The Dynamics of Venture Capital Contracts

Transcription

1 The Dynamics of Venture Capital Contracts Carsten Bienz & Julia Hirsch a,b Center for Financial Studies and Goethe University Frankfurt This version: May 11, 2005 Preliminary Version. Please do not quote without permission. Abstract We analyze staging in the venture capital industry using a hand-collected sample of contract data on 464 investment rounds into 290 entrepreneurial firms from Germany. Both the decision for staging per se and for a specific mode of staging (pure milestone financing, pure round financing, mixes) are analyzed. We show that the decision for staging per se is determined by the degree of uncertainty and asymmetric information whereas the use of a specific form of staging is determined by the expected distribution of bargaining power between the contracting parties and the predictability of the development process when new funding becomes necessary. Keywords: venture capital, corporate governance, moral hazard, renegotiation, contract theory, empirical contract theory, contract econometrics JEL classification: G24, G32, G34. a Address: Carsten Bienz: Graduate Program Finance and Monetary Economics, Schumannstraße 60, Uni-Postfach 64, Frankfurt am Main, Germany. Frankfurt, Germany. Tel: +49 (0) bienz@wiwi.uni-frankfurt.de, Julia Hirsch: Goethe University, Schumannstraße 60, Uni-Postfach 64, Frankfurt am Main, Germany. Tel: +49 (0) jhirsch@wiwi.uni-frankfurt.de b This paper was presented at SITE-SSE and LSE. We would like to thank Renée Adams, Eric Berglöf, Mike Burkart, Antoine Faure-Grimaud, Guido Friebel, Mariasunta Giannetti, Per Strömberg, Uwe Walz and Volker Zimmermann (KfW) for valuable comments. Finally, we would also like to thank all KfW employees at KI for their support and especially Andreas Weber (KfW) for his help. All errors remain ours. Carsten Bienz gratefully acknowledges financial support by German Research Foundation (DFG).

2 1 Introduction Recently, the interest for venture capital as a specialized intermediary for the financing of young innovative firms has grown considerably. Traditionally, innovative firms lack financing as they face severe problems of asymmetric information and moral hazard as well as a high degree of uncertainty in combination with wealth constrained entrepreneurs. The venture capital industry, however, has developed several mechanisms to deal with these problems using highly elaborated contracts 1. In this paper we look at one major mechanism namely staging. Staging denotes a special form of financing where not all the capital necessary to finance the project is paid out up front but instead several tranches are paid to the firm dependent on its development. Thus staging refers to a gradual release of funds over time which aims at attenuating the severe problems related to asymmetric information and uncertainty. At the same time, staging may also cause additional problems such as contracting and delay costs, window dressing as well as hold-up problems. So the decision for staging requires the contracting parties to balance potential costs and benefits. Moreover, the decision for staging offers the opportunity to decide about its specific form. Often, the exact conditions under which capital will be supplied are left open and are not determined ex-ante. Then renegotiation becomes an issue of central importance. On the other hand, there exists the possibility to define specific milestones that determine a stepwise funding process. In this case, renegotiation possibilities are excluded 2. Generally, economists reckon renegotiation to be a double-edged sword. While some studies stress the benefits of renegotiation (Hermalin and Katz (1991) [15]), others show that under certain conditions, renegotiation may cause inefficiencies (Fudenberg and Tirole (1990) [10]). The aim of this paper is to analyze staging and the different modes of staging in venture capital contracts empirically against the insights of the venture capital and the renegotiation literature. So our aim is not only to test specific theories about venture capital but also to confront financial contracting theory with the real world. We show that the decision for staging per se is determined by the degree of uncertainty and asymmetric information whereas the use of a specific form of staging is determined by the expected distribution of bargaining power between the contracting parties when new funding becomes necessary and the predictability of the development process. There exists a growing empirical literature on venture capital issues. In a pioneering work, Kaplan and Strömberg (2003) [16] look at one other major mechanism that venture 1 For a more complete introduction into venture capital see for example Paul Gomper s and Josh Lerner s Venture Capital Cycle [13]. 2 Renegotiation may only take place when milestones are not reached but new funding is provided anyhow albeit under different conditions. 1

3 capitalists employ: the separation of cash-flow and control rights. Their article is part of a strain of literature that mainly focuses on security design. Moreover, most of these studies are based on US data (see Sahlman (1990) [23] and Gompers (1997) [12], for example). Some important exceptions are Kaplan, Martel and Strömberg (2003) [18] whose study is based on a set of 23 countries, the study of Cumming (2002) [6] and the work of Bottazzi, Da Rin and Hellman (2004) [4] for the European market. The only exhaustive study that exists for the German market is the one of Bascha and Walz (2002) [2] which analyzes security choice and is based on survey data. To our knowledge, there exist only three empirical studies on staging. First, in his 1995 Journal of Finance article, Paul Gompers [11] looks at the structure of staged investments. He focuses on the determinants of funding size and funding duration for round financing, but he does not analyze the decision for staging per se nor the decision for the different forms of staging. Second, the papers of Kaplan and Strömberg from 2003 [16] and 2004 [17] examine the determinants of round and milestone financing but do not analyze the decision for one of the two forms. Moreover, all three studies work with data sets from the US venture capital industry. Therefore, the contribution of this paper is twofold: it will be the first paper analyzing the decision for the use of staging per se as well as the decision between the different forms of staging, and it aims at closing the existing gap of empirical research for non-us venture capital industries. We are not aware of other empirical papers testing the effect of possible renegotiation on contract design. However, with respect to the actual decision to renegotiate or not, Guasch, Laffont and Straub [14] present and estimate a model of renegotiation for South-American procurement contracts. Their variable of interest is the actual decision for renegotiation initiated by the contracting firm. They do not test how potential renegotiation affects contract design but rather investigate the decision for renegotiation given actual contracts. Thus our approach is complementary to theirs. The importance of this approach is highlighted by Salanie [24], who provides a general overview over empirical tests of contract theory. This paper is part of a larger research cooperation between the Center for Financial Studies (CFS) at Goethe University in Frankfurt and KfW (formerly Kreditanstalt für Wiederaufbau) in Frankfurt. The underlying data set is a proprietary data set from KfW that exhibits unique characteristics. Contrary to the majority of empirical studies, the data set is not based on survey data but the information was gathered directly from the contracts. Moreover, it is based on all documents concerning a specific deal, i.e. the business plan, the balance sheet, the term sheet, the shareholder s agreement, the bylaws of the corporation or company, additional agreements, procedural rules and key employment contracts. In addition, it covers a large time period which lasts from 1991 until

4 Finally, it constitutes a representative sample of the German venture capital industry as it is a random sample of all VC projects supported by KfW which is involved in a substantial part of all German venture capital investments. The rest of the paper is organized as follows. The next section reviews the theoretical literature. We will focus both on the specific venture capital literature on staging and on the renegotiation literature in order to deduce testable hypotheses about the decision for or against staging and the choice of the form of staging. The third section describes the underlying data set and gives some descriptive statistics about the situation of the venture capital industry in Germany. In the fourth section, we will present our empirical results. Section five concludes. 2 What does theory tell us? In this section, we will look at the theory behind staging. Staging means that an investor does not pay out all the necessary funds for the investment upfront but in several tranches. There exist two different forms of staging - milestone and round financing - which we will analyze in subsection 2.2. In what follows, we will have a closer look on the determinants of the decision for staging in general from a theoretical point of view. 2.1 Why do we use staging? Staging implies that some of the funds deemed necessary for the firm in order to complete its project are withheld. This mechanism has three major implications. First of all, as not all capital is provided up front, staging gives the investor implicitly the right to decide about the continuation or liquidation of the firm. This exit option of the investor saves resources because infeasible (negative NPV) projects are identified and liquidated early. Obviously, this advantage is the more pronounced the higher the existing ex-ante uncertainty about the feasibility of the project. Second, in an asymmetric information context, staging can be interpreted as a signaling mechanism. An entrepreneur who has private information about the quality of his project is able to use staging as a signal of firm quality. Good firms face a low risk of liquidation, whereas the liquidation risk is especially high for low quality projects. This issue is modelled by Dessein (2004) [8] who develops a theory of control where a control transfer to the investor serves as a signal of favorable information and congruence of objectives between the investor and the entrepreneur. He shows that investor control is increasing in the ex ante information asymmetries and in the ex post uncertainty of the project. One possibility of implementing investor control in his model is short-term financing which can also be interpreted as staging. Thus, staging is the more convenient, 3

5 the higher the degree of asymmetric information ex ante and the more difficult future monitoring ex post. Third, staging can act as a commitment device for the entrepreneur not to renegotiate the initial contract. This effect does not depend on the liquidation risk related to staging but rather on the fact that injecting less capital at the beginning of the relationship implies less sunk capital and consequently limited renegotiation possibilities for the entrepreneur. Neher (1999) [19] models precisely this mechanism. He shows that the entrepreneur cannot commit credibly to work and not to renegotiate if the venture has no collateral and the only asset is the entrepreneur s human capital itself. Furthermore he shows that this holdup problem can be solved by staging because the stepwise provision of capital is related to a gradual embodiment of the entrepreneur s human capital in the firm. Thus staging should be the more advantageous, the more crucial the entrepreneur s human capital for the success of the project and the less tangible the assets. We conjecture that the higher the advantages of staging, the more probable its use and thus, we resume the previous arguments in the following hypothesis: Hypothesis 1 The higher the uncertainty, the higher the degree of asymmetric information, the more difficult future monitoring, the more crucial the entrepreneur s human capital and the less tangible the firm s assets, the more probable is staging. Staging, of course, has also some shortcomings. First of all, staging may cause delays as firms have to commit time to the negotiation process. This lag caused in implementing the project can induce additional costs, such as delays in the development process or in market entry, lost economies of scale or cost overruns, which must then be outweighed by the benefits of staging. Second, Cornelli and Yosha (2003) [5] show that the entrepreneur may react to the prospect of early liquidation with an attempt to manipulate available information. This phenomenon, called window dressing, reduces the positive effects of staging and tends to be the more probable the softer the information available. We conjecture that the higher the disadvantages of staging, the less probable its use and thus, we state the following hypothesis: Hypothesis 2 The higher the delay costs of staging and the higher the danger of window dressing, the less probable is staging. Finally, staging also changes the entrepreneur s incentives. On the one hand, the liquidation risk increases his incentives and so the project s overall probability of success. Wang and Zhou (2002) [25], for example, show that the termination threat induces the entrepreneur to work more in order to ensure further financing by the investor. Therefore, in their model, the effort level of the entrepreneur is always higher under staged financing. But on the other hand, staging can also induce opportunistic behavior by the investor, 4

6 which in turn reduces the entrepreneur s ex ante incentives. The negative incentive effect of staging must then be traded off against the above mentioned positive incentive effect. This trade-off is described in a broad variety of papers. Rajan (1992) [21], for example, analyzes the choice between arm s length financing (what can be related to upfront financing in our framework) and insider financing (what can be related to staging) and shows that the choice is related to a trade-off between ex post flexibility and ex ante efficiency. In the same spirit, Edlin and Hermalin (2000) [9] model the hold-up problem and the following under-investment in effort by the entrepreneur, on the one hand, and a positive incentive effect, called threat-point effect, on the other hand. They ascribe this positive incentive effect to a strengthening of the entrepreneur s bargaining position due to the availability of an outside option which in turn becomes more probable when the entrepreneur exerts more effort. In a similar way, Bigus (2002) [3] points out that the hold-up problem is the less pronounced, the smaller the loss of the entrepreneur when the relationship ends and a new investor must be found. He argues that the loss can be reduced by reenforcing patent protection as to avoid idea stealing by the investor. As it is indeed the investor s renegotiation possibility that causes this trade-off, it can also be analyzed from the point of view of the broad literature on renegotiation. In a ground-breaking article Aghion, Dewatripont and Rey (1994) [1] point out that the underinvestment problem normally related to renegotiation (and therefore to staging) can be overcome by an adequate design of the renegotiation process 3. This implies that the extent of both effects not only depends on project and financing relationship characteristics but also on the concrete design of the staging mechanism. Therefore, we will analyze the different modes of staging in the next subsection. 2.2 When do we observe what form of staging? There exist two different ways of implementing staging - round financing and milestone financing - that differ considerably. With round financing, every new tranche is negotiated separately when the venture needs further funding 4. Milestone financing, on the other hand, specifies certain contingencies that the firm has to achieve in order to obtain new funds. These conditions are determined ex ante, i.e. before the initial contract is signed. Examples are the amount of revenues realized, the number of patents filed or the development of prototypes. Provided that the milestones of the initial agreement have been 3 They show that only two conditions must be fulfilled in order to guarantee an efficient outcome: on the one hand, the initial contract must specify a default option in case renegotiation fails and, on the other hand, it must assign all the bargaining power to one party. The default option is determined by the initial contract, especially by the covenants which attribute specific rights to the investor in certain circumstances. 4 Thus no further funding is also an option. 5

7 reached, the VC has the contractual obligation to release the funds to the firm 5. Consequently, one obvious advantage of milestone financing is the fact that it prevents hold-up by the investor and eliminates the ex ante inefficiency caused thereby. This solution mechanism is modelled by Nöldecke and Schmidt (1995) [20], for example. In their model, the under-investment problem is overcome if the parties write a simple option contract that is very closely related to milestone financing: the entrepreneur has the right but not the obligation to exert certain effort levels and reach the specified milestones 6. If the milestones are reached, the investor must pay out the next tranche of capital as determined in the initial contract. If the milestones are not reached, the default point is specified by the action chosen by the entrepreneur, i.e. the result achieved. If the difference in the outcome between the two cases is larger than the effort costs of the entrepreneur, higher effort becomes worthwhile. So milestone financing is the more profitable, the more pronounced the negative incentive effects due to possible hold-up by the investor. As mentioned above, the negative incentive effects increase with a decline in the entrepreneur s outside financing options (see Edlin and Hermalin (2000) [9], for example). Another advantage of milestone financing has been raised by Cuny and Talmor (2004) [7]: the greater flexibility of milestone financing in adjusting the parties claims across different states. With milestone financing, one contract covers multiple states of the world simultaneously and thus, claims need not to be priced fairly ex post but only ex ante, i.e. before knowing the outcome. The higher flexibility is especially valuable if either the contracting parties preferences differ across states or particular states require different incentive mechanisms. Cuny and Talmor (2004) [7] also mention a possible disadvantage of milestone financing. Milestone financing promises the whole investment (contingent on milestones) ex ante, while with round financing, there exists only a commitment to finance the current investment round. This implies that the investor s claim should be considerably larger under milestone than under round financing. However, a larger (equity) claim by the investor reduces the entrepreneur s incentives. Thus, the larger the total investment, the larger the negative incentive effect due to milestone financing 7. A further obvious caveat with respect to milestone financing is that milestones must be available and enforceable. There are only a few studies that consider explicitly the 5 If any of the predetermined conditions are not met, no further funding will take place unless both parties negotiate new terms for further funding. 6 Note that the exercise of any standard option contract is followed by a guaranteed delivery, while in our case some residual uncertainty is required because otherwise the moral hazard problem becomes irrelevant. 7 This problem only arises with contracts that assume that the VC s equity stake is fixed initially and any additional capital is paid into the firm s capital reserves. While this construct is frequently found in practice, other constructs that avoid this problem by adjusting the VC s equity share accordingly are also encountered in practice. Thus the magnitude of this issue is not quite clear. 6

8 nature of signals available and their relation to contracts. Repullo and Suarez (2004) [22], for example, look at the optimal capital structure with staging and find that if signals are verifiable, completely contingent contracts are feasible. We will blind out this aspect for the moment and assume that milestones are always available 8. To sum up, we can say that the decision between milestone and round financing is characterized by the trade-off between ex ante efficiency and ex post flexibility which in turn is shaped in particular by the entrepreneur s outside financing option. Thus we can state the following hypothesis: Hypothesis 3 The better the entrepreneur s outside financing option, the less pronounced the heterogeneity between the contracting parties and the higher the total investment amount, the more probable is round financing with respect to milestone financing. 3 The Data Set We use a proprietary data set compiled from data of KfW in Frankfurt, Germany. KfW is Germany s largest public agency active in the venture capital area. KfW never invests directly in any of the portfolio firms but supports the firms indirectly by promoting the investment of the VC. Its various VC programmes have supported the funding of more than a thousand start-up companies since the late 1980s. Against the background of limited human resources, we were decided to draw a random sample of 300 portfolio companies from the data base of all supported firms. In order to avoid possible biases in our sample, we categorized each portfolio company into one of three classes with respect to their investment date and eight classes with respect to the programme or programme combination used by the VC. We then drew proportionally from each category. Table 1 gives an overview. Unfortunately, the data for 10 portfolio companies could not be evaluated, so our random sample finally consists of 290 portfolio companies that were financed in 464 investment rounds starting in 1991 until For each investment round, we evaluated the company s business plan in order to get balance sheet data, information with respect to the market position of the company and details about the project financed. Moreover, the term sheet and the shareholder s agreement provided us with detailed information about the security used, the timing and conditions of the investment, the syndication of the investment, control and information rights of the venture capitalists (VCs) and exit covenants. We complemented this data set with information about the venture capitalist who applied at KfW for support for his portfolio firm, such as its type, origin and industry focus. 8 See section 4.2 for details on signal availability. 7

9 In what follows, we will describe the data set in more detail and introduce the variables necessary for our regressions. First of all, we have information about the project and the respective portfolio company. The variable AGE represents the age of the firm when the corresponding financing round was closed. Moreover, we observe the firm s industry: LIFE- SCIENCE, INTERNET, IT/TELECOM, TRADITIONAL HIGH-TECH INDUSTRIES and OTHER INDUSTRIES are all dummy variables that indicate the project s industry. Furthermore, we observe the firm s development stage in each financing round. On the one hand, we know whether the firm has finished its product tests, if it has already a finished product, if the firm holds any patents or if it even has reference customers. We define a dummy named PATENTS taking value one if the firm holds any patents or its patents are pending when the financing round was closed. On the other hand, we have information about the development stages as defined by the German Venture Capital Association. We distinguish seed and start-up firms and expansion and later stage firms: the DUMMY EARLY STAGE indicates whether the firm ranks to the first group or not. Furthermore, we often have the firms balance sheet data at the date of each financing round: we know if the firm has any revenues (if this is the case the dummy REVENUES takes value 1) and whether its balance sheet is audited or not (if this is the case the dummy AUDITED BALANCE SHEET takes value one). Moreover, the FIXED ASSET RATIO (FAR) indicates the ratio of fixed assets to balance sheet total whereby we use balance sheet data of the year preceding the year of closing of the corresponding financing round 9. Finally, we know if the portfolio company has had contact with further investors apart from the VC we are looking at. We define a dummy INTERMEDIARIES that takes value one if the portfolio company has received bank, angel or other VC finance before the first round of VC financing we are looking at or if VC financing takes place via a syndicate of various VCs. Second, we have information, about the investment conditions. We classify each VC according to his type in three categories named INDEPENDENT VC, PUBLIC VC and OTHER VCs. The latter category includes both bank(-dependent) and corporate VCs 10. Additionally, we know whether the VC s are specialized in specific industries or development stages. Moreover, we observe the total amount invested, the financing instrument used and the timing of the investment. As the aim of this paper is to analyze the determinants for staging and its different 9 As we have many missing values in our sample, we adopt the following procedure. For all firms in a first financing round with an age of less than one year at the date of contracting and an investment phase of seed or early, we set the fixed asset ratio to zero. If we lack information for higher rounds, we use the same ratio as in the round before. If this ratio is not available, we code both as missing value. Additionally, we do not resort to the preceding round in the case of second rounds where we coded the first round data to be zero. 10 We also include the business angels in our sample in this category. 8

10 forms, it is important to exploit all the available information related to staging. Therefore, we define three dummy variables for the occurrence of staging and its different forms. Whereby the first dummy STAGING captures only if the project is financed in several steps or not, the other two dummies look at the specific form of this stepwise capital infusion: ROUND takes value one if staging is made in several independent rounds and MILESTONE if future capital infusions are contingent on specific known milestones, milestones are not only law milestones and at least 20% of the capital infusion is dependent on the achievement of the defined milestones 11. We then define the subcategories PURE MILESTONES, PURE ROUNDS and MIXES. PURE MILESTONES take value one if milestone financing occurs and we know that no round financing takes place. Analogously, we define PURE ROUNDS. MIXES, on the other hand, takes value one if we know that milestone and round financing are used simultaneously. As concerns milestone financing, we also gather the type of milestones employed. Finally, we observe the year when the financing round is closed and define three time dummies. PERIOD 1 takes value one if the financing round was closed during the early period of relatively low venture capital activity, namely before 1998, PERIOD 2 if it was closed during the boom, i.e. between 1998 and 2000 and PERIOD 3 if it was closed after a period of relative decline and reorganization of the venture capital industry. Last but not least, we have information about the entrepreneurs running the portfolio firm. We know if any of the founders has a PhD or higher degree of education (then, the dummy variable RESEARCH DEGREE takes value one), we observe whether any of the founders has a background in engineering or natural sciences (in this case, the dummy variable SCIENCE BACKGROUND takes value one) and we know whether we face a repeat entrepreneur, i.e. someone who has already run a firm (this is captured by the dummy variable REPEAT ENTREPRENEUR). In order to account for the value of the entrepreneur s human capital within the firm, we construct the variable E EXPERT. We know that the entrepreneur s role is crucial for the technological development of the product - especially if he has an advanced scientific background and the development is highly complex. Thus, we define the dummy E EXPERT in such a way that it takes value one if the firm belongs to a high tech industry, if the entrepreneur holds a research degree and if the firm s product has not finished any tests, i.e. if the product development process is just about to start. Analogously, we construct further indicator variables that we deem necessary to test our hypotheses. First, we create a proxy for the degree of asymmetric information between both parties named AI. We think that the degree of asymmetric information depends heavily upon the amount of verifiable information available for the project. Thus, we 11 For robustness we also considered to make at least 30% of the capital contingent, but the results are qualitatively similar. Thus we opt for the 20% cut-off value. 9

11 define a categorical variable by summing up four dummy variables which describe, in our opinion, essential steps in reducing information asymmetry: the dummy AUDITED BALANCE SHEET that indicates whether the balance sheet has been audited or not, the dummy FINISHED PRODUCT which signals the existence of a product, the dummy REFERENCE CUSTOMERS that indicates the existence of any reference customers and finally, the dummy BREAK EVEN that takes value one if the firm has reached its breakeven point. While we could have included more factors in this definition, one problem we have is that we loose observations due to missing data. We thus strive for an optimal balance between measuring the degree of asymmetric information and data availability. 12 Finally, we construct two measures that aim at determining the current position of the firm in its development process: MARKET takes value one if the firm has a finished product and revenues. This proxy indicates that the introduction of the product in the market has taken place. The dummy PRODUCT is somewhat similar to the dummy about the expertise of the entrepreneur with the important difference that it covers only aspects of the product development per se and does not look at the entrepreneur s abilities. The dummy takes value one if the product development process is just starting, i.e. if there does not exist a finished product, there do not exist any reference customers and product tests have not been successfully completed yet. In what follows, we want to give a more detailed overview about our sample. Therefore, in a first step, we will describe the basic descriptive statistics. The average amount invested per financing round is about 5.4 million euros and the portfolio companies are in average 5.08 years old when they receive VC financing for the first time. The medians are considerably smaller (1.3 million euros and 2 years), an indication for outliers. At this point, one can already infer that the percentage of early stage financing is quite high in our sample. Indeed, 11.3% of the financing rounds correspond to seed financing and 61.5% to start-up financing whereas only 20.7% of the financing rounds are related to expansion and 6.6% to later stage 13. The portfolio companies are active in a broad range of industries: 20.7% in the field of life-science, 36% in the sector of IT, telecommunications and software development, 9.7% belong to the internet sector, 19.6% are active in traditional high-tech industries and 14% could not be classified in neither of these sectors but are rather less R&D intensive. Second, we want to give some insights about the different types of financial instruments used. As a broad range of different combinations of financial instruments achieves the same allocation of cash-flow and control rights, it is important to analyze these instruments 12 We run robustness checks using a scaled down version of the AI variable leaving out the variable for REFERENCE CUSTOMERS. Our results stay qualitatively the same, however. 13 These percentages refer to the financing rounds for which we have available data for the respective criterion. 10

12 properly. Therefore, we classify each financial instrument used in a single financing round into four broad categories: pure equity, pure debt 14, debt-equity mixes and convertibles. A more detailed differentiation takes into account the existence of liquidation preferences or alternatively the extent of the debt component in debt-equity mixes. Each instrument is thus classified along the five characteristics upside cash flow rights, downside protection, change of control, cash flow rights at exit and voting rights. This classification procedure allows a better interpretation of the securities used and guarantees better comparability. We see that whereas debt and different types of equity play an important role (23 % and 29% respectively), debt-equity mixes are used most frequently, i.e. in 38% of all financing rounds. Only 8% of all firms are financed through convertible instruments. In order to control for possible substitutability between the usage of debt and staging, we define a dummy DEBT COMPONENT which takes value one if we have a strong debt component, i.e. if we have pure debt, nonstandard debt, convertible mixes and US style convertibles. Last but not least, table 3 describes the observed staging behavior in our sample. In almost 70% of the analyzed observations staging is used as a mechanism whereby pure rounds are the most frequently used design form (58%), followed by mixes (24%) and pure milestones (18%). We can also detect differences in the staging behavior across time and VC types. As shown in table 3, there is a slight decline in the use of staging during the boom period - this may be attributed to the reduced bargaining power of the VCs - and an increase in period 3 slightly beyond the level of period 1. By looking at the different staging forms however, we recognize pronounced changes: whereas in period 1 only 7% of staging was in the form of pure milestones, 7% was in the form of mixes and 86% was in the form of pure rounds, this changes dramatically up to now: staging in the form of pure milestones increased up to 25%, staging in the form of mixes increased to 30% and staging in the form of pure rounds decreased to 45%. As concerns the behavior of the different VC types, we see that independent VCs use staging more often than other VC types which in turn use staging more often than public VCs. VCs also differ in their use of the modes of staging. Independent VCs use rounds in more than half of all staged firms, but use relatively more mixes than pure milestones. Public VCs use rounds in almost half of all staged firms, but they do use relatively more pure milestones than mixes. Other VC types use rounds in 75% of all staged firms and in the other cases they use - similarly to public VCs- more pure milestones. Before looking at our results, some further remarks are necessary. First, a small remark on the timing of our variables is due. All balance sheet data as well as exogenous factors such as market size or degree of asymmetric information is information which is known to the VC and the entrepreneur before they negotiate their contract. 14 Debt is actually subordinated debt, thus it is only senior to the equity in the firm. 11

13 A second remark is due to the varying size of the sample. Unfortunately, not all the desired information is always available when collecting the data. Though, we do not see a systematic selection bias problem because there are several reasons for missing data. On the one hand, data may be missing for very young firms in the seed stage, but on the other hand, we often had also the most exhaustive term sheets for these firms. In order to further rule out these possible problems, we will only use variables that are available for the majority of firms. Finally, we want to note that our dummy STAGING takes value one whenever we know that a further financing round is expected. This implies that we have information even for the most recently financed projects. Regarding earlier financing agreements, however, staged projects are also those for which a further financing round became necessary but was not expected at the beginning. Therefore, the probability of staging may be overestimated, as may be the probability of round financing. Still, we think that this eventual overestimation should not be considered as being too alarming because this phenomenon was not very common. Additionally, it is quite difficult to distinguish between the fact that further financing rounds were really unexpected or were just not mentioned in the original contract. 4 Empirical Evidence 4.1 The determinants for staging We begin by evaluating hypotheses 1 and 2 that refer to the determinants of the staging decision. Therefore, in a first step, we will present descriptive statistics and in a second step, we will test the hypotheses by using different microeconometric methods, especially univariate probit models. In a last step, we will run robustness checks for our results Descriptive results Our descriptive results, shown in table 4, give us a first impression. In almost 70% of all observations, staging is used. Hypothesis 1 states that the higher the uncertainty, the higher the degree of asymmetric information and the more difficult future monitoring, the more probable is staging. It is obvious that uncertainty, the degree of asymmetric information and the difficulty of future monitoring, all increase, on average, with younger firms or firms in early investment stages. Our data confirms this conjecture as we observe substantial differences between staged and non-staged observations concerning these variables. Firm age is significantly lower when staging is used: whereas non-staged firms have an average age of more than nine years, staged firms are only three and a half years old 12

14 on average. Though this result may be driven by potential outliers, it is confirmed by our investment phase dummy variables: among staged firms, there are significantly more start-up firms and significantly less expansion and late stage firms. Moreover, the mean of our proxy variable for the degree of asymmetric information is significantly higher for non-staged firms. This indicates that the degree of asymmetric information is significantly lower for non-staged than for staged firms. In fact, all four components of AI, namely the dummy variables for the firm breaking-even, an audited balance sheet, a finished product and reference customers, differ significantly between staged and non-staged firms. Thus, the more information becomes available, the less probable is staging. Finally, the dummy REVENUES is statistically lower for staged than for non-staged firms. This, again, underlines the importance of the asymmetric information aspect. The second part of hypothesis 1 states that the more crucial the entrepreneur s human capital and the less tangible the firm s assets, the more probable is staging. As we see in table 4, the fixed asset ratio does not differ significantly between staged and non staged firms. With respect to human capital, we notice that the mean of the research degree dummy differs significantly between staged and non-staged firms. The variable is significantly higher for staged firms. This confirms the hypothesis: if the entrepreneur holds a PhD or higher degree, i.e. if his human capital is more crucial, staging is more probable. However, the importance of the entrepreneur s human capital may not only depend on the absolute amount of human capital but but also on its value for the firm. This aspect is reflected by the industry dummies and the PRODUCT dummy that takes value one if the product development is about to start: its mean is higher for staged than for non-staged firms. As the entrepreneur s human capital is especially important during the product development process, this provides further evidence for our hypothesis. The same is true for the industry dummies: whereas life-science firms represent only about 14% of non-staged firms, they amount to 24% of staged firms; traditional high-tech firms represent about 26% of all non-staged firms and only 17% of staged firms; and finally, other firms cover approximately 21% of non-staged firms and only 10% of staged firms. This shows that firms in R&D intensive industries are staged more often and again confirms hypothesis 1. Hypothesis 2 states that the higher the delay costs and the higher the danger of window dressing, the less probable is staging. Delay costs can occur due to various factors we cannot control for given our data. With respect to window dressing, we conjecture that window dressing should be easier in very young firms where only soft information is available; respectively, window dressing should be harder to accomplish in older firms where audited financial statements exist. But this means that in situations where the danger of window dressing is high, the benefits of staging are high too; and if the danger 13

15 of window dressing is low, the benefits of staging are low too. Unfortunately, we cannot control for these underlying countervailing effects but only for the net effect of staging. Our results so far suggest that the net benefits of staging are higher for younger firms. Therefore, in what follows, we will restrict our analysis to hypothesis 1. Still, one has to keep in mind that we control, indeed, for the net benefits of staging, i.e. that the benefits outweigh the disadvantages. Finally, the control variables also give interesting insights into staging behavior. First of all, VCs use staging significantly less when they use a financing instrument with a strong debt component. This provides evidence that staging may be partly used as a substitute to debt 15. Second, the percentage of rounds that were realized in period 2 is significantly higher for non-staged firms than for staged firms; the reverse is true for rounds realized in period 3. Finally, the average percentage of independent VCs is significantly higher for staged than for non-staged firms whereas the reverse is true for public VCs Regression results Our empirical strategy is based on the following model: Staging i = f i (Uncertainty, Asymmetric Information, Human Capital, Controls) + ɛ i, where { 1 : if firm i is staged Staging i = 0 : if firm i is not staged While finding adequate controls is relatively straightforward, differentiating between uncertainty and asymmetric information poses a relatively large challenge. As shown in the summary statistics, there are several potential indicator variables: firm age, the firm s development stage, the revenue dummy and the AI indicator. While all these variables capture several elements related to uncertainty and information differences, none is able to capture all of them. Thus, we are given a set of possible variables that, on the one hand, are quite heterogeneous in the economic effects they capture but, on the other hand, are certainly not mutually exclusive as they are partial substitutes for each other. This view is also reconfirmed by the correlations of these variables (see table 5 and 6). What we will do throughout this paper is to run the same standard specification with each proxy separately, thus making sure that our findings are robust to the type of uncertainty found. We now will present some first results with respect to our hypotheses by running univariate probit regressions with error terms clustered at the firm level. The results corresponding to the first part of hypothesis 1 can be found in table 7. As pointed out, we include different proxy variables in order to control for the degree of uncertainty, the 15 See subsection 3 for further comments on this issue. 14

16 degree of asymmetric information and the difficulty of future monitoring. First of all, we regress age on the dummy for staging - as we conjecture that information problems are more pronounced for younger firms. Indeed, AGE is significant at the 5% level showing that the older the firm, the less probable becomes staging. As there may be big differences between the development stages of equally old firms, we replace AGE by our early stage dummy. Now the result is even more pronounced: the dummy is positive and significant at the 1% level. This means that the probability for staging is 19.4% higher for early stage firms. In a second step, we focus more on the asymmetric information aspect and use AI as an exogenous variable. Again the result is significant at the 1% level: the higher the degree of asymmetric information the more probable becomes staging. Using the components of AI on an individual basis yields further insights: in fact, the result seems to be driven by only two components of AI, namely the break-even dummy and the dummy for finished products. Both are significant at the 1% and 10% level respectively. Finally, we include the dummy for revenues as an indicator for information problems into our regressions. This variable accentuates the fact that uncertainty decreases substantially once the firm s product is on the market, i.e. when revenues are realized. This indicator is not significant, however. It is also interesting to have a look at the control variables. Surprisingly, the dummy variables for the different VC types are never significant. The period 3 dummy is significant in all regressions except model 4 whereas the period 1 dummy is only significant in model 5 and only at the 10% level. All together, this means that staging has become more probable nowadays with marginal effects around 10% (regarding models one to three). As concerns our industry dummies, the life-science dummy is positive and significant throughout all regressions and the IT/telecom dummy in all regressions expect the first one. Again, the marginal effects are quite high with values between 15% and 24% for life-sciences and 13% and 20% for IT/telecom firms. Finally, the internet dummy is significant in the last three regressions: the probability for staging is about 18% higher in this industry. To measure the importance of the entrepreneur s human capital and the tangibility of assets, we also run several regressions. In a first step, we include the dummy variable E EXPERTISE (variant a) and, in a second step, the different dummy variables concerning the entrepreneur s human capital (variant b), i.e. the dummy REPEAT EN- TREPRENEUR, the dummy SCIENCE BACKGROUND and the dummy RESEARCH DEGREE 16. We run these two regressions for three different specifications. First of all, we include the fixed asset ratio in order to test our hypothesis. Whereas the fixed asset ratio is not significant in the first variant, it is significant at the 10% level in the second one. This result is not robust, however, because when we include further variables like 16 Alternatively, we include the dummy PRODUCT but it never turns out to be significant. 15

17 AGE the fixed asset ratio is no longer significant. The human capital variables are never significant. Second, we include the categorical variable AI and the dummy early stage as proxies for uncertainty respectively. Both dummies are still significant at the 1% level 17. As concerns our control variables, we get mixed results: the dummy LIFESCIENCE is significant in all regressions of variant a but for variant b only in conjunction with the AI variable. With this specification, the dummies INTERNET and IT/TELECOM are significant in variant a too. Finally, the dummy IT/TELECOM is also significant in variant a of the early stage dummy regression and both in variant a of the FAR regression. These results are not robust, however. The period 3 dummy is, to the contrary, significant in all specifications and all variants except for variant b in the third specification. To conclude, we can say that the regression results are in line with the descriptive statistics. They confirm the first part of hypothesis 1 that states that the higher the uncertainty, the higher the degree of asymmetric information and the more difficult future monitoring, the more probable staging is. Just as in the t-tests, all our proxies are highly significant in each specification of our regressions. The importance of the entrepreneur s human capital, on the contrary, is partly confirmed by the dummy RESEARCH DEGREE in our descriptive statistics, but none of the human capital variables is significant in the regressions. The same is true for the fixed asset ratio. Thus the second part of hypothesis 1 that states that staging should be more probable, the more important the entrepreneur s human capital and the less tangible the assets cannot be confirmed with our data. This result should be handled with precaution, however, due to the high amount of missing values as well as the limitations of our indicator variables Robustness checks As shown in subsection 1, the use of a strong debt component seems to have an influence on the staging behavior. We cannot include this variable in the above mentioned regressions because debt is endogenous as it is a contractual element just like staging and thus it is determined simultaneously to staging. In order to cope with this disadvantage and at the same time do justice to possible substitutability effects, we will run robustness checks for our previous results by splitting our sample in one part that includes all observations that have a strong debt component and a second part with all observations that do not have a strong debt component. As concerns the first part of hypothesis 1, in the previous subsection we showed that firms are more probable to use staging when they are younger, in early investment phases and have a higher degree of asymmetric information. All these issues are confirmed by 17 We also run the regressions with the variable AGE and the dummy REVENUES which yielded us the following results: AGE turns out to be still significant and the REVENUES is still insignificant in both variants. 16

18 our descriptive statistics for the two subsamples (see tables 9-10). In both subsamples, non-staged firms are significantly older than staged firms and the mean of the dummy early stage is significantly higher for the groups of firms that use staging than for those which do not 18. Moreover, firms that generate revenues are significantly less inclined to use staging when they belong to the subsample with a strong debt component. Finally, the mean of our proxy variable for asymmetric information is significantly higher for non-staged firms than for staged firms in both subsamples. In fact, in the subsample of observations with a strong debt component, this result is driven by the component variables AUDITED BALANCE SHEET, BREAK-EVEN, FINISHED PRODUCT whose means differ significantly whereas as to the subsample without a strong debt component only the means of the latter two components differ. As concerns the second part of hypothesis 1, i.e. the importance of the entrepreneur s human capital and the tangibility of assets for the staging decision, we get mixed results. Interestingly, the fixed asset ratio seems to be an important determinant only in combination with the use of a strong debt component. With respect to the human capital variables, we know from the results for the complete sample that only the research degree dummy and the dummy PRODUCT played a role in determining the decision for staging. Now, the first variable plays only a significant role for the subsample without a strong debt component. The mean of the dummy PRODUCT, on the other hand, only differs significantly for the subsample of observations with a strong debt component. Finally, as concerns the industry dummies, the firms of other industries tend to use significantly less staging and life-sciences firms tend to use significantly more staging in the subsample of observations without a strong debt component. Our regression results confirm these observations. When we look at the regression results for the first part of hypothesis 1, our proxy variables measuring the degree of uncertainty and asymmetric information are highly significant in both subsamples (see tables 11 and 12 for regression results) - only in model 1 of the strong debt component subsample, AGE is insignificant. Moreover, a time effect can be found for the observations with a strong debt component, and we can observe an industry effect for the observations without a strong debt component: with life-science and IT/telecom firms, the use of staging is more probable 19. All these effects, except the last mentioned industry effects, are robust when including the human capital variables (see tables 13 and 14) 20. In model 1b of the subsample with a strong debt component, the AI variable is again not significant, but when the component variables of AI are included, the dummy AUDITED BALANCE 18 This fact is also reflected in the individual dummies START-UP and LATE. 19 We get significant results in all regressions except model 5 and in model 3-5 respectively. 20 We also run the regressions with AGE and REVENUES whereby the first variable remains significant in all but one regression and the latter one is still insignificant. 17

19 SHEET turns out to be highly significant. In this case, the dummy RESEARCH DEGREE turns out to be significant too. The same is true when including AGE instead of the AI component variables. In all other specifications none of the human capital variables is significant, however. Interestingly, the fixed asset ratio turns out to be highly significant when using a strong debt component: the higher the fixed asset ratio, the less probable is staging. This means that financial aspects seems to play a more important role in this case Summary To conclude this subsection, we find that the decision for staging is determined by the degree of uncertainty and asymmetric information the investor is confronted with. Throughout all our robustness checks, the proxy variables such as AGE, DUMMY EARLY STAGE and AI are highly significant. As concerns the importance of the tangibility of assets, we can state that it seems to play a minor role. The fixed asset ratio is only significant for the subsample of observations with a strong debt component which means that investors who use a strong debt component seem to care more about financial indicators. With respect to our human capital variables, we get mixed results. Whereas the descriptive statistics partly point out such an effect, it is not reconfirmed in the regression results. As mentioned above, this last result should be handled with precaution, however, due to the limitations of our proxy variables. 4.2 How does milestone financing work? Whereas round financing does not demand any specific requirements, milestone financing calls for a complete contingent contract to be feasible. Signals to write this complete contract on are a necessary precondition for milestone financing. Therefore, before we analyze the decision between the different staging modes in more detail in the next section, we will now investigate the specifics of milestone financing. The essential requirement for milestone financing is the definition of suitable milestones. These milestones must be able to describe the expected development of the firm in the near future. Therefore, the kind of milestones used should depend heavily on the development stage of the firm. We distinguish four different types of milestones in our data set: product milestones such as the realization of a prototype or the successful accomplishment of product tests, financial milestones such as a minimum sales level as well as other firm specific milestones which could not be classified in neither of these two categories and therefore contain a broad range of different signals. One example may be the hiring of a specific type of manager. Finally, we classify all mixtures between different signal types as a separate category. 18

20 A first impression on the use of milestones can be found in table 15. In this table we see that product milestones are used more frequently in earlier development stages than financial milestones that are used in later development stages of the firm. In economic terms this makes sense, as only developed firms are able to generate cash-flows that financial signals can be based on. On the other hand it is difficult to differentiate between the use of signal mixes and firm specific milestones. In detail, we see that we get quite robust results for product and financial signals throughout the descriptive statistics (see table 15) and the univariate probit regressions we run (see tables 16 and 17). As to product signals, these are used when the degree of asymmetric information is relatively high: the mean of all four component variables (BREAK EVEN, AUDITED BALANCE SHEET, FINISHED PRODUCT and REFERENCE CUSTOMERS) is significantly smaller when product signals are used 21. This means that product milestones are more probable if the firm has not yet reached break even, does not have an audited balance sheet, does not hold a finished product and still does not have any reference customers 22. In our regressions with the components of AI (model 4), only the coefficient of the dummy AUDITED BALANCE SHEET turns out to be negative and significant. This is quite intuitive as this is certainly one of the main prerequisites in order to be able to use financial milestones. This asymmetric information issue is also reflected by the significant AGE variable in model 1 of our regression, the significant PRODUCT in model 5 and the significant negative coefficient of the dummy MARKET 23 that indicates that firms that have already entered the market tend to use less product milestones. On the contrary, financial signals are rather used with older firms, in advanced development stages when the degree of asymmetric information is lower, i.e. when break-even is reached, when there exists already a finished product or even reference customers. Similarly, the average percentage of start-up stage firms is lower and the average percentage of later stage firms is higher. This means that the product development process has ended or must at least be quite advanced. All these indicator variables are highly significant both in our regressions and in our descriptive statistics. For our control variables, we find quite robust time and industry effects with respect to financial signals: they have been used more frequently in period 1 and less with life-sciences industries. This may, however, only reflect the underlying characteristics of our data set and the fact that life-science firms are confronted with a rather long development process. Finally, for product signals, we do not get any robust effects in our control variables. The remaining two signal type categories, namely other firm-specific milestones and 21 The same is true for the finished product test dummy. 22 Moreover, the higher mean of the dummy research degree may show that more R&D intensive industries are more probable to have firms in the product development process and tend so to use more product milestones. 23 We do not report the results due to the limited space. 19