MSc Business Economics, Finance track
Master thesis
Can seed investors identify good opportunities?
Marijke van Ruijven - 10248595 June 2016
Statement of Originality
This document is written by Marijke van Ruijven, who declares to take full responsibility for
the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
This thesis investigates, by using an investor-company matched panel dataset, whether and how the characteristics, such as skills and preferences, of individual investors are related to the performance of entrepreneurial firms. The dataset enables to analyse investor fixed effects, which is a function of unobserved characteristics of investors. The performance is measured in market value return, which is only observable for start-ups that did an IPO and there is, therefore, a potential sample selection bias. This study, while controlling for the sample selection by using the Heckman model, provides evidence that not all investors are the same and that some have better skills in identifying good opportunities and adding value to a project. I also find that greater investor fixed effects are related to larger amounts invested and less syndication. Even after controlling for the funding round, I show that the best investors appear more often at later stages and in tech companies.
Table of contents !
I. Introduction!...!5!
II. Literature Review!...!7!
III. Data and Descriptive Statistics!...!13!
3.1 Variable Construction!...!13!
3.2 Data on Start-up Companies and Investors!...!15!
3.3 Investor-Company Matched Panel Dataset!...!17!
IV. Empirical Method!...!18!
4.1 Investor Fixed Effects!...!18!
4.2 Sample Selection!...!19!
4.3 Estimating Market Capitalization!...!21!
4.4 The Worst and the Best Investors!...!21!
4.5 Different Types of Investors!...!22!
V. Results!...!24!
5.1 The Probability of an IPO!...!24!
5.2 Market Capitalization!...!25!
5.3 Investor Heterogeneity!...!26!
5.4 The Worst and the Best Investors!...!27!
5.5 Relationship between the Risk and Returns!...!28!
5.6 Different Types of Investors!...!29!
VI. Conclusion!...!31!
References!...!34!
Tables!...!38!
I. Introduction
Small firms and new businesses have become an increasingly important component of economic growth in developed countries and there has been a dramatic increase in the amount of capital allocated to the private equity market (Denis, 2004). The ability to access capital is one of the most important issues for entrepreneurial firms and since debt financing is in most of the cases not possible, entrepreneurs have to rely on outside equity financing (Denis, 2004). Besides this huge demand for finance, there is limited supply on the other side. An explanation for this is not that the multiple sources of equity financing are not wealthy enough, but that the investments in the start-up phase of entrepreneurial firms are simply too risky. Although the different sources vary in the type of companies in which they invest, in the type of funding, in their control rights and way of monitoring, in their cost of capital and in their liquidity (Winton and Yerramilli, 2008), there still exists a funding gap.
Examples of different types of investors are venture capitalists (VCs henceforth) and angels. VCs invest on behalf of the limited partners of a capital fund and angel investors refer to high net worth individuals that invest their own funds in a small set of companies. These types of investors have different characteristics and play different roles in the firms in which they invest (Kaplan and Stromberg, 2001b). Wong (2002) provides, for example, evidence that angel investments are typically smaller, are concentrated in younger companies, and take place at an earlier stage in the company’s life cycle compared to VC investments.
The aim of this thesis is to investigate what types of investors are behind the most valuable entrepreneurial firms. In contrast to other papers that do research on the performance of angels and VCs (Chahine et al., 2007; Wiltbank et al., 2009; Nofsinger and Wang, 2011), this study uses an investor-company matched panel dataset. This dataset contains 399 U.S. start-up companies that are matched with 332 unique investors, with more than half identified as tech companies. The dataset is restricted to investors that invest in at least two companies, which enables to identify their fixed effects. The significance of these fixed effects is tested to verify whether individual investors matter for the success of start-ups. Each individual investor will have a different effect on the future market value of a start-up based on their observed and unobserved characteristics, such as skills and preferences.
The results provide evidence that indeed not all the investors are the same and that some have better skills in identifying good opportunities and adding value to a project. While testing the effects of different types of investors, the results show that, in each category (i.e. entering zone, level of syndication, investor type and investment size), some types of
investors matter more than others. This suggests that under a competitive capital market, where anyone can invest anywhere, not only the quality of the project itself would matter, but also the skills and preferences of investors.
The quality of a project, measured in future market value, is, however, only observable for start-ups that did an IPO. The dataset is thus also restricted to start-ups that did an IPO, which are often the best performing companies. This sample selection has a potential bias because there might be a relationship between the probability of an IPO and the performance of a start-up company (i.e. the error terms of both regressions might be correlated). This potential bias is circumvented by using the Heckman model, which adds the inverse Mill’s ratio, calculated from the probit regression that estimates the probability of an IPO, as additional explanatory variable.
Studying the distribution of the effects of investors’ skills and preferences enables to track down the worst and the best investors across different companies over time. After controlling for the funding round, I find that the best ones invest more in tech companies and that the worst ones appear more often at earlier stages. It also reveals that the distribution is not symmetric and suggests that there just occur more real good investors than bad investors. However, the low Sharpe ratios indicate that there is a lot of risk associated with the investments and that there are fewer investors that have high returns because they are smart.
By sorting the individual investors in types of investors, based on the funding round in which they appear, the level of syndication, the investor type, and the investments size, this research can study and compare the different types of investors in terms of risk and return. The return can be seen as the average ability of an individual investor and not as the return of a specific investment. Analysing the relationship between risk and returns enables to see whether the phenomenon in the capital asset pricing theory that predicts that a high beta (risk) goes along with a high return (Moskowitz and Vissing-Jorgensen, 2002; Sharpe, 1964), holds or not. This is important for the investors, so that they know whether investing in, for example, an earlier stage, and thus taking more risk, pays off or not.
This study shows that this is not the case, as it finds no significant correlation between the risk and returns of early stage investors, which is thus inconsistent with the capital asset pricing theory. It suggests that taking more risk does not increase the average ability of an investor and that it thus does not pay off, which might explain the existence of the funding gap in the capital market. In fact, the results evoke that later stage investors, who encounter less information asymmetry, have special skills in identifying good opportunities.
whether it pays off for investors to syndicate their investment or not. In essence, the pooling of investments not only increases the availability of capital for current and follow on funding, it also spreads the risk and brings more expertise together. It eases information asymmetries between investors and entrepreneurs, which should improve the selection process (Gompers, 1995), and is a potential way to reduce the funding gap. However, the results reveal, even after controlling for the funding round, that investors that do not syndicate their investment have higher returns and suffer less risk. This thus indicates that stand-alone investors are better in screening projects and suggests that the level of experience plays an important role. In fact, more experienced investors do more stand-alone investments (Lerner, 1994) and are better in obtaining credible signals, and they might therefore have better skills in identifying good opportunities. So the level of syndication itself does not only matter for the improvement of the selection process, but the level of experience as well.
Moreover, the outcomes show that the expected values added by investors that have a large investment size are greater compared to the ones that invest smaller amounts, while the risk associated with those investments are smaller. Because this study uses the market value return as dependent variable and controls for the funding round and the total round size, it does not simply implicate that a larger investment amount results in a higher value of equity. In fact, it insinuates that the high observed value of equity for large investors is caused by greater investor fixed effects that are related to larger amounts invested. These greater effects go along with less volatility, which enables to conclude that these investors are better in identifying good opportunities compared to investors that invest smaller amounts.
The rest of this thesis is organized as follows. Section II provides a brief summary and discussion of previous research on different types of investors, what decisions they have to make and how they circumvent the uncertainties. The different data sources that are used in this study are presented in section III. This section also describes the construction of the dataset and defines the main constructed variables of interest. Based on the existing literature, the empirical hypotheses are derived in Section IV. The methodology to test for these hypotheses is provided in this section as well. Section V shows the results of these tests and Section VI summarizes and discusses the findings and offers some concluding remarks.
II. Literature Review
In the entrepreneurial finance literature, the existence of a funding gap is widely recognized, in which the market is characterized with huge demand for finance on the one side and a
limited supply on the other side (Lam, 2010). An explanation for this is that although the investors on the supply side are often wealthy enough to do investments, there are some constraints on their ability to invest. Mason and Harrison (2002b), for example, find evidence that angel investors do not see enough deals that meet their investment criteria, that most of the received investment proposals are of poor quality, and that they are often not able to negotiate acceptable terms and conditions. In fact, especially seed investors require high returns for their early stage investments, so that they can be compensated for the high risks that are associated with those investments. This is validated by Manigart et al. (2002), who show that VCs require returns between 36% and 45% for their early stage investments, which is significantly higher than later stage investments. This already suggests that the differences in variation and magnitude of the required return depends not only on the type of investor, the investors’ strategies and the decisions on how to deal with uncertainties, but also at what stage the investment is made.
The reason for this is that the risk distribution changes during the stage wise development of start-up companies (Ruhnka and Young, 1987). In fact, according to Dean and Giglierano (1990), investors divide the start-ups’ life cycle into five stages: founders’ round, seed round, second round, mezzanine round, and public company. Because investors make their decisions based on their experience, preferences and tolerance of risk, different types of investors will occur in different rounds. Aram (1989), for example, finds evidence that investors who are more committed to early stage financing are more likely to have been entrepreneurs themselves. However, these investors have lower annual income and lower net worth compared to later stage investors. He also finds that investors that are more committed to technology based companies play a higher-risk, higher-return investment game and that those investors look more to business service professionals as referrals instead of friends. Because technology based companies have an unattractive risk-reward equation, Mason and Harrison (2004), conclude that VCs are reluctance to invest in such companies.
VCs thus differ in their preferences compared to other investors, which affects their decisions and performances. A study that measures the performance of VCs is the one by Cochrane (2001). He uses the standard measure of performance in the VC market, which is the internal rate of return (IRR). The measure of the ex-ante returns to a potential investor in this research is, however, upward biased, because the return to IPO only measures the winners. Cochrane (2001) overcomes this selection bias with a maximum likelihood estimate that identifies and measures the probability of an IPO. With the selection bias correction, he finds a geometric mean log return of about 15% with a -7.1% log CAPM intercept and an
arithmetic average return of about 53% and a standard deviation of 107%. This indicates that VC investments are like stock options, with small chances of a enormous payoffs.
VCs thus experiment with their investments and are sometimes lucky and at other occasions not. The research of Nanda and Rhodes-Kropf (2013), studies whether there are certain times when VC investors are more willing to experiment than others and find that start-ups that receive their initial funding in more active investment periods (i.e. hot markets), are significantly more likely to go bankrupt than those funded in periods when fewer start-up firms were funded (i.e. cold markets). They test whether the same investors change the types of firms they fund in hot and cold markets and find evidence that the increased failure rates in hot markets seem to be driven by within-VC variation, so by investor fixed effects. They thus conclude that investor fixed effects matter.
Another research that draws this conclusion and thus also emphasizes on the importance of investor fixed effects is that of Werth and Boeert (2013). They include the investor fixed effects for a robustness check and find evidence that business angels differ with respect to their attitude towards risks and decision-making as well as in their general ability to source investments and to connect with others. They also state that these investor-specific characteristics may have a positive effect on the performance of their respective portfolio companies. This research thus studies the effects of business angel networks and contributes to the research on the angel investor market.
Although this market contributes, in total, a higher amount of capital to entrepreneurial firms compared to the VC market (Wong, 2002), there is much less research done in this field. The paper of Mason and Harrison (2002a) provides the first attempt to analyse the returns to angel investments and according to them is looking on a deal- by-deal basis and simply measuring the returns in terms of achieved multiples and the length of the holding period, the most appropriate way to evaluate the performance of investments by angels. The paper finds that the distribution of returns is highly negatively skewed, with 34% of exits at a total loss, 13% at a partial loss or break-even, but with 23% showing an IRR of 50% or above. This suggests that angels are more concerned in avoiding bad investment than finding superior ones.
Different investors are thus concerned about different things and thus make other decisions on, for example, in which company and in what round they should invest, whether they should invest in subsequent rounds and what share of equity they should hold (Dean and Giglierano, 1990). Both Mason and Rogers (1997) and Feeney et al. (1999) structure this decision-making process in three stages: screening, evaluation and negotiation. In the first
stage, investors use their investment criteria to consider whether the investment opportunity fits their preferences and strategy. When the investment opportunity satisfies the investment criteria, the investors move on to the next stage in which they evaluate the intrinsic merits (Mason and Harrison, 2002b). In the final stage the investors negotiate with the entrepreneurs over the terms and conditions of the investment (Mason and Rogers, 1997).
Those three stages can be difficult according to Fried and Hisrich (1994), who mention that the decisions investors have to make go along with significant adverse selection risk. Investments are, in fact, illiquid and the success highly depends on a small group of managers/entrepreneurs. This group is able to engage in opportunistic behavior after the investment is made due to information asymmetries (Sahlman, 1988). It is therefore important that the initial decision of investors whether to invest or not is a good decision (Fried and Hisrich, 1994). Hence the performance of investors, as discussed by Gorman and Sahlman (1989), depends on their choices and the researchers state that investors, such as VC firms, make portfolio decisions in which the uncertainties of small and young start-up companies are traded off against the potential returns from the investment. Because the uncertainties are often so high, the trade-offs are based on preferences and rules of thumb that the investors have developed (Dean and Giglierano, 1990).
This is in line with the research of Ruhnka and Young (1991), who suggest that VC investment decisions are primarily subjective assessments. Besides the argument that the investment decisions are complicated because they remain hostage to unanticipated competitors, market shifts and financial cycles, they also mention the high level of uncertainty as argument. The reason why the uncertainty is so high is that small start-up companies are characterized by information opaque (Berger and Udell, 1998). Small companies, for example, do very often not enter in publicly visible or widely reported contracts with their suppliers, employees and customers. Moreover, small companies do not have access to public markets and thus do not trade in securities that are continuously priced. For those reasons it is hard for the start-up companies to credibly convey their quality (Berger and Udell, 1998), which makes it difficult for investors to make good decisions.
There are, however, some strategies that help the investors to identify and to limit the risks (Ruhnka and Young, 1991). One strategy is to require the start-up company to provide a detailed business plan that identifies the major uncertainties involved. A second risk-reduction strategy is incremental funding, whereby the start-up company only receives its follow on funding if it achieves key development benchmarks in the sequential period (Ruhnka and Young, 1987). Consequently, the investor reduces its funding commitment and
only needs to provide additional funding if the start-up is doing well. Another strategy is to diversify the risk by including ten or more different investments in a portfolio, so that the unsystematic risk is reduced, or by including investments at different stages, so that the liquidation risk is reduced (Norton and Tenenbaum, 1993).
However, Bygrave (1987, 1988) argues that maintaining a high level of specialization is more useful for controlling risk compared to diversification. He also points out that specialization is not only helpful for controlling risk, but also for gaining access to networks, information and deal flow from other investors. To exploit their technical and product expertise, investors thus should, according to Bygrave (1987, 1988) have portfolios that are less diversified across companies, industries and financing stages. This hypothesis is favoured in the research of Norton and Tenenbaum (1993), which studies the two investment strategies (i.e. diversification versus specialization) by using the responses to a survey of VCs. They find evidence that investors in seed stage investments are less diversified across different companies and industries and that the investors specialize in a certain financing stage rather than diversify their investments across different stages.
Besides the decision on whether to diversify or specialize, investors also need to make decisions on how much control they should take in the company. The study of Wiltbank et al. (2009) compares the performance of investors with different control strategies and concludes that angels, who emphasize non-predictive control, experience a reduction in investment failures without a reduction in their number of successes. The control strategy of investors is related to the degree to which the investors are active in the firms and in the way they select and monitor their investments (Prowse, 1998). The selection of the investments is also related to the investor’s preferences. Nofsinger and Wang (2011), for instance, show empirical results that VCs rely in their selection on the experience of entrepreneurs and the quality of investor protection and that angels are more likely to have a social relationship with the entrepreneur, and thus have information about that person’s skills and character. This study thus compares different types of investors, such as angel investors and VCs, based on the observed characteristics of the investors and investments.
Another research that studies those two types of investors is that of Hellman and Thiele (2015), which builds on the view that entrepreneurs first receive angel and then VC funding. The researchers see the two types of investor as friends because they need each other. Angels, for example, need successive funding but are in most of the cases not able to provide that by themselves and thus rely on VCs. On the other side, VCs need angels for their own deal flow and for the reduction of information asymmetry (Admati and Pfleiderer, 1994).
The complementary role of the two investors was already shared in the research of Berger and Udell (1998), who state that angel contracts are often only constructed if a possible future VC is anticipated. However, according to Hellman and Thiele (2015), angels complain in practice that VCs offer unfairly low valuations, which decreases the incentive for angels to invest in the seed stage of start-ups. The VCs thus abuse their market power according to the angels. Despite the debate whether they are ‘friends’ or ‘foes’, both types of funding are subject to moral hazard. The moral hazard problems are likely to occur when the entrepreneur needs a relative large amount of external finance compared to his/her own wealth, which is at risk via pledges of personal collateral or guarantees (Berger and Udell, 1998). So especially the allocation of funds and the learning process of high-growth, high-risk companies, which need a lot of external finance and only possess a small amount of insider finance, are subject to moral hazard (Bergemann and Hege, 1998). To control and mitigate the risk, investors can stage their financing (Bergemann and Hege, 1998; Gompers, 1992, 1995; Neher, 1999). Another complementary mechanism that also effectively control for agency problems is contracting. According to Van Osnabrugge (2000), who compares angel investors and VCs, both types of investors reduce the agency risk at all the stages of the investment process by using contracts. However, he finds evidence that the two investors differ in the way they place emphasis on the contracts. Angels, for example, use the incomplete contracts approach and assume that writing a contract is costly due to transaction costs, bounded rationality and asymmetric information, and is therefore incomplete. For that reason they focus more ex post investment and renegotiate after new information is revealed. On the other hand, VCs incur screening costs to reduce the asymmetric information between the principal and agent and thus use a more ex ante approach (Van Osnabrugge, 2000).
Another way to resolve for information asymmetry and the associated adverse selection problem is syndication (Lerner, 1994). Syndication refers to investors that are looking for other investors to complete their investments in companies and can either take place in the first round or in the subsequent rounds (Casamatta and Haritchabalet, 2007). Besides the increased availability of capital for current and follow on funding that is provided by syndication, it also spreads the risk and brings more expertise together. It is a chance for investors to double check their own decisions (Perez, 1986) and is thus a way to gather information on investment opportunities. The reduction in information asymmetry improves the selection process and lead to the termination of less promising start-ups (Gompers, 1995). Both Lerner (1994) and Casamatta and Haritchabalet (2007) explain that investors choose their financial partner based on the level of experience and the funding round.
Experienced investors, for example, syndicate in the first round in most of the cases only with investors with similar levels of experience, while in later rounds this level of experience is less important. This suggests that experienced investors invest less in the first round than inexperienced investors because it is harder to find co-investors with the same level of experience. This is in line with the findings of Aram (1989), which show that investors with a fewer number of different types of co-investors tend to have more personal experience. Another reason, provided by Casamatta and Haritchabalet (2007), is that it is more costly for experienced investors. In fact, they need to execute more accurate appraisals because they have more to lose (e.g. reputation).
Despite the expectation that more experienced investors, who obtain better signals (Lerner, 1994) and do more stand-alone investments (Hopp and Rieder, 2006), receive higher returns, Brander et al. (2002) find that syndicated investments exhibit higher returns and higher volatility compared to stand-alone investments. However, this research does not take into account the level of experience and Hege et al. (2006) find, while using a different performance measure, no significant relationship between the size of the syndicate and the level of excess returns. Cumming (2006) does find a significant relation between the probability of syndication and the investment size and concludes that syndication is more likely to happen the greater the total value of the investment. Besides this relation, the size of investment is also positively related to the likelihood of a successful exit (Guo et al., 2015) and thus the performance of the start-up company. In fact, according to Gue et al., (2015) the higher the investment size, the greater the likelihood of a successful exit and in the sample that only contains successful exits they find evidence that the greater the total investment value, the higher the likelihood of an IPO exit.
III. Data and Descriptive Statistics 3.1 Variable Construction
The new constructed variables are briefly explained in this section and are provided in Table 1. The constructed dependent variable, for example, is based on the new generated variable (ty_ipo) that indicates the fiscal year of the company two years after the IPO. The logarithm of the market value (csho times prcc_f) of those fiscal years can be seen as the market value two years after the IPO (mv). The logarithm of the cumulative funding amount at the time of each round is indicated as cum and is the sum of all previous received amounts of funding for
each company. The dependent variable in this study (i.e. market value return (mv_return)) is measured by the market value (mv) divided by the amount invested up to that point (cum). This rough measurement of investor returns put more weight to the first round and thus controls for the different weights of each funding round. The dependent variable ipo is used in the first stage probit regression, as will be explained in section 4.2, and is a dummy variable that is equal to one if the company did an IPO and zero otherwise and is based on ipo_date.
Besides these dependent variables, this research also constructs new independent variables such as the companies’ stage and funding amounts. Investments are made in the company’s seed stage of its start-up financial cycle if the funding round type is either ‘seed’ or ‘angel’. If the investments are not made in this stage, but the company has just one or two funding rounds (rounds), the investments are assumed to be made in the early stage and the dummy for early stage is equal to one. This is also the case if the company has more than two funding rounds, but the funding round code is either A or B at the time of investment. Investments that are made in a company that has more than two funding rounds and that are not identified as seed or early stage investments are indicated as being made in the later stage.
The total funding amount of each company (totalfund) is measured by the logarithm of the total funding amount and the size of the total funding of each company in each funding round (size) is measured by the logarithm of the total raised amount in each round. Both variables are measured in USD. The average total funding per investor in each company (avg_fund) is measured by the total funding (totalfund) divided by the total number of investors in each company. The average total funding per investor in each company in each round (avg_size) is measured by the round size (size) divided by the total number of investors in each funding round (investor). The funding rounds are identified by the investment date (fund_at) (i.e. the investments made on the same date in the same company are grouped as a funding round). The dummy for syndication (syn) is equal to one if there is more than one investor in a funding round and zero otherwise.
From the perspective of the investor, some additional variables are created such as the number of companies in which each investor invests (occur) and the number of total investment of each investor (invest). Also a dummy (loc) is created that is equal to one if the investor operates in the same region as the company in which he or she invested.
This research also controls for the age (age_ipo) in years of the company at the IPO, measured by the difference in the founded date (found_at) and the IPO date (ipo_date), and the age (age_round) in years of the company at each funding round (fund_at minus found_at). To extend this control variable, the company’s age (age_f(l)f) at the first (last) received fund
is generated (f(l)fund_at minus found_at). However, the probit regression uses the current age1 (age) of the company as control variable. The dummy variable that is equal to one if the company operates in the technology industry and zero otherwise, controls for the industry. The tech companies are identified based on their SIC codes (sic) in the same way as Loughran and Ritter (2004) did.2
3.2 Data on Start-up Companies and Investors
The data on start-up companies and investors comes from CrunchBase, which can be seen as a free database of technology companies, people, and investors that anyone can edit. Any addition made by professionals in the technology community goes through an approval process before being made available online. This approval process is executed by TechCrunch, which is the most influential technology blog in the United States, and the database can therefore be seen as reliable (Werth and Boeert, 2013).
The first dataset contains information on start-up companies and consists of 66,368 companies from 137 countries starting from 1900. It contains information on location, industry, status, number of funding rounds, total raised money, founding date and the dates of the first and last funding round of the companies. The second dataset contains information on investors and consists of 30,205 unique investors with a total of 168,635 investments that are linked to 44,579 unique companies, with the first investment made in 1977. Data on the type of funding, the type of investor and the investment amounts are available in this dataset. The two datasets are merged so that it is known for every company by which investors it is funded. The sample dataset is not restricted to certain countries and thus contains worldwide firms and investors. It contains 44,407 companies, from 124 different countries. However, 22,977 companies operate in the U.S. and thus represent more than half of the dataset. Most of the companies, 34,297, are still privately held by founders and initial investors, 4,941 companies were acquired, 948 did an IPO and 4,221 closed down. Almost all of the unique investors from the second dataset are perfectly matched to companies from the first dataset, which results in a total number of 30,149 investors who invest on average 5.65 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
1
As of 17th May 2016.
2!Tech companies are defined as those in SIC codes 3571, 3572, 3575, 3577, 3578 (computer hardware), 3661, 3663, 3669 (communications equipment), 3671, 3672, 3674, 3675, 3677, 3678, 3679 (electronics), 3812
(navigation equipment), 3823, 3825, 3826, 3827, 3829 (measuring and controlling devices), 3841, 3845 (medical instruments), 4812, 4813 (telephone equipment), 4899 (communications services), and 7371, 7372, 7373, 7374, 7375, 7378, and 7379 (software).!
times in 4.40 different companies, shown in the first panel of Table 2.
The high number of observations and the dataset structure enable the use of fixed effects in order to study the unobserved time-invariant specific characteristics of investors such as skills and preferences. To estimate the investor fixed effects, all the investors that did not invest in multiple companies are dropped out. The descriptive statistics of this restricted sample are shown in the second panel of Table 2. The restricted sample consists of 40,764 companies with a total of 10,953 unique investors. Around 39 percent of these companies received their first funding in the seed stage of their financial cycle, which means that the funds are in most of the cases provided by family or friends of the entrepreneur or by business angels. In fact, according to Berger and Udell (1998), the seed stage is associated with the development of a formal business plan and the company is at that point very small, young and opaque and, therefore, must rely on initial insider finance, which are funds provided by the start-up team itself, family and friends, on trade credit and/or on angel investors.
The companies raised in total a median amount of 4.2 million dollars, with on average 1.97 funding rounds. Around 38 percent of these funding rounds consist of more than one investor, with an average of 2.28 investors per round and a median amount of $3.5 million that is raised in each funding round. 41,108 investments are made in the seed stage of the company and these investments together raised an amount of $12.5 billion, which is around 1.91 percent of the total amount raised. 74,062 investments are made in the early stage and together account for 57.8 percent of the total raised amount of dollars. The remaining investments are made in later stages. Companies that are still privately held by founders and initial investors amounts to 31,311 in this restricted sample, while 4,823 of the companies were acquired, 896 did an IPO and 3,734 closed down.
The median age of the firm at the time that it receives its first funding and last funding is 1.54 years and 3.04 years, respectively. The negative minimum of the age of the first and last round indicate that some companies receive their funding before the company is officially founded. With the new incorporated restriction that allows only investors that invest in more than one company, the number of investments and the number of different companies in which they invest heavily increases to 13.66 investments and 10.36 different companies.
Financial data comes from COMPUSTAT North America Monthly Updates - Fundamentals Annual. This database is chosen because of limited data availability in global databases and because CrunchBase only provides complete coverage of the United States technology start-up market and not of the global technology start-up market (Werth and Boeert, 2013). The new sample is thus restricted to the location of the firm and only contains
companies that operate in North America. The COMPUSTAT dataset represents all U.S. publicly listed companies from 1977 until today and consists of 32,907 companies. The starting date of 1977 is chosen because at that time the first investment that is reported in CrunchBase was made. The duplicates in this dataset are dropped out.
3.3 Investor-Company Matched Panel Dataset
Before the CrunchBase and COMPUSTAT datasets can be merged, some additional pre-processing is done so that the merge becomes more successful. The company names in both datasets are converted to names without capital letters and eliminated for everything behind the last ‘,’ in the company name and for ‘inc’, ‘inc.’, ‘llc’, ‘ltd.’, ‘s.a.’, ‘s/a’, ‘sa’, ‘plc’, ‘plc.’, ‘limited’, ‘ag’, ‘s a’, ‘corp.’, ‘corp’, ‘p l c’, ‘p.l.c.’ ‘ab’, ‘a/s’, ‘a.s.’. This process cleans the company names and makes fuzzy matching easier with more accurate results. 28.4 percent of the companies in the restricted sample set are merged with the COMPUSTAT dataset, which results in a dataset of 1,138 companies on which financial data is available. The companies are matched with 1,977 unique investors.
This study uses the logarithm of market capitalization two years after the IPO divided by the amount invested up to that point as dependent variable. The market capitalization is used to evaluate the performance since this research is interested in the actual size of the company given by the market. During the IPO, the market value of a firm is undervalued and there seems to be a decline in operating performance (Jain and Kini, 1994). This underpricing is however a short term phenomenon (Ritter, 1991). By taking the market value of firms two years after the IPO, the effect of the IPO itself will be reduced and this will thus give a more accurate estimate of the true value (performance) of the firm. The dataset is thus restricted to companies for which the market value two years after the IPO is available.
This results in a dataset that contains 399 companies that are matched with 332 unique investors, with more than half identified as tech companies. The descriptive statistics of this final sample set are shown in Table 3 and reveal that the median logarithm market value of the companies two years after the IPO amounts to 19.39. However, this value has a wide spread (i.e. 14.21 to 26.11). The percentage of firms that are in the seed stage of their financial cycle at their last received funding declined from 32 to 1 percent compared to the unmatched dataset, which is due to the fact that companies in their seed stage are in most of the cases not yet eligible to do an IPO. Most of those companies are thus excluded, because
this sample set requires firms that did an IPO so that financial information is available.
The companies raised, compared to the $4.2 million in the previous unmerged dataset, in total a median amount of around 56 million dollars, with on average 2.90 funding rounds. This signs that companies need to raise a large amount of funding before an IPO can take place. The small increase in the average funding rounds indicates that the funding rounds for those companies raise more money and/or are larger (i.e. consist of more investors). Indeed, the median amount that is raised in each funding round increased from $3.5 million to $20 million. However, only 42 percent of the funding rounds consist of more than one investor, with an average of 2.40 investors per round, which is just a slight increase compared to the unmerged dataset and thus suggests that funding rounds are not larger.
The median age of the firm at the IPO is 7.88 years and the median age of the firm at the time of a funding round is 5.35 years. However, both variables highly vary and have large spreads because of the large time span in this sample. Out of the 332 investors, only 20 percent of the investors invest in the same region and the average number that an investor invests amounts to 5.69 in, on average, 3.81 different companies. The occurrences have, however, wide ranges from 2 up to around 45 investments and 43 different firms.
IV. Empirical Method
The aim of this thesis is to investigate, by using an investor-company matched panel dataset, whether and how the characteristics of individual investors are related to the performance of entrepreneurial firms. The dataset enables to analyse investor fixed effects, which is a function of unobserved characteristics of investors. Each individual investor will have a different effect on the future market value of a start-up based on their observed and unobserved characteristics, such as skills and preferences. The combination of the point estimates and the variances of the fixed effects will tell whether some investors have special skills or not. In addition, different types of investors are studied to see which investor have more impact or, in other words, what types of investors are better (i.e. have special skills) in identifying good opportunities that are behind successful start-ups.
4.1 Investor Fixed Effects!
The first objective is to test whether the investor’s characteristics affect the investor’s ability to screen projects and thus whether they matter for the success of start-ups. Based on the
research of Berstrand and Schoar (2003), Nanda and Rhodes-Kropf (2013) and Werth and Boeert (2013), who all conclude that individual fixed effects matter, the first hypothesis is formulated:
Hypothesis 1: Characteristics, such as skills and preferences, of investors do matter and have a significant effect on the market value of entrepreneurial firms.
The dependent variable that is used in this research is the market value two years after the IPO, measured by the logarithm of market capitalization (number of common shares outstanding multiplied by the fiscal year closing price), divided by the amount invested up to that point. The investor fixed effects thus measure the expected value added by the investors. The significance of these fixed effects is tested by a partial F-test, in which is tested whether the fixed effects differ from each other or not, to verify whether individual investors matter for the success of start-ups. To estimate the effects of the unobserved characteristics of investors, a set of investor dummies is created that contains for each investor a dummy that is equal to one if the investor occurs in the sample set and zero otherwise. To conduct the partial F-test, two regressions are executed. The first regression refers to the restricted regression and does not include the point estimates of the effects of investors’ characteristics, while the second regression does include the point estimates and is labelled as the unrestricted regression. To calculate the F-statistic, the following formula is used:
In which SSRc is the residual sum of squares obtained from the restricted regression, SSRu is the residual sum of squares obtained from the unrestricted regression, r is the number of restrictions, n is the number of cross sectional units and k is the number of independent variables. A high F-statistic suggests that at least one investor is systematically differentin the way that he/she screens projects and that the null hypothesis that every investor adds the same expected value to the company thus cannot be rejected.
4.2 Sample Selection
The dependent variable, as previous mentioned, is the market value two years after the IPO divided by the amount invested up to that point. The market value is, however, only
( 1) c u u SSR SSR r F SSR n k − = − −
observable for start-ups that did an IPO and there is, therefore, a potential sample selection bias. In fact, the error term of the regression on the market value might be correlated to the one from the selection equation, which is in this case the equation that estimates the probability of an IPO. One way to circumvent this bias is to use a maximum likelihood procedure to measure the probability of going public (Cochrane, 2001). Another way, which is used in this research, is using the Heckman model (Heckman, 1979). This model uses two stages to address and solve the non-linearity due to the sample selection.
Specifically, in the first stage a probit regression is executed in which the dependent variable is a dummy that is equal to one if the firm did an IPO during the sample period and zero otherwise. The regression is done on the full sample that still contains all the companies from the CrunchBase dataset and not on the final restricted sample set (i.e. selected sample) that only consists of companies that are matched with COMPUSTAT and for which the market value is observable. This research assumes that the likelihood of an IPO (and thus the likelihood of the market value being observed) is a function of the age of the company (age), the number of funding rounds of the company (rounds), the total received amount of funding of the company (totalfund) and whether the company operates in the technology industry or not (tech). The first stage probit regression of the Heckman model looks like:
!"#! = ! !! +!!!!"#! +!!!!!"#$%!!+!!!!!"#$!!"#!+ !!!"#ℎ! + !!
The inverse Mills’ ratios are generated from the estimations of this probit model and will be used in the second stage, in which it is included as an additional explanatory variable to correct the selection bias. The inverse Mills’ ratio is the ratio of the probability density function over the cumulative distribution function of a distribution and the inclusion of this variable in the OLS regression provides consistent, asymptotically efficient estimates for all the parameters. The formula to calculate the inverse Mills’ ratio is as follows:
!! = !
!(!!!) Φ(!!!)
In which the (!!!) represents the probit estimates of the selection equation, ! the standard normal density function and Φ!the standard normal cumulative distribution function.
4.3 Estimating Market Capitalization
The second stage regressions in the Heckman model are the unrestricted and restricted regressions, as mentioned in section 4.1, in which the unrestricted regression includes the investor fixed effects and the restricted regression does not. In the unrestricted regression, the market value return is a function of round effects, year effects, investor fixed effects, some control variables and the inverse Mills’ ratio. The model looks like:
!!"! = !!!!!!+ !!!+ !!+ !! + !!!!+ !!"!
In which !!"! represents the market value return of company ! in round ! at year !. !!"# is a vector of control variables and!!!!and !! denote the round and year effects, respectively. !! is the sum of all the coefficients for the investor dummies in which ! belongs to !!", which is
the set of investors in company ! by round !. ! is the correlation coefficient between the error terms of the selection equation and this equation, !! is the inverse Mills’ ratio, and ε!"#!is the
error term. The control variables on company level and funding round level are the age of the company at each round (age_round), the total amount of funding at each round (size), the industry in which the company operates (tech) and a set of dummies that indicate the round number (round2, round3, round4 and round4+). The standard errors are clustered by
company to account for autocorrelation and thus allow for intragroup correlation. This relaxes the requirement that all the observations need to be independent (i.e. it only allows for
independent observations across companies but not necessarily within companies).
4.4 The Worst and the Best Investors
The second objective of this research is to investigate the magnitude of the expected value added by investors and to identity successful and less successful investors. This is done by looking at the distribution of the investor fixed effects. The investors in the upper tail of the distribution can be seen as the top investors, with special skills, and are thus assumed to be the best in identifying good opportunities. The investors in the lower tail of the distribution are less successful and are assumed to be the worst in identifying good opportunities. However, the distribution can show a somewhat distorted picture. The point estimates of the investor fixed effects in the lower and upper tails might be encountered with high standard errors. The distribution of the t-statistics of the investor fixed effects is therefore drawn and enables identifying the significant fixed effects based on any significance level.
The descriptive statistics of the top investors are compared with those of the worst investors. To see whether the two groups differ from each other, a t-test that tests for differences in means for two different groups (i.e. the worst investors and the best investors) is conducted. In the case of company entrance, for example, the means of the variable that indicates at which round an investor enters the company (i.e. round) are compared. The formula that is used to conduct the t-test is as follows:
! = ! ! − ! !!!
!!+ !!!
!!
A p-value smaller than 0.05 indicates that the t-test is significant at a five percent level and that the null hypothesis that there is no difference between the two population means can be rejected. The means of every variable from the descriptive statistics are compared, and thus for every variable a t-test is conducted, so that it reveals on what aspects the worst and best investors are similar or differ from each other. This research hypothesizes that the means of the worst and the best investors are different from each other.
4.5 Different Types of Investors
The last objective of this research is to compare different types of investors, such as angels, VCs, early stage investors, syndicated investors etc., to see whether some of them have better skills in identifying good opportunities. Based on the commonly held view, derived from the capital asset pricing theory (Sharpe, 1964), that an investor needs to be compensated for taking on risks, the following hypothesis is formulated:
Hypothesis 2: It pays off for investors to invest in an earlier stage, and thus taking more risk. This hypothesis is thus also based on the view that investments made in earlier stages of start-up companies are more uncertain. Nofsinger and Wang (2011), for example, state that VC investors have an advantage over angel investors in overcoming information asymmetry and the moral hazard problems because they have more information about the entrepreneur and the company. A reason for this is, according to them, that by the time of VC involvement, the start-up company has demonstrated the viability of the business and the use of previously obtained funds (possibly from angel investors). This information is usually unavailable during
the initial start-up phase and thus makes the investments made in that phase more risky. One way to reduce the volatility of an investment is, as mentioned in section 2, syndication. Syndication will spread the risk, will bring more expertise together and will reduce the information asymmetry, which improves the selection process and lead to the termination of less promising companies (Gompers, 1995). Based on this and the previous mentioned commonly held view that predicts that high risk goes along with high returns, the last hypothesis is derived:
Hypothesis 3: Investors that are classified as syndicated investors are less volatile and exhibit lower returns compared to stand-alone investors.
These two hypotheses will be tested by first dividing the investors in different types of investors. This analysis is divided in four categories (i.e. entering zone, level of syndication, investor type and investment size), in which in each category the investors are grouped differently. In each category the average fixed effect for every type is calculated.
In the entering zone category there are three types of investors: seed investors, early stage investors and later stage investors. In this category however, the investors are also grouped based on whether they enter the company in the first round or not. Based on this criteria there exist just two types of investors: first round investors and no first round investors. The level of syndication category consists of two investors as well: syndicated investors and stand-alone investors. In the investor type category there are four types of investors: angels, debt financiers, early stage VCs and later stage VCs. In the last category the investors are grouped based on whether their investment size is below or above the benchmark of $500,000 (or the median investment size of the sample), so that there exist again two types of investors: small investors and large investors.
The first objective is to test in each category the significance of the average fixed effects per type to verify whether individual investors matter for the success of start-ups. This is tested by a partial F-test, in which is tested whether the fixed effects differ from each other or not. To conduct this partial F-test, a restricted and an unrestricted regression are executed, as explained in section 4.1. The second objective is to see how heterogeneous the average fixed effects per type are. Looking and comparing the distributions of the fixed effects per type of investor in each category will reveal for which type the values are more skewed to the right or left and more concentrated around the mean.
The last objective is to test the relationship between the risk and returns of the average fixed effect per type. The mean and standard errors of the fixed effects are therefore plotted.
The return can be seen as the average ability of an individual investor and not as the return of a specific investment. Comparing the means and standard errors of different investor types will give insight about what type is more volatile, and whether higher volatility will also lead to higher returns, and will reveal information about whether some investors have special skills or not. Investor types that have higher returns and lower standard errors compared to other types of investors are assumed to have better skills in identifying good opportunities.
V. Results
5.1 The Probability of an IPO
To test whether the investor’s characteristics, such as skills and preferences, are related to the performance of a start-up company and thus whether they affect the investor’s ability to screen projects, the Heckman model is used. This model is used because the performance is measured in market value return, which is only observable for start-ups that did an IPO. Because it is unlikely for start-ups that are not performing well to choose to go public, the sample only contains companies that are performing well and thus do not reflect the true environment. In order to get a true reflection of the effect of investor’s characteristics on the performance of start-ups, first the probability of an IPO, and thus the probability of an observed market value, is estimated. The probability of becoming public increases when the total amount of received funding increases, when the company gets older and when the company operates in the technology industry, as shown in the first column of Table 4. The number of funding rounds, however, has no significant effect on the probability of an IPO. To see whether there is indeed a relationship between the probability of an IPO and the performance of a company, a likelihood-ratio test is conducted. This test examines whether the correlation coefficient between the errors terms of both regressions (i.e. rho) is equal to zero. If the correlation, in this case -0.934 as reported in Table 4, is nonzero, it means that doing an IPO affects the performance of a company and that, due to this relation, the estimates of the performance do not reflect the actual values. The outcomes of this test (i.e. a χ2 of 223.55 and a p-value smaller than 0.0001) indicate that there is indeed a relationship and that the estimates of the performance are thus biased. For that reason, the Heckman model is used because it controls for the sample selection that causes biased estimates.
The Heckman model consists of two stages, as mentioned in section 4.2, in which in the first stage a new variable (i.e. inverse Mills’ ratio) is created. This variable can be
calculated, for every company, from the probit regression that estimates the probability of an IPO. This variable is added in the second stage regressions as additional independent variable to control for the nonlinearity caused by the sample selection. The first stage thus needs to be performed first in order to move on to the second stage.
5.2 Market Capitalization
In the second stage, the expected value that an investor, due to its skills and preferences, adds to the performance of start-ups is estimated. The outcomes of this regression are reported in the third column of Table 4 and reveal that the age of the company and the total amount of funding at each round have a significant negative effect on the market value return. This means that the older a company becomes and the greater the amount of funding in a round is, the lower the investor’s return on his/her investment will be. Whether the company operates in the tech industry and in which round an investor enters the company, have however no significant effect on the performance.
To test whether the expected value that an individual investor adds matters more than others, a partial F-test is performed. To conduct this test, as mentioned in section 4.1, two regressions are necessary: an unrestricted regression and a restricted regression. The outcomes of the two regressions are reported in the second and the third column of Table 4. The second column represents the restricted regression, in which the investor fixed effects are not included, and the third column shows the unrestricted regression. The F-statistic that is calculated based on these two regressions amounts to 16.401 with a p-value that is smaller than 0.0001, which allows rejecting the null hypothesis that all the expected values added by investors are the same. This suggests that investors are not homogenous and that the unobserved characteristics of investors both economically and statistically matter, which is line with the first hypothesis (i.e. characteristics, such as skills and preferences, of investors do matter and have a significant effect on the market value of entrepreneurial firms).
The previous test tests for all the investors without making a distinction between investors. However, this study is also interested in the effect of different types of investors, as explained in section 4.5. Therefore, the average fixed effect per type is estimated and for each category a partial F-test is conducted to test whether these estimates differ from each other or not. The regression outputs for the different categories (i.e. entering zone, level of syndication, investor type and investment size) are reported in Table 5. The first column represents the restricted regression that does not contain investor fixed effects. This restricted
regression is associated with the unrestricted regressions of the entering zone category, and does not include the funding rounds as control variables to prevent multicollinearity. Column five again represents a restricted regression that does not include investor fixed effects, but now controls for the entering zone of the investors. This regression is the restricted regression for the other three categories.
The outcomes of the partial F-tests are reported in Table 6 and verify that, in each category, some average fixed effects per type matter more than others for the success of start-up companies.
5.3 Investor Heterogeneity
This study is thus interested in how heterogeneous the investor fixed effects are. This can also be studied by looking at the distribution of the investor fixed effects, which is shown on the left side of Figure 1. This figure shows that the distribution is not symmetric and is skewed to the right (i.e. skewness is 0.657), with a median (-0.182) smaller than the mean (-0.100). There are some outliers in this sample, as the distribution has a spread from -4.601 to 6.848, while the 10th and 90th percentile just amount -1.707 and 1.411, respectively. So an explanation for why the median and mean differ from each other can thus be that there just occur more real good investors than bad investors. In fact, the investors in the upper tail can be seen as the real good investors.
The left distribution in Figure 1 can, however, show a somewhat distorted picture, because the point estimates of the effect of investors’ characteristics in the upper and lower tail might be encountered with high standard errors. The distribution of the t-statistics of the investor fixed effects is therefore drawn, showed on the right side of Figure 1, and enables identifying the significant point estimates based on any significance level. This figure is again somewhat skewed to the right, with a skewness value of 0.396, and a mean (-0.165) that is just a slightly bit higher than the median (-0.221).
The t-statistics can be interpreted as the Sharpe ratio for each investors and the decrease in skewness suggests that there are less real good investors or more bad investors. The Sharpe ratio estimates the risk-adjusted performance and helps to explain whether the excess returns are due to smart investment decisions or due to too much additional risk. The low Sharpe ratios indicate that there is a lot of risk associated with the investments and that there are fewer investors that have high returns because they are smart.
per type in each category, shown in the Figures 2 to 8, are compared. From those figures the conclusion can be made that none of the distributions are symmetric and that the one for investors that enter later in the company are skewed to the right or less skewed to the left compared to early stage investors, that the distribution of stand-alone investors is skewed to the right as well and that the one of investors with large investment size is skewed to the left. The distributions of those three types of investors are in each case more concentrated around the mean. These findings suggest that all the investor types have dissimilar elements and that some fixed effects thus mater more than others.
5.4 The Worst and the Best Investors
The second objective of this research is to investigate the magnitude of the investors’ effects and to identity successful and less successful investors. This is again done by looking at the distributions of the investor fixed effects and the associated t-statistics, which are shown in Figure 1. The point estimates of the expected value added by investors with t-statistics that are in the upper or in the lower tail of the distribution (i.e. below the critical value of -1.96 or above 1.96) are assumed to be significant at a five percent level. This means that these large estimates are not caused by too much additional risk.
This study uses the 90th percentile as benchmark for the upper tail, which means that investors that have a significant point estimate of the effect of their unobserved characteristics that is greater than the 90th percentile (i.e. greater than 3.563) are assumed to be in the upper
tail. In this sample five investors are labelled as best investors and together they invest 19 times in the sample set. The investors in the lower tail are less successful and are assumed to be the worst in identifying good opportunities. Investors with a significant point estimate of their fixed effects that is smaller than the 10th percentile (i.e. smaller than -2.955) are assumed to be in the lower tail. In this case, there are four investors that are assumed to be the worst one and they together occur 16 times.
The 75th and 25th percentiles are used for a robustness check. Using this benchmark results in eleven investors that can be seen as the best ones, because their point estimates are greater than 1.725, and they together invest 38 times in the sample. There are twelve investors with point estimates smaller than -1.773 and those investors are thus assumed to be less successful and together they invest 39 times.
The characteristics of the investors that are assumed to be the best are summarized in the second (and fourth) column of Table 5. The top investors are compared with the worst
investors. The fifth row of Table 5 provides evidence that the best investors invest more in tech companies compared to the worst investors. This difference is significant at a five percent level (i.e. the p-value is 0.03). However, while using the 25th and 75th percentile as benchmark to identify the worst and the best investors, it is just significant at a ten percent level. This suggests that the difference, in terms of the industry in which the investors invest, is stronger between the very best and the very worst investors.
From row six in Table 5 it can be seen that the worst investors invest, on average, more often at an earlier stage of the company compared to the best ones. This difference is significant at a ten percent level for both benchmarks. This result, together with the outcomes provided in the seventh row, also suggests that the best investors start investing in later rounds compared to the worst ones. However, the difference between both groups of investors is only significant while using the 25th and 75th percentile as benchmark. A possible explanation why this result is only significant while using that benchmark and not while using the other benchmark, is because the 25th and 75th percentile benchmark allows for more
investors and thus more observations.
Table 5 also reveals that the best ones invest in companies that, on average, have a higher market value return compared to the worst ones. In fact, the difference in the means of market value returns of the worst and best investors differ significantly at a five percent level, as shown in both the third column and the sixth column. The market value return, as already explained in section 3.1, is a function of the market value and the amount invested up to that point and thus puts more weight to early stage investments. Together with the evidence that the best investors enter the company at a later stage, the conclusion can be drawn that the higher market value returns of the best investors are caused by higher market values and not because the investments were made in early stages.
5.5 Relationship between the Risk and Returns
To test the relationship between the risk and returns of investor fixed effect, the means and the logarithm of the standard errors are plotted. Figure 9 shows the relationship between the risk and returns of all the investors, while Figure 10 to 15 show the relationship with a distinction between investors. Except for Figure 12, all the figures show that there exists (almost) no relationship between the risk and returns. The average ability of an investor is thus not affected by a change in volatility, which contradicts the capital asset pricing theory (Sharpe, 1964).
