reporting effect from the utility effect, I use a mediation model wherein changes in the product satisfaction ratings serve as a proxy for the utility effect.

In the early-access crowdfunding setting, the primary channel for funders to commu-nicate with the entrepreneurs is through the entrepreneur’s Twitter account. I utilize

(24)

this characteristic of the setting to develop a new measure for market pressure based on the subset of tweets that both relate to the state of the project and contain a direct or indirect request. This subset of tweets is identified by a machine learning algorithm trained on a sample of 5,000 manually classified tweets that I obtain by deploying a custom classification tool on Amazon Mechanical Turk. I use this algorithm to classify the entire sample of 220,000 tweets directed toward entrepreneurs, which yields a sub-sample of 105,000 tweets that plausibly impose market pressure on the entrepreneur. On average, this translates to an entrepreneur receiving around 56 market pressure tweets from funders in each 10-week period.

The results show a negative association between the number of reporting events in a period and the level of market pressure imposed on the entrepreneur. This result is economically meaningful, because one more reporting event in a 10-week period is associated with, on average, 6 fewer market pressure tweets, which reflects a reduction of 17% relative to the median. I obtain a similar result when I use the stricter changes specification. Increasing the number of reports in a 10-week period by 1 report is associated with a reduction in market pressure of around 9 tweets, which corresponds to a reduction of 25% relative to the median. This result is driven primarily by the progress reporting part of the product updates, as the mediation analysis shows that only a small fraction of the relationship is driven by a utility effect. Additional tests show that this association is stronger when accompanied by higher quality reporting events, is not influenced by the frequency of unverifiable additional announcements, and is weaker during periods with a large presence of newly joined market participants. Finally, a latent dirichlet allocation (LDA) model trained on the market pressure tweets reveals that the reporting frequency primarily affects pressure relating to the product’s quality and feature set but has little effect on pressure relating to the release schedule of the product’s development.

(25)

de-veloping a project-period specific measure for the level of market pressure based on textual communications from market participants by training a machine learning algo-rithm using Amazon Mechanical Turk. This empirical approach is largely new to the literature but highlights and encourages classifying textual data using training sam-ples of appropriate sizes in a cost-efficient and objective manner, which is becoming increasingly important.

2.2. Background and Literature Review

2.2.1. Crowdfunding

The existing literature shows that obtaining early-stage funding through traditional channels such as bank loans or large equity investors can be troublesome for some en-trepreneurs (e.g.,Berger and Udell, 1995; Cosh, Cumming, and Hughes, 2009). Many entrepreneurs are therefore increasingly turning to crowdfunding as an alternative source of financing. Crowdfunding suggests that early stage funding is obtained directly from a large group of individuals who each are willing to support the entrepreneur by pledging a relatively small amount of money (Belleflamme, Lambert, and Schwien-bacher, 2014; Belleflamme, Omrani, and Peitz, 2015). This crowd-based funding ap-proach helps the entrepreneur to obtain not only financing at an early stage but also feedback and ideas from these individuals, which can help to reduce the uncertainty around product demand that is inherent in starting a new business (Agrawal, Catalini, and Goldfarb, 2014; Strausz, 2017; Xu, 2017; Chemla and Tinn, 2018).

(26)

such as electronic books, online courses, and video series.

The popularity of these crowdfunding models is often considered puzzling from an economic and legal perspective owing to the strong presence of moral hazard and in-formation asymmetries (e.g.,Strausz, 2017; Gutirrez Urtiaga and Saez Lacave, 2018). This is particularly true for reward crowdfunding. First, the entrepreneurs in these markets tend to be young and relatively inexperienced, this makes it difficult for fun-ders to evaluate the reputation of an entrepreneur.3 _{Second, funders in these markets}

are often considered to primarily be consumers which are not traditionally exposed to such agency frictions. The theoretical papers by Strausz (2017) and Chemla and Tinn (2018) model this reward crowdfunding scenario and conclude that these funders are a new type of stakeholder that is best described as a consumer exhibiting “investor like” behavior. The incentives of these funders are different compared to traditional in-vestors, however, the underlying agency frictions are similar. Empirical studies on the Kickstarter platform find evidence consistent with these theoretical predictions (e.g., Barbi and Bigelli, 2017; Courtney, Dutta, and Li, 2017; Kuppuswamy and Bayus, 2018).

2.2.2. Disclosures

The agency frictions inherent to crowdfunding highlight the importance of mecha-nisms that reduce the information asymmetry between the entrepreneur and the fun-ders (Strausz, 2017). Courtney et al. (2017) show that signals such as third-party endorsements and the entrepreneur’s prior crowdfunding experience have a positive effect on the projects’ likelihood of attaining funding. Besides these indirect signals, the entrepreneur can also reduce information asymmetry directly by providing more and better disclosures to the funders. Several papers have documented that Kickstarter projects with longer project descriptions on average obtain more project funding (Barbi and Bigelli, 2017; Kuppuswamy and Bayus, 2018; Cascino et al., 2018). Madsen and McMullin (2018) extend these results by analyzing the “Risks and Challenges” sec-tion and documenting that projects with higher quality risk disclosures receive more funding. Furthermore, Cumming, Hornuf, Karami, and Schweizer (2017) show that projects with poorly worded and confusing campaign pitches have a higher likelihood of being ex post identified as fraudulent.4

3 _{“Developer Satisfaction Survey 2017” by the International Game Developer Association.}_https: //www.igda.org/page/dss2017

4 _{On a related note, the role of proprietary information in providing credibility to a voluntary}

(27)

Prior literature also documents the importance of reporting frequency in equity market scenarios. There is plenty of research that looks at the effect of reporting frequency on the level of information asymmetry in traditional capital market settings (Cuijpers and Peek, 2010; Fu, Kraft, and Zhang, 2012). More relevant are the papers by Block, Hornuf, and Moritz (2018) and Hornuf and Schwienbacher (2018) as they show that entrepreneurs that provide more updates have a higher likelihood of obtaining funding in equity crowdfunding markets. However, little is known about the role of reporting frequency in reward crowdfunding markets.

Early-access crowdfunding is particularly suitable to study the role of reporting fre-quency in reward crowdfunding markets. The concept of disclosure on the Kickstarter platform is largely determined by the up-front campaign description as the funding is concentrated at the beginning and the end of the campaign (Kuppuswamy and Bayus, 2018). In the early-access crowdfunding variant, however, the funding distribution is more constant, as the funding period is open throughout the development process. As a result, the information asymmetry for an early-access project depends not only on the initial up-front disclosure but also on the quality and frequency of intermediate product updates on the progress of the project’s development. This characteristic of the early-access model provides one of the first opportunities to study the effects of re-porting frequency in a reward crowdfunding scenario. While not a natural experiment, there is quasi-exogenous variation in the reporting frequencies both across projects and over time. This variation is driven by the new and unexplored nature of the early-access model combined with the inexperience of the entrepreneurs, this results in a wide range of reporting frequencies being explored. Combining this quasi-exogenous variation with a wide array of control variables and project fixed effects enables me to evaluate the unconfounded impact of higher reporting frequencies.

(28)

increased difficult to raise future funding, to entrepreneurs when they provide updates at a higher frequency that are not up to par with the expectations of the funders. For each entrepreneur there will be an appropriate disclosure frequency where these advantages and disadvantages are balanced, however, these entrepreneurs are unlikely to know what this frequency is and will thus experiment with various frequencies in a quasi-exogenous way.

2.2.3. Market Pressure

Another primary determinant of the crowdfunding investment decision, besides disclo-sures by the entrepreneur, is “herding” behavior based on the actions of incumbent funders (Hildebrand et al., 2017). Courtney et al. (2017) find that the sentiment of incumbent funders is one of the primary determinants of the likelihood that new fun-ders will opt to fund a Kickstarter project. Furthermore, Hornuf and Schwienbacher (2018) and Hildebrand et al. (2017) show that similar behavior occurs in both equity crowdfunding and lending-based crowdfunding settings. This herding behavior is rele-vant because it gives the entrepreneur a strong incentive to care about communications from incumbent funders, as they are a primary determinant of his or her ability to attract future funding. This is particularly relevant in the early-access crowdfunding model because the entrepreneur’s development decisions are expected to be partially driven by pressure imposed by the market (i.e., the incumbent funders). I refer to this phenomenon as “market pressure” and define it as “communications from funders to the entrepreneur that are intended to steer the entrepreneur’s decisions in a specific direction”.

Market pressure could have either a positive or a negative effect on the decision making quality of the entrepreneur. From the perspective of moral hazard, market pressure might improve the decision making quality. As described in the seminal work of Jensen and Meckling (1976), agency frictions can be resolved if the market has a way to mon-itor and control the behavior of the manager. Communications (i.e., a “voice”) from the market to the entrepreneur can operate as a governance mechanism, as it helps to align the market’s interest with that of the entrepreneur (Armstrong et al., 2010). Several papers empirically document this in the traditional capital market by showing that communicating with management constitutes an effective instrument for moni-toring and controlling the manager (Brav, Jiang, Partnoy, and Thomas, 2008; Dimson, Karaka¸s, and Li, 2015; McCahery, Sautner, and Starks, 2016; Harford, Kecsk´es, and Mansi, 2018).

(29)

product development because the time value of money suggests that the up-front in-vestment becomes more expensive the longer the funder must wait for the final product. As a result, a misalignment may develop between the entrepreneur’s decision horizon and the funders’ myopic horizon. In general terms, myopia refers to the overvaluation of short-term outcomes and the undervaluation of long-term outcomes (Fishburn and Rubinstein, 1982). Misalignment stemming from market myopia is relevant because exposure to such myopic preferences can induce the entrepreneur to engage in myopic development decisions. In that case, market pressure driven by market myopia might decrease decision-making quality. Such market myopia, when reporting on the progress of product development, is also observed in the context of R&D announcements by bio-pharmaceutical companies (e.g.,Mc Namara and Baden-Fuller, 2007). In a more general sense, this idea of market induced managerial myopia is popularized in the seminal work of Stein (1988), Stein (1989), and Froot, Perold, and Stein (1992).

2.3. Hypotheses Development

The primary objective of this paper is to study how reporting frequency interacts with the level of market pressure imposed on the entrepreneur in a crowdfunding setting. This interaction is relevant to studying early-access crowdfunding because the inter-mediate product updates act as reporting events that are likely to introduce disclosure effects that could either increase or decrease the level of market pressure (e.g.,Gigler et al., 2014; Edmans et al., 2016). I expect the level of market pressure to be a first-order determinant of managerial decision-making in the crowdfunding setting, yet it is not ex ante clear how this market pressure is influenced by the primary disclo-sure choice (i.e., reporting frequency) of an entrepreneur pursuing financing through early-access crowdfunding.

(30)

commu-able information would increase funders’ propensity to strategically communicate with the entrepreneur to steer his or her decision-making. However, as discussed in Demsetz and Lehn (1985) and Bushman et al. (2004) this relationship is more nuanced than that. Demsetz and Lehn (1985) argue that active monitoring by market participants is most important in settings characterized by low transparency. This prediction derives from the fact that there is less need for market participants to exert market pressure in a scenario where the entrepreneur has committed to less self-motivated decision-making by increasing transparency (Demsetz and Lehn, 1985; Bushman et al., 2004; Armstrong et al., 2010). Following a similar logic, in scenarios of low transparency, there is more room for moral hazard and in turn more necessity for market pressure with the purpose of reducing agency frictions. As a result, I predict that higher re-porting frequencies in the crowdfunding setting will reduce funders’ propensity to exert market pressure on the entrepreneur as a way of resolving agency frictions.

However, from the perspective of myopic market preferences, I expect that a higher reporting frequency will increase the level of market pressure imposed on the en-trepreneur. Based on the conditions formalized by Froot et al. (1992), there is an indirect and a direct channel through which short-term oriented market participants can influence the decision making of the entrepreneur. Indirect myopic market pressure occurs when the entrepreneur cares about the public behavior of incumbent funders, seeing it as a primary determinant for future funding (Hornuf and Schwienbacher, 2018). Direct myopic market pressure occurs when market participants alter the sen-sitivity of the entrepreneur to their myopic preferences by varying the frequency with which they communicate their preferences to the entrepreneur (Froot et al., 1992). In a capital market context, various papers empirically confirm these predictions. Jacobson and Aaker (1993) and Asker, Farre-Mensa, and Ljungqvist (2015) both document that market-induced myopia is expected to be larger in scenarios where the manager is more exposed to the market participants’ horizon. Regarding market participants influenc-ing the sensitivity of the manager to their preferences, the results by Bushee (1998) indicate that the institutional investors’ horizon preference influences the amount of R&D a manager will invest in a product.

The aforementioned direct channel of myopic market pressure can be influenced by the reporting frequency in several ways.5 _{The first way works through the effect that}

reporting frequency has on the self-selection of funders with certain horizons. Basic

5 _{Various papers, such as Hermalin and Weisbach (2012), Gigler et al. (2014), and Edmans et al.}

(31)

economic theory suggests that impatient funders prefer a higher update frequency of updates, and vice versa, as short-term oriented funders discount longer update inter-vals more strongly (e.g.,Thaler et al., 1997; Wagenhofer, 2014). Furthermore, it is also possible that the reporting frequency alters the horizon of incumbent funders directly. Experimental evidence suggests that the evaluation frequency (which is a product of the reporting frequency) imposed on individuals alters their risk preference and more specifically their myopic tendencies (Gneezy and Potters, 1997; Thaler et al., 1997; Van Der Heijden et al., 2012). Besides having an effect on the funder horizon, re-porting frequency is also expected to affect funders’ propensity to communicate their preference to the entrepreneur. Each reporting event is expected to trigger a period of heightened attention, temporarily increasing the amount of communication. Theoret-ically, this expectation is one case of the broader “agenda-setting hypothesis”, which predicts that the frequency of reports on an issue determines the level of attention that issue receives (Cornelissen, 2011; Sayre, Bode, Shah, Wilcox, and Shah, 2010; Conway, Kenski, and Wang, 2015). Moreover, higher reporting frequencies expose the funders to more of the outcome variability that increases the likelihood that some of the reports will not meet expectations. Based on the results of Guo et al. (2015) and Casas-arce et al. (2017), I expect funders to asymmetrically overvalue these below expectation reports, which increases their propensity to communicate their dissatisfaction to the entrepreneur.

Main prediction

Combining the effect on market pressure that is driven by agency frictions with the effect on market pressure that is driven by myopic preferences yields conflicting ex-pectations regarding the direction of the relationship between reporting frequency and market pressure. I expect market pressure driven by agency frictions to decrease with higher reporting frequencies, whereas I expect market pressure driven by my-opic preferences to increase with higher reporting frequencies. In a more general sense, this dynamic is similar to the trade-off between the positive effects (reduction in cost of capital) and the negative effects (inducing managerial short-termism) of increased disclosure as modelled by Edmans et al. (2016). To reflect this trade-off, I frame Hy-pothesis 1 in a non directional manner.

(32)

2.3.1. Moderating effects

Reporting quality

Providing information in a timely manner is, as previously discussed, important for es-tablishing the information environment that is necessary for reducing agency frictions. However, the overall quality of the information environment is influenced also by the quality of the reporting events (i.e., the degree to which the expectations of the mar-ket are updated via the reported information). In the capital marmar-ket context, this is often referred to as the trade-off between the timeliness of information and the quality (i.e., the “representational faithfulness”) of information (e.g.,Doyle and Magilke, 2013). Boland, Bronson, and Hogan (2015) and Lee, Mande, and Son (2015) provide empir-ical evidence of this trade-off by showing that accelerated filings are likely associated with lower levels of reporting quality. More generally, Casas-arce et al. (2017) show in a management accounting context that there can also be an upper bound to the reporting frequency, whereby the additional processing costs of more granular infor-mation start to outweigh the benefits of providing timelier inforinfor-mation. Based on this trade-off, I would expect that reporting frequency and reporting quality act as substi-tutes, whereby an increase in the reporting quality weakens the relationship between reporting frequency and market pressure. From the perspective of agency frictions, however, there might also be a complementary relationship between the reporting fre-quency and the reporting quality. This is particularly relevant to discussions about the effects of a change in reporting frequency. Increasing the number of reporting events in combination with increasing the quality of those reporting events is substantially more costly and should thus increase the signaling strength of these reporting events. Since it is not ex ante clear whether the relationship between changes in reporting frequency and changes in market pressure is stronger or weaker when accompanied by a change in the reporting quality, I frame Hypothesis 2 in a non directional manner.

Hypothesis 2. The association between a change in the reporting frequency and a change in the level of market pressure is influenced by changes in the average reporting quality.

Frequency of unverifiable announcements

(33)

Gigler (1994), a market participant will incorporate announcements into their decision-making only if these communications are costly to the entrepreneur. The empirical evidence on this prediction in the capital market context is mixed. For example, the literature on managerial use of social media for additional communications shows that it can sometimes improve the information environment (Chen, Hwang, and Liu, 2013; Blankespoor, Miller, and White, 2014) and sometimes it serve only a strategic purpose (Jung, Naughton, Tahoun, and Wang, 2017). In the crowdfunding context, there is little to no enforcement or regulation, making it difficult for an entrepreneur to credibly commit to unbiased additional announcements. Moreover, Ben-Rephael, Da, Easton, and Israelsen (2017) document that in the context of Form 8-K filings, the decreased information benefits of additional communications are particularly profound for less sophisticated market participants. Based on these analytical and empirical results, I predict that the frequency of an entrepreneur’s unverifiable announcements will not influence the relationship between reporting frequency and market pressure.

Hypothesis 3. The association between reporting frequency and level of market pres-sure is not influenced by the frequency of unverifiable announcements.

2.4. Setting and Empirical Design

2.4.1. Data

I collect publicly available data from the most popular early-access platform, Steam Early Access. The Steam platform is the largest digital distributor of games in the world, with more than 125 million registered users and more than 15 million concur-rent users during peak hours. In 2013, the Steam platform started allowing software entrepreneurs to register their projects for the Steam early-access program. This pro-gram enables users of the platform to crowdfund a video game project by prepurchasing the game in return for early-access to all development versions (i.e., pre-alpha, alpha, and beta) and the final version.

(34)

peri-number of funders for a project is around 80,000, and the mean price is around $19. This yields the conservative estimate that an average project will raise raise around US $1.5 million via this early-access crowdfunding platform.

2.4.2. Construct operationalization

Reporting frequency

A reporting event is defined as an update to the software that is substantial enough to at least partially meet the funders’ expectations (i.e., substantial enough for the funders to update their expectations). I identify a reporting event based on two criteria: (1) there is an announcement on the official project page containing at least one keyword referencing a new version or update, and (2) the reporting event is accompanied within three days by a change in the file size of the software.6 _{The first criterion guarantees}

that the reporting event is visible to funders and a change in the software’s file size guarantees that the reporting event has substance. Data on these announcements and reporting events is gathered directly from the Steam API and from the project news feed on the Steam platform. The reporting frequency is operationalized by the number of reporting events in a time interval.

One concern with this approach is that each product update will not only adjust the funders’ beliefs about the development progress but also provide funders with a utility from receiving new product features. The primary mechanism of interest in this study is the reporting effect (i.e., the reduction in information asymmetry) and not the utility effect. To rule out the alternative explanation that market pressure is associated with the number of product updates owing to the utility effect, I use a mediation model. In this mediation model, I use the number of negative reviews that the project receives during a given period as a proxy for the utility effect. Any extra utility derived from additional product updates should be strongly correlated with a reduction in the number of negative product reviews. This mediation analysis approach allows me to separate the relationship between the frequency of product updates and market pressure into the reporting effect and the mediated effect that flows through changes in product satisfaction.

Market pressure

Measuring the level of market pressure is empirically challenging. In the early-access setting, the primary channel for funders to communicate with the entrepreneur is through the entrepreneur’s social media accounts, in particular his or her Twitter

6 _{The keywords used are: “version”, “release”, “patch”, “change log”, “changelog”, “update”,}

(35)

account. I utilize this characteristic of the setting to develop a new measure for market pressure based on the subset of tweets received by an entrepreneur during a given period that are plausibly imposing pressure. I consider a tweet to contribute to the market pressure experienced by the entrepreneur when it fulfills two criteria: (1) the tweet must relate to the current and/or future state of the project, and (2) the tweet must contain a direct and/or an indirect request. An example of a direct request would be a funder asking the entrepreneur to focus their development attention on a particular aspect of the project, whereas an example of indirect request would be a complaint or a general inquiry about the project’s development status. The second criterion is of particular importance, as a tweet without an explicit or implicit request is unlikely to steer the entrepreneur’s decision-making in a specific direction, which is a key aspect of my theoretical definition of market pressure. I provide a small selection of market pressure Twitter messages in appendix A.

I classify tweets based on both criteria using an objective procedure that involves training a machine learning algorithm on a large sample of tweets manually classified by Amazon Mechanical Turk (i.e., M-Turk) workers. I manually identify the Twitter account of each entrepreneur, and all Twitter messages to and from these accounts are gathered in a programmatic manner from the Twitter API and the Twitter advanced search functionality. This yields a sample of roughly 220,000 tweets directed toward the entrepreneurs. A training sample of 5,000 tweets was randomly selected and classified by M-Turk workers to identify whether each tweet relates to the development of the project and whether it contains an explicit or an implicit request. Each M-Turk worker was paid US $0.20 for every 5 tweets they classified. In order to validate the accuracy of this procedure, I also classify 1,000 tweets manually and cross-validate this sample with the M-Turk classification. This cross-validation shows a disagreement of between 3% to 4%, which provides reassurance that the M-Turk workers understand the task and are properly incentivized. A detailed description of the classification instrument and procedure is available in appendix B.

(36)

Moderating variables

I measure the reporting quality of a reporting event based on the length of the an-nouncement posted on the official project page. This approach is similar to the ap-proaches used by Cascino et al. (2018) and Madsen and McMullin (2018). I opera-tionalize the length of an announcement using a tokenizer and counting the number of tokens (i.e., words) in the announcement. I chose this method because the number of tokens in the announcement message is more easily compared across different projects and reporting events that are alternatives such as the file size change of the prod-uct update. I operationalize the frequency of additional unverifiable announcements by identifying the number of announcements on the official project page that are not accompanied by a new product update.

Control variables

I expect that a period characterized by high levels of dissatisfaction among funders is likely to increase market pressure. To control for this, I include a variable that accounts for the number of negative reviews posted on the project page for a given period. Changes to the funderbase of a project can also influence the level of market pressure imposed on the entrepreneur. To account for this, I include a control variable based on the change in ownership for a given period. Because I expect the general attention a product receives on social media to influence the project’s visibility, I include a control variable based on the number of YouTube videos posted about a project during a given period. Because I expect the level of active users of the software being developed to influence the likelihood that a funder will decide to express concerns to the entrepreneur, I include a control variable to account for the median playtime in a given period. Finally, I expect an entrepreneur that is more active on social media to attract more market pressure via social media, because their social media becomes a more effective communication channel for funders wanting to influence the entrepreneur. To control for this, I include the number of tweets from the entrepreneur in a given period.7

2.4.3. Empirical models

Hypothesis 1

To test Hypothesis 1, namely, the association between the reporting frequency and the level of market pressure, I use the following model:

7 _{The various statistics required to calculate these control variables are collected from the Steam}

(37)

M arketP ressurei,t = β0+ β1RepF reqi,t−1+ β {Controls}i,t−1

+ P rojectF E + T imeF E + (2.1)

where i indicates the project, t indicates the period, M arketP ressurei,t indicates the

number of market pressure tweets directed to the entrepreneur for project i and period t, RepF reqi,t−1 indicates the number of reporting events for project i in period t− 1,

P rojectF Eis a dummy variable for indicating the project i, and T imeF E is a dummy variable for controlling for life-cycle fixed effects t. The primary coefficient of interest is β1. Given that M arketP ressurei,t is a count variable, all models are estimated with

both an OLS and a Poisson regression.8

This model is similar to the model employed by Fu et al. (2012), the difference being that I use a different period length. The length of a period in Fu et al. (2012) is 12 months, which is not appropriate for the early-access crowdfunding setting, as the reporting frequency for a given project tends to change within such a long period. Finding the appropriate period length is an important empirical choice, as a period length that is too long will absorb most of the variation, whereas a period length that is too short will result in a noisy estimation. My decision criterion for the period length is to pick a length as long as possible up to the point where the estimation starts to lose power. I estimate the main model for various period lengths ranging from 2 weeks to 16 weeks in 2-week increments, and the results are displayed in Graph 1. The results indicate that the model starts to lose statistical power at a period length of around 12 weeks. I therefore choose a period length of 10 weeks for my analyses.

There is substantial variation in the level of market pressure across the various projects and periods. To study how changes in the reporting frequency are associated with changes in market pressure, I use a changes version of the first model, wherein each variable is corrected for the average value in the preceding three months:

∆M arketP ressurei,t = β0+ β1∆RepF reqi,t−1+ β {∆Controls}i,t−1

+ T imeF E + (2.2)

Moderating effects

(38)

∆M arketP ressurei,t = β0+ β1∆RepF reqi,t−1+ β2∆RepQuali,t−1+ β3∆Interactioni,t−1

+β _{{∆Controls}}i,t−1+ P rojectF E + T imeF E +

(2.3) To test Hypothesis 3, namely, the effect of the frequency of unverifiable announcements on the association between reporting frequency and market pressure, I use the following model:

M arketP ressurei,t = β0+ β1RepF reqi,t−1+ β2F reqAddAnni,t−1+ β3Interactioni,t−1

+β _{Controls}i,t−1+ T imeF E +

(2.4) where i indicates the project, t indicates the period, M arketP ressurei,t indicates

the number of market pressure tweets directed toward the entrepreneur for project i and period t, RepF reqi,t−1 indicates the number of reporting events for project i and

period t_{− 1, RepQual}i,t−1indicates the average announcement length i in period t− 1,

F reqAddAnni,t−1 indicates the number of additional unverifiable announcements i in

period t_{− 1, P rojectF E is a dummy variable to indicate the project i, and T imeF E} is a dummy variable to control for life-cycle fixed effects. The primary coefficient of interest is β3.

2.5. Results

2.5.1. Descriptive statistics

Table 1 provides detailed descriptive statistics for the observations I use to test the various models. Table 1 Panel A relates to the sample used for models 1, 4, and 5. Table 1 Panel B contains the descriptive statistics for the changes specification used for models 2 and 3. The main sample consists of 883 observations for 144 projects, each observation being a 10-week project period. Owing to the nature of a changes specification, the sample described in Table 1 Panel B consists of 449 observations, each observation being the change relative to the average of the previous 3 periods.9

9 _{Inferences remain similar when calculating the changes using only the previous period, however}

(39)

The primary dependent variable M arketP ressure is the number of market pressure tweets received in a 10-week period. Table 1 Panel A shows that there is substantial variation in the level of market pressure across different periods. At the median level, an entrepreneur receives 16 market pressure tweets in a 10-week period, with a mean level of 56 tweets. This translates to around 2 tweets per week at the median level (6 at the mean). Some project periods receive a very high number of tweets; the entrepreneur in the period with the most tweets received 2,241 tweets during the 10-week period. The descriptive statistics in Table 1 Panel B show that the level of market pressure varies substantially across periods of the same project. The average median change is 3 tweets, which corresponds to 17% of the period median number of tweets.

The primary independent variable RepF req is the number of reports issued by the entrepreneur in a 10-week period. As expected, Table 1 Panel A shows that the re-porting frequency varies across project periods. On average, an entrepreneur will issue between 1 and 2 reports per 10-week period. In the span of a year, this translates to 7 to 8 reports on average, or an average of 1 report every 7 weeks. Some entrepreneurs, however, choose to report much more frequently, up to as much as once a week. There is also variation in the reporting frequency between periods of the same project. As displayed in Table 1 Panel B, some entrepreneurs choose to increase or decrease their frequency by as much as 5 reports compared to the previous three periods.

Table 1 Panel A shows, with regard to the other variables, that a project receives, on average, 48 negative reviews in a 10-week period but that this number varies strongly between project periods. An entrepreneurs’ Twitter account sends, on average, 88 tweets during each 10-week period, which confirms that Twitter is a relevant and active channel for two-way communication between the entrepreneur and his or her funders. Finally, each project receives, on average, 79,000 new funders, 11 YouTube videos, and a median per-funder playtime of 20 hours per week at the high end of the distribution.10 _{Regarding the moderating variables, an average reporting event}

is accompanied by an announcement that is, on average, 284 words long and has a standard deviation of 410 words, and an entrepreneur creates, on average, 4 to 5 unverifiable additional announcements per 10-week period.

[Table 1 Panel A and Table 1 Panel B about here] specification.

10 _{The actual median playtime will be lower as the data source only collects playtime data for the}

(40)

2.5.2. Main results

Table 2 shows the results for my main models: Table 2 Panel A corresponds to the results for model 1, and Table 2 Panel B corresponds to the results for model 2. The level of market pressure is a count variable, so model 1 is estimated using both Poisson and OLS regressions. Panel A of Table 2 contains the following columns. Column 1 and Column 2 contain the full model without fixed effects, tested using Poisson and OLS regressions. Respectively; columns 3, 4, and 5 contain the various combinations of fixed effects estimated using Poisson regressions. Finally, Column 6 contains the full specification with both project and life-cycle fixed effects estimated using an OLS regression. Panel B of Table 2 contains the following columns. Column 1, 2, and 3 contain the main changes model with various combinations of control variables excluding any fixed effects. Column 4 contains the full changes model with project life-cycle fixed effects. All models are tested with standard errors that are corrected for heteroscedasticity and are clustered at the project level.

The results of Table 2 Panel A indicate that there is a negative association (-0.059 p = 0.065 — -2.65, p = 0.05) between the number of reporting events in a period and the number of market pressure tweets that the entrepreneur receives. This result provides support for my first hypothesis, which states that there is an association between reporting frequency and the level of market pressure. Furthermore, the direction of the association is consistently negative throughout the various specifications, which indicates that in the early-access crowdfunding setting, the reporting frequency is primarily associated with market pressure that is driven by agency frictions. Inferences remain similar when all the control variables, project fixed effects, and project life-cycle fixed effects are included. This result is economically meaningful, as one more reporting event in a 10-week period is associated with, on average, 3 fewer market pressure tweets, or a reduction of 17% relative to the median number of market pressure tweets. A similar result is obtained when using the stricter changes specification displayed in Table 2 Panel B. The results of the changes model indicate a similar negative association (-4.74, p = 0.05) between a change in the number of reporting events and a change in the level of market pressure. Reducing the number of reports in a 10-week period by 1 report is associated with a reduction in market pressure of around 5 tweets, which corresponds to a reduction of 29% relative to the median number of market pressure tweets.

(41)

mediation model wherein I use the number of negative reviews that the project receives during a given period (i.e., the product satisfaction) as a proxy for the utility effect. I run this mediation model for both the levels and changes specifications; the results are displayed in Figure 2 and Figure 3. The results in Figure 2 indicate that product satis-faction is not significantly associated with the reporting frequency (0.63, p = 0.60) but is associated with the level of market pressure (0.23, p = 0.000). More importantly, the effect of reporting frequency on market pressure that is explained through the effect on product satisfaction is not significant (0.14, p = 0.60). Similar inferences are obtained when running the mediation model for the changes specification; the indirect effect ex-plained through the effect on product satisfaction is not significant (-0.02, p = 0.94). These mediation results provide reassurance that the negative association between the reporting frequency and market pressure is plausibly driven by the reporting effect of these product updates.

[Table 2 Panel A and Table 2 Panel B about here]

2.5.3. Moderating results

Table 3 shows the results for model 3, which is used to test whether the association be-tween a change in the reporting frequency and a change in the level of market pressure is moderated by the average reporting quality of the reporting events. The coefficient for ∆N umber of Reportst−1 is no longer statistically significant (-3.65, p = 0.18). The

coefficient for ∆Reporting Qualityt−1 is not statistically significant (-0.01, p = 0.26),

which indicates that a change in the reporting quality is by itself not associated with a change in the market pressure. The interaction term (-0.01, p = 0.04) indicates that the average reporting quality amplifies the association between reporting frequency and market pressure. Given the insignificance of ∆Reporting Qualityt−1 this result

suggests that a change in reporting frequency only affects the market pressure when it is also accompanied by a change in reporting quality. Simultaneously increasing the quality of an average report with 100 words (35% of mean number of words) increases the association between reporting frequency and market pressure by an average of 1 tweet (around 33% of the main effect size, which is 3 tweets). This result is consistent across all specifications and confirms a complementary relationship between reporting quality and reporting frequency. Overall, these results support my second hypothesis that changes in the reporting quality amplify the association between a change in the reporting frequency and changes in the level of market pressure.

(42)

number of unverifiable additional announcements in the same period. The coefficient for N umber of Reportst−1 remains negative and statistically significant (-3.67, p =

0.05) but does increase in magnitude compared to the results in Table 2 Panel A. The coefficient for N um Announcementst−1 is not statistically significant (-1.17, p =

0.15). This indicates that, by itself, having more extra announcements is not associated with lower market pressure. The interaction term is not significant (0.15, p = 0.38). These results suggest that the frequency of unverifiable additional announcements does not influence the association between the reporting frequency and the level of market pressure.

[Table 4 about here]

2.6. Additional analyses

2.6.1. Effect of recently joined funders

An interesting phenomenon in the crowdfunding setting is the large fluctuation in the amount of newly joined funders across periods. The presence of a large number of newly joined funders might have implications for the relationship between reporting frequency and the level of market pressure. On the one hand, it is possible that these newly joined funders will exhibit a higher than average amount of attention, which will increase the scrutiny placed on the entrepreneur. The analytical model of Hirshleifer and Teoh (2003) highlights that the presence or absence of attention from market participants has important implications for the degree to which, and if so how, they react to reporting events. Based on this “market distraction hypothesis”, I expect that periods with a large number of newly joined funders will exhibit a stronger negative relationship between reporting frequency and market pressure. However, Porter and Smith (1995) also suggest that market participants with limited or no experience might not be able to come to rational expectations, with the result that these inexperienced market participants exhibit a form of “temporary myopia”. Newly joined funders, for example, might have irrationally high expectations for the product updates, which would increase their propensity to impose market pressure on the entrepreneur. This “temporary myopia hypothesis” suggests that the presence of more newly joined back-ers will weaken the negative relationship between reporting frequency and market pressure.

(43)

own-ership statistics via the Steam API and various community websites and operationalize it as the average percentage change in ownership for the three 10-week periods before the reporting period. Using this new variable, I test the following model:

M arketP ressurei,t = β0+ β1RepF reqi,t−1+ β2N ewF underT rendi,t−1

+β3Interactioni,t−1+ β {Controls}i,t−1+ T imeF E +

(2.5)

The results of model 5 are shown in Table 5. Owing to how the N ew F under T rendt−1

variable is created, the number of observations drops to 645, but the results in Col-umn 4 show that the main result still holds (-5.17, p = 0.06). ColCol-umn 3 shows that N umber of Reportst−1 remains negative and statistically significant (-9.18, p= 0.004)

when N ew F under T rendt−1 is included, but the magnitude is substantially higher

than the coefficient in Column 4. The coefficient for N ew F under T rendt−1 is

sta-tistically significant and negative (-16.38, p = 0.01), indicating that the presence of recently joined funders is by itself associated with lower levels of market pressure. This relationship suggests that an average increase in ownership of 10% across the previous three periods is associated with 2 fewer market pressure tweets. The interaction term is positive and statistically significant (5.39, p = 0.02) throughout all specifications, which implies that the negative relationship between reporting frequency and market pressure is lower during periods with a large presence of newly joined funders. Untab-ulated results show that the inferences remain the same when Poisson regressions are used instead of OLS regressions. In summary, this result suggests that newly joined funders temporarily exhibit a higher degree of myopia, which pushes the relationship between reporting frequency and market pressure in the positive direction owing to how higher reporting frequencies are expected to interact with the myopia of the mar-ket participants.

2.6.2. Cross-sectional splits

2.6.2.1. Effect of variation in monitoring incentives

(44)

This is not the case for multiplayer projects, which become useless without an active user base. As such, I expect that the funders of multiplayer projects will have stronger incentives to provide monitoring than will funders of single-player projects. Columns 1 and 2 of Table 6 show the results of a comparison between single-player and multiplayer projects. I determine whether a project is single-player or multiplayer based on the tags provided on the project page. The results show a significantly negative reporting frequency effect for multiplayer projects (-3.40, p = 0.03) but no significant effect for singleplayer projects (-0.57, p = 0.72). This finding indicates that the result is driven primarily by funders of multiplayer projects, as they have stronger incentives to monitor the entrepreneur and thus are more likely to respond to the frequency of reporting.

2.6.2.2. Effect of funders’ free-riding propensity

A potential downside to the crowdfunding model is that effective monitoring might be difficult owing to widespread free-riding behavior among funders (Belleflamme et al., 2015). Kuppuswamy and Bayus (2018), for example, document that Kickstarter projects with a large amount of initial support receive less additional funder support in later stages of the campaign. Based on this prior evidence, I expect that the relation-ship between the reporting frequency and the level of market pressure will be weaker for projects whose funders are more likely to exhibit free-riding behavior. I develop a proxy for cross-sectional variation in the free-riding propensity based on the amount of funding raised during the first two periods of the project (i.e., the first 20 weeks). Columns 3 to 5 of Table 6 contain the results for projects with varying levels of initial success. The results show a significant reporting effect for projects with an average initial funding success (-2.76, p = 0.04) but an insignificant effect for projects that have either low (0.20, p = 0.78) or high (-4.85, p = 0.32) initial funding success. The insignificant coefficient for high initial success projects is consistent with prior evidence that free-riding behavior can impede the effectiveness of monitoring in crowdfunding markets.

Essays on reporting and information processing

Essays on reporting and information

processing

Promotiecommissie:

Acknowledgments

Contents

List of Figures

List of Tables

Chapter 1

Chapter 2

Reporting Frequency and Market

Pressure in Crowdfunding

2.1.

Introduction

2.2.

Background and Literature Review

2.2.1.

Crowdfunding

2.2.2.

Disclosures

2.2.3.

Market Pressure

2.3.

Hypotheses Development

2.3.1.

Moderating effects

2.4.

Setting and Empirical Design

2.4.1.

Data

2.4.2.

Construct operationalization

2.4.3.

Empirical models

2.5.

Results

2.5.1.

Descriptive statistics

2.5.2.

Main results

2.5.3.

Moderating results

2.6.

Additional analyses

2.6.1.

Effect of recently joined funders

2.6.2.

Cross-sectional splits

2.6.3.

Market Pressure Heterogeneity and LDA