ArticlePDF Available

Abstract and Figures

We examine the trade-off between efficiency and equality within the context of entrepreneurial founding teams. Using a formal theory where founders may have preferences over relative outcomes, we derive predictions about the antecedents and consequences of dividing equity equally among all founders. Using proprietary survey data, we empirically test the predictions. Our central finding is that teams that split equity equally are less likely to raise funds from outside investors. The relationship appears not to be causal, but instead driven by selection effects across heterogeneous teams with varying degrees of inequality aversion. This paper was accepted by David Hsu, entrepreneurship and innovation.
Content may be subject to copyright.
NBER WORKING PAPER SERIES
THE FIRST DEAL: THE DIVISION OF FOUNDER EQUITY IN NEW VENTURES
Thomas F. Hellmann
Noam Wasserman
Working Paper 16922
http://www.nber.org/papers/w16922
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
April 2011
We would like to thank David Hsu, Bill Kerr, Josh Lerner and seminar participants at the Arizona
State University, NBER, University of Arizona, University of British Columbia, University of Toronto,
and University of Cape Town for their helpful comments. We thank Xiang Ao, Andrew Hobbs, Bill
Holodnak, Aaron Lapat, Furqan Nazeeri, and Mike DiPierro for their valuable research support. Financial
support from the Harvard Business School and SSHRC is gratefully acknowledged. Author names
are in alphabetic order; both authors contributed equally to this paper. Updated versions are available
at http://strategy.sauder.ubc.ca/hellmann/. All errors are ours. The views expressed herein are those
of the authors and do not necessarily reflect the views of the National Bureau of Economic Research.
NBER working papers are circulated for discussion and comment purposes. They have not been peer-
reviewed or been subject to the review by the NBER Board of Directors that accompanies official
NBER publications.
© 2011 by Thomas F. Hellmann and Noam Wasserman. All rights reserved. Short sections of text,
not to exceed two paragraphs, may be quoted without explicit permission provided that full credit,
including © notice, is given to the source.
The First Deal: The Division of Founder Equity in New Ventures
Thomas F. Hellmann and Noam Wasserman
NBER Working Paper No. 16922
April 2011
JEL No. G30,L26,M13
ABSTRACT
This paper examines the division of founder shares in entrepreneurial ventures, focusing on the decision
of whether or not to divide the shares equally among all founders. To motivate the empirical analysis
we develop a simple theory of costly bargaining, where founders trade off the simplicity of accepting
an equal split, with the costs of negotiating a differentiated allocation of founder equity. We test the
predictions of the theory on a proprietary dataset comprised of 1,476 founders in 511 entrepreneurial
ventures. The empirical analysis consists of three main steps. First we consider determinants of equal
splitting. We identify three founder characteristics –idea generation, prior entrepreneurial experience
and founder capital contributions – regarding which greater team heterogeneity reduces the likelihood
of equal splitting. Second, we show that these same founder characteristics also significantly affect
the share premium in teams that split the equity unequally. Third, we show that equal splitting is associated
with lower pre-money valuations in first financing rounds. Further econometric tests suggest that,
as predicted by the theory, this effect is driven by unobservable heterogeneity, and it is more pronounced
in teams that make quick decisions about founder share allocations. In addition we perform some counterfactual
calculations that estimate the amount of money ‘left on the table’ by stronger founders who agree to
an equal split. We estimate that the value at stake is approximately 10% of the firm equity, 25% of
the average founder stake, or $450K in net present value.
Thomas F. Hellmann
Sauder School of Business
University of British Columbia
2053 Main Mall
Vancouver, BC V6T 1Z2
CANADA
and NBER
hellmann@sauder.ubc.ca
Noam Wasserman
Harvard Business School
Rock Center 210
Boston, MA 02163, USA
nwasserman@hbs.edu
[1]
Section1:Introduction
Sooner or later every entrepreneurial team has to face a tough decision of how to divide founder shares.
This can be difficult because it requires founders to assess the relative value of each other’s
contributions. A simple solution is to value all members equally; this avoids making value judgments
and requires minimal negotiation. However, an equal split of founder equity may not always be
appropriate, for some founders may feel like there are contributing relatively more and thus expect to
receive more shares. This paper examines the division of founder shares in entrepreneurial ventures,
focusing on the decision of whether or not to divide the shares equally among all founders.
Founding teams adopt a variety of approaches to splitting the ownership among themselves. For
instance, at Pandora Radio (originally known as Savage Beast), an online music company that was
founded by three people who had few a priori differences between them regarding their expected
contributions to the venture, the team split the equity quickly and equally (Wasserman et al., 2008b). At
Ockham Technologies, where there were significant differences in the backgrounds of the three
founders, the team engaged in a detailed negotiation over equity stakes and ended up deciding on an
unequal split (Wasserman, 2003b). In this paper we provide a theory and some empirical evidence to
explain under what circumstances founders choose to split their shares equally, the process of how they
reach that decision, and how this decision may be related to subsequent outcomes.
Given the paucity of prior research on this topic, we develop a simple theoretical model of the central
issues involved in founder equity splits. The model includes negotiation frictions, based on the cost (or
discomfort) of valuing relative differences. It generates three sets of empirical predictions: (i) larger
teams and teams with more heterogeneous founders are less likely to agree on an equal split; (ii) the
same founder characteristics whose team-level heterogeneity predicts fewer equal splits, also affect
individual share allocations in case of unequal splitting; and (iii) equal splitting is associated with lower
valuations, for reasons discussed below.
We test these predictions on a unique, proprietary dataset comprised of 1,476 founders in 511 private
ventures. A major barrier to studying the allocation of founder equity has been lack of data; standard
datasets lack information about founders’ initial ownership of their ventures, at least in part because of
the sensitivity of such data, and fail to include such factors as the duration of the negotiation over equity
[2]
splits. To get past these barriers, we use data from the annual CompStudy survey of Technology and
Life Sciences ventures in North America. The CompStudy survey was first conducted in 2000, and over
the last decade, has become a standard annual reference for private-venture boards and CEOs regarding
executive-compensation benchmarks. In 2008 and 2009, the CompStudy survey added detailed
questions about founding teams and their equity splits. These data serve as the core of the empirical
section of this paper, and enable us to begin opening the black box of founders’ equity allocations.
As a starting point we compare the actual percentage of founder shares received against the benchmark
of an equal split. We define share premium = equity share – (1/N), where N is the number of founders in
the venture. Figure 1 shows the distribution of share premia in our data. The most striking fact about this
distribution is the spike at zero. Indeed, approximately one third of all founding teams simply decide to
split the equity equally. Our objective is thus to understand the underlying reasons for this curious base
finding.
Our empirical analysis is comprised of three main steps, closely following the theory’s empirical
implications. The first step is to explain the determinants of equal splitting. The theory suggests that
larger teams are less likely to split the equity equally, and the data clearly supports this hypothesis. The
theory suggests that equal splitting is less likely when there is more heterogeneity within the founding
team. Our data allow us to consider four founder characteristics: years of work experience, prior
founding experience (a.k.a. serial entrepreneurs), whether they contributed to the founding idea, and
capital invested in the venture. We find that greater team heterogeneity in entrepreneurial experience,
idea generation and capital contributions predict a lower probability of equal splitting. The theory also
predicts more equal splitting when negotiation costs are high. We think of these not only in terms of
direct cost (time and resources spent on negotiating) but mostly in terms of indirect costs, especially in
terms of social barriers to evaluating differences. This social-cost interpretation is supported by a finding
that teams where founders are related through family are more likely to do an equal split. Teams with
more experienced founders are also less likely to split the equity equally.
The second step of the analysis focuses on the subset of teams that split the equity unequally, to analyze
the determinants of founders’ share premia. Our theory predicts that the same founder characteristics
whose heterogeneity measures affect unequal splitting should also affect the size of the share premium
among unequal splitters. Indeed, we find that prior entrepreneurial experience, contributing to the
founding idea and investing capital are all associated with higher share premia.
[3]
The third step of the empirical analysis examines how equal splitting is related to valuation. We find a
negative relationship between equal splitting and the pre-money valuation at the time of a first round of
outside financing. Our theoretical model predicts such a negative relationship, not because of a direct
causal effect but because of a “stakes effect” (founders are less hesitant to initiate negotiation when the
stakes are high) and what we call a “negotiator effect” (better entrepreneurs are keen negotiators, both
with outsider stakeholders and with each other). In this case a team’s refusal to negotiate an unequal
split may reveal an underlying weakness, a lack of entrepreneurial negotiation skills. Negotiating skills
are obviously not directly observable, but we consider an indirect test, where we decompose the decision
to split the equity into an expected and an unexpected component, arguing that the negotiator effect
should work through the unexpected component. We indeed find that the relationship between equal
splitting and valuation is driven by the unexpected component, and that the expected component has no
explanatory power. This is consistent with unobserved heterogeneity, such as a negotiator effect.
We also examine the speed with which a team negotiates its division of equity. In our sample, 47% of all
teams report that they agreed on an equity split within a day or less. Our theoretical model suggests that
quick negotiations are associated with equal splitting, something we indeed find in the data.
Furthermore, the theory suggests that the negative relationship between equal splitting and valuation
should be concentrated among quick negotiators, which is supported by the empirical evidence. This last
finding is particular striking, because it shows that processes deep inside the entrepreneurial venture
might still be related to financial outcomes such as firm valuation.
We also consider the possibility that some teams have a distinct preference for equal splitting. This
could be because of a desire to balance control rights, especially avoiding one founder’s obtaining a
majority of shares. Or it could be that founders have ‘other-regarding’ preferences (sometimes also
called altruism or a preference for equality; see Fehr and Schmidt (2006)) that value equality by itself.
We first show that such benefits of equal splitting are easily incorporated into our theoretical framework.
We then provide some preliminary evidence that suggests that equal splitting is positively correlated
with a preference to balance founder representation on the board of directors, consistent with the control
argument. It is also positive correlated with equality of founder salaries and bonus targets, which is
consistent with theories of other-regarding preferences.
In the final section of the paper we perform some counterfactual calculations to gauge how much money
is at stake when a team decides to forgo negotiations and accept an equal split. Any such calculations
[4]
require bold assumptions, so we provide a variety of alternative approaches. Overall we find that the
decision to allocate founder shares equally does not appear to be economically trivial.
Although we are not aware of any other paper that looks inside the black box of founder equity splits,
our analysis still builds on a variety of prior literatures. The recent finance literature has emphasized that
in order to understand a firm’s financial structure, it is important to go back to the very beginnings of the
firm (Kaplan et al., 2004; Lemmon et al., 2010). The work of Robb and Robinson (2009) also
recognizes the importance of insiders (i.e., founders) in the capital structure of entrepreneurial firms.
Hauswald and Hege (2006) examine ownership and control rights for joint ventures between established
firms, and find a high incidence of equal share divisions.4 The problem of dividing shares is closely
related to the “division of a pie” problem that has been studied extensively in game-theoretic literature
(e.g., Binmore et al., 1986; Rubinstein, 1982). Interestingly, some of the recent game-theoretic literature
examines why identical parties may still not always agree on an equal division (Ashlagiy et al., 2008),
whereas we are concerned with the question of why non-identical founders agree on an equal division. A
small economics literature also discusses equal compensation for unequal agents (Bose et al., 2010;
Encinosa et al., 2007). Finally, there is an organizational literature that examines the social factors that
affect the formation of founding teams (e.g., Ruef et al., 2003).
The remainder of this paper is structured as follows. In section 2 we develop a simple theoretical model
and discuss its empirical implications. In section 3 we describe the CompStudy survey from which the
data are taken, and explain the variables used in the empirical analyses. Section 4 examines the
determinants of equal splitting. Section 5 analyzes determinants of individual share premia. Section 6
studies the relationship between equal splitting and valuations. Section 7 considers potential non-
pecuniary benefits of equal splitting. Section 8 estimates the value at stake using counterfactual
calculations. Section 9 briefly concludes.
Section2:TheoreticalMotivation
In this section we discuss the theoretical foundation and develop a simple formal model. Rather than
seeking to develop a complex theory, we try to develop the simplest possible model to guide and
4Forrelatedworkonjointventures,seealsoRobinson&Stuart(2007);Dyer,Singh,&Kale(2008);andGulati&Wang
(2003).
[5]
interpret the empirical evidence. Suppose there are N founders, indexed by i=1,…,N. Founders need to
decide how to split their venture’s equity. They have imperfect information about the exact value each
founder contributes to the venture, and face the following simple choice. Either they negotiate equity
stakes, which requires that each founder incurs private costs k, or they all forgo those costs and simply
split the equity equally. We assume that without negotiation costs, an equal split is the only focal point,
i.e., it is not possible to negotiate an unequal split without incurring negotiation costs. We justify this by
saying that any agreement on an unequal split requires founders to agree about how to value their
respective differences, which is likely to involve some costly negotiation. The private costs k should be
interpreted broadly, to include not only the time and resources devoted to evaluating differences among
founders, but also the psychological costs and social impediments of discussing who contributes more
value.5
There are three dates in the model. At date 0, the two founders have to decide whether or not to evaluate
their differences. If they evaluate them, they incur the costs k, learn about the respective differences, and
negotiate an equity split at date 1. If they don’t evaluate their differences at date 0, they simple agree to
an equal split at date 1. At date 2 the value of the venture, denoted by π, is revealed.
Consider the negotiation at time 1. For simplicity we assume that the founders bargain according to the
Nash-Shapley bargaining value, although nothing depends on this. After incurring costs k, founders have
common knowledge about their relative value-added to the venture, and split the equity according to Si =
1/N + Δi, where Si denotes the percentage share received by the individual-founders and Δi is the share
premium, resulting from the Nash-Shapley value. We will discuss the determinants of Δi in the empirical
part of the paper.
If founders incur k, Δi becomes observable to all at date 1. At date 0, however, the founders only
observe an imperfect signal, given by δi= Δi+εi. δi represents each founder’s initial belief about his/her
own value contribution, prior to going through an extensive process of communicating and evaluating
this belief with his/her partner(s). We assume that εi are drawn independently from a common
distribution. To obtain simple analytical expressions we use the uniform distribution over the interval [-
u,u], where uMax|Δi|. Note that this distribution satisfied the condition E(εi)=0, so that each partner has
5Anotherpossiblecostofnegotiatingisthatfoundersantagonizeeachothertothepointofbreakingup.Inthatcasethe
costofnegotiationincludesatermλπ/N,whereλistheprobabilityofteambreakupandπtheexpectedvalueofthe
venture.
[6]
an unbiased belief about his/her relative value-added. A recent literature argues that entrepreneurs are
optimists (Puri et al., 2007). It is straightforward to extend the model to allow for optimistic founders
who have biased beliefs about their value contribution. This would require a distribution with E(εi)>0,
and imply that optimistic entrepreneurs are more likely to demand negotiation, and therefore less likely
to split the equity equally.
We assume that at date 0, each partner has a choice of either accepting an equal split, which has the
advantage of forgoing the evaluation costs k, or demanding a negotiation about equity splits. Even if
only one partner demands negotiation, all have to incur k. The decision to demand negotiation is
therefore given by a trade-off between the expected gains from negotiation versus the cost of bargaining.
Formally, a partner asks for negotiation whenever δiπ > k, where π represents the founders’ (rational)
expectations of the future value of their company.6 Using δiπ < k εi<(k/π)–Δi and using the properties
of the uniform distribution, the probability that founder i does not ask for negotiation is given by
[u+(k/π)–Δi]/2u. The probability of equal splitting at date 0 is given by the probability that none of the
founders ask for negotiation, given by θ = Prob[εi<(k/π)–Δi for all i] = Πi=1,...,N[u+(k/π)-Δi]/(2u)N.
The equilibrium of this model is as follows. At date 0, each partner observes his/her signal δi. They split
the equity equally with probability θ. However, with probability 1-θ, at least one of the partners
demands negotiation. In that case, all partners incur evaluation costs k. They discover their true
differences Δi and split the equity according to Si = 1/N + Δi.
We now use this simple model to derive some empirical implications. Our discussion explicitly
acknowledges that as econometricians we only observe a subset of the information available to the
agents in the model. Moreover, to empirically explain equal splitting we could trivially assume that there
exists a large portion of teams with Δi=0. However, this amounts to assuming rather than explaining
equal splitting. Instead, we assume that there is a non-degenerate distribution of Δi‘s, so that the case of
Δi=0 is a measure zero event. This means that our model only predicts equal splitting for teams that do
not incur evaluation costs. We will return to this assumption later.
Consider first the role of team size. In our model, when one founder demands negotiation, the entire
team ends up having to engage in the negotiation. Intuitively this suggests that larger teams are less
6 Notethateachfounderobserveshissignalδibutnothispartner’ssignalδj.Atthetimeofinitiatinganynegotiationor
communication,s/hecanthereforeonlyconsiderhis/herownsignal.
[7]
likely to end up with an equal split. Consider the effect of adding a N+1th founder to a team of N. For
simplicity, suppose that the additional founder does not perturb the relative value-added of the other
founders, i.e., ΔN+1=0. In that case, using obvious notation, we have θN+1 = Πi=1,...,N+1[u+(k/π)-
Δi]/(2u)(N+1) so that θN+1/θN=[u+(k/π)]/2u<1. Moreover, because of the strict inequality, this result
continues to hold in a neighborhood of ΔN+1=0. For teams with comparable levels of heterogeneity we
state the first Empirical Implication.
Empirical Implication 1 (Team size): Larger teams have a lower probability of equal splitting.
Next we turn our attention to the negotiation costs k. We immediately note that the higher k, the greater
the probability of equal splitting.7 Again, as econometricians we cannot directly observe the value of k,
but we may be able to identify some covariates that we denote by KT – the subscript T refers to the fact
that this variable varies at the team level.
Empirical Implication 2 (Evaluation costs): If there are team-specific covariates KT that are positively
related to the evaluation costs k, then higher values of KT are associated with a higher probability of
equal splitting.
Let us now examine the role of founder heterogeneity. For this consider any distribution of Δi’s and
consider a change where one founder’s Δi increases while that of another founder, who has a smaller Δi,
decreases.8 Without loss of generality we focus on founders 1 and 2 with Δ1 > Δ2 and we denote the
increase in heterogeneity by some small σ > 0. Using obvious notation, we have Δ1 = Δ1 + σ > Δ1 > Δ2
> Δ2σ = Δ2. Straightforward calculations reveal that dθ/dσ < 0.9 This says that greater heterogeneity
among founders (as measured by an increase in σ) decreases the probability of equal splitting. The Δi’s
are again not observable to the econometrician. Let Xi be an observable covariate that varies at the
individual-founder level and let HT(Xi) be a measure of the heterogeneity of this covariate within the
team. Suppose that the covariate Xi is relevant to the share premium Δi then we obtain the following
prediction.
7Formallywenotethatsign[dθ/dk]=sign[dlog(θ)/dk]=sign[(1/π)]>0.
8Increasesinheterogeneityinvolvingmorethantwofoundersfollowasimilarlogic.
9 Evaluatingdθ/dσnearσ=0wenotethatsign[dθ/dσ]=sign[dlog(θ)/dσ]=sign[(u+(k/π)‐Δ2+σ)1(u+(k/π)‐Δ1‐σ)1]=sign[Δ2
Δ1]<0.
[8]
Empirical Implication 3 (Team heterogeneity): The greater the heterogeneity HT(Xi), the lower the
probability of equal splitting θ.
In addition to making predictions for the probability of equal splits, our model also makes predictions
about the share premium Δi.10
Empirical Implication 4 (Share premium): If for a covariates Xi the heterogeneity HT(Xi) lowers the
probability of equal splitting θ, then Xi also affects the individual share premium Δi, and vice versa.
Empirical Implication 4 says that the covariates that affect the share premium at the individual level are
the same as the covariates whose heterogeneity affects the probability of equal splitting at the team
level.
Finally we explore the relationship between the equity split decision and the value π of the company.
Empirically we will measure π through a company’s financial valuation. Our model suggests two
reasons why equal splitting may be associated with lower valuations. First, it is easy to see that there is a
direct negative effect of π on θ.11 This is because teams that expect to have more profitable ventures are
less likely to agree on an equal split. We call this the “stakes effect,” and it says that founding teams
with greater expectations have stronger incentives to negotiate their initial allocation of equity.
In addition to the “stakes” effect, we consider a second effect that we call the “negotiator” effect. The
simple intuition is that keen negotiators make better entrepreneurs and also bargain harder among
themselves. Put differently, entrepreneurs who obtain better deals by negotiating hard with their
customers, suppliers, employees or investors may also be more inclined to negotiate among themselves.
This suggests that negotiation costs are negative related to team quality. More formally, we assume that
there is an unobservable quality parameter (μ, interpreted as the willingness to negotiate hard) that
generates higher profits (i.e., π is increasing in μ) and at the same time is associated with lower
negotiation costs (i.e., k is decreasing in μ).
10Notethatbydefinitionwehave∑i=1,…,NΔi=0sothatΔimeasuresrelativevalueadded.Asaconsequence,allfounder
specificcovariatesXishouldbeexpressedinrelativeterms.
11Formally,sign[dθ/dπ]=sign[dlog(θ)/dπ]=sign[∑‐(k/π2)] < 0.
[9]
Note that neither the stakes nor the negotiator effects imply a causal relationship between equal splits
and lower valuations. In fact, the stakes effect reflects what econometricians refer to as reverse causality,
whereas the negotiator effect is a case of unobserved heterogeneity.
Empirical Implication 5 (Valuation). Equal splitting is associated with lower valuations because of a
“stakes effect” where teams that expect a higher valuation have stronger reasons to negotiate an
unequal split; and because of a “negotiator” effect where k is a decreasing function of team quality μ,
so that an equal split becomes a sign of lower team quality.
Our final set of empirical implications concerns the importance of negotiation itself. Our basic model
set-up suggests that quick (slow) negotiation is associated with equal (unequal) splitting. Finding
empirically a positive correlation between equal splitting and quick negotiation would thus confirm the
fundamental premise of the model.12 Moreover, combining this insight with Empirical Prediction 5
suggest that equal splitting with or without lengthy negotiations represent two distinct economic
outcomes. In particular, the negative association between equal splitting and valuation that emerges
from Empirical Implication 5 should only pertain to quick equal splitting, but not to equal splitting after
lengthy negotiations.
Empirical Implication 6 (Negotiation speed and valuation): If an equal split emerges from a quick
negotiation it is associated with a lower valuation than if it emerges from a lengthy negotiation process.
Section3:Dataandvariables
3.1:Datasources
The data for this paper come from the annual CompStudy survey of private American ventures, for
which one author is the lead investigator (see Wasserman, 2003a, 2006; 2011 for more details). The first
CompStudy survey was conducted in 2000 with 211 private information technology ventures (broadly
defined, including telecommunications). Two years later a parallel survey of life sciences ventures was
12Narrowlyspeakingthemodelactuallypredictsaperfectcorrelation,somethingwewouldnotexpectempirically.An
imperfectcorrelationbringsupaninterestingtheoreticalquestionofhowtointerprettheoffdiagonalelements.Forteams
thatnegotiatequicklybutagreeonanunequalsplitthenaturalinterpretationisthatfoundersalreadyhadadetailed
understandingoftheirrelativestrengthspriortoinitiatingthenegotiationprocess.Forteamsthatundergoalengthy
negotiationprocess,attheendofwhichtheyagreetoanequalsplit,itappearsthatthereisadistinctbenefittodoingso.
InSection7wefurtherformalizethis.
[10]
added, and since then, annual surveys of both industries have been conducted.13 The list of target
companies is generated by combining the list of private companies included in the VentureXpert
database with the membership lists of local technology associations (e.g., the Massachusetts Technology
Council). Invitations are sent to the CEOs and CFOs of those companies, and participants are
encouraged to provide additional names of companies that meet the survey selection criteria.14 To
encourage participation in the survey, participants are offered a free copy of a detailed “CompStudy
Compensation Report” that is based on the survey results and made available only to participants. The
report includes detailed position-by-position breakdowns of salaries, bonuses, and equity holdings for
the eleven most common C-level and VP-level positions in private ventures. The breakdowns provide
compensation benchmarks by industry segment, geographic location, company size and age, financing
rounds, founder versus non-founder status, and other metrics collected in the survey. Over the last
decade, CompStudy’s annual compensation reports have become a standard reference within the top
management teams of private American ventures and for the board members and investors who are
involved with those ventures.
The dataset used in this paper combines the Technology and Life Sciences surveys from 2008-2009,
while controlling for the industry of each participant. A major benefit of conducting annual surveys, and
of collecting one’s own data, is that each year the researcher can add new questions to tackle emerging
research questions that aren’t addressed by existing datasets (e.g., about the equity-split negotiation
process) or are highly confidential (e.g., about the percentage of equity received by each founder). For
the 2008 CompStudy survey, we added detailed questions about each founding team, its prior work
experience and relationships, and the equity split within the team. These questions were repeated in the
2009 survey. The dataset for this paper combines the survey responses from both 2008 and 2009, in both
the technology and life sciences industries. Across those two surveys, we received responses from a total
of 576 multi-founder teams. Dropping 65 repeat respondents in 2009, we arrive at our full dataset of 511
ventures and 1,476 founders (an average of 2.9 founders per venture).
13Thesetwoindustriesarethepredominantonesforhighpotentialventures.Forinstance,duringtheyearscoveredbythe
CompStudysurveys(20002009),informationtechnologyventuresaccountedforalmost30%ofallIPOs,lifesciencesand
healthcareforanother18%,thefinancialindustryforalmost12%,andnootherindustrywasabove8.3%.Similarly,these
twoindustriesreceivefarmoreventurecapitalthananyotherindustries.
14TheCEOsandCFOsoftheventuresweretargetedduetothesensitivityofthequestionsandthebreadthofcorporate
knowledgerequiredtocompletingthesurvey.
[11]
Each year, survey response rates vary between 10%-20%, higher than the typical response rates for
surveys of similar firms and targeting similar levels of executives (e.g., Graham et al., 2001). The
surveys are conducted online so that fields can be validated as they are being entered. When possible,
data are cross-checked with publicly-available information sources to validate the accuracy of the
submissions. Each year, the survey data are checked for representativeness against the VentureXpert
population. Regarding geographic distributions and industry segments, the dataset is a representative
sample of private high-potential ventures in the technology and life sciences industries within the United
States. Regarding age of venture, the dataset contains slightly younger companies given that
VentureXpert only includes ventures that have raised institutional capital while our dataset also includes
pre-funding ventures (9% of our sample).
Compared to a sample of public companies, this dataset should be far less susceptible to survivor biases.
Our survey design allows us to capture companies at a very early stage, but it is still not possible to
sample companies at birth, or before. Indeed, our analysis starts at a point in time where teams have
been formed, so that we cannot analyze the process of how teams are formed. Clearly team formation is
not random, but our research question does not require random assignment of founders to teams. On the
contrary, we are interested in examining how those founders who are actually starting a company split
the equity.
The standard concern with not sampling at birth is the possibility that the sample suffers from some
survivorship bias. Given the unique nature of our data, no out-of-sample comparison is feasible.
However, we can perform a within-sample test for survivor bias. Specifically, we analyze whether in our
sample older ventures differ from younger ventures with regards to our two core dependent variables,
the team-level equal-split dummy and (the absolute value of) the individual-level share premium
received by each founder. We use t-tests of median split sub-samples of the dataset, splitting the
ventures into 252 “older ventures” and 259 “younger ventures.” If there is strong survivor bias in the
sample, we would expect to see differences in these variables between the younger ventures and the
older ventures, but we find no such bias. For the younger ventures, 32.0% had split the equity equally
(standard deviation of 0.468); for the older ventures, 34.9% had split the equity equally (standard
deviation of .478), with the t-test for difference being insignificant. Likewise, a t-test of the absolute
value of the founder-level share premium (for the younger ventures, a mean of 9.2 and standard
[12]
deviation of 11.3; for the older ventures, a mean of 10.2 and standard deviation of 13.0) found no
significant difference between older ventures and younger ventures.
Another sample selection issue concerns the question of exactly who gets counted as a founder. For our
main analysis we use the self-reported data, but as a robustness check we also examined whether our
main results could be affected by some unusual team definitions. There are two types of gray areas in
the data. First, some founders join a while after founding, making them look more like non-founding
executives. Second, some founders receive very low equity stakes, making them look more like
employees. As a robustness check we removed from the sample all teams that had (i) any “founders”
who joined more than two years after the first founder, and/or (ii) any founders with an equity stake that
was more than 70% below the equal stake.15 We found that this did not change the main insights of the
analysis.
3.2: Empirical variables
Appendix 1 summarizes our core variables, their definitions, and the specific survey question used for
each variable. Table 1 shows the descriptive statistics. Panel A shows the statistics for the team-level
variables, and also separates the sample into teams that split the equity equally versus unequally. Panel
B shows the descriptive statistics for the individual-founder variables, also separated into teams that split
equally versus unequally. Table 2 reports the pair-wise correlation between the main variables of
interest, Panel A at the team level and Panel B at the individual-founder level.
Our core dependent variables capture the allocation of equity within the founding teams. Our raw data
include for each co-founder the specific percentage of equity received. Our initial analyses focus on the
determinants of whether teams split the equity equally versus unequally. We use a binary variable of
whether the team split the equity equally (i.e., all founders received the same percentage) or unequally.
This binary variable is the basis for estimating the probability of an equal split, as captured by θ in the
theory model. At the individual-founder level, to analyze the determinants of the percentage received by
each founder we use the share premium, defined as the actual equity stake minus the equal stake (which
is 1/N where N is the number of founders in the venture). This measures the deviation from the focal
15The70%thresholdappliestoanyfounderthatreceiveslessthan15%inateamoftwo,lessthan10%inateamofthree,
orlessthan7.5%inateamoffour.Wealsovariedthisthresholdandobtainedverysimilarresults.
[13]
outcome of an equal split, and corresponds to the Δi’s in the theory. Figure 1 shows the distribution of
the share premium across the entire sample.16
Another important dependent variable is the pre-money valuation received by the venture during its first
round of outside financing. While the post-money valuation is a measure of the total company value
after completion of the round, the pre-money valuation is a more appropriate measure of the value
actually captured by the founders.
Most economic theories would argue that the allocation of equity is a forward-looking decision, so that
the optimal allocation depends largely on parameters that affect the fundamental problem of moral
hazard in teams (Holmstrom, 1982). Unfortunately these parameters, such as the relative effort elasticity
and the relative productivity of the different founders, are empirically hard to observe. Instead we have
to contend with more objective measures of founder differences that are observable at the time of
founding. While these variables may appear to be backward looking (e.g., founder X has prior
entrepreneurial experience) they typically contain relevant forward looking information (e.g., founder X
is more productive because of prior experience).
The independent variables include individual-founder and team-level variables. At the individual-
founder level, we have four main variables: work experience, entrepreneurial experience, ideas and
capital. Those correspond to the Xi’s in the theory model.
Evans and Leighton (1989) suggest that education has greater returns in self-employment than in wage
work, suggesting an important role for human capital in entrepreneurial firms. In addition, prior
entrepreneurial experience may be particularly important (Gompers et al., 2010). Thus, our individual-
founder variables include two measures of human capital: the founder’s years of work experience prior
to founding the current venture, and whether the founder had prior founding experience (“serial
entrepreneur”).
16Becauseweareinterestedintheinitialdecisiontosplitequitywedonotexamineanysubsequentchangesinfounder
shares.Atthetimeofinitialcontractingfounderscanalsospecifysocalledvestingclauseswheresomeofthefounder
sharesareearnedovertimeonthebasisofpredefinedmilestones,butwheretheinitialequitystakereflectsthefull
potentialsharethatcanbeearnedbyeachfounder.Oursurveyidentifieswhichteamsadoptfoundervesting(though
withoutdetailsontheschedulesandconditions).Theadoptionoffoundervestingschemesisnotsignificantlycorrelated
withequalsplitting.
[14]
Our third individual-founder variable captures whether the founder had been responsible for the idea on
which the venture was founded (“idea person”). There is some debate about the value of bringing an
idea into a team. The famous Arrow paradox (1962) argues that it is difficult to capture the value of an
idea, because prior to disclosing an idea, no one is willing to pay for an unknown idea, and after
disclosure the information has been transmitted so there is no further need to pay for it. The null
hypothesis is thus that ideas do not affect the division of equity among founders. However, this
reasoning only acknowledges the backward looking component of idea generation. One resolution of the
Arrow paradox, suggested by Arora (1996), is that the idea generator also has complementary skills and
non-codifiable knowledge that makes him/her uniquely valuable to the implementation of the idea.
Under this hypothesis we would expect the idea generator to command a higher share premium because
of forward-looking components, such as the creativity and implementation skills that idea generators
bring into the founding team.
Our final individual-founder variable captures the financial contributions made by the founders to the
venture. For each founder, the survey asks whether the amount of capital contributed falls into one of
five categories: $0k, $1k-$25k, $26k-$100k, $101k-$500k, and More than $500k.17
An executive’s economic gains may be influenced by the executive’s position in the organization (e.g.,
Lazear et al., 1981). Thus, we also consider data about the positions held by each of the founders within
the venture, such as CEO, Chairman, CTO, or other.18
The main team variables include the size of the founding team, as well as the team-level versions of the
individual-founder variables. This includes the team means of the individual-founder variables listed
above, which we think of as negotiation cost covariates KT. For example, firms with more experienced
entrepreneurs may have a lower cost of negotiation, because their experience has hardened them to the
17Thereisnoperfectwayofconvertingthisordinalmeasureintoanumericalmeasureofcapital.Tobepragmatic,weuse
themidpointswhenavailable,otherwisethelowestpoint,resultinginthefollowingcategories(in$million):0,0.012,
0.063,0.3and0.5.Wetriedseveralvariations,suchasvaryingtheamountforthetopcodedcategory,usingnatural
logarithmsorusingcategoricalvariables15,andfoundthattheresultswereverysimilaracrossallthesepermutations.
18Thisvariablecapturesthemainpositionbyafounder,butitdoesnotprovidecompleteinformationonboard
representationfortworeasons.First,thesedummiesaremutuallyexclusive,sothatafounderwouldonlyreportbeingthe
chairoftheboardifs/hedidnotholdanyexecutivepositionsuchasCEO.Second,itdoesnotmeasureboardparticipation
otherthanchairoftheboard.Asaconsequencewewilluseadifferentpartofthesurveytoobtaininformationonboard
participation,asdescribedbelow.
[15]
fact that negotiation is part of entrepreneurial life. As a measure of team heterogeneity (i.e., the HT(Xi))
we focus on the coefficient of variation of the individual-founder variables.
We also consider some social impediments to negotiation. Organizational sociologists, such as Uzzi
(1997) and Granovetter (1985; 2005), have shown how economic transactions are embedded in the prior
social relationships of the participants in the transaction, with different relationships resulting in
different transaction outcomes. Subsequent work has shown that, in particular for founding teams, prior
relationships can have powerful impacts on the team’s early evolution (e.g., Ruef et al., 2003;
Wasserman et al., 2008a). In our study, the founding team’s prior relationships are categorized into prior
work experience together, prior founding experience together, prior friends but not coworkers, and
related to each other at time of founding. In our theoretical framework, we think of these relationship
variables impacting negotiation costs, i.e., we can think of them as part of the covariates KT.
Our empirical analysis also includes standard controls for industry, geography and date. Industry
dummies capture whether the venture is a technology venture versus a life-sciences venture.19 The
venture’s location was captured by dummies for the two “entrepreneurship hubs,” in California and
Massachusetts, and one for non-US participants, namely a small subset of participants from Canada.
Year dummies were used in the equity-split models to capture the year in which the equity was split, and
in the financing models for the year in which the first outside round was raised.
The survey captures the elapsed time over which the founders negotiated the equity split, categorized by:
1 day or less, 2 days to 2 weeks, 2 weeks to 2 months, 2 months to 6 months, and More than 6 months.
47% of all teams agreed on an equal split after 1 day or less. This naturally captures our notion of
forgoing negotiation, so that our analysis uses a dummy variable called “quick negotiation” that takes
the value 1 if a team agreed on an equal split after 1 day or less, 0 otherwise.
We also use some additional variables for extensions and robustness checks. We examine four
additional measures of team equality, concerning salaries, bonuses, board of directors and CEOs. The
CEO measures come from the usual founder section of the survey, but the three remaining measures
come from different parts of the survey. This data has three important limitations. First, all the
compensation and board data is measured at the time of the survey, not at the time that the team
19Moregranularindustryleveldummieswerealsoanalyzed,butdidnotchangetheresultsinanymodels.
[16]
allocates founder equity. Second, the compensation and board data identifies whether an executive is a
founder or not, but due to anonymity concerns it cannot be linked to the individual founder data. As a
consequence we can only link the compensation data to the founder data at the level of the firm. Third,
the compensation data is only available for 279 out of 511 companies (54.6%), and the board data for
445 companies (87.1%).
We use the salary (bonus) data to create a simple dummy variable that takes the value 1 whenever all
founders receive the same salary (bonus), 0 otherwise. The CEO (board) variable is based on whether
control is given to a single founder or not. Specifically, the Balanced-CEOs dummy takes the value 1
either if none of the founders assume the position of CEO, or else if two founders share the CEO title; 0
otherwise. Only 6 out of 511 teams (1.2%) have Co-CEOs, 83 teams (16.2%) have no CEOs. 80 teams
(15.7%) have no founders on the board, 126 teams (24.6%) have two or more founders on the board.
Section4:Determinantsofequalsplitting
Table 3 reports the results from a series of Probit regressions where the dependent variable measures
whether teams split the equity equally. The first model features industry, geography and year controls, as
well as the team size variable. We note that the larger the founding team, the more likely it is to split the
equity unequally (p<0.01). This result remains robust across all specifications, and is consistent with
Empirical Implication 1. In terms of the control variables, the industry dummies, as well as the year
dummies which are not reported for brevity’s sake, are never significant. For the geographic controls,
we find that Canadian teams are more likely to split the equity equally (p<0.1).
The second model of Table 3 adds team-level variables that are based on the individual-founder
characteristics – entrepreneurial and work experience, idea generation and founder capital. A higher
mean level for the former variables reflects greater team experience, which might lower the cost of
negotiation. For example, if one of the founders is a serial entrepreneur, it is likely that his/her prior
experience with splitting equity will make the negotiation process easier. Empirical Prediction 2
suggests that higher mean values reduce the probability of equal splitting. Table 3 shows that this
argument holds for one of the four variables: teams with higher average work experience are less likely
to split the equity equally.
[17]
Table 3 also considers the coefficient of variation as a heterogeneity measure for these variables.
Empirical Prediction 3 suggests that heterogeneity should reduce the incidence of equal splitting. Table
3 shows this to be true for three out of four variables: the mix of serial entrepreneurs (p<0.1), idea
people (p<0.05), and founder capital invested (p<0.01).
The third model of Table 3 adds some social determinants, focusing on the prior relationships among
founders. These variables are measured at the team level. The main result is that teams where founders
are related through family are more likely to split the equity equally (p<0.1). This suggests that close
social ties may create a barrier to valuing relative differences among founding team members.
We performed a variety of robustness checks. Using a Logit instead of a Probit model does not affect the
patterns of results. It may be argued that instead of looking at the mean of the experience variables one
should consider the maximum. For example, it may be less important that all founders are serial
entrepreneurs, all that matters is that at least one of them is. We therefore reran Table 3 using maxima
instead of means, but found again that the only significant variable concerned work experience. Finally,
for the three categorical variables (serial entrepreneurs, ideas and capital), instead of using the
coefficient of variation, we considered using entropy as an alternative measure of heterogeneity. The
results were very similar, except that the p value for the entropy measure of idea heterogeneity fell into
the range of 0.12 to 0.14.
Section 5: Determinants of individual share premia
Empirical Prediction 4 concerns the division of equity among non-equal splitters. We focus on four key
determinants of the share premium, namely whether a founder has prior start-up experience (i.e., serial
entrepreneur), years of prior work experience, whether the founder contributed to the founding idea, and
the amount of founding capital provided. Because our theory shows that only relative differences within
teams matter, we de-mean all of the founder characteristics.
Model 1 of Table 4 reports the results from an OLS regression, where the standard errors are clustered
by team. One econometric challenge is that all the share premia within a team necessarily add up to zero.
This conflicts with the standard assumption of independently distributed errors. A similar problem
occurs with the estimation of market shares (which always sum to one). The standard solution is to drop
[18]
one observation per market (or team in our context), which solves the linear dependency (Gaver et al.,
1988).20 Given the small size of teams, estimation results can be sensitive as to which founder is
dropped. We therefore re-estimate the model multiple times, dropping one founder per team at random.
Model 2 reports the results of a bootstrapped OLS regression with one million iterations of randomly
dropping one founder per team.
The results in models 1 and 2 are highly consistent. We find that the share premium is higher for serial
entrepreneurs (p<0.01), for the idea person (p<0.01), and for founders who invest more founding capital
(p<0.01). However, prior years of work experience do not have a significant impact on the share
premium. This is consistent with Empirical Implication 4, because the three founder characteristics that
affect the share premium are exactly the same three characteristics whose heterogeneity at the team level
predicts unequal splitting.
In addition to the four founder characteristics, other aspects may affect the division of shares. Of
particular interest are the managerial positions assumed by the different founders. In Model 3 and 4 of
Table 4 we add controls for whether each founder is CEO, Chair or CTO. We find a significant share
premium for the CEO (p<0.01) and for being the chair of the board of directors (p<0.01). Being a CTO
has a positive but smaller effect on the share premium (p<0.01). The inclusion of founder roles may also
influence the strength of the coefficients for founder characteristics. We note that the coefficients for
founder ideas and founder capital remain highly significant. The coefficient for serial entrepreneurs is
smaller and becomes marginally insignificant in Model 3. Moreover, in that model the negative
relationship between experience and share premia becomes marginally statistically significant (p<0.1).
We perform several robustness checks for this model. First, it is possible to add team-level controls such
as team size. In unreported regressions we added a full set of dummy variables to account for every
possible team size. We found that all the dummies were insignificant and all the main coefficients
retained very similar point estimates and significance levels. Upon reflection, this should not be a
surprise, since the additional controls estimate differences in the average premium when we know that
the average premium is zero by construction. The same can be said for any other team-level controls.
Indeed, we also reran the model using team fixed effects and found again that the estimates were very
20Severalothereconometricmethodshavebeendevelopedtohandletheaddingupconstraintinmarketsharedata.
However,thesemethodscannotbeappliedtoourdata,sincetheyrequireafixednumberoffirmspermarket,whereasin
ourdatatherearevariablenumbersoffoundersperteam.
[19]
similar. Second, we augmented the sample to also include all equal splitters and found that while the
coefficients were naturally smaller, they still retained their statistical significance. Third, we reran all
regressions using the relative share premium, defined as (Share – Equal Share) / Equal Share, and
obtained similar results.
There are many ways in which the analysis of share premia could be extended, and we plan to further
explore these in future work. However, what matters for this paper is Empirical Prediction 4, which
states that those founder variables whose heterogeneity measure affect the probability of equal splitting
should also have a first-order effect on the share premium. The results from Table 4 support this
prediction.
Section6:Therelationshipbetweenequalsplittingandvaluation
6.1 Base specification
To examine Empirical Implication 5, we explore the connection between equity splits and financing
outcomes. As our financing outcome, we focus on (the natural logarithm of) the venture’s pre-money
valuation. This captures the value of the venture at the time of the investment round, but before the
capital has been added to the value of the venture. The pre-money valuation is computed by multiplying
the total founder shares with the price paid by the first round investors, thus measuring the value of the
founders’ stakes. We limit the analyses to first rounds that occur within three years of the equity split, in
order to minimize the chances that intervening events might obfuscate the linkages between the
founders’ equity-split and financing events.21
The main independent variable of interest is whether or not a team splits the equity equally. The analysis
of Table 5 also includes base controls for team size, industry and geography. In terms of time controls,
we use financing year fixed effects to control for market conditions at the time of financing. In addition,
we control for the time between when the equity was split and when the venture raised its first outside
round. It is also possible that the team and social factors analyzed in Table 3 have a direct impact on the
valuation. In the second model of Table 5 we include them as additional controls.
21Inunreportedrobustnesschecks,wevariedthiswindowoftimeineachdirection,withnochangesintheresults.
[20]
The main result from Table 5 is that equal splits have a negative and significant effect on valuation
(p<0.05). This is consistent with Empirical Implication 5. In terms of magnitudes, the average pre-
money valuation of equal splitters is close to $5M, compared to $5.5M for unequal splitters. We obtain
very similar magnitudes when evaluating the regression coefficients at the mean. Broadly speaking,
equal splitters are associated with a valuation discount of approximately 10%.
The number of observations in Table 5 drops by 42% for three reasons: the valuation is known but
occurs beyond the 3 year horizon (9% of firms), the valuation is unknown (25% of firms), or the firm
never raised any outside financing (8%). We first verified that the results continue to hold when we add
the valuations that occur beyond the three years horizon. We then consider some additional tests for
whether the remaining missing observations affect the main result. We consider the correlation between
equal splitting and observing a valuation but find no significant relationship. We then estimate a
Heckman framework where the outcome regression uses the pre-money valuation as dependent
variables, just as in Model 1 of Table 5. In the selection equation the dependent variable equals 1 if the
valuation is observed, 0 otherwise. The selection equation can be specified with or without the
companies that never obtain outside financing. The independent variables are the usual controls from
Model 3 of Table 3 (using Model 1 or 2 does not affect the results). We admit that we do not have a
clever instrument for the selection equation, but note that statistical identification comes from the fact
that the selection equation uses founding year controls, whereas the outcome regression uses financing
year controls. We find that the effect of equal splits on valuations remains very similar, always retaining
statistical significance. Moreover, the estimate of ρ, which measures the correlation of error terms across
the selection and outcome equation, is statically insignificant, suggesting no significant selection effects.
Thus we cannot detect any systematic biases between those companies that do or do not report their
valuations.
6.2. Unobserved heterogeneity
Table 5 establishes a relationship between equal splits and pre-money valuation, but we do not claim
that such a relationship is causal. Our data does not contain a natural experiment. More important, our
theory emphasizes a non-causal relationship in the first place. In particular, the “negotiator” effect
identified in our theory concerns unobserved heterogeneity.
[21]
We propose an additional test to further investigate this. If there is unobserved heterogeneity, we would
expect it to show up in the error term of the Probit model in Table 3. For example, willingness to
negotiate would increase the error term which leads to a higher probability of an equal split. While the
true error is unobservable, we can proxy it with the difference between the realization (p = 0 or 1) and
the predicted probability (̂). We call u=p-̂ the unexpected component of the equal split, and note that
higher values of u imply greater reluctance to negotiate.22 We then decompose the coefficient of equal
splits into its expected (̂) and unexpected (u) components and rerun the model of Table 5. Table 6
reports the results, showing that the unexpected component is negative and statistically significant, while
the expected component is highly insignificant. This is consistent with unobserved heterogeneity.
Obviously we cannot say what exactly the unobserved heterogeneity is, but our theoretical model
identifies one possible source: If some teams have better quality because of some unobservable keenness
to negotiate, then they are likely to both obtain better valuations and negotiate equity shares among
themselves.
6.3.Theroleofnegotiationspeed
Empirical Implication 6 considers the role of the negotiation process. The theory suggests that quick
negotiation should be associated with equal splitting. Moreover, while quick equal splitters should have
a lower valuation than non-equal splitters, this need not be true for slow equal splitters. Put differently,
in the case of quick negotiations, our model makes a clear prediction, whereas in the case of long
negotiations, it is ambiguous whether equal splits should be associated with higher or lower valuations.
To empirically examine the role of negotiation speed, we first examine the relationship between quick
negotiation and equal splits. From Table 1, Panel B we note that the two variables are positively
correlated (p<0.01). In an unreported regression we confirm that this correlation continues to hold in a
multivariate setting where we also control for all the independent variables for variables used in Table
22Togettheintuition,considerateamwithhighlyheterogeneousobservablecharacteristics.FromTable3wewould
expectthisteamtochooseanunequalsplit,i.e.,̂islow.Supposenowthatthisteamsurprisesusanddoesanequalsplit
(p=1),suggestingstrongaversiontonegotiate.Inthatcasewefindthatuislarge(closeto1).Conversely,consideravery
homogenousteam,sothatwemightexpectanequalsplit,i.e.,̂ishigh.Iftheteamsurprisesusanddoesanunequalsplit
(p=0),wesuspecthighwillingnesstonegotiate,andwefindthatuissmall(closeto‐1).
[22]
3.23 This validates the fundamental premise of our theory that equal splitting is associated with quick
negotiations.
To test Empirical Implication 6 we then examine whether the relationship between equal splits and
valuation is affected by the speed of negotiation. In Table 7 we find that the negative relationship
between equal splits and valuation appears to be driven by the quick negotiators: equal splitters who
negotiate quickly have a negative and statistically significant coefficient (p<0.05), whereas equal
splitters who reach an agreement after lengthy negotiations have a negative but insignificant coefficient.
Note that the difference between the two coefficients is not statistically significant, so that we should
remain somewhat cautious about the strength of the differential effect. Still, this provides at least some
suggestive evidence of the differences between fast and slow equal splitters. This is consistent with
Empirical Implication 6.
Section 7: Non-pecuniary benefits of equal splitting
7.1. Theoretical considerations
So far in our analysis the main benefit of equal splitting is the avoidance of negotiation costs. There may
also exist non-pecuniary benefits to equal splitting. Our empirical analysis already revealed that some
teams agree to an equal split after lengthy negotiation. The results from Tables 5 and 7 do not indicate
any valuation benefits of equal splitting, but there may be other ‘unobservable’ or ‘non-monetary’
benefits that we will now consider.
To extend our theory, suppose there is a non-pecuniary benefit to equal splitting, denoted by b, which
for simplicity is the same for all founders. After negotiation costs are sunk, all founders agree that an
equal split is preferable to an unequal split whenever Max(Δi)b, otherwise they settle on the Nash-
Shapley solution. Prior to incurring negotiation costs a founder demands negotiation whenever δiπ >
k+b. Thus the probability of equal splitting becomes θ = Πi=1,...,N[u+((k+b)/π)-Δi]/(2u)N. It is immediate
that the higher the benefit of equal splitting, the higher the probability of equal splitting, both at the ex-
post stage (slow negotiations) and the ex-ante stage (quick negotiations). This simple model extension
23Theonlyotherindependentvariablethatissignificantforexplainingnegotiationspeedisteamsize,withlargerteams
requiringmoretimefornegotiation(p>0.01).
[23]
shows that non-pecuniary benefits of equal splitting should not be viewed as an alternative theory, but as
an extension of our basic model. 24
7.2: Equal splitting and control
Let us consider the benefit of balancing control rights. In addition to allocating cash flow rights, shares
also allocate voting rights.25 The interesting difference is that there may be a threshold effect, where the
value of owning shares jumps discreetly when a founder (or a subgroup of founders) crosses the
majority threshold. Some teams may thus choose an equal split in order to avoid giving too much control
to any one founder. Hauswald and Hege (2006) emphasize this argument in their analysis of joint
ventures.
The benefit of balancing voting rights is not directly observable, so we consider a number of indirect
tests. We ask whether teams that split the equity equally also manifest a desire to balance control in
other dimensions. Entrepreneurial teams allocate voting control through founder shares, but they also
allocate control when they make decisions about executive positions and participation on the board of
directors. We ask whether equal splitters are more likely to have executive teams where no single
founder is chosen to be the CEO or to sit on the board of directors. The CEO-Balance (Board-Balance)
variable equals 1 whenever either none or more than one founder is CEO (sits on the board).26
Table 2 shows that CEO-Balance is positively correlated with equal splitting, although the correlation is
statistically insignificant (p<0.2). However, Board-Balance is strongly correlated and statistically
significant (p<0.01). The t-tests in Table 1 reveal the same relationships.
To further validate this relationship, we consider Probit regressions using the independent variables from
Model 3 of Table 3 (using Models 1 or 2 yields similar results) plus the equal split dummy. Again we do
not impose a causal interpretation on the estimated coefficients, but merely want to verify that the pair-
wise correlation survives in a multivariate environment. The results are reported in Table 8. We find that
24Notealsothatthissimplemodelextensionprovidesajustificationwhyinequilibriumanequalsplitmaystilloccurwith
slownegotiations,somethingthatisameasurezeroeventinthebasemodel.
25Throughoutthissectionweassumethatallfoundersharescarrythesamevotingrights.Whileoursurveydoesnotask
aboutvotingrights,oneauthor’sextensivefieldresearchsuggeststhatunequalvotingrightsamongfoundersareextremely
uncommon.
26Section3.2discussesthelimitationsofourempiricalmeasure.
[24]
the equal split coefficient is positive and significant (p<0.05) for Board-Balance, and positive but
marginally insignificant (p<0.13) for CEO-Balance.27 Overall there is credible evidence that equal
splitters are more likely to also balance board control, but only weak evidence that they balance CEO
control.
We called the benefit of balancing control a “private” benefit, implying that it has no direct impact on
valuation. In unreported regressions we examine whether our measures of balanced control affect
valuation by including them in the valuation model of Table 5. Indeed we find that they always remain
insignificant.28
7.3:Equalsplittingandcompensationequality
Non-pecuniary benefits may involve things other than control, so we now extend the argument for a
more general preference for equality. A recent literature explores the economic importance of “other-
regarding preferences,” a.k.a. inequity aversion or altruism (see Bartling et al., 2007; Fehr et al., 1999,
2006). Moreover, organizational scholars have long argued that equality within teams may promote
better team cohesion and greater cooperation, especially within teams that operate under “social logics”
(e.g., Adams, 1965; Leventhal, 1976). All these arguments suggest that some teams perceive other
benefits of choosing equal allocations.
27Asdiscussedinsections3and5,wealsoobservewhetherapersonischairmanoftheboardatthetimeoffounding.We
canthusconstructasimilarmeasureofbalancedpowerwhereweconsiderbothCEOsofboardchairsaspowerpositions,
andaskaboutthebalanceofpower.Wefindthatthisaugmentedbalancemeasureisalsouncorrelatedwithequalsplitting.
28Tolookforfurtherevidenceofcontrolbalancing,weaskwhetherthebenefitofbalancingvotingrightsdependsonteam
size.Inteamsoftwo,anydeviationfromequalsplittingimpliesthatthedominantfounderobtainsavotingmajority.In
teamsofthree,asmalldeviationfromequalsplittingimpliesthatthedominantfoundercancolludewithoneotherfounder
toobtainasupermajority.Inteamsoffour,asmalldeviationfromequalsplittingimpliesthatthedominantfoundercan
colludewithoneotherfoundertoobtainasimplemajority.Etc.Themaininsightfromthisisthatthebenefitofbalanced
votingcontrolisparticularlyhighinteamsoftwo,whereanydeviationfromequalsplittingautomaticallygivesasingle
foundervotingmajority.Weusethisinsighttoempiricallyexaminewhetherequalsplittingismoreprevalentamongteams
oftwo.WereruntheregressionsofTable3(notreportedhere)addingadummyforteamswithexactlytwofounders.This
dummyispositiveandstatisticallyhighlysignificant(p<0.01),indicatingthatteamsoftwoareindeedmorelikelytosplitthe
equityequally.Thisevidenceissupportiveofthenotionthatthedecisiontosplittheequityequallyispartiallydrivenbya
desiretobalancevotingcontrol.Moreover,thisdesiretobalancecontroldoesnotaffecttheotherdeterminantsofequal
splitting,indicatingthatbalanceofcontrolisacomplementaryandnotanalternativeexplanation.(Notsurprisingly,the
onecoefficientthatisaffectedbytheinclusionofthe‘teamoftwo’dummyistheteamsizecoefficientitself.Itremains
negativebutbecomesinsignificant,implyingthattheteamsizeeffectisdrivenbythehigherincidenceofequalsharing
amongteamswithtwofounders.)
[25]
The empirical challenge is that preferences for equality are not directly measurable. However, one
implication from these theories is that preferences for equality should not only affect the division of
equity, but other team decisions, too. We already saw that equal splitters are more likely to balance
certain control functions, most notably board participation. We now ask whether equal splitters are also
more likely to adopt equal compensation packages. We focus on two central aspects of compensation,
namely founder salary and target bonus.29
Table 2 shows that there is a positive and statistically highly significant (p<0.01) correlation between
equal splitting and equal salaries, similar for the t-test in Table 1. Again we also consider a Probit
regression for equal salaries, using the independent variables of Model 3 of Table 3 (using Models 1 or 2
yields similar results), plus the equal split dummy. Table 8 shows that that the equal split coefficient is
positive and highly significant (p<0.01), confirming that the correlation between equal splits and equal
salaries continues to hold in a multivariate environment. We also obtain very similar results for our
measure of equal bonus targets. These tests indicate that equal splitting is positively correlated with
other dimensions of compensation equality, supporting the notion that teams that allocate founder shares
equally also exhibit a more general preference for team equality. Note also that these findings refute a
potential alternative hypothesis that founder systematically trade-off equal equity allocations against
unequal compensation packages.
Section8:Estimatingthevalueatstake
When founders agree to an equal split, how much money is at stake? In some sense, this question cannot
be answered, because all we know is that teams that split the equity equally chose to do so. Any counter-
factual estimates require bold assumptions and need to be interpreted with caution. We nonetheless find
it worthwhile to provide a quantitative approximation of how costly the decision to split the equity
equally might be.
Appendix B provides a detailed explanation of our methods to generate counterfactuals. In essence, we
use the regression for the share premia among unequal splitters (Model 3 of Table 4) to generate out-of-
sample predictions for the expected premium in the sample of equal splitters. We argue that this value
can be thought of as an implied transfer of shares, because the decision to split the equity equally means
29Again,Section3.2discussesthelimitationsofourempiricalmeasure.
[26]
that founders gave up the opportunity to receive their predicted share premia. To estimate average and
median values of the implied transfer for each team, we focus on the absolute value of the predicted
share premium. We then multiply this with a suitably chosen pre-money valuation to obtain the absolute
premium value. This provides a dollar amount for the money at stake in the decision of whether to agree
to an equal split. We also compare this premium value against the typical value of a founder stake.
Our counterfactuals estimate what equal splitters might have done if they had chosen to split the equity
unequally, in the way that typical unequal splitters would have done. In terms of our theory, we can
think of this as an artificial lowering of the negotiation costs k, sufficient to induce at least one founder
to ask for negotiation. Our counterfactuals hold the fundamental profitability (π) of the company
constant. Practically, we do not change the valuation of equal splitters, i.e., we assume that their
valuation discount, found in Table 5, remains intact.
Table 9 presents results from the counterfactual exercise. It reports the mean and medians for three
distinct calculations. The first pair of columns shows the estimates for the share premia. In the second
pair of columns we multiply the share premia with the undiscounted pre-money valuation to obtain a
simple estimate of the premium value. The third pair of columns uses a pre-money valuation discounted
at 12.75%; the appendix provides an explanation for this choice of discount rate.
The first row shows the predictions for equal splitters, the second row for unequal splitters. We note that
the mean values are well above the median values. This is typical for value distributions of
entrepreneurial companies, where the majority of companies have moderate valuations, but a few
companies have large positive values that raise the average well above the median.
The third row shows the actual premium values for unequal splitters. Comparing the second and third
rows, we note a considerable difference between the actual and predicted share premium. The reason for
this discrepancy is that the linear prediction model, while providing unbiased estimates of the average
premium, severely underestimates the absolute premium. In the appendix we explain a method-of-
moments rationale of using a stretch factor. While this factor preserves the mean of the predicted value,
it also matches the mean (or median) absolute deviation of the predicted value to the mean (or median)
absolute deviation of the actual value, within the sample of unequal splitters. The fourth row reports the
mean and median stretch factor. In the fifth row we apply the stretch factor to the absolute predicted
[27]
share premia among equal splitters. We find that the stretch factor has a large impact on the estimated
premium values, raising the discounted premium value above $500K.
Overall we note that the range of estimates for the value at stake with equal splitting varies from a low
of $175,945 (median of the discounted value prediction) to a high of $788,637 (mean of the stretched
value prediction). While we do not need to take a stance on which of these predictions is the most
reasonable, we believe that the main insight from Table 9 is that the values at stake are substantial.30
Another way of assessing the value at stake is to compare the predicted premium values to the total
value held by a typical founder. The sixth and seventh rows of Table 9 report the total value of shares
held by the average or median founder. Rows eight to ten then show the relationship between the
predicted premium value and this total share value. For equal splitters we find that the forgone premium
value is worth between 17% - 40% of the total value of shares.
The results of Section 6 already showed that equal splitting is associated with a valuation discount,
amounting to approximately 10% of the valuation. The results in this section do not even consider this
discount, but identify additional costs of equal splitting. At the risk of oversimplifying we average the
basic and stretched estimates, and find that the economic value at stake seems to amount to
approximately 10% (± 2%) of the firm equity, 25% (± 5%) of the average founder stake, or $450K (±
$120K) in net present value.
Section9:Conclusion
This paper is concerned with the first financial arrangement within a new firm, namely the division of
founder shares. It opens the black box of financial relationships within founding teams, something that
has received little attention in the prior literature. Arguably the division of equity is one of the key
decisions taken by founder teams, yet we find a surprisingly high incidence of equal splitting. We
develop a simple theory where founders have a choice between accepting an equal split without having
to negotiate, or undertaking costly negotiations to come up with a differentiated allocation of equity
shares. The theory generates several empirical predictions that are borne out in the data. Moreover,
simple calculations suggest that the amount of money at stake is far from trivial.
30Appendix2discussessomeadditionalrobustnesschecks.
[28]
Future research might look into other aspects of the financial contracts among founders. One important
area of research is examining to what extent founder financing is used as a substitute for external
financing (see also Robb and Robinson (2009)). Another interesting set of issues revolves around the
evolution of founder equity shares, and the use of founder vesting schedules that make the allocation of
equity shares contingent on milestones. More generally, we believe that there is a benefit to exploring
the financing arrangements among the founders themselves.
[29]
Figure1.HistogramofSharePremia
This figure shows the distribution of the share premia among the 1476 individual founders. The share premium is defined as the percentage equity
share of a founder minus the equal share, given by 1/N where N is the number of founders in a team.
0.1 .2 .3 .4
Fraction
-40 -20 020 40 60
premium
[30]
Table1.DescriptiveStatistics
This table provides descriptive statistics for all the variables used in the analysis. Variables are defined in Appendix 1. The table reports the number of observations, mean value and standard deviations in the full sample, in
the subsample of teams that split the equity equally, and in the subsample of teams that split the equality unequally. Panel A features all variables that vary at the level of individual-founders, Panel B those that vary at the
level of the team. (D) means that the variable is a dummy variable; (L) means that the natural logarithm of the variable is reported. The last column reports the results of t-tests for the difference between the equal and unequal
sample, where ***, ** and * indicate statistical significance at confidence levels of 99%, 95% and 90% respectively.
PanelA:IndividualfounderLevel
All Founders Equal Split Unequal Split t-test
Obs. Mean Std. Dev. Obs. Mean Std. Dev. Obs. Mean Std. Dev.
Share premium 1476 0.000 15.592 428 0.000 0.000 1048 0.000 18.507
Serial entrepreneur (D) 1476 0.318 0.466 428 0.311 0.463 1048 0.322 0.467
Prior years of work experience 1476 17.018 9.450 428 15.752 9.885 1048 17.534 9.222 ***
Idea person (D) 1476 0.232 0.422 428 0.217 0.413 1048 0.238 0.426
Founder capital invested 1476 0.055 0.113 428 0.059 0.111 1048 0.054 0.113
CEO position 1476 0.294 0.456 428 0.332 0.471 1048 0.279 0.449
**
Chairman position 1476 0.045 0.208 428 0.044 0.206 1048 0.046 0.209
CTO position 1476 0.146 0.353 428 0.187 0.390 1048 0.129 0.335
***
[31]
Table1.DescriptiveStatisticscontinued
PanelB:TeamLevel
All Teams Equal-split teams Unequal-split teams t-test
Obs. Mean Std. Dev. Obs. Mean Std. Dev. Obs. Mean Std. Dev.
Equal split 511 0.335 0.472 171 1.000 0.000 340 0.000 0.000 NA
Team size 511 2.888 1.259 171 2.503 0.877 340 3.082 1.374 ***
Team's CV of serial entrepreneurs 511 0.659 0.799 171 0.443 0.688 340 0.768 0.830 ***
Team's CV of work experience 511 0.336 0.366 171 0.298 0.378 340 0.356 0.358 *
Team's CV of idea people 511 0.408 0.694 171 0.258 0.551 340 0.484 0.746 ***
Team's CV of founder capital invested 511 0.447 0.650 171 0.213 0.479 340 0.564 0.692 ***
Team's mean # of serial entrepreneurs 511 0.337 0.353 171 0.334 0.388 340 0.339 0.335
Team's mean work experience 511 16.795 7.958 171 15.779 8.785 340 17.306 7.469 **
Team's mean # of idea people 511 0.239 0.346 171 0.213 0.351 340 0.252 0.343
Team's mean founder capital invested 511 0.058 0.095 171 0.060 0.101 340 0.057 0.092
Prior work experience together (D) 504 0.714 0.452 168 0.673 0.471 336 0.735 0.442
Prior founding experience together (D) 498 0.189 0.392 167 0.174 0.380 331 0.196 0.398
Friends-not-coworkers before founding (D) 496 0.375 0.485 167 0.347 0.478 329 0.389 0.488
Related to each other (D) 494 0.107 0.310 167 0.138 0.346 327 0.092 0.289
Quick negotiation (D) 446 0.469 0.500 153 0.608 0.490 293 0.396 0.490 ***
Time to outside finance 400 1.375 2.581 130 1.097 1.447 270 1.951 3.969 **
Pre-money valuation (L) 298 1.219 1.053 92 1.013 1.270 206 1.312 0.928 *
Geography: Canada (D) 511 0.047 0.212 171 0.064 0.246 340 0.038 0.192
Geography: California (D) 511 0.321 0.467 171 0.304 0.461 340 0.329 0.471
Geography: Massachusetts 511 0.170 0.376 171 0.164 0.371 340 0.174 0.379
Industry: IT (D) 511 0.564 0.496 171 0.596 0.492 340 0.547 0.499
Industry: Life Sciences (D) 511 0.313 0.464 171 0.304 0.461 340 0.318 0.466
CEO-Balance (D) 511 0.174 0.380 171 0.205 0.405 340 0.159 0.366
Board-Balance (D) 445 0.463 0.499 147 0.565 0.498 298 0.413 0.493
**
Equal salary (D) 279 0.301 0.460 94 0.457 0.501 185 0.222 0.416
***
Equal Bonus Target (D) 279 0.573 0.495 94 0.670 0.473 185 0.524 0.501
**
[32]
Table2.Correlations
Thistablereportsthepiecewisecorrelationcoefficientsforallthevariablesusedintheanalysis.VariablesaredefinedinAppendix1.PanelAfeaturesallvariablesthatvaryat
thelevelofindividualfounders,PanelBthosethatvaryattheleveloftheteam.(D)meansthatthevariableisadummyvariable;(L)meansthatthenaturallogarithmofthe
variableisreported.***,**and*indicatestatisticalsignificanceatconfidencelevelsof99%,95%and90%respectively.
PanelA:IndividualfounderLevel(n=1,476)
(1.) (2.) (3.) (4.) (5.) (6.) (7.) (8.)
1. Equal split (D)
2. Share premium 0.00
3. Serial entrepreneur (D) -0.01 0.17***
4. Prior years of work experience -0.09*** 0.05* 0.25***
5. Idea person (D) -0.02 0.14*** 0.14*** 0.11***
6. Founder capital invested 0.02 0.18*** 0.20*** 0.23*** 0.06**
7. CEO position (D) 0.05* 0.35*** 0.18*** 0.02 0.12*** 0.07**
8. Chairman position (D) 0.00 0.12*** 0.17*** 0.22*** 0.09*** 0.20*** -0.14***
9. CTO position (D) 0.07** -0.06* -0.02 -0.07** -0.02** -0.06** -0.27*** -0.09***
[33]
Table2.Correlationscontinued:PanelB:TeamLevel(n=511)
-1 -2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14 -15 -16 -17 -18 -19 -20 -21 -22 -23 -24 -25 -26
-0.22
***
-0.19 0.33
*** ***
-0.08 0.13 0.11
*****
-0.15 0.10 0.09
*** ** **
-0.26 0.08 0.19 0.09 0.12
*** * *** ** **
-0.12 0.15 -0.12 .12*
** *** ** **
-0.09 0.09 -0.29 0.26
** ** *** ***
0.29 0.11 0.16
*** ** ***
-0.11 0.19 0.17 0.24
** *** *** ***
0.24 0.09 -0.10 0.12 0.10
******* ****
0.08 0.46 0.16 0.12 0.16 0.18
* *** *** ** *** ***
0.12 -0.16
** ***
.12* 0.12 0.13 0.12
** ** ** **
0.20 -0.21 -0.12 -0.08 0.11
*** *** ** * **
16. Time to outside finance 0.16 -0.09 -0.11 -0.02 -0.02 0.00 -0.09 -0.06 -0.03 0.16 -0.01 -0.02 0.08 0.03 0.18
** ** *** ***
-0.13 0.13 -0.11 0.08

** ** *
0.09 -0.08 0.03 -0.17
** * **
-0.05 -0.15
***
0.06 0.14 -0.10 -0.31
** ** ***
-0.08 0.09 -0.22 -0.20 -0.08 .11* -0.05
** ** *** *** * **
-0.10 0.23 0.28 -0.09 0.02 -0.77
** *** *** ** ***
0.06 0.04 0.08 -0.02 -0.07 0.01 -0.01 0.06 0.01 0.08 -0.04 -0.02 0.00 0.03 -0.01 0.11 -0.02 -0.03 -0.04 0.04 -0.13 0.16
* * ** ** ***
0.14 0.07 -0.04 0.04 0.01 -0.11 0.02 -0.08 -0.01 -0.01 0.05 0.04 0.03 -0.04 -0.09 0.03 -0.01 0.05 0.08 -0.14 0.03 -0.04 0.01
** ** * ***
0.24 -0.19 -0.11 0.04 -0.02 -0.11 -0.04 -0.10 -0.08 0.05 -0.10 -0.08 -0.02 -0.07 0.10 -0.06 -0.07 -0.01 0.01 -0.07 0.13 -0.12 0.08 0.26
*** *** * * * ** ** ***
0.14 -0.08 -0.01 0.01 0.13 -0.10 0.03 -0.03 -0.07 0.04 -0.04 -0.05 -0.01 -0.06 0.03 0.03 -0.07 -0.09 0.08 0.00 0.10 -0.14 0.11 0.19 0.53
** ** * ** * *** ***
X
X
23. CEO-Balance (D)
24. Board-Balance (D)
25. Equal salary (D)
26. Equal Bonus Target (D)
1. Equal split (D) X
X
X
X
X
0.03 0.03 0.02 0.02 0.01 -0.01
-0.01 X
22. Industry: Life Sciences (D) -0.01 0.06 0.01 0.07 0.02 0.03 -0.01
-0.01 0.01 0.03 -0.05 -0.01 0.04
-0.06 0.00
21. Industry: IT (D) 0.05 -0.01 -0.04 -0.01 -0.03 -0.07
-0.02 -0.01 -0.04 0.03 -0.04 -0.04
20. Geography: Mass. (D) -0.01 0.02 0.02 0.00 0.02 -0.02 -0.05 0.04
-0.05 0.05 -0.05 0.05 X
0.01 -0.05 -0.02 0.00 0.01 0.02
0.02 X
19. Geography: California (D) -0.03 0.03 -0.02 -0.03 -0.01 -0.04
-0.02 0.06 -0.03 -0.02 -0.01 X18. Geography: Canada (D) 0.06 0.07 0.01 0.03 0.04 -0.02 0.00 0.07
0.02 -0.05 X
0.06 0.04 0.08 0.09 0.08 0.05
17. Pre-money valuation (L) 0.08 0.00 -0.06 -0.09
0.00 0.04 0.07 0.01 0.03 X15. Quick negotiation (D) -0.05 -0.07 -0.05 0.00
-0.02 -0.03 0.00 -0.04 X14. Related to each other (D) 0.07 -0.07 -0.01 -0.04 0.02
0.07 0.06 X13. Friends-not-coworkers before founding (D) -0.04 0.02 0.02 0.03 0.05 0.02 -0.05 0.00
12. Prior joint founding exp’c (D) -0.03 0.06 -0.04 -0.04 -0.02 X
11. Prior work exp’c together (D) -0.07 0.05 0.03 0.02 0.02 X
10. Team's mean founder capital invested 0.02 -0.06 -0.02 -0.06 0.05 X
9. Team's mean # of idea people -0.05 -0.05 -0.01 -0.04 0.00 X
8. Team's mean work experience 0.06 0.05 0.04 X
7. Team's mean # of serial entrepreneurs -0.01 -0.02 X
6. Team's CV of founder capital invested X
5. Team's CV of idea people -0.02 X
4. Team's CV of work experience X
3. Team's CV of serial entrep’s. X
2. Team size X
[34]
Table3.TheDeterminantsofEqualSplitting
This table reports the estimates from three Probit regressions where the dependent variable, called Equal split, takes the
value 1 whenever a team splits the equity equally, 0 otherwise. The unit of analysis is a team. All independent variables
are defined in Appendix 1. All four models control for team size (i.e., the number of founders), for geographic controls
(California, Massachusetts, Rest of US and Canada) and for industry controls (information technology, life sciences and
other) and for a set of dummy variables for the year when the founder split the equity. The four founder characteristics are
whether a founder is a serial entrepreneur (i.e., has prior founding experience), the number of years of work experience
before founding the venture, whether a founder came up with the idea on which the venture was based, and the amount of
initial founding capital provided by the founder. Model 3 add the average team values and the heterogeneity measures for
each of the four founder characteristics. Model 3 further adds four dummy variables about the prior relationships among
the founders, namely whether any of the founders had worked together before founding the venture, whether any of the
founders had founding a prior venture together, whether any of the founders had been friends but not co-workers before
founding the venture, and whether any of the founders were related to each other. (D) means that the variable is a dummy
variable; (L) means that the natural logarithm of the variable is reported. The table reports the coefficient estimate and its
associated robust standard errors. ***, ** and * indicate statistical significance at confidence levels of 99%, 95% and 90%
respectively.
[35]
Dependentvariable:Model1Model2Model3
EqualsplitCoef.(S.E.) SigCoef.(S.E.) SigCoef.(S.E.) Sig
Teamsize‐0.317

(0.07)*** 0.223

(0.07)***‐0.212

(0.07)***
Team'smean#ofserialentrepreneurs0.198

(0.18)0.217

(0.21)
Team'smeanworkexperience‐0.017

(0.01)*‐0.019

(0.01)**
Team'smean#ofideapeople‐0.131

(0.20)‐0.102

(0.20)
Team'smeanfoundercapitalinvested0.735

(0.69)0.790

(0.72)
Team'sCVofserialentrepreneurs‐0.169

(0.09)*‐0.162

(0.09)*
Team'sCVofworkexperience‐0.182

(0.20)‐0.199

(0.21)
Team'sCVofideapeople‐0.222

(0.11)**‐0.209

(0.11)**
Team'sCVoffoundercapitalinvested‐0.589

(0.11)***‐0.608

(0.12)***
Priorworkexperiencetogether(D)0.026

(0.15)
Priorfoundingexperiencetogether(D)‐0.074

(0.20)
Friendsnotcoworkersbeforefounding(D)‐0.018

(0.14)
Relatedtoeachother(D)0.402

(0.21)*
Geography:Canada(D)0.525

(0.29)*0.498

(0.30)*0.520

(0.31)*
Geography:California(D)‐0.007

(0.14)‐0.070 0.15 ‐0.097

(0.15)
Geography:Massachusetts0.009

(0.17)‐0.039

(0.17)‐0.006

(0.17)
Industry:IT(D)0.199

(0.19)0.262

(0.20)0.242

(0.21)
Industry:LifeSciences(D)0.157

(0.20)0.263

(0.23)0.280

(0.24)
SplityeardummiesYesYesYes
Constant0.379

(0.27)0.779

(0.33)**0.752

(0.36)**
NumberofObservation(Teams)511 511 511
Prob>χ20.0000 0.0000 0.0000
PseudoR20.0591   0.1376   0.1422  
[36]
Table4.AnalysisofSharePremia
ThistablereportstheestimatesfromOLSregressionswherethedependentvariableisthesharepremium,whichmeasuresthedifferencebetweenthepercentageofequityreceivedbyafounder
andthepercentageofequityreceivedunderanequalsharingrule.Theunitofanalysisistheindividualfounder,andthesampleincludesonlythoseteamsthatdidnotsharetheequityequally.All
independentvariablesaredefinedinAppendix1.Allmodelsusefourfoundercharacteristics,namelywhetherafounderisaserialentrepreneur(i.e.,haspriorfoundingexperience),thenumberof
yearsofworkexperiencebeforefoundingtheventure,whetherafoundercameupwiththeideaonwhichtheventurewasbased,andtheamountofinitialfoundingcapitalprovidedbythefounder.
Models3and4alsoincludedummyvariablesmeasuringfounderroles,namelywhetherthefounderisaCEO,ChairmanoftheboardorCTO.Allindependentvariablesaredemeanedattheteam
level.(D)meansthatthevariableisadummyvariable.Models1and3reportstheresultsfromOLSregressions.Models2and4reportstheresultsfromonemillioniterationsofOLSregressionswith
bootstrappedstandarderrors,whereonefounderperteamwasrandomlydroppedineachiteration.Thetablereportsthecoefficientestimateanditsassociatedrobuststandarderrors.***,**and*
indicatestatisticalsignificanceatconfidencelevelsof99%,95%and90%respectively.
DependentvariableModel1:OLSModel2:BootstrappedModel3:OLSModel4:Bootstrapped
SharepremiumCoef.(S.E.) SigCoef.(S.E.) SigCoef.(S.E.) SigCoef.(S.E.) Sig
Serialentrepreneur(D)8.877 (2.28) ***7.785 (2.23)***2.806 (2.13)4.449 (2.18)**
Workexperience‐0.113 (0.13) -0.132 (0.12)‐0.239 (0.13)* -0.179 (0.11)
Ideaperson(D)14.986 (2.71) ***9.975 (2.83)***9.011 (2.50)***6.647 (2.64)**
Foundercapitalinvested71.988(13.34) ***65.326 (15.48)***54.243 (12.58)***49.205 (14.42)***
CEOposition(D)0.000(0.00)***-0.408 (0.58)19.860 (1.82)***13.999 (1.79)***
Chairmanposition(D)21.504 (5.01)***17.182 (5.26)***
CTOposition(D)7.559 (2.14)***6.323 (2.03)***
Constant0.000 (0.00)***0.071 (0.00)

NumberofObservation1048708 1048 708
NumberofTeams340340 340 340
Prob>F0.00000.0000 0.0000 0.0000
PseudoR20.1896 0.1764   0.3393   0.2845  
[37]
Table5.TheEffectofEqualSplittingonValuation
ThistablereportstheestimatesfromtwoOLSregressionswherethedependentvariableisthe
naturallogarithmofthecompany’spre‐moneyvaluation,measuredatthetimeofthefirstexternal
financinground,providedtheroundoccurredwithinthreeyearsofthedateoftheequitysplit.The
unitofanalysisisateam.AllindependentvariablesaredefinedinAppendix1.Thekeyindependent
variableisEqualsplit,whichisadummyvariablethattakesthevalue1wheneverateamsplitsthe
equityequally,0otherwise.Allmodelscontrolforteamsize(i.e.,thenumberoffounders),forthe
timeelapsed(measuredinyears)betweenthedateoftheequitysplitandthedateofthefirst
externalfinancinground;forgeographiccontrols(California,Massachusetts,RestofUSand
Canada)andforindustrycontrols(informationtechnology,lifesciencesandother)andforasetof
dummyvariablesfortheyearwhenthefinancingroundoccurred.Thefourfoundercharacteristics
arewhetherafounderisaserialentrepreneur(i.e.,haspriorfoundingexperience),thenumberof
yearsofworkexperiencebeforefoundingtheventure,whetherafoundercameupwiththeideaon
whichtheventurewasbased,andtheamountofinitialfoundingcapitalprovidedbythefounder.
Model2includestheaverageteamvalueforeachofthefourfoundercharacteristics,aswellas
heterogeneitymeasuresforthefourfoundercharacteristics,namelythecoefficientofvariation
withinateam.Itfurtheraddsfourdummyvariablesaboutthepriorrelationshipsamongthe
founders,namelywhetheranyofthefoundershadworkedtogetherbeforefoundingtheventure,
whetheranyofthefoundershadfoundingapriorventuretogether,whetheranyofthefounders
hadbeenfriendsbutnotco‐workersbeforefoundingtheventure,andwhetheranyofthefounders
wererelatedtoeachother.(D)meansthatthevariableisadummyvariable;(L)meansthatthe
naturallogarithmofthevariableisreported.Thetablereportsthecoefficientestimateandits
associatedrobuststandarderrors.***,**and*indicatestatisticalsignificanceatconfidencelevels
of99%,95%and90%respectively.
[38]
Dependentvariable:Model1Model2
Premoneyvaluation(L)Coef.(S.E.)SigCoef.(S.E.) Sig
Equalsplit‐0.296
(0.14)**‐0.320
(0.15)**
Teamsize0.074
(0.05)0.131
(0.07)**
Timeuntilfirstinstitutionalroundoffinancing0.135
(0.08)0.120
(0.09)
Team'smean#ofserialentrepreneurs0.010
(0.20)
Team'smeanworkexperience0.012
(0.01)
Team'smean#ofideapeople0.331
(0.19)*
Team'smeanfoundercapitalinvested0.951
(0.59)
Team'sCVofserialentrepreneurs‐0.075
(0.09)
Team'sCVofworkexperience‐0.100
(0.23)
Team'sCVofideapeople‐0.203
(0.09)**
Team'sCVoffoundercapitalinvested0.034
(0.09)
Priorworkexperiencetogether(D)0.047
(0.14)
Priorfoundingexperiencetogether(D)‐0.138
(0.16)
Friendsnotcoworkersbeforefounding(D)‐0.225
(0.12)*
Relatedtoeachother(D)0.043
(0.24)
Geography,industryandfinancingyear
dummiesYesYes
Constant1.146
(0.30)*** 0.718
(0.40)*
 
NumberofObservation(Teams)298 286
Prob>F 0.0001 0.0000
RSquared0.1646   0.2159 
[39]
Table6.ExpectedandUnexpectedComponentsoftheEqualSplitEffect
This table reports the estimates from two OLS regressions where the dependent variable is the natural logarithm of the company’s pre-money
valuation, measured at the time of the first external financing round, provided the round occurred within three years of the date of the equity split.
The unit of analysis is a team. All independent variables are defined in Appendix 1. The key independent variables are the expected and
unexpected components of the equal split decision. The expected component is given by the predicted probability of an equal split, derived from
Model 4 of Table 3. The unexpected component is given by difference between the equal split dummy variable and the predicted probability of an
equal split, derived from Model 4 of Table 3. All models control for team size (i.e., the number of founders), for the time elapsed (measured in
years) between ate of the equity split and the date of the first external financing round; for geographic controls (California, Massachusetts, Rest of
US and Canada) and for industry controls (information technology, life sciences and other) and for a set of dummy variables for the year when
the financing round occurred. The four founder characteristics are whether a founder is a serial entrepreneur (i.e., has prior founding experience),
the number of years of work experience before founding the venture, whether a founder came up with the idea on which the venture was based,
and the amount of initial founding capital provided by the founder. Model 2 includes the average team value for each of the four founder
characteristics, as well as heterogeneity measures for the four founder characteristics, namely the coefficient of variation within a team. It further
adds four dummy variables about the prior relationships among the founders, namely whether any of the founders had worked together before
founding the venture, whether any of the founders had founding a prior venture together, whether any of the founders had been friends but not co-
workers before founding the venture, and whether any of the founders were related to each other. (D) means that the variable is a dummy
variable; (L) means that the natural logarithm of the variable is reported. The table reports the coefficient estimate and its associated robust
standard errors. ***, ** and * indicate statistical significance at confidence levels of 99%, 95% and 90% respectively.
Dependentvariable:Model1Model2
Premoneyvaluation(L)Coef.(S.E.) SigCoef.(S.E.) Sig
Expectedcomponentofequalsplit‐0.136 (0.43)‐0.627 (1.09)
Unexpectedcomponentofequalsplit‐0.326 (0.15)**‐0.31 (0.16)*
ControlsfromModel1ofTable5 Yes    
ControlsfromModel2ofTable5 Yes
NumberofObservation(Teams)286286
Prob>F0.00020.0000
RSquared0.1612 0.2161 
[40]
Table7.TheImpactofNegotiationLength
This table reports the estimates from two OLS regressions where the dependent variable is the natural logarithm of the company’s pre-money
valuation, measured at the time of the first external financing round, provided the round occurred within three years of the date of the equity split.
The unit of analysis is a team. All independent variables are defined in Appendix 1. The key independent variable are Equal and quick split,
which is a dummy variable that takes the value 1 whenever a team splits the equity equally and reaches an agreement within a day or less, 0
otherwise; Equal and slow split, which is a dummy variable that takes the value 1 whenever a team splits the equity equally but does not reach an
agreement within a day or less, 0 otherwise; Unequal and quick split, which is a dummy variable that takes the value 1 whenever a team splits the
equity unequally and reaches an agreement within a day or less, 0 otherwise. All models control for team size (i.e., the number of founders), for
the time elapsed (measured in years) between ate of the equity split and the date of the first external financing round; for geographic controls
(California, Massachusetts, Rest of US and Canada) and for industry controls (information technology, life sciences and other) and for a set of
dummy variables for the year when the financing round occurred. The four founder characteristics are whether a founder is a serial entrepreneur
(i.e., has prior founding experience), the number of years of work experience before founding the venture, whether a founder came up with the
idea on which the venture was based, and the amount of initial founding capital provided by the founder. Model 2 includes the average team
value for each of the four founder characteristics, as well as heterogeneity measures for the four founder characteristics, namely the coefficient of
variation within a team. It further adds four dummy variables about the prior relationships among the founders, namely whether any of the
founders had worked together before founding the venture, whether any of the founders had founding a prior venture together, whether any of the
founders had been friends but not co-workers before founding the venture, and whether any of the founders were related to each other. (D) means
that the variable is a dummy variable; (L) means that the natural logarithm of the variable is reported. The table reports the coefficient estimate
and its associated robust standard errors. ***, ** and * indicate statistical significance at confidence levels of 99%, 95% and 90% respectively.
Dependentvariable:Model1Model2
Premoneyvaluation(L)Coef.(S.E.) SigCoef.(S.E.) Sig
Equalandquicksplit‐0.395 (0.20)**‐0.469(0.21)**
Equalandslowsplit‐0.228 (0.22)‐0.293(2.23)
ControlsfromModel1ofTable5 Yes    
ControlsfromModel2ofTable5 Yes
NumberofObservation(Teams)275 275
Prob>F0.0001 0.0000
RSquared0.2408   0.2440 
[41]
Table8:EqualSplits,BalanceofControlandCompensationEquality
ThistablereportstheestimatesfromfourProbitregressionswherethedependentvariablesare:CEO‐Balance,whichtakesthevalue1ifnoneofthefoundersassumethepositionofCEO,oriftwoormore
founderssharetheCEOtitle;0otherwise.Board‐Balance,whichtakesthevalue1ifnoneofthefounderssitontheboard,oriftwoormorefounderssitontheboard;0otherwise.EqualSalary,whichtakesthe
value1iftheremainingfounders(atthetimeofthesurvey)receivedequalsalaries;0otherwise.EqualbonusTarget,whichtakesthevalue1iftheremainingfounders(atthetimeofthesurvey)receivedequal
bonuses;0otherwise.Theunitofanalysisisateam.AllindependentvariablesaredefinedinAppendix1.ThekeyindependentvariableisEqualsplit,whichisadummyvariablethattakesthevalue1whenevera
teamsplitstheequityequally,0otherwise.Allmodelscontrolforteamsize(i.e.,thenumberoffounders);forgeographiccontrols(California,Massachusetts,RestofUSandCanada)andforindustrycontrols
(informationtechnology,lifesciencesandother)andforasetofdummyvariablesfortheyearwhenthefoundersplittheequityforasetofdummyvariablesfortheyearwhenthefinancingroundoccurred.The
fourfoundercharacteristicsarewhetherafounderisaserialentrepreneur(i.e.,haspriorfoundingexperience),thenumberofyearsofworkexperiencebeforefoundingtheventure,whetherafoundercameup
withtheideaonwhichtheventurewasbased,andtheamountofinitialfoundingcapitalprovidedbythefounder.Theregressionincludestheaverageteamvalueforeachofthefourfoundercharacteristics,as
wellasheterogeneitymeasuresforthefourfoundercharacteristics,namelythecoefficientofvariationwithinateam.Itfurtheraddsfourdummyvariablesaboutthepriorrelationshipsamongthefounders,
namelywhetheranyofthefoundershadworkedtogetherbeforefoundingtheventure,whetheranyofthefoundershadfoundingapriorventuretogether,whetheranyofthefoundershadbeenfriendsbutnot
co‐workersbeforefoundingtheventure,andwhetheranyofthefounderswererelatedtoeachother.(D)meansthatthevariableisadummyvariable;(L)meansthatthenaturallogarithmofthevariableis
reported.Thetablereportsthecoefficientestimateanditsassociatedrobuststandarderrors.***,**and*indicatestatisticalsignificanceatconfidencelevelsof99%,95%and90%respectively.
Dependentvariable:CEOBalance(D)BoardBalance(D)EqualSalary(D)EqualBonusTarget(D)
 Coef.(S.E.) SigCoef.(S.E.) SigCoef.(S.E.) SigCoef.(S.E.) Sig
Equalsplit0.238(0.16)0.358(0.15)**0.851(0.20)***0.586(0.20)***
Teamsize0.03(0.06)0.103(0.07)‐0.176(0.13)‐0.062(0.08)
Team'smean#ofserialentrepreneurs‐0.106(0.25)0.336(0.22)‐0.394(0.32)0.148(0.30)
Team'smeanworkexperience‐0.001 0.01 ‐0.008(0.01)‐0.011(0.01)0.001(0.01)
Team'smean#ofideapeople0.05(0.21)‐0.057(0.21)‐0.230(0.32)‐0.335(0.27)
Team'smeanfoundercapitalinvested0.9920(0.72)0.135(0.71)3.035(1.06)***2.352(0.98)**
Team'sCVofserialentrepreneurs0.174(0.09)‐0.067(0.09)‐0.093(0.13)‐0.007(0.12)
Team'sCVofworkexperience‐0.256(0.22)0.208(0.20)0.311(0.30)0.434(0.29)
Team'sCVofideapeople‐0.145(0.12)0.052(0.10)0.223
(0.15)0.540(0.14)***
Team'sCVoffoundercapitalinvested0.058(0.12)‐0.220(0.11)**‐0.096(0.16)‐0.287 0.14 **
Priorworkexperiencetogether(D)‐0.136(0.17)0.130(0.16)‐0.182(0.23)‐0.176(0.21)
Priorfoundingexperiencetogether(D)0.035(0.21)0.136(0.19)‐0.106(0.30)‐0.312(0.27)
Friendsnotcoworkersbeforefounding(D)‐0.043(0.15)0.047(0.14)‐0.122(0.20)‐0.111(0.19)
Relatedtoeachother(D)0.188 0.23 ‐0.250(0.22)‐0.645(0.31)**‐0.386(0.29)
Geography,industryandsplityeardummiesYESYESYESYES
Constant‐1.105(0.38)‐0.058(0.37)‐0.958(0.51)‐0.644(0.49)

NumberofObservation(Teams)490428273273
Prob>F 0.1830.0030.0000.001
RSquared0.0782 0.0911 0.2073 0.1821 

[42]
Table9.EstimatesofImpliedTransfersamongFoundersthatSplitEqually
Thistableprovidescounterfactualcalculations,discussedinSection8.Appendix2containsadetaileddiscussion.
CounterfactualsSharesValueDiscountedvalue
 (Row)MeanMedianMeanMedianMeanMedian
Absolutepremiumcalculation
Predictionforequalsplitters17.94%7.47%$374,860$207,150$333,491$175,945
Predictionforunequalsplitters28.98%8.01%$518,588$295,292$479,327$274,227
Actualforunequalsplitters313.66%10.00%$733,550$338,333$677,371$317,079
Stretchedcalculation
Stretchfactor41.521.25
Predictionforequalsplitters‐stretched512.07%9.32%$788,637$368,745$570,065$258,678
Totalvalue
Equalsplitters639.95%50.00%$1,965,174$1,100,000$1,736,192$1,000,000
Unequalsplitters732.44%30.00%$1,763,421$1,040,000$1,628,499$963,018
Premiumasafractionoftotalvalue
Predictionforequalsplitters819.87%14.94%19.08%18.83%19.21%17.59%
Predictionforequalsplitters‐stretched930.21%18.64%40.13%33.52%32.83%25.87%
Actualforunequalsplitters1042.11%33.33%41.60%32.53%41.59%32.93%
Numberofobservations
Numberoffounders‐equalsplitters428235235
Numberoffounders‐unequalsplitters1048643643
Numberofteams‐equalsplitters1719292
Numberofteams‐unequalsplitters340206206
[43]
Appendix 1. Variable Definitions
Variable Description Question in Survey
Equity split
Founder’s share
premium
The percentage of equity received by each
founder as a result of the equity-split
negotiation, minus the amount that would
have been received if the equity had been
split equally
“% of company's equity received at
time of initial equity split” and
“Number of people who founded your
company”
Equal (unequal)
split
Dummy variable for whether all founders
received (did not receive) the same amount of
equity
(Calculated from Premium variable)
Duration of
negotiation
The elapsed time over which the founders
negotiated the equity split, categorized by: 1
day or less, 2 days to 2 weeks, 2 weeks to 2
months, 2 months to 6 months, More than 6
months
“How much time did the founders
spend negotiating the initial equity
split?”
Quick
negotiation
Dummy variable for whether the founders
negotiated the equity split quickly, using
various cutoffs for short durations
(Calculated from Duration variable)
Slow
negotiation
Dummy variable for whether the founders
negotiated the equity split for a long time,
using various cutoffs for long durations
(Calculated from Duration variable)
Individual-level Variables
Serial
entrepreneur
Whether the founder had prior founding
experience; dummy variable
“Previously founded another
company?”
Prior years of
work
experience
Number of years of prior work experience
before founding the venture
“Years of work experience before
founding this company”
Idea person Whether the founder came up with the idea
on which the venture was based; dummy
variable
“Founder whose idea it was to begin
this venture”
Founder capital
invested
Amount of initial founding capital provided
by the founder, categorized by: $0k, $1k -
$25k, $26k - $100k, $101k - $500k, More
than $500k
“Amount of founding capital
contributed by this founder”
CEO position Whether the founder received the CEO
position at time of founding
“Initial position in the company [CEO
choice]”
Chairman
position
Whether the founder received the Chairman
position at time of founding
“Initial position in the company
[Chairman choice]”
CTO position Whether the founder received the CTO
position at time of founding
“Initial position in the company [CTO
choice]”
Team-level Variables
Team size Number of people in the founding team “Number of people who founded your
company”
Prior work
experience
together
Whether any of the founders had worked
together before founding the venture
“Before founding this company, how
many of the founders had previously
worked together?”
[44]
Variable Description Question in Survey
Prior founding
experience
together
Whether any of the founders had founding a
prior venture together
“Before founding this company, how
many of the founders had founded
another company together?”
Friends-not-
coworkers
before founding
Whether any of the founders had been friends
but not co-workers before founding the
venture
“Before founding this company, how
many of the founders were friends but
not co-workers?”
Related to each
other
Whether any of the founders were related to
each other
“Before founding this company, how
many of the founders were related to
each other?”
Pre-money
valuation
The venture’s pre-money valuation in its first
round of outside financing, if raised within 3
years (note: sensitivity to this cutoff tested
for robustness)
“Round 1: Pre-investment valuation
($M)” [also includes questions about
participants in the round]
Round 1 date Date that first outside round of financing
closed (if round raised)
“Round 1: Approximate date of
completion”
Geography, Industry, and Year Dummies
Geography
dummies
Where the venture was located “State in which your company is
headquartered”
Industry
dummies
The venture’s industry: Technology vs. Life
Sciences (also collected: primary and
secondary industry segments)
“Please select one primary as well as a
secondary business segment if
applicable”
Split-year
dummies
Dummy variables capturing the year in which
the founding team split the equity
“When did the founders initially split
the equity?”
Financing-year
dummies
Dummy variables capturing the year in which
the venture raised it first round of outside
capital
“Round 1: Approximate date of
completion”
Balanced Control and Compensation Assessments
CEO-Balance Dummy variable that equals 1 if none of the
founders assume the position of CEO, or if
two or more founders share the CEO title; 0
otherwise.
(Calculated based on “Initial position
in the company” answers across the
founding team)
Board-Balance Dummy variable that equals 1 if (at the time
of the survey) none of the founders sit on the
board, or if two or more founders sit on the
board; 0 otherwise.
(Calculated based on answers to “Does
this executive sit on the board” across
the team)
Equal-Salaries Dummy variable that equals 1 if the
remaining founders (at the time of the
survey) received equal salaries; 0 otherwise.
(Calculated based on earliest “Annual
base salary” answers across the team)
Equal-Bonus-
Target
Dummy variable that equals 1 if the
remaining founders (at the time of the
survey) received equal bonuses; 0 otherwise.
(Calculated based on earliest “Cash
bonus received” answers across the
team)
[45]
Appendix 2. Description of Counterfactual Methodology
Our counterfactual calculations assume that, contrary to their revealed choice, teams that split
equally negotiate a non-equal split, using the same ‘principles’ as unequal splitters. In terms of
theory, our counterfactual exercise effectively looks at a scenario where we replace the true
negotiation costs k with a counterfactual negotiation cost k that is sufficiently low so that at least
one founder always wants to negotiate. Put differently, our counterfactual calculation examines
what would happen if a team with an unobservable preference for equal splits were to change its
mind and suddenly exhibit a preference for negotiating an uneven allocation of equity.
If equal splitters were to split the equity unevenly, some founders would receive more and others
fewer shares. We call an implied transfer the difference between the value of shares that a
founder receives under an equal split minus the value of shares that the founder would have
received under the counterfactual of an unequal split. The implied transfers within a team always
sums to zero, because one founder’s implied gain is another founder’s implied loss. To obtain an
idea of the size of these transfers we focus on the absolute values of the implied transfers.31
To construct a counterfactual share premium we use the information from the sample of non-
equal splitters to make an out-of-sample prediction for the equal splitters. Specifically, we use
the linear predicted values of Model 3 in Table 4 to construct predicted values for the sample of
equal splitters.
To calculate the absolute premium value, we consider both undiscounted and discounted
valuations. The specific choice of discount rates is always a contentious issue. We use a
pragmatic approach, noting that the valuation model of Table 5 effectively estimates a discount
rate. The coefficient for the time-to-finance variable is 0.12. Since the valuation is measured as a
natural logarithm, this implies a continuous time discount rate of 12%, or an annual discount rate
of Exp(0.12) 12.75%.
31Toprovideanexample,considerateamoffourfoundersthatsplittheequityequally,eachreceiving25%of
founderstock.Supposeourcounterfactualcalculationssuggestedthatunderanunequalsplit,thefounders
received35%,30%,22%and13%.Theaverageabsolutepremiumwouldbe(10%+5%+3%+12%)/4=7.5%,
suggestingthatrelativetotheunequalsplit,theequalsplitimplicitlyredistributes7.5%oftheequityperfounder.
Ifthecompanyhadapremoneyvaluationof$10M,theabsolutepremiumvaluewouldbe$750K.
[46]
Table 8 shows that the average (median) absolute share premium amounts to 7.94% (7.47%) of
the company. We compare this counterfactual value for equal splitters with the results for the
non-equal splitters, where the average (median) is 8.98% (8.01%). For the sample of non-equal
splitters, we can also calculate the actual average (median) absolute premium, given by 13.66%
(10%) as shown in Table 8. We note that the actual premium is considerably higher than the
predicted premium. The predicted value of the OLS model correctly estimates the mean of the
distribution, which is always zero in this model. However, due to its “averaging logic,” the
predicted values from the OLS model are likely to underestimate the variance of the underlying
distribution. As a consequence it might be argued that they underestimate the implied transfers.
A “method-of-moments logic” would suggest using a prediction model that fits not only the
mean of the distribution, but also its dispersion. One can construct several such estimators, but
for brevity’s sake we focus on one such approach. We note that the ratio of the actual and
predicted average (median) absolute share premium is 1.52 (1.25), as shown in the fifth row. We
therefore propose a “stretched” linear prediction model where the predicted values are multiplied
by the stretch factor of 1.52 (or 1.25 for median). Since the expected values in our model are
always zero, the same remains true for the “stretched” linear prediction model, thus guaranteeing
that the predicted share premium remains unbiased. In addition, the stretched predicted share
premium has the property that, in the sample of non-equal splitters, the predicted absolute share
premium exactly equals the actual absolute share premium. Table 8 shows how this impacts the
average (median) absolute predicted share premium for equal splitters. The predicted average
(median) share premium rises to 12.07% (9.32%). As a consequence, the predicted average
(median) premium value rises to $570,065 ($258,678).
Rows 6 and 7 in Table 8 report the total shares and total share value for the average (median)
founder, broken down by equal versus non-equal splitters. These numbers in turn allow us to
provide an idea of how big the implied transfers are. In particular we can compare the predicted
absolute share premia against the total shares held by a typical founder. These calculations are
shown in rows 8, 9 and 10.
We performed three additional sets of calculations not reported in Table 8. First we examined
how sensitive the results were to the discount rate. We doubled the discount rate to 0.24 in log
terms (or Exp(1.24) 27.13%) and found that the predicted premium values fell by another 10-
[47]
12%. Second, we consider the impact of missing valuations. We used a simple linear prediction
model to generate predicted valuations for all companies that obtained some outside financing.
Specifically we regressed the valuation on all control variables of Model 2 in Table 5, except
that, due to missing financing dates for companies with unreported valuations, we dropped the
time to finance variables and replaced the financing year dummies with founding year dummies.
We then redid all the calculations but found that the predicted absolute discounted premium
value was within 1% of the estimates reported in Table 8.
Our counterfactual calculations rest on several strong assumptions. By construction our predicted
values take into account the observable differences between equal and unequal splitters.
However, our predictions do not take into account unobservable differences between these two
subgroups. Put differently, the assumption that equal splitters would split the equity using the
rules of the non-equal splitters allows us to make the counterfactual comparison. However, it is a
strong assumption, so we caution against too literal an interpretation of these counterfactual
calculations.
[48]
References
Adams JS. 1965. Inequity in social exchange. In L Berkowitz (Ed.), Advances in Experimental
Social Psychology, Vol. 2: 267-299. Academic Press: New York
Arora A. 1996. Contracting for tacit knowledge: The provision of technical services in
technology licensing. Journal of Development Economics 50(August): 233-256
Arrow KJ. 1962. Economic welfare and the allocation of resources for inventions. In R Nelson
(Ed.), The Rate and Direction of Inventive Activity: Economic and Social Factors. Princeton
University Press: Princeton, NJ
Ashlagiy I, Karagözoglu E, Klaus B. 2008. A noncooperative support for equal division in estate
division problems. Harvard Business School Working Paper 09-069
Bartling B, von Siemens F. 2007. Equal sharing rules in partnerships. Working Paper, University
of Munich
Binmore K, Rubinstein A, Wolinsky A. 1986. The Nash bargaining solution in economic
modelling. Rand Journal of Economics 17(2): 176-188
Bose A, Pal D, Sappington D. 2010. Equal pay for unequal work: Limiting sabotage in teams.
Journal of Economics & Managerial Strategy 19(1): 25-53
Dyer JH, Singh H, Kale P. 2008. Splitting the pie: Rent distribution in alliances and networks.
Managerial and Decision Economics 29: 137–148
Encinosa WE, Gaynor M, Rebitzer JB. 2007. The sociology of groups and the economics of
incentives: Theory and evidence on compensation systems. Journal of Economic Behavior
and Organization 62(2): 187-214
Evans DS, Leighton LS. 1989. Some empirical aspects of entrepreneurship. American Economic
Review 79: 519-535
Fehr E, Schmidt KM. 1999. A theory of fairness, competition, and cooperation. Quarterly
Journal of Economics 114(3): 817-868
Fehr E, Schmidt KM. 2006. The economics of fairness, reciprocity and altruism: Experimental
evidence and new theories. In S Kolm, JM Ythier (Eds.), Handbook of the Economics of
Giving, Altruism, and Reciprocity, Vol. 1. North-Holland
Gaver KM, Horsky D, Narasimhan C. 1988. Invariant estimators for market share systems and
their finite sample behavior. Marketing Science 7: 169-186
Gompers P, Lerner J, Scharfstein D, Kovner AR. 2010. Performance persistence in
entrepreneurship and venture capital. Journal of Financial Economics 96(1)
Graham J, Harvey C. 2001. The theory and practice of corporate finance: Evidence from the
field. Journal of Financial Economics 60: 187-243
Granovetter M. 1985. Economic action and social structure: The problem of embeddedness.
American Journal of Sociology 91(3): 481-510
Granovetter M. 2005. The impact of social structure on economic outcomes. Journal of
Economic Perspectives 19(1): 33-50
Gulati R, Wang L. 2003. Size of the pie and share of the pie: Implications of structural
embeddedness for value creation and value appropriation in joint ventures. Research in the
Sociology of Organizations 20: 209-242
[49]
Hauswald RBH, Hege U. 2006. Ownership and control in joint ventures: Theory and evidence.
Mimeo, HEC Paris
Holmstrom B. 1982. Moral hazard in teams. Bell Journal of Economics 13: 324-340
Kaplan SN, Stromberg P. 2004. Characteristics, contracts, and actions: Evidence from venture
capitalist analyses. Journal of Finance 59: 2173-2206
Lazear EP, Rosen S. 1981. Rank-order tournaments as optimum labor contracts. Journal of
Political Economy 89: 841-864
Lemmon M, Roberts M, Zender J. 2010. Back to the beginning: Persistence and the cross-
sectional distribution of capital structure. Journal of Finance Forthcoming
Leventhal GS. 1976. The distribution of rewards and resources in groups and organizations. In L
Berkowitz, E Walster (Eds.), Equity Theory: Toward a General Theory of Social
Interaction: 92-133. Academic Press: New York
Puri M, Robinson DT. 2007. Optimism and economic choice. Journal of Financial Economics
86(1): 71-99
Robb A, Robinson DT. 2009. The capital structure decisions of new firms. Available at SSRN:
http://ssrn.com/abstract=1345895
Robinson D, Stuart T. 2007. Financial contracting in biotech strategic alliances. Journal of Law
and Economics 50(3)
Rubinstein A. 1982. Perfect equilibrium in a bargaining model. Econometrica 50: 97-110
Ruef M, Aldrich HE, Carter N. 2003. The structure of founding teams: Homophily, strong ties,
and isolation among U.S. entrepreneurs. American Sociological Review 68: 195-222
Uzzi B. 1997. Social structures and competition in interfirm networks: The paradox of
embeddedness. Administrative Science Quarterly 42(1): 35-67
Wasserman N. 2003a. Founder-CEO succession and the paradox of entrepreneurial success.
Organization Science 14(2): 149-172
Wasserman N. 2003b. Ockham Technologies: Living on the razor's edge, Harvard Business
School case no. 804-129:
Wasserman N. 2006. Stewards, agents, and the founder discount: Executive compensation in
new ventures. Academy of Management Journal 49: 960-976
Wasserman N. 2011. Founding Dilemmas. Princeton University Press: Princeton, NJ
Wasserman N, Marx M. 2008a. Split decisions: Social and economic effects on founding team
turnover. Academy of Management Annual Meeting
Wasserman N, Maurice LP. 2008b. Savage Beast (A) & (B). Harvard Business School case nos.
809-069, 809-096
... One exception is Hellmann and Wasserman (2016) who model this process and test it using survey data. A difficulty of equity splits is an ability to correctly predict each founder's relative value; moreover, a split that does not allocate equal equity among all founders signifies that some founders may be more valuable than another founder (cf. ...
... For example, Breugst, Patzelt, and Rathgeber (2015) develop a case study of eight teams and find that there is substantial variation in the perceived distributive justice of equity splits across and within teams in their sample. In their survey data of innovation-driven ventures, Hellmann and Wasserman (2016) find that 32 percent of founding teams equally split venture equity and that 42 percent of teams decide on this split within one day. Furthermore, they find that this equal split is correlated with a lower likelihood of raising outside capital. ...
... (i) Relationships It is important for cofounders to judiciously decide the kind of startup cofounders they should take along, with the diversity of skillset perspective, (ii) Roles It covers the decision regarding the division of key responsibilities, assignment of designations, and the allocation of decision-making powers amongst cofounders, and (iii) Rewards This emphasizes on the division of ownership equity amongst cofounders, as much as the distribution of resultant benefits amongst them. A misalignment in the three Rs can result in a conflict (Hellmann & Wasserman, 2017;Wasserman, 2013). ...
Article
Technology startups are exposed to multiple challenges. One of the key challenges is conflict, as it has a decisive role in the technology startup evolution. This study examines the effect of conflict on the following two counts: How do conflicts differentiate the success or failure of technology startups? How do conflicts impact the startup lifecycle comprising multiple stages of formation? The studies on conflict are few in the context of an emerging economy like India. This study explores the role of conflict by gathering primary data from 151 cofounders (101 who have experienced failure and 50 who are successful and continuing their operations) from India's six leading technology startup hubs. The presence of cofounder conflicts or Investor conflicts increases the odds of failure of technology startups. 1. The cofounder(s) should navigate and resolve the potential conflict issues related to relationships, roles, and rewards. Besides ensuring the cofounder's agreement, having an effective issue resolution mechanism is required. 2. The strategic intervention of investors coupled with the proper governance structure can lend a helping hand in minimizing conflicts. The epiphanies from cofounder(s) perspectives offered practical suggestions for conflict resolution between cofounders and investors.
... The average team size was 2.88 (SD = 1.18) and ranged from two to eight members. This size is consistent with previous studies on early ventures (e.g., 2.8 in Hellmann & Wasserman, 2017;or 2.3 in Nuscheler, Engelen, & Zahra, 2019) and roughly matches the team size of 2.4 members reported in a representative survey of startups in Germany (Ripsas & Tröger, 2015). On average, the teams had worked together for 2.92 years (SD = 1.28) in their current compositions. ...
Article
Full-text available
Greed is an excessive form of self-interest that leads individuals to pursue material wealth and immaterial desires with little or no concern for the well-being of others. While the literature suggests that greed results in managers acting unethically, a situational strength perspective suggests that this may not always be the case. Therefore, this paper aims at understanding how the new venture context, particularly the role of a team member's affective and cognitive trust in the new venture team, shapes the manifestation of greed in entrepreneurs’ unethical pro-organizational behavior—behavior that allows new venture team members to advance their ventures despite the violation of social norms. Consistent with our theorizing, we find that new venture team members’ affective and cognitive trust in their teams shape the relationship between greed and unethical pro-organizational behavior in opposing ways. Particularly, higher levels of greed are more likely to be connected to unethical pro-organizational behavior when a member's affective trust in the team is high and cognitive trust in the team is low. Our study offers implications for the entrepreneurship and management literatures, alongside implications for practice.
... Moreover, because past performance may influence both R&D decisions and subsequent survival, we created a dummy variable (past performance) that equals 1 if a firm's net profit in the previous year was positive and 0 otherwise (Xu et al., 2019). The capital structure of equity ownership is an important factor for new ventures (Hellmann & Wasserman, 2017), so we controlled for CEO equity, measured as the percentage of equity held by an owner-CEO. Strategic alliances and R&D cooperation foster innovation efficiency and performance (Haeussler et al., 2012;Ortega-Argilés et al., 2005), so we included cooperation, a dummy variable that equals 1 if the focal venture had cooperated with universities, government agencies, or other firms and 0 otherwise. ...
Article
Full-text available
Plain English Summary While capital is essential for new ventures to innovate and survive, is having more of it always good? Our research shows “No” because more money may “spoil the child” by reducing the benefits that new ventures enjoy from R&D investment. We analyzed 791 new technology ventures across six years and found evidence of a side effect of munificent financial resources, such that when ventures have high levels of financial munificence, they garner fewer survival benefits from increasing R&D. This side effect is weakened when ventures have CEOs who are more experienced, highly educated, or female. These findings extend previous research on the limitations of financial munificence by showing its negative moderating effect on the R&D–survival relationship. For entrepreneurs and venture capitalists in the industry, we advise caution regarding the role of abundant financial resources in new ventures.
... Conditional on this entry decision,Hellmann and Wasserman (2017) highlight the important role of the founding teams' first split of equity for startup outcomes. ...
... Finally, decisions on new product development are likely influenced by other stakeholders such as employees, investors, or board of directors where terminating a project could mean significant layoffs or a drastic decline in stock price/firm value. Thus, we encourage future research to employ more fine-grained manipulations such as teams with and without equity ownership (Hellmann & Wasserman, 2017), ventures with various stakeholder powers (Hampel et al., 2020), resource scarce versus resource abundant ventures (Hanlon & Suanders, 2007), pre-revenue versus post-revenue ventures (Marvel et al., 2021), family versus non-family ventures (Ko et al., 2021), and/or corporate versus noncorporate ventures (Covin et al., 2018) as they examine entrepreneurial persistence with losing projects. ...
Article
Full-text available
Persisting with a losing project (i.e., a new product development project facing superior competition) is a social endeavor that can increase the costs of failure to the entrepreneur and other stakeholders. Yet, it tends to be explained almost exclusively in terms of intrapersonal predictors, such as the sunk cost fallacy. This paper examines whether, how, and under which conditions interpersonal influence, such as the intensity of a team’s recommendation to persist with a losing project, encourages entrepreneurs to persist. Drawing from the psychologies of escalation and self-regulation, we build a model of entrepreneurs’ undue persistence that we test through experimental design and conjoint analysis. We find that an entrepreneur’s decision to persist with a losing project is determined partly by the team’s recommendation to persist and that the strength of this effect varies across entrepreneurs based on their self-regulation and experience.