Eponymous Entrepreneurs⇤
Sharon Belenzon† Aaron K. Chatterji‡ Brendan Daley§
Fuqua School of Business Duke University
November 11, 2014
Abstract We demonstrate that firm eponymy—the familiar convention of firms being named after their owners—is linked to superior performance. We propose a novel ex- planation, referred to as “utility amplification,” and develop a corresponding signaling model. The model generates three main empirical predictions: (1) The incidence of eponymy will be low; (2) Eponymous firms will outperform other firms; (3) These e↵ects will be intensified when the entrepreneur’s name is rare. Using unique data on over 485,000 firms from Europe and the United States, we find support for all of these predictions. Several extensions and robustness checks are considered.
Keywords: Entrepreneurship, Signaling, Firm Names
⇤The authors thank seminar participants at Harvard, MIT, Stanford, Northwestern, NBER, UCLA, and UVA for their useful comments and suggestions. †[email protected] ‡[email protected] §[email protected] 1 Introduction Many firms are eponymously named; that is, they bear the name(s) of their founding owner(s). Leveraging a unique dataset, this paper demonstrates that eponymy is linked to superior firm performance. For instance, controlling for other characteristics, eponymous ventures generate, on average, a 3.2 percentage-point higher return on assets (ROA), which is approximately one-third the magnitude of the sample mean ROA. Further, and perhaps counterintuitively, we propose that non-pecuniary considerations may be a large driver of the eponymy-performance relationship. Succinctly put, we propose that eponymy creates a stronger association between the entrepreneur and her firm that amplifies the utility or disutility of favorable or unfavorable market outcomes, respectively. We refer to this phenomenon as “utility amplification,” and intend it to capture any non- pecuniary benefits (or costs) to the entrepreneur of having a favorable (or unfavorable) impression of her firm tied more closely to herself.1 Consequently, high-ability entrepreneurs are more drawn to eponymy than are low-ability ones. While we believe this explanation to be novel, it clearly fits in a recent literature arguing that non-pecuniary considerations likely play a significant role in entrepreneurship. Hamilton (2000) and Moskowitz and Vissing-Jørgensen (2002) propose that there are likely substantial non-pecuniary benefits to entrepreneurship, given the large wage di↵erential between self- employment and paid employment and that returns to private equity (where entrepreneurs typically invest their private holdings in a single company) are no greater than public equity. Further, Pugsley and Hurst (2012) report survey evidence that non-pecuniary motivations are major drivers in entrepreneurial decision-making and could account for why the majority of small businesses do not grow. In a similar vein as our results, they find that entrepreneurs who report non-pecuniary motives have slightly higher survival rates than those who report pecuniary motivations. To formalize this explanation, we introduce a model, characterize its equilibrium, and derive its testable implications. We then demonstrate that the evidence is consistent with the predictions of the theory, using novel data on over 420,000 companies and 5.5 million individuals from Europe and a smaller dataset of over 60,000 firms from the United States.
Eponymous Firms Selecting the name of their firm is an important and highly visible choice all new-business owners must make. As a Wall Street Journal article memorably stated, “For entrepreneurs,
1While our primary interpretation is that these benefits/costs are non-pecuniary, more generally they could be any benefits/costs distinct from firm-performance measures. For example, future business or em- ployment opportunities for the entrepreneur may be influenced by this impression (see Section 2).
1 the importance of picking the right name for a company may rank second only to naming a child. (And it’s a lot more expensive to change.)”2 Not only is there a proliferation of practitioner guides for choosing a business name, an entire industry of naming consultants exists solely for this purpose. Interestingly, one point of seeming consensus among them is that naming the firm after the owner is not advisable because doing so indicates a lack of creativity and reduces resale value.3 Nevertheless, eponymy is certainly a familiar naming strategy. Well-known examples in- clude Dow Chemical, Gucci, Guinness, Hewlett Packard, Hess, Johnson & Johnson, Kroger, Porsche, Proctor & Gamble, Ryanair, Walgreens, and many others. In our dataset, approxi- mately 13% of firms bear the name of their majority owner. Although not all entrepreneurs think carefully about the name of the firm (Hewlett and Packard reportedly flipped a coin to see whose name would come first4), other naming stories indicate that names, particularly eponymous names, matter. Marvin Bower of McKinsey & Company explicitly chose not to put his own name on the firm after the death of Mr. McKinsey. He noted, “I didn’t want anybody dictating to me how I was going to spend my time. So I had no interest in calling it Bower & Co., or even McKinsey-Bower. I wanted my freedom.”5 To our knowledge, previous academic research has not addressed this well-known naming strategy.
Eponymy, Utility Amplification, and Firm Performance Our model builds on a traditional signaling framework with some important variations. The entrepreneur can engage in activities, eponymy being the prime example, that a↵ect the degree to which the firm is associated with the entrepreneur herself. Our key, novel assumption is that higher levels of signaling activity (i.e., a stronger association between the firm and the entrepreneur herself) are not directly costly, but instead “amplify” the utility or disutility of the favorable or unfavorable market outcomes, respectively. In Section 2 we further discuss the interpretation of this assumption as well as its support in the prior literature. Intuitively, this amplification e↵ect makes eponymy a more attractive strategy for high-ability entrepreneurs, who expect better market outcomes, than for low-ability ones. We demonstrate that our model has a unique stable equilibrium, and that it is partial
2Bounds, Wendy. “How to Choose a Company Name: A 12-Point Test,” Wall Street Journal, June 5, 2008 (http://blogs.wsj.com/independentstreet/2008/06/05/how-to-choose-a-company-name-a-12-point-test). 3For example, please see http://www.businessnamingbasics.com/namedevelopment/naming-business- oneself-easy/ (last accessed November 11, 2014). 4Burrows, Peter. “Hewlett & Packard: Architects of the Info Age,” BusinessWeek, March 28, 2004 (http://www.businessweek.com/stories/2004-03-28/hewlett-and-packard-architects-of-the-info-age) 5Huey, John. “How McKinsey Does It,” Fortune, November 1, 1993.
2 pooling when high-ability entrepreneurs are relatively rare (as appears to be the case in our data). In terms of eponymy, high-ability entrepreneurs engage in it, while low-ability ones mix between doing so and not. In equilibrium, low types trade o↵the boost in perception that eponymy (i.e., pooling with high types) brings, with the amplified disutility in the (likely) event of unfavorable market outcomes. The model then makes three main empirical predictions: (1) The incidence of eponymy will be low; (2) Eponymous firms will perform better than other firms; and (3) These e↵ects will be intensified when the entrepreneur’s name is rare. We believe (3) to be a particularly discriminating test of our model: for entrepreneurs with rarer names, the link between eponymy and performance will be stronger, but they will be less likely to engage in eponymy. While other ex-ante plausible explanations for a link between eponymy and performance exist, explaining this potentially counterintuitive pattern is a more di cult task. We find empirical support for all three predictions. First, we confirm that eponymy is indeed rare in our large sample of public and private firms across several nations. Sec- ond, we empirically document an association between eponymy and higher profitability and higher returns on assets. Third, we find the relationship between eponymy and performance is strongest in rare names (e↵ectively disappearing in common names), but also that the incidence of eponymy is lowest with rare names, exactly as our signaling model predicts. Furthermore, we conduct several additional analyses designed to reveal whether the mechanism we propose is plausible. First, since our argument is based on eponymy sig- naling privately known ability, it is encouraging that we find our empirical results to be strongest for young firms, of which likely little is known. Second, our theory predicts that the eponymy-performance relationship should be strongest in industries with greater perfor- mance dispersion and in industries where market information more accurately reflects the entrepreneur’s skills, both of which we find empirical support for. Finally, we check the sensitivity of our main results with numerous robustness checks, including accounting for di↵erences in ownership structure, name switching, ownership changes, serial entrepreneurs, and the ethnic background of business owners.
Broader Implications for the Study of Entrepreneurship Beyond the specifics of our findings, this work may have broader implications for the study of entrepreneurship. New and small firms are often closely held and rely primarily on the contributions of their founders. Yet founders vary in their underlying ability as well as in their commitment to the enterprise. These attributes constitute important information for customers, investors, and suppliers, but are generally di cult to observe. Our model
3 formalizes a connection between firms and founders and explains how firm names can operate as a credible signal of ability (though not a perfectly revealing one). Future work could build on this model to better understand how other indicators may shed light on the quality and commitment of entrepreneurs. Additionally, while other characteristics associated with superior entrepreneurial perfor- mance have been documented, they rarely represent choices available to all entrepreneurs.6 Hsu and Ziedonis (2012) point out that despite a growing literature on entrepreneurship, there is scarce evidence on strategic actions new firms can actually undertake to gain a foothold in the market. This work, however, explores a choice that, presumably, every en- trepreneur faces: selecting the name of the firm. Not only is the naming choice universal, it is arguably a decision the owner is most likely to make on her own (as opposed to later decisions in which investors and employees may be involved). Finally, the naming decision is also unlikely to directly influence the underlying ability of the entrepreneur and the quality of her idea. These characteristics make the name of the firm particularly interesting for the study of entrepreneurship.
The remainder of the paper is organized as follows. The next subsection discusses the related literature. Section 2 presents and analyzes our signaling model (proofs are located in Appendix A). Section 3 provides an overview our data and the econometric specification. Section 4 tests the predictions of the model. Section 5 contains robustness checks. Section 6concludes.
1.1 Related Literature This paper contributes to three areas of the literature: entrepreneurship, signaling, and names (and their intersections). Scholars of entrepreneurship have observed that information asymmetries are especially pernicious for new business ventures. For example, an entrepreneur’s new business will likely have di culty attracting investment and customers if no mechanism is available by which she can transmit information about her ability/quality (Shapiro, 1983; Amit, Glosten, and Muller 1990; Shane and Cable 2002). In this paper, we propose that eponymy acts a strategic choice of the entrepreneur that ameliorates some of this informational asymmetry. Specifically, eponymy functions as a signal (a la Spence 1973, 1974; Milgrom and Roberts 1986; Bagwell and Riordan 1991) of the entrepreneur’s quality or skill. In terms of technique,
6Examples include prior industry experience (e.g., Chatterji 2009) and prominent a liations (Stuart, Hoang, and Hybels 1999).
4 the model builds on tools developed in Daley and Green (2014). Within the signaling literature, our theory is distinguished by the manner in which the signal (i.e., eponymy) a↵ects the payo↵s of the sender (i.e., the “utility-amplification” assumption). There is an academic literature on names across several di↵erent disciplines (Bertrand and Mullainathan 2004; Fryer Jr. and Levitt 2004; Simcoe and Waguespack 2011), and on firm names in particular (Ingram 1996; Tadelis 1999, 2002; Lee 2001; Glynn and Abzug 2002). In common with the present paper, some of this work also explores how variation in naming strategies is related to firm performance. For example, Ingram (1996) argues that hotel chains that name all of their properties after the corporate parent (e.g., Marriott) do so to establish a credible commitment to quality service, and finds that firms using this naming strategy survive longer. In perhaps the most related paper to ours, McDevitt (2013) builds and tests a model in which plumbers name their firms to signal quality. However, the preferences of both senders and receivers leading to his results are quite di↵erent from ours. Specifically, low-quality plumbers use names that appear early in the alphabetized list to attract customers who are unwilling to invest much time in searching for the right contractor, whereas high-quality plumbers rely on repeat business and word-of-mouth. In addition, McDevitt (2013) uses all (approximately 2,300) plumbing firms in Illinois as of 2008 for its main analysis. In contrast, we employ data on over 485,000 firms from Europe and the U.S. across a wide variety of industries. In a paper on individual names, Fryer Jr. and Levitt (2004) endeavor to explain striking empirical patterns in distinctively black names in the United States. They consider whether signaling theory could be applied. For example, they posit that giving your child a dis- tinctively black name could potentially be a costly signal to others about your a nity for the black community. They argue that the empirical results are more consistent with other explanations, such as those related to social identity. A significant portion of the literature concerns itself with name changes and the market for names. Tadelis (1999, 2002) describes how the market for names arises, sustains, and provides long-term incentives throughout the business owner’s career, a fundamental issue in the literature on career concerns (e.g., Fama 1980; Kreps 1990; Holmstr¨om 1999). The logic in these models relies on the key assumption that the firm’s name and the owner’s identity can be separated. In our study, some owners choose to explicitly attach their identity to the firm’s name, which would likely have new implications in Tadelis’s models, a topic we leave for future work. Other papers have documented strategic reasons for name changes. McDevitt (2011) finds that low-performing plumbing firms are more likely to change their names, whereas a
5 set of papers in the marketing and finance literatures explore whether firm name changes create value (for a summary, see Cooper, Dimitrov, and Rau 2001). Some evidence suggests names can influence stock prices (Horsky 1987; Cooper et al. 2001; Lee 2001) and mutual fund inflows (Cooper, Gullen, and Rau 2005). In sum, we aim to add to these literatures by both building a novel model and leveraging a unique dataset to explain and measure the incidence of eponymy, and its link to firm performance, across hundreds of thousands of public and private firms.
2 A Signaling Model of Eponymy as Utility Amplification
In our signaling model, the sender is an entrepreneur with type ✓,whichiseitherH or L (mnemonic for high or low). To fix language, we refer to ✓ as the ability of the entrepreneur and also the quality of her firm. For specificity, the receivers are potential customers, though one could consider the larger set of market participants that form an impression about the firm’s quality, including suppliers, lenders, competitors, regulators, etc.7 We use the term the market synonymously with receivers.Theentrepreneurprivatelyknows✓ and receivers share a commonly known prior µ =Pr(✓ = H) (0, 1). 0 2 We first describe the sequence of the game in full and then remark about our assumptions and their interpretations. The entrepreneur engages in a signaling activity, s, that she selects from the interval [s, s], where 1 s < s.Themarketobservesthechoiceofs,as well as an additional indicator of the firm’s quality, g.Thisindicatorisarandomvariable taking values in l, h according to Pr(g = h ✓ = H)=Pr(g = l ✓ = L)=p ( 1 , 1). { } | | 2 2 Based on its observation of both s and g,themarketformsitsoverallimpressionofthe firm, ↵ =Pr(✓ = H s, g). The entrepreneur’s utility will depend on this impression. Let | V (↵)=↵ (1 ↵), that is, the ↵-weighted convex combination of 1 and 1. Our main, and novel, assumption is that the entrepreneur’s utility scales with her level of signaling: U(s, ↵)=sV (↵). In other words, the choice of higher s increases both the utility of favorable impressions and the disutility of negative ones. We now discuss the modeling choices.8
7For example, an entrepreneur whose firm sells wares intended for conspicuous consumption cares not only about its impression with potential customers, but also about how it is perceived by individuals unable to purchase the good but who are among the “audience” of its consumers. 8For signaling to even be possible, the entrepreneur must have some private information to convey. For both simplicity and keeping with the signaling literature, we refer to ✓ as the entrepreneur’s ability. However, it can be less stark than this. Qualitatively, nothing is altered if the entrepreneur has only an imperfect, but meaningful, initial assessment of her ability—including the possibility that the entrepreneur may be “over-confident” (see Camerer and Lovallo (1999) for evidence and Santos-Pinto (2012) for a model that accommodates this feature).
6 The market impression: That the entrepreneur cares directly about the market impression, ↵,isareduced-formassumptioncommoninthesignalingliteraturestartingwithMailath (1987). We assume that for any chosen signaling level, the entrepreneur prefers a more favorable impression for two reasons. First, the prices the firm can charge, the volume it can expect to sell, and ultimately its profitability and performance are increasing in the market impression. Second (and more novel), there are additional factors, beyond firm performance, for which the sender’s utility is increasing in ↵.Weprimarilyinterpretthesefactorsasnon- pecuniary considerations, as we discuss below, and their impact is accentuated by higher levels of the signal.
Eponymy as utility amplification:Ourmainassumptionisthatthesignalingactivity,s, “amplifies” the utility or disutility of a market impression. The interpretation is that s corresponds to the degree to which the entrepreneur ties the identification of the firm (and its market impression) to her person. One way to unpack the sender’s utility function is:9
U(s, ↵)=V (↵) +(s 1) V (↵) (1) firm level of non-pecuniary performance association factors | {z } | {z } | {z } A simple example of the non-pecuniary factors could be that the founder feels pride/shame if the market believes her (as opposed to an unnamed owner of the firm) to be of high/low ability; according to the adages, an eponymous entrepreneur may “bring honor to her name,” but also risks “besmirching” it.10 More generally, utility amplification is the assumption that the owner of a firm with a favorable market impression would prefer to be more closely associated with the firm, and, conversely, that an owner of a of a firm with an unfavorable market impression would prefer to be less closely associated with the firm. While non- pecuniary factors seem prime candidates for generating such a preference, more market-based explanations include the possibility that the impression of the firm may lead to better or
9The precise functional form of V is chosen for tractability. For a simple example that generates such a V , suppose the firm faces the demand curve Q(P ↵)=ap↵ bP , has a fixed cost of 1, and has zero marginal a2↵ | 2 costs. Then firm profits are 4b 1, which is V when a =8b. Formally, though, results are unchanged for @U any linear V , so long as @s is negative when ↵ is below a threshold and positive when ↵ is above it. For example, nothing is altered by adding a constant to the sender’s utility function if dealing only with positive utility values is preferable. Finally, results would be qualitatively similar if both appearances of V in (1) were replaced with (possibly distinct) bounded, strictly increasing, and di↵erentiable functions. 10There is ample evidence that individuals care directly about others’ impressions, even if these impressions are (arguably) only tenuously linked to back to them. From Leary and Kowalski’s (1990) extensive survey: “However, the fact that people often are concerned with how others perceive them, even when no immediate or future outcomes depend on the impressions they make [emphasis added], suggests that other factors may motivate impression management.” See also Andreoni and Petrie (2004), Bohnet and Frey (1999), Frey (1978), Ho↵man et al. (1994), Ho↵man et al. (1996), and Rege and Telle (2004).
7 worse future business or employment opportunities for the eponymous entrepreneur.11 We have modeled the set of possible signals as an interval, [s, s], with the interpretation that there can be varying degrees to which the impression of the firm is identified with the entrepreneur’s person.12 However, as Proposition 1 formalizes, in equilibrium, the sender will always select either s or s,andthatasimplermodelinwhichsignalingwereabinarychoice, with the clear interpretations of “non-eponymy” or “eponymy,” would yield precisely the same predictions. We feel that, in addition to accommodating more real-world possibilities, allowing for the richer set of signals actually clarifies the model’s properties (see Section 2.1 below) and puts more distance between its assumptions and conclusions.
The additional indicator:Theadditionalindictor,g,standsinforanyadditionalinformation regarding the firm’s quality besides the entrepreneur’s choice of s.Forinstance,inthecase of a new restaurant, this information would include its review in the local newspaper. Of course, this additional information likely does not all appear in one-shot—for example, con- sumer experiences accumulate and word-of-mouth spreads over time. In reality then, market information and firm performance evolve jointly throughout the life of the firm. However, the works of Alos-Ferrer and Prat (2012) and Daley and Green (2014) show that a fully dynamic, infinite-horizon modeling of this phenomenon, while more analytically complex, does not meaningfully alter the incentives for the initial signaling decision. One final feature to note is that in contrast to standard signaling models, the utility of the entrepreneur depends only on s and ↵,andnoton✓.Crucially,however,becausethe market’s impression of the firm depends on g, in addition to s,theentrepreneur’sexpected utility prior to the realization of g does depend on her ability. This fact becomes apparent as we analyze the game below.
2.1 Preliminary Analysis Our solution concept is Perfect Bayesian Equilibrium, henceforth equilibrium.13 In any equilibrium, after s is chosen, but before g is realized, the market updates from its prior,
11The implicit assumption in this argument is that receivers can tell when a firm is eponymously named. Of course, if a firm name is potentially eponymous (i.e., appears to be a personal name), perhaps not all receivers would know whether it is actually eponymous or not. However, we argue that many consumers will know, and more are likely to find out over time. We therefore operate under the assumption that eponymy is identifiable to receivers. Consistent with this, empirically, we find that eponymous firms perform better than those bearing the names of non-owners (see Section 5.1). 12In light of (1), one can interpret s = 1 as no association between the identity of the entrepreneur and the firm (in which case her utility is based only on firm performance), and s as the minimum association between the entrepreneur and the firm. Clearly then, s 1, but they can be distinct if some degree of association is unavoidable. 13Our notion of Perfect Bayesian Equilibrium is that of Fudenberg and Tirole (1991, pp. 331-333).
8 1 1 2 1 2 2 1
1 0 1
1 0 1 1 Α 2 Μ 2 Μ 2
!1 0 !1
!1 !2 !2 2 0 0 0 ! s s s s s s s s s (a) The (s, ↵)-indi↵erence (b) The (s, µ)-indi↵erence (c) The (s, µ)-indi↵erence curves of the entrepreneur. curves of a low-type entrepreneur. curves of a high-type entrepreneur.
Figure 1: Plots of indi↵erence curves for utility/expected-utility levels 2, 1, 0, 1, 2 using 3 3 { } parameters p = 4 ,s= 2 , and s =7.
µ ,toaninterim belief, denoted µ(s)=Pr(✓ = H s). Then, upon observing g,themarket 0 | updates its belief from µ(s)to↵(µ(s),g). Notice that, given any interim belief µ,thesecond update is merely a straightforward application of Bayes rule:
µp µ(1 p) ↵(µ, h)= and ↵(µ, l)= . (2) µp +(1 µ)(1 p) µ(1 p)+(1 µ)p It follows that, given arbitrary values for s and µ(s), the entrepreneur’s expected utility is
u✓(s, µ(s)) = Eg[sV (↵(µ(s),g)) ✓]. |
As is typical in signaling models, analysis and intuition are aided by studying the indif- ference curves of the sender. Given the analysis above, the indi↵erence curves of interest are the level sets of the entrepreneur’s expected utility function, u✓(s, µ). Figure 1 illustrates the key features of the entrepreneur’s indi↵erence curves. Panel (a) shows the entrepreneur’s indi↵erence curves over (s, ↵), which, recall, are independent of type. For any fixed s,utilityisincreasingin↵. However, for fixed ↵, utility is increasing in 1 s if and only if ↵> 2 .Thisisthemanifestationoftheassumptionthatalargers amplifies the utility (or disutility) of a market impression. In the extremes, the best outcome for the entrepreneur is to have chosen the maximal s coupled with the market being convinced she is high ability, whereas the worst outcome also involves having chosen the maximal s but coupled with the market being convinced she is low ability. Panels (b) and (c) depict the indi↵erence curves over (s, µ)forthelow-andhigh-ability
9 entrepreneur, respectively. Because ↵ is derived from µ via Bayes rule, these panels retain the features just described, but with two key di↵erences. First, there is a level e↵ect: relative to panel (a), the indi↵erence curves are shifted up for the low type (panel (b)) and shifted down for the high type (panel (c)). This is due to their respective expectations about the realization of g. Fixing a belief level b (0, 1), be it interim or final, because the realization 2 of g is informative, uL(s, b)
Fact 1 The (s, µ)-indi↵erence curves satisfy the single-crossing property.
2.2 Equilibrium and Empirical Properties As is common in signaling models, the game has many equilibria. It may be worth noting, however, that none of them are separating.Thisfollowsfromtheatypicalstructureof the sender’s utility function. In a typical signaling model, separation occurs in equilibrium because the high type chooses a signaling level great enough that the low type finds it too costly to imitate, even though doing so would “fool” the receivers. In our model, selecting alarges is “costly” to the low type only if there is su cient chance that the market will ultimately settle upon an unfavorable impression of the firm. If an equilibrium called for types to separate, imitating the high-ability entrepreneur would therefore be costless, because doing so would completely convince the market of the entrepreneur’s high ability. Hence, deviation by the low-ability entrepreneur would be profitable, contradicting the existence of a separating equilibrium. In the theory of signaling games, there is a long tradition of using refinements to narrow the set of equilibria. The most well-known and commonly used of these refinements are those related to the concept of strategic stability (Kohlberg and Mertens 1986), which focus
10 on eliminating o↵-path beliefs thought to be unreasonable.14 In keeping with this tradition, we focus on equilibria that are stable in that they satisfy the D1 criterion. Intuitively, the refinement can be interpreted as requiring that, following an o↵-path (i.e., unexpected) choice of s by the entrepreneur, the market believes the deviating entrepreneur to be of whichever type is more willing to undertake this deviation.15 As we have just seen, the model’s single-crossing property means the high-ability entrepreneur is more willing than the low to associate the firm with herself by choosing higher signaling levels. In our model then, the refinement will eliminate equilibria in which the high type selects a signaling level s