4.

RELATION TO FINANCIAL CONTRACTING THEORIES

In this section, we interpret our results relative to financial contracting theories. We focus on
three general classes of theories: (1) classical principal-agent; (2) control theories; and (3) debt
theories.We consider the extent to which the contract provisions we have examined are consistent
with the assumptions and predictions of the theories. We augment this discussion and analysis
with cross-sectional regressions when appropriate. We conclude this section by discussing the
results for the general classes of theories in relation to more recent VC specific theories.
4.1. Allocation of cash flow rights and pay-performance incentives
4.1.1. Theoretical predictions. The first set of theories deals with the allocation of cash
flow rights between the VC and entrepreneur. The classical principal-agency approach, pioneered
by Holmstr¨om (1979), assumes that the agent’s effort is unobservable to the principal. Signals,
such as firm output or profits, however, are correlated with effort and can be contracted on. The
optimal incentive contract ensures that the agent puts in enough effort by making the agent’s
compensation dependent on the outcome of the signals. In the basic model, and in the absence
of risk-aversion, the investor maximizes the sensitivity of the agent’s compensation to the signal.
Moreover, it is in the investor’s interest to make the entrepreneur’s compensation contingent on
as many verifiable signals correlated with effort as possible (the “Informativeness Principle” of
Holmstr¨om (1979); see also Harris and Raviv (1979), Innes (1990)).
The result that the entrepreneur’s pay-performance sensitivity should be maximized may
not hold if the entrepreneur is risk averse or if multitasking problems exist. Compensation
contingent on performance is costly because a risk-averse entrepreneur requires a higher
16 REVIEW OF ECONOMIC STUDIES
level of compensation to offset the extra risk. Prendergast (1999) mentions several potentially
testable predictions: (1) entrepreneurs who are more risk averse will have less pay performance
sensitivity. (2) Noisier performance signals impose more risk on the entrepreneur, making
pay performance sensitive compensation more expensive. Less noisy performance measures
should be given relatively higher weight in compensation. (3) The higher the positive effect
on performance for a given effort by the entrepreneur, the higher the pay-performance
sensitivity should be. (4) If entrepreneurs differ in ability, performance sensitive contracts
will be relatively more attractive to better entrepreneurs (holding risk-aversion constant). This
gives rise to selection effects (see Lazear, 1986). Investors can screen out entrepreneurs by
offering high-powered compensation contracts because only high-ability entrepreneurs will
accept such contracts. Hence, the higher the asymmetric information regarding the ability of the
entrepreneurs, the higher the pay performance sensitivity in the contracts offered by the investor.
Multi-tasking models (Holmstr¨om and Milgrom (1991), Baker (1992)) assume that there
are several activities for which the entrepreneur needs to exert effort. This leads to a problem
of potential “gaming”, where the entrepreneur will only exert effort in the activity whose
signal is most rewarded in the compensation contract. As a result, overall incentives have to
be muted in the optimal compensation contract, and the principal will tend to rely more on
subjective performance evaluation. This suggests that compensation should be less dependent on
performance and more dependent on subjective performance measures in situations with gaming
potential.
4.1.2. Empirical results. The descriptive analysis of cash-flow rights in Table 2 seems
largely consistent with both the assumptions and predictions of the classical principal agent
approach. In our financings, the entrepreneur gets a large fraction of equity in the firm and
the entrepreneur’s equity stake increases with performance. Moreover, Table 3 shows that the
contracts often condition the entrepreneur’s equity compensation on a multitude of financial and
non-financial signals, consistent with the “Informativeness Principle”.
To test the models, we would ideally plot the entrepreneur’s compensation as a function
of the venture’s value, and examine what independent variables determine the shape of this
function. The pay-performance sensitivity of the founder’s compensation is increasing both in
the function’s slope—the fraction of the value promised to the entrepreneur at a given value—
and its convexity—the increase in the fraction of the value given to the entrepreneur as value
increases.
Because it is hard to capture these aspects in one single measure and no measure is perfect,
we examine several. First, we look at the percentage of the residual cash flow rights allocated
to the founder (FRCFs). This captures the compensation function slope given that the venture is
worth more than the liquidation claims. This percentage, unfortunately, is not ideal, because it is
also potentially a function of the (unobservable) quality of the entrepreneur.
Second, we look at the change in the fraction of FRCFs as a function of vesting. As this
fraction becomes more state-contingent, the compensation function becomes more convex and
the pay-performance sensitivity increases. We separate contingencies into the sensitivity due to
performance vesting—financial or non-financial—and the sensitivity due to time vesting of the
founders’ shares. This is a relevant separation in theory as time vesting can be interpreted as
a form of subjective performance evaluation. With time vesting, the founder’s compensation is
contingent on the board’s decision to retain the founder, rather than on explicit benchmarks. This
might be expected to be more likely when multi-tasking is present. 11 We measure the change
11. This also is a prediction from the control theory of Hart and Moore (1994) that we describe in the following
section. Vesting makes it more costly for the founder to leave the company which mitigates a potential hold-up problem.
KAPLAN & STRO¨MBERG 17
in FRCFs both on an absolute scale and relative to the level of the FRCF. We think the relative
change in FRCF is more indicative of incentives because it measures the fraction of the founders’
equity that is at risk, therefore, controlling for unobservable quality.
Third, we look at the degree to which the founder has claims to the liquidation proceeds.
Giving the founder part of the liquidation cash flow rights lowers the convexity of the
compensation function and leads to lower pay performance sensitivity.
The key independent variables in our analyses are pre-revenue, repeat entrepreneur, and
months since the first VC round. Pre-revenue equals one if the company had no revenues at
the time of financing, and zero otherwise. It seems likely that uncertainty and asymmetric
information are greater for pre-revenue firms. It also is arguable that incentive problems are
more severe for early stage ventures since the entrepreneur’s effort is relatively more important.
As Myers (2000) assumes, this likely would be the case of the development of a new technology
or concept. Once the technology or concept has been created or validated, the founders are more
replaceable with outsiders.
Repeat entrepreneur takes the value of one if the founders previously founded a company
that was either taken public or acquired by a public company, and zero otherwise. This provides
a measure of the extent to which VC may have prior information on the quality of the founders.
Such information reduces the likelihood of adverse selection and the uncertainty of the venture. 12
If this is important, we would expect repeat entrepreneurs to have lower powered incentives. On
the other hand, repeat entrepreneur also is likely to be correlated with higher outside wealth and
lower risk aversion. If repeat entrepreneur measures differences in risk aversion, we would expect
repeat entrepreneurs to have higher powered incentives.
Months since the first VC financing round is another measure of asymmetric information.
VC uncertainty about management quality and asymmetric information should be highest in the
first VC financing round and should decrease thereafter.
Table 4 presents the results of our cross-sectional regression analyses. The regressions
include controls for industry volatility, industry research and development (R&D), industry
size, company location, and VC.13 We also control for the amount of funds invested by
the VC by including in the accumulated VC investment as a control variable. 14 Finally, we
attempt to include a control for performance. Although we have incomplete data on accounting
performance, we have more complete data on financing terms. Accordingly, we measure
performance based on the annual return on the securities since the initial investment. When
the company performs poorly, the VCs will pay less for the securities acquired in the round.
A problem with using this return as an independent variable is that the round price is endogenous
and related to the percentage of equity allocated to the VC in the round. To address this, we
include dummies for whether the return is in the upper or lower quartile rather than including the
return directly.
The first regression in Table 4 examines the determinants of the fraction of equity (cash flow
rights) allocated to the entrepreneur. The only significant result is that as the VC-entrepreneur
relationship progresses, the founder’s equity stake declines. This is consistent with classical
principal agent theory—lower powered incentives are required as asymmetric information
declines. This also may be a function of the need to provide newly hired managers with equity
incentives as the company grows. Query 3
The second and third regressions of Table 4 analyse the determinants of absolute and
relative sensitivity to explicit performance benchmarks. Explicit performance-based equity
12. We use the company business plan, the VC analysis, and company web sites to obtain this information.
13. The table provides a more detailed description of the control variables in the regressions.
14. Because this is potentially endogenous, we instrument this variable with median industry capital expenditures
and yearly time fixed effects using two-stage least squares.

compensation is used more in pre-revenue ventures, less for repeat entrepreneurs, and earlier in
VC-founder relationships. All three of these results are consistent with asymmetric information
and moral hazard models where pay performance sensitivity increases when uncertainty about
venture quality is higher. The negative coefficient on repeat entrepreneur suggests that repeat
entrepreneur picks up information effects rather than risk aversion effects. Also, the negative
coefficient on the return variable indicates that the use of explicit performance benchmarks
increases with poor performance.
The results imply economically meaningful effects. The relative (absolute) sensitivity to
explicit performance benchmarks is 4·7% (1·6%) greater in pre-revenue companies and 2·2%
(0·7%) lower when repeat entrepreneurs are present. These compare to mean values for relative
and absolute sensitivities of 3·5% and 1·1%, respectively.
The fourth and fifth regressions analyse the determinants of absolute and relative pay
performance sensitivity due to time vesting. Time vesting is significantly higher for pre-revenue
ventures and, in the relative regressions, for early VC-founder relationships. The relative and
absolute sensitivities to time vesting for pre-revenue companies are large at 23·6 and 8·5%,
respectively. Again, this is supportive of asymmetric information and moral hazard models.
Interestingly, the signs on the industry ratios are opposite the ones in the performance
benchmarks regressions. The contracts in high volatility, high R&D, and smaller, presumably
less established industries seem to use time vesting rather than explicit performance benchmarks
to induce pay-performance sensitivity. One interpretation, consistent with theory, is that these
are environments where explicit performance signals are noisier measures of true performance.
Because of this, explicit performance benchmarking will be more costly, both because of
managerial risk-aversion as well as multi-tasking or gaming problems.
Finally, in the sixth regression, we use a logit specification to analyse the determinants of
founder liquidation rights.15 While pre-revenue is not significant, both repeat entrepreneur and
months since first VC round are positive and significant. Hence, when the founder has been
successful in the past, and as the VC learns more about the firm over time, the founder has
more liquidation rights. Again, this is consistent with less pay-for-performance as asymmetric
information declines. Again, this result is not consistent with repeat entrepreneur as a measure
of (less) risk aversion.
The results in Table 4 control for VC and location fixed effects. The use of performance
benchmarks varies significantly among VCs. Also, explicit performance benchmarks are used
significantly less in California ventures. Hence, contract design may be affected by different
contracting “styles” for different VCs and markets.
To sum up, the regressions in Table 4 are largely supportive of classical principal agent
theories and their screening implications. As uncertainty about the quality of the venture and
the founder increases, the VC increases the pay performance sensitivity by making the founder’s
cash flow compensation increasingly convex in performance through more explicit performance
compensation, more time vesting, and fewer liquidation cash flow rights. Also, as explicit
performance signals become noisier measures of true performance, the contracts substitute
explicit performance benchmarks with more vesting and lower liquidation cash flow rights.
4.2. Control theories: cash flow verifiable but not actions
4.2.1. Theoretical predictions. Board rights and voting rights give the controlling party
the right to decide on any action that is not pre-specified in the original contract. Such rights
15. The logit regression excludes observations for VCs that did not allocate founder liquidation rights in any of
their investments. The results are qualitatively similar when we exclude the VC dummies and include all observations.
20 REVIEW OF ECONOMIC STUDIES
are valuable in an incomplete contracting world, when it is not feasible or credible to specify all
possible actions and contingencies in an ex ante contract. Incomplete contracting and control
rights were introduced by Grossman and Hart (1986) and Hart and Moore (1988, 1990).16
These theories change the traditional principal-agent model assumptions by assuming that
actions are observable, but not verifiable. Output and monetary benefits may or may not be
contractible. As a result, control rights that determine who chooses which action to take will
be important.
Two important papers that take this approach are Aghion and Bolton (1992) and
Dewatripont and Tirole (1994). In Aghion and Bolton (1992), the project yields both monetary
benefits that are verifiable and transferable to outside investors, and private benefits or actions
that are non-verifiable and go only to the entrepreneur. The magnitude of these benefits, in turn,
depends on what (non-verifiable) action is taken with respect to the project. This introduces
a conflict of interest. Aghion and Bolton (1992) show that as the external financing capacity
of the project increases (i.e. the higher the profitability of the project and the lower the
conflicts of interest), control moves from more investor control to more entrepreneur control.
In particular, for projects with high external financing capacity, the entrepreneur should always
have control. As external financing capacity decreases, there should be state-contingent control
similar to a debt contract that transfers control to investors only in bad states of the world.
Finally, for projects with low external financing capacity, the investor should always be allocated
control.
Dewatripont and Tirole (1994) build on Aghion and Bolton (1992) by focusing on the
optimal correlation between control rights and cash flow rights. In their model, the entrepreneur
always prefers to take the riskier action (e.g. continuing the project), even though it will be ex post optimal to
take the less risky action (e.g. liquidate the project) when the project is doing badly.
They show that when performance is poor, the party in control should have a cash flow claim
that is concave in performance (such as debt), while when performance improves more control
should be transferred to a holder of a convex cash flow claim (such as equity).
4.2.2. Empirical results. The extensive use of control rights in our sample is broadly
consistent with the assumptions and predictions of these theories. State-contingent board and
voting control rights also are common features in our contracts. This contracting is more
elaborate than that in ordinary debt contracts that only give liquidation rights in case of default on
a promised payment. As Table 3 shows, control can be made contingent on financial performance
or on non-financial events independently of the division of cash-flow rights. These results are
broadly consistent with Aghion and Bolton (1992). We now examine the predictions of these
theories in more detail by analysing the cross-section.
The control theories predict that as agency problems and financial constraints become more
severe, the contracts should change from entrepreneur control to state-contingent control to full
VC control. We interpret the situation where neither the VC nor the founder is in control as
similar to state-contingent control. For example, in boards where outside, jointly appointed,
board members are pivotal, it seems plausible that these members will vote with the VC as
founder performance declines. On the other hand, we assume that when the VC explicitly gets
control in the bad state, through a default board provision or state-contingent voting control, the
VC has relatively more control compared to cases where the outsiders are pivotal. Therefore,
for a given financing round, we let the board/voting control variables equal zero if the founder
always controls a majority of the seats/votes, one if neither the VC nor the founder has a majority
16. Also, see Hart (1995) and Hart (2001) for an overview of these theories.

Two-stage least squares and ordered logit regressions of VC board control, voting control, and automatic conversion
provisions on various independent variables for 213 investments in 119 portfolio companies by 14 VC partnerships.
Investments were made between 1987 and 1999. The degree of board and voting control variables take the value
of 0 if the founder always has control, 1 if neither the VC nor the founder has control, 2 if the VCs have control
only in the bad state, and 3 if the VCs always have control. Pre-revenue takes the value of one if the venture has no
revenues at the time of financing, and zero otherwise. Repeat entrepr. takes the value of one if the founder’s previous
venture was able to go public or was sold to a public company. Industry volatility is the value-weighted volatility of
the log stock return for the firm’s industry according to the Fama and French (1997) industry classification. Industry
fixed assets is the median fixed to total assets for all (public as well as closely held) firms in the venture’s four-digit
SIC industry according to Onesource. Industry size is equal to log sales, in millions, in the venture’s four-digit SIC
industry according to the 1997 U.S. census. VC cumulative investment is instrumented using two-stage least squares
with industry capital expenditures to sales as well as year dummies. Return takes the value of minus one if the ratio of
per share round price to first round price is in the lowest quartile, one if it is in the highest quartile, and zero otherwise.
White (1980) robust standard errors are in parentheses. Asterisks indicate statistical significance at the 1% ___; 5%__,
and 10%_ levels.
of the seats/votes in the bad state, two if the VC controls a majority in the bad state only, and
three if the VC controls a majority of the seats/votes in the good and bad states.17
The first two regressions presented in Table 5 use ordered logits to analyse board rights
and voting rights, respectively. VCs are more likely to have board and voting control in prerevenue
ventures, as the time since the first financing round increases, and in industries with
higher volatility and more fixed assets. VCs are more likely to have board control when returns
are high and less likely to have voting control with repeat entrepreneurs.
The coefficients are economically as well as statistically significant. For example, VC voting
control (group 3) is almost 30% more likely in pre-revenue companies while the other three states
17. The classification of the situation in which neither party is in control in the bad state is arguably ambiguous.
We have run alternative regressions (not in the table) in which we classify these situations as entrepreneur control (0) or
with the VC control in the bad state (2). Our results are qualitatively identical.
22 REVIEW OF ECONOMIC STUDIES
are each roughly 10% less likely. VC voting control is 20% less likely in companies with a repeat
entrepreneur while neither control in the bad state (group one) is 10% more likely.
These results indicate that when the uncertainty about the venture and the quality of the
founder is higher, the VC is allocated more control. In contrast, VC control increases as the VC–
entrepreneur relationship progresses. Although the uncertainty about the venture should decrease
over time, the VC also invests more funds. The founder’s stake declines over time as the VC
receives more control and cash flow rights in exchange for the additional funds invested.18 This
is consistent with Aghion and Bolton (1992) who predict that the investor should be allocated
more control as the financing constraint becomes more binding.
Of the other industry characteristics, the amount of fixed to total assets is significantly
positively related to VC voting control.19 One explanation for this is that such companies are
less reliant on the intangible human capital of the original founder, making intervention by the
VC more efficient.
The last regression uses the price at which the VC’s securities are automatically converted
into common equity as a measure of VC control. The lower this price is, the more states in which
the VC has pre-committed to give up its control and liquidation rights. As a dependent variable,
we normalize the automatic conversion price by dividing by the round price and then taking logs.
The automatic conversion price is higher for pre-revenue ventures and lower for repeat
entrepreneurs. Again, this is supportive of the prediction that as uncertainty about the quality
of venture increases, the VC demands more control. Among the other industry controls, the
automatic conversion price is lower for ventures in larger, more established industries. Finally,
as one might expect almost by construction, the relative automatic conversion price decreases as
the round price increases (as measured by return). As with cash flow rights, the degree of board
and voting control varies significantly across different VCs. Also, the degree of board control is
higher in the companies based in the Northeast U.S.
Together, the results imply that for ventures with greater initial uncertainty about viability,
the VC receives more board and voting control, and demands stronger performance (through the
automatic conversion price) before ceding control. This is broadly consistent with Aghion and
Bolton (1992). When uncertainty is high, conflicts are more likely to arise between the VC and
the founder regarding issues whether the manager should be replaced or the business should be
continued. Hence, the VCs need to be allocated control in more states in order for their investment
to be ex ante profitable. Query 5
4.3. Debt theories and the allocation of liquidation rights
4.3.1. Theoretical predictions. Many financial contracting theories predict that the
investor should hold a debt-like claim. As mentioned in Section 4·1·1, the security design theories
based on classical principal agent theory (such as Innes, 1990) show that giving investors a senior
claim is useful for incentive purposes as it makes the manager’s residual claim more sensitive
to performance. Similarly, signaling theories such as Myers and Majluf (1984) and Duffie and
Demarzo (1999), show that in an asymmetric information setting, the manager can signal that
18. When we include the log accumulated VC financing in the ordered logit regression, it is strongly positive and
significant. The months since first VC round variable is still significant. The amount of financing is endogenous, and
cannot be instrumented in an ordered logit framework, hence, we do not include this specification in the table. We have
also run 2SLS regressions on the difference between the fraction of VC and founder votes or board seats, instrumenting
log VC investment with year dummies. Log VC financing is positive and all other results are similar.
19. When we replace fixed to total assets with R&D to sales, the coefficient is negative. Because the two variables
are highly negatively correlated, we do not include both simultaneously in the regressions. The result does not change
when we include cumulative VC investment. We do not include it in the reported regression because it is endogenous to
VC control and cannot be instrumented in the ordered logit framework.
KAPLAN & STRO¨MBERG 23
success is more likely by offering the investor a senior claim that receives all of the value in case
of failure. According to these models, VC liquidation rights should be stronger when there is
greater uncertainty about venture quality and founder ability.
Apart from seniority, however, the other important characteristic of debt is the ability to take
control and liquidate the firm when performance is bad. This can be interpreted as another way
of allocating the investor state-contingent control. One criticism, however, of this explanation
for debt—see Hart (1995)—is that changes in control in Aghion and Bolton (1992) do not
necessarily coincide with default on a contracted payment, which is a central feature of realworld
debt contracts. This feature of debt can be derived in a model where not only actions, but
also profits and cash flows are non-observable or non-verifiable. In this world, there is no way to
stop the entrepreneur from stealing the firm’s profits.
There are two main strands to this literature, which we denote “stealing models”. Costly
state verification (CSV) models—e.g. Townsend (1979) and Gale and Hellwig (1985)—assume
that profits are completely unobservable, unless a verification cost is paid. The other strand—e.g.
Hart and Moore (1998), Bolton and Scharfstein (1990) and Fluck (1998)—assumes that profits
are observable but not verifiable to outsiders and courts. The optimal financial claim in both
approaches is a debt-like claim in which (1) the entrepreneur promises a fixed payment to the
investor; and (2) the investor takes control of the project and liquidates the assets if the payment
is not made. Hence, these theories are consistent both with the seniority and the default aspects
of debt contracts. Bolton and Scharfstein (1990) also show that withholding future funding can
play a role similar to demanding repayment in forcing liquidation. Query 6
In these models, the value investors can realize when they seize assets limits the liquidation
claim that can be promised to outside investors. Hence, liquidation claims should be greater
when assets are more tangible and when the founder’s human capital is less crucial. Because the
liquidation threat always exists in the stealing models, there are no cross-sectional predictions
regarding the investor’s ability to force liquidation in the bad state. Given that the models rely
on the assumption of non-verifiability of cash flow, however, one possible prediction would be
where cash flow is harder to verify, the ability to liquidate should be more important and we
should observe stronger VC liquidation rights.
4.3.2. Empirical approach. In Section 3, we showed that the allocation of liquidation
rights is an important feature of the venture contracts. First, in all of our observations but one,
the VC is senior to common equity in liquidation. The seniority of the VC claim is consistent
with classical moral hazard theories, signaling and screening theories, as well as the “stealing”
theories. Second, consistent with the “stealing theories”, the VCs do have some power to
liquidate upon default on a contracted payment.
We note, however, that “stealing” theories do not explain a number of our results including
the allocation of residual cash-flow rights and the frequent contingent contracting on financial
performance measures. Such features are ruled out by assumption because cash flow is not
verifiable.20,21
In Table 6 we analyse the cross-sectional determinants of VC liquidation rights in more
detail. In the first regression, we analyse the determinants of the size of the VC’s liquidation
20. They are still consistent with the control theory of Aghion and Bolton (1992). See Hart (2001).
21. There are “stealing” models that explain the use of outside equity. In the models of outside equity financing in
Myers (2000) and Fluck (1998), the liquidation right is replaced by the right to fire management. This occurs if dividends
are too low. Here it is essential, that outside equity investors have this right, which in turn implies that they need to control
a majority of the board/votes. Moreover, in these models, the firm has to pay out dividends. Both of these assumptions
are frequently violated in our venture capital financings. In a significant fraction, the VC lacks board/voting control.
Moreover, few ventures pay any cash dividends.
24 REVIEW OF ECONOMIC STUDIES
claim. We use a dummy variable that equals one if the liquidation claim exceeds the VC’s
cumulative investment (through cumulative dividends, participating preferred, etc.) and zero
otherwise.22 Because the VC’s liquidation claim varies little between rounds, we report results
using the first financing round for each firm. The results with the full sample are almost identical.
Overall, the results are weak. There is no evidence that the VC’s liquidation claim is larger
when asymmetric information problems are more severe, because volatility, pre-revenue, and
repeat entrepreneur are not significant. Collateral value, measured as fixed to total assets, is not
significant either. Only the debt capacity of the venture, measured as the industry median longterm
debt ratio, is related to the VC liquidation claim.
The remaining regressions of Table 6 analyse the VC’s ability to liquidate. We first analyse
whether redemption rights are present in the VC contracts. Again the results are weak. In the
second regression, redemption rights are unrelated to the pre-revenue and the repeat entrepreneur
variables.
Even though redemption rights are the part of the VC contracts that most resemble debt,
there are other ways that a VC can force a liquidation of badly performing firms. The most
important mechanism is through staging of the investment.23 We distinguish between two
different forms of staging: ex ante (or within-round) and ex post (or between-round) staging.
In an ex ante staged deal, part of the VC’s committed funding is contingent on financial
or non-financial performance milestones. This essentially gives the VC the right to liquidate the
venture in the bad state of the world. Even many VC financings are not explicitly staged ex ante,
most of them are implicitly staged ex post, in the sense that even when all the funding in the
round is released immediately, future financing will be needed to support the firm until the IPO.
By providing less funding in a given round, and hence shortening the time until the next financing
round, the VC increases the ability to liquidate the venture if performance is unsatisfactory. 24
Ex ante staging within a given round, however, arguably makes this liquidation right stronger
and more explicitly related to performance.
The fourth and fifth regressions in Table 6 analyse between-round staging by explaining the
time until the next financing round. The repeat entrepreneur dummy is positively related to the
time between rounds, indicating that previously successful founders receive more funding in a
given round, reducing the VC’s liquidation threat. In contrast with Gompers (1995), most of our
industry variables are insignificant.25 The exception is the industry long-term debt ratio, which
is negative and significant. Hence, the use of between-round staging seems to be complementary
with the use of debt in the industry.
The three final regressions address ex ante staging by analysing the fraction of funds
committed in a given round that are provided up front. Repeat entrepreneurs receive more of
their funding up front, although the variable is not significant when VC dummies are included.
Again, the use of ex ante staging is positively related to industry debt. In addition, when the
round return is higher, more funds are provided up front and are less contingent on performance.
The coefficient on the time since the first VC round is negative, showing that ex ante staging is
more common in earlier rounds.
To sum up, our results on liquidation rights and claims are mixed. The results on redemption
rights and staging, although somewhat weak, at least suggest that the VCs increase their ability
to liquidate when dealing with less proven entrepreneurs. There does not seem to be any strong
22. Separate regressions on cumulative dividends and participating preferred stock generate similar results.
23. See Bolton and Scharfstein (1990). Neher (1999) provides a model of staging based on Hart and Moore (1998).
24. Gompers (1995) analyses ex post staging, using VE data. Time between financing rounds decreases with
industry R&D intensity and market-to-book ratios; it increases with industry tangible asset ratios.
25. None of the measures Gompers (1995) uses—the fixed asset ratio, R&D to sales, or market to book ratio—are
significant in our regressions. Part of this could be an issue of power, however, since Gompers has a larger sample.

relation between liquidation rights and the tangibility of assets or with the degree to which cash
flows are uncertain and hard to verify. However, the strength of VC liquidation rights and claims
increases with industry debt. To the extent that industry debt is a better measure of collateral than
tangible assets, this result is consistent with the stealing theories.
4.4. Venture-capital specific theories
While the theories analysed above are general financial contracting models, a number of recent
theoretical papers focus specifically on venture capital contracts. Most of these papers try to
explain the use of convertible securities in venture capital financings (based on the results in
Sahlman, 1990).
The first group of theories is motivated by empirical evidence that, apart from the
mere provision of funds, VCs exert substantial effort in monitoring and aiding their portfolio
companies.26 Motivated by this observation, these theories, including Casamatta (2000), Dessi
(2001), Inderst and M¨uller (2001), Renucci (2000), Repullo and Suarez (1998) and Schmidt
(1999), model the VC–entrepreneur relationship as a double moral-hazard problem, where both
the entrepreneur and the VC have to be given incentives to provide costly effort. Query 7
Consistent with our empirical evidence, these models predict that both the entrepreneur and Query 8
the VC should be allocated residual cash flow rights.
These theories have more mixed success in explaining the liquidation cash flow rights
that come with the convertible preferred securities. In Repullo and Suarez (1998) the optimal
contract calls for giving the initial VC investors a warrant-like claim. This avoids inefficient
liquidations for projects that turn out to be marginally profitable after the first round of financing.
This intuition is inconsistent with our findings because the liquidation rights that come with a
convertible or participating preferred in practice would worsen the inefficient liquidation problem
in their model.
In the models of Casamatta (2000), Dessi (2001), Renucci (2000), and Inderst and M¨uller
(2001), VC liquidation rights emerge as part of the optimal contract, but only for parts of the
parameter space. For example, in Casamatta (2000), only as the external financing constraint
becomes more binding does the VC’s claim resemble a convertible and the entrepreneur’s
resemble common stock. For low levels of external financing need, the entrepreneur gets a
convertible and the VC common stock.27 Given that (1) VCs are allocated liquidation cash flow
rights in our sample (212 out of 213 financings), and (2) that liquidation claims do not vary
much with our proxies for the severity of the agency problem (see Table 6), the cross-sectional
predictions do not seem consistent with the data.28
Schmidt’s (1999) theory relies heavily on the VC contract being a standard convertible
security, rather than participating preferred or a combination of straight preferred and common.
Given the widespread use of participating preferred in particular (39% of our sample), this is a
serious shortcoming of Schmidt’s model.
A second theoretical approach, represented by Cornelli and Yosha (1998), motivates
convertible securities as a way of stopping the entrepreneur from manipulating interim
26. See Hellman and Puri (2000) and Kaplan and Str¨omberg (2001, 2002).
27. Similarly, in Inderst and M¨uller (2001) VCs only get convertibles when there is low competition among VC
funds, and in Dessi (2001) convertibles are only optimal under very specific assumptions regarding collusion between
the VC and entrepreneur against uninformed third-party investors.
28. Still, there are two observations that give some support to Casamatta’s (2000) model. First, as we show in
Table 8, below, the strength of the VC’s liquidation claim increases in later financing rounds, i.e. as the VC invests
additional funds. Second, empirical evidence on angel financings (see e.g.Wong, 2001) shows that angel investments are
smaller in magnitude and more likely to involve common stock.
KAPLAN & STRO¨MBERG 27
performance.29 In their model, entrepreneurs have the incentive to manipulate interim
performance signals upward to secure additional rounds of VC financing. A properly designed
convertible contract avoids this problem by making conversion undesirable for the entrepreneur
(i.e. by setting a low conversion price). Because a high performance signal increases the
probability of a VC conversion that dilutes the entrepreneur’s equity stake, the entrepreneur will
refrain from such manipulation. Although the use of convertibles is consistent with our evidence,
the model relies on the convertible contract maturing before the true performance of the venture is
revealed. Because the real-world convertibles we observe have infinite maturity (in the sense that
they cannot be called by the company), and typically only convert at a successful IPO or sale, this
explanation does not seem consistent with our data. Moreover, our observation that VC residual
cash flow rights decrease with interim performance signals goes against the model’s predictions.
A third theoretical approach, represented by Bergl¨of (1994) and Hellman (2001), derives
convertibles as a way to avoid inefficient venture exit decisions. In Bergl¨of’s model, convertibles
enable VCs and entrepreneurs to capture a maximum amount of rents when selling the venture
to third-party buyers. The convertibles give control rights to the entrepreneur in the good state
which enables him to get fully compensated for his private benefits, and control to the VC in the
bad state which compensates the VC from being expropriated by the new majority owners after
the sale. Although Bergl¨of’s prediction of the allocation of control rights is consistent with our
evidence, the allocation of cash flow rights is not. His model predicts that all cash flow rights are
allocated to the VC, and hence does not capture the fact that cash flow rights are also allocated
based on firm performance. Moreover, the expropriation assumption is questionable in that it
relies on the VC staying as a minority shareholder after a trade sale.
Building on Aghion and Bolton (1992), Hellman (2001) argues that convertibles solve a
conflict between the VC and the entrepreneur regarding exit choice. The conflict arises because
a minimum amount of equity has to be given up to the entrepreneur in the case of an IPO,
while no such restriction exists in a trade sale, making VCs prefer trade sales ceteris paribus. The
optimal contract calls for leaving control with the entrepreneur as long as the VC breaks even, and
allocating cash flow rights to the VC, subject to leaving the minimum stake to the entrepreneur in
the case of an IPO. The model is successful in explaining a number of features of VC contracts,
such as participating preferred (which will be optimal when the financing constraint is tight),
automatic conversion at an IPO (which is necessary to guarantee the entrepreneur’s minimum
IPO share), and transfer of control from the entrepreneur to the VC as the financing constraint
tightens (as in Aghion and Bolton). Moreover, if we assume that the financing constraint tightens
in later financing rounds (as more funds are invested), the finding in Table 2 that both the use
of participating preferred and VC control increase in later rounds is consistent with Hellman’s
predictions. One shortcoming, however, similar to that of the double moral-hazard models, is that
VCs are predicted to hold only common stock for parts of the parameter space. Still, Hellman
(2001) shows that by augmenting Aghion and Bolton with a role for cash flow incentives (which
Hellman models in reduced form by the minimum entrepreneur IPO stake), a number of features
of the VC contracts can be explained simultaneously.
5. COMPLEXITY OF REAL WORLD CONTRACTS: COMPLEMENTARITY AND
DYNAMICS
In the previous section, we examined venture capital contracts in the light of standard
financial contracting theories. It appears that real world contracts are more complex than
29. The intuition of their model is somewhat similar to the risk-shifting motivation for convertibles of Green (1984)
and Brennan and Schwartz (1988).
28 REVIEW OF ECONOMIC STUDIES
the theories predict. First, control rights, cash flow incentives and liquidation rights are all
used simultaneously. Moreover, control rights are multidimensional, with several different
types of control being allocated between VCs and entrepreneurs, and switching gradually
with performance. Second, the contractual relationship between the VCs and the entrepreneurs
evolves; each new round involves a new set of contract terms with previous contracts potentially
being renegotiated. In this section, we describe these features in more detail in order to provide
stylized facts for future theoretical work.
5.1. Complementarity and substitutability of contract provisions
We use cluster analysis to divide the contracts into five generic groups to analyse the
simultaneous use of different governance mechanisms—pay-performance compensation, board
and voting control, and liquidation rights. The hierarchical cluster method30 we use forms groups
that maximize the Euclidian distance between matrices of dummy variables measuring twelve
main contractual characteristics for the companies in each group. Although cluster analysis is
a somewhat subjective methodology because it depends on the number of groups and input
variables used, the results using different specifications are qualitatively similar. We restrict the
sample to first VC financing rounds when available (89 cases) or, otherwise, from second round
investments when available (seven cases) in order to purge the comparison from effects due to
the subsequent evolution of the contracts.
Table 7 reports the results. The order of the five clusters increases in the degree of residual
cash-flow rights given up by the founder. The ordering also corresponds roughly to the degree
of board and voting control allocated to the VC, with cluster 1 (5) being the firms with the most
(least) VC control. The first three clusters exhibit the highest degree of VC control and also
contain a significantly higher fraction of pre-revenue firms. Hence, the first three clusters are
presumably associated with a higher degree of uncertainty regarding the viability of the venture.
We make a number of observations regarding the interplay among the different rights. First,
voting and board control are positively correlated. When VCs control the board (cluster 1), they
typically also have a voting majority; when founders control the board (cluster 5), the founder
group tends to control a voting majority. Not surprisingly, these two control mechanisms are more
complements than substitutes. Also, cash flow rights and control rights largely go together. The
fraction of equity held by the founder is significantly higher for cluster 5 compared to cluster 1.
Second, although voting, board and residual cash flow rights are correlated, the correlation
is far from perfect. Cluster 2 firms have VC voting control, but no or only state-contingent board
control; cluster 4 firms exhibit a high fraction of founder voting control without board control.
This further supports the result that control is more multi-dimensional and continuous than most
financial contracting theories assume.31 Moreover, the fraction of votes and board seats relative
to cash flow rights vary between clusters and are typically different from one. In general, the VC
and founder hold more control rights relative to their residual cash flow rights, at the expense of
third parties such as employees.
Third, the founder’s pay performance sensitivity, measured by the convexity induced by
performance benchmarks and vesting, is lower when the founder controls the venture. In
cluster 1, the founder’s cash flow rights are reduced by roughly 20% if performance benchmarks
are not met, and by another 20% upon leaving the firm; in contrast, the pay performance
convexity of clusters 4 and 5 is close to zero.32 Hence, consistent with Dewatripont and Tirole
30. See Aldenderfer and Blashfield (1984) for details on this procedure.
31. One exception is Kirilenko (forthcoming).
32. In clusters 4 and 5, the sensitivity to vesting is even slightly negative. This is due to vesting of employee and
other managers. If these do not vest the founder ends up with a higher fraction of the residual cash flow rights.

Table compares mean (median) frequencies and values of various contract provisions and firm characteristics between
five clusters of portfolio companies. Data is taken from first VC financing rounds when available (89 cases) otherwise
from second round investments (seven cases). 23 firms where data on first or second round investments were not
available were excluded from the sample. Clusters are obtained from a hierarchical cluster analysis maximizing
between-group variation measured by Euclidian distance with respect to the following 12 dummy variables: (1) founder
cash-flow rights depend on performance benchmarks; (2) founder cash-flow rights depend on vesting; (3) VC has
voting majority; (4) founder has voting majority; (5) voting majority is state-contingent; (6) VC controls board; (7)
founder controls board; (8) neither VC nor founder controls board; (9) board majority is state-contingent; (10) VC has
redemption rights; (11) VC financing commitment is staged ex ante; (12) VC liquidation claim is larger than investment
commitment. Asterisks indicate jointly statistically significant differences between clusters at the 1% ___; 5% __; and
10% _ levels, using a Kruskal–Wallis _2 test.
(1994) high-powered cash flow compensation and VC control are largely complementary. 33
33. See also Baker and Gompers (1999) who analyse equity ownership of IPO firms. They find that CEOs of
VC-backed firms hold a smaller fraction of total equity but have a higher elasticity to performance due to options.

The firms in cluster 3 represent an intermediate case, with founder residual cash flow rights
dependent on vesting, but not on explicit performance benchmarks. In addition, neither the
VC nor the entrepreneur has control, with state-contingent voting control and outside directors
pivotal in the board. This type of contract turns out to be particularly common in prerevenue,
R&D ventures.34 One potential explanation could be that these ventures are particularly
dependent on the skills of the original founder, increasing potential hold-up problems along the
lines of Hart and Moore (1994). The high degree of vesting makes it costly for the founder to
leave the firm prematurely, mitigating the hold-up threat during the time the venture is most
dependent on his or her skills. The open control, in turn, protects the entrepreneur from being
held up by the VC, since outside board members are unlikely to vote to replace the founder unless
performance is truly inferior.
Fourth, liquidation claims and redemption rights are largely independent and not
systematically related to control rights and residual cash flow rights.
Fifth, the size of the VC commitment and the release of funds are related to control. For
firms in clusters 1 and 2, where the VCs have the most control, the total commitment of VC funds
is higher, with a large fraction contingent on milestones. Hence, VC control is complementary
with ex ante staged, longer-term contracts that give the VC a greater ability to force the firm into
liquidation by withholding funds.
Sixth, the degree to which contracts are written contingent on actions and performance
differs across clusters. Non-financial performance, actions, and founder employment
contingencies are more common in the clusters where the VC is in control. Moreover, these
ventures are more often pre-revenue, where the future performance uncertainty is the highest.
Hence, in the ventures where observability and verifiability problems might be expected to be
particularly severe, we observe the highest degree of contracting on events that the incomplete
contracting literature describes as non-verifiable. The examples in Table 3 suggest, however, that
some of these actions and non-financial performance measures might indeed be difficult to verify
objectively.
There are two possible explanations for the complementarity between VC control and
contingencies. Because the contingencies may be difficult to verify, the VC will only be
comfortable using them when the VC has control and, therefore, a greater ability to decide
whether the contingency is met.35 An alternative interpretation is that because of the large
uncertainty and difficulty in writing all contingencies, control is given to the VC. To keep the
VC from taking too much advantage of that control, some contingencies are written that protect
the entrepreneur.
5.2. Dynamic evolution of contracts
So far we have largely ignored the fact that different financing rounds for the same company are
related. Accordingly, in Table 8, we report cash flow rights, voting rights, board rights, liquidation
rights, and other terms as a function of the financing round. We distinguish between financings
in which future financing is contingent on performance (ex ante staging) and those that are
not. Finally, to control for the fact that we do not have complete observations of all financing
rounds for every firm, the last column shows changes in terms between rounds where we have
observations for two consecutive VC financings.
34. The R&D to sales ratio is significantly higher for this group compared to the others (significant at the 5%-level
using a Mann–Whitney test). We have omitted R&D to sales from the table due to space considerations.
35. This suggests that there is the possibility of the VC behaving opportunistically towards the founder in
interpreting the contracts. It is likely that the reputational concerns of the VC mitigate such problems.
KAPLAN & STRO¨MBERG 33
Table 8 indicates that founders’ cash flow, voting, and board rights decline over financing
rounds while VC rights increase. The degree of state-contingent control also decreases over
time in favour of more VC control. The results for board rights are similar, although the rate
at which VCs increase their board control is slower. The increase in VC cash flow and control
rights over financing rounds is consistent with the VC demanding more equity and control as
compensation for providing additional funding. Hence, a less successful venture will see control
being transferred from the founder to the VC through two mechanisms: through explicit statecontingent
control, specified ex ante in the contracts in a given round, and through dilution of
control as the VC has to provide additional subsequent financing. The table also shows that while
the use redemption rights does not change much over time, the size of the VC liquidation claims
(relative to VC investment) does increase somewhat between rounds.
Comparing the ex ante staged rounds with the other first round observations, we observe
some interesting differences. Naturally, these contracts commit a significantly larger amount of
financing, of which on average half is released subject to future performance. Consistent with
the cluster analysis, the VCs are significantly more often in control, both in terms of votes and
board seats, and more future contingencies are specified in advance. We believe that the choice
between ex ante staged, longer-term contracts, vs. short term, “ex post” staged contracts, is an
interesting issue for future research.
Finally, we also provide evidence that contract terms are sometimes renegotiated. In
about 30% of subsequent financing rounds (34 cases), some contractual rights from a previous
financing round are renegotiated as a part of the new contract. The most commonly renegotiated
terms are the automatic conversion price (typically increasing it; 11 cases) and the VC liquidation
claim (changing dividends or participation; 10 cases). Other renegotiated terms include changing
the redemption maturity, waiving funding milestones, or changing vesting provisions and
performance benchmarks (each present in five cases). 36 Since our data are incomplete, these
numbers likely underestimate the frequency of renegotiation.37 This gives additional support
to the incomplete contracting theories which rely heavily on the presence of bargaining and
renegotiation (see e.g. Hart, 1995). The fact that contractual provisions are not always enforced
does not necessarily imply that they are irrelevant. Even in those instances, the initial contracts
are likely to be important in determining the outside options for any subsequent bargaining over
rights in later rounds.
6. CONCLUSION
In this paper, we have compared the characteristics of real world financial contracts between
VCs and entrepreneurs to their counterparts in financial contracting theory. The distinguishing
characteristic of VC financings is that they allow VCs to separately allocate cash flow rights,
board rights, voting rights, liquidation rights, and other control rights. We describe and
measure these rights. We then interpret our results in relation to existing financial contracting
theories. Overall, the theories do rather well, particularly, the classical principal agent theories
(e.g. Holmstr¨om, 1979) and control theories (e.g. Aghion and Bolton, 1992).
Still the results suggest there is room for additional theory. Our results clearly show that
real world contracts are more complex than existing theories predict. For example, Section 5
indicates that the allocations of cash flow and control rights and the use of contingencies are
related in systematic ways. We hope that these results and others in the paper provide stylized
facts for future theory.
36. Note that a renegotiation may involve several different provisions.
37. In particular, for the ex ante staged financings, we usually have data from the initial contracting stage only and
do not know whether some terms were subsequently renegotiated.