Additional Robustness Checks

In this subsection we describe the results from six additional robustness checks (Models 5 –10) that are presented in Table 21.8 . As well, we discuss a number of additional robustness checks that were carried out but not explicitly reported. So that we can include dummy variables for exit years, Models 5 –10 exclude buyback and non exits from the sample. Model 5 presents a standard specification without buybacks and nonexits that is comparable to Model 1 without such exclusions. The regression results are similar and continue to support our central hypothesis. Model 6 provides a similar regression, but excludes investment years 1995 to 1998, because we want to check if contracts are related to exits in a way that is driven by the time period spanned by the data. For instance, if VCs in Europe were less sophisticated in the mid-1990s, and thus wrote less detailed contracts, and since market conditions enabled different exit opportunities in the Internet bubble period, then contracts might be connected to exits for reasons unrelated to the hypotheses discussed in subsection 21.2.1. However, the estimates in Model 6 indicate that the data do not support this alternative explanation for the results. The relation between VC control rights and acquisitions continues to hold for the subsample that excludes investment years 1995 to 1998. The results also hold for other subsamples that exclude other periods that are not explicitly presented here. These results are available on request. Regression Models 5 –8 in Table 21.8 indicate a negative relation between the right to replace the CEO and write-offs. That is, detailed VC contracts tend to enable the VC to prevent“ bad ” outcomes. In Model 7, for example, the right to replace the founding entrepreneur specifically reduces the probability of a write-off by 31.7%, and each additional control right studied (drag-along, redemption, and antidilution) reduces the probability of a write-off by 18.2%. This finding is consistent with Lerner (1995) and Gompers and Lerner (1999a). It is also consistent with the Kaplan et al. (2007) result that international VC funds that do use contracts with strong VC control rights are more likely to survive the Internet bubble crash after 2000. Our results suggest that strong VC rights are more likely to protect a VCs interest and force an acquisition. Weak control rights are more often associated with IPOs, but are also associated with write-offs and a non-covered invest- ment. For instance, there were 15 pure common equity investments in the data with absolutely no control rights above and beyond those held by the entrepreneur. Among those, six were IPOs, six were write-offs, and three were not yet exited in 2005. We also note that the trend in the European data is more often toward using convertible securities, both in the data introduced in this chapter (Table 21.2 ) and in the Kaplan et al. (2007) data. These trends also are consistent with the use of greater downside protection after the crash of the Internet bubble on April 14, 2000, which caused a major downturn in the market. To control for the possibility of endogeneity of contracts vis-à-vis exits, Model 7 excludes the preplanned exits in the data. In the full sample of 223 investments, 70 of the investments (31.4%) indicate a degree of preplanned exit behavior at the time of investment (and do not necessarily indicate such plans to the investee); 55 of the 70 (79%) are exited investments in our sample as of 2005. This evidence on preplanning behavior suggests the possibility of endogeneity vis-à-vis contracts and exits, although the VCs indicate that the preplanned exit outcome was by no means certain at the time of investment. Among the preplanned exits, the investors indicate that their (preplanned) strategy turned out to have the desired result only 53% of the time. Regression Model 7 in Table 21.8 indicates that the results are robust to excluding preplanned exits. This finding is strongly consistent with alternative controls for endogeneity in Table 21.6 . In Models 8 and 9, we examine the subsample of only seed investments i.e., start-ups at the time of first VC investment, and seed and expansion investments, respectively. These are important to show that the results are not driven by investments that were close to exiting at the time of first investment. 6 Model 10 examines only the funds that provided all of their investments as of 2002, which means we had to reject the observations of 15 funds. This robustness check is important to show that the results are not affected by VCs withholding information on some of their investments, such as the poorer performing ones. All of the estimates consistently support the central propositions that relate contracts to exits. In other models (not reported but available on request), we consider dummy variables for funds, rather than countries. The results show that fund effects are not driving the results. However, we note that we could not use dummy variables for 12 funds because of collinearity problems. Further, in other models, which are available on request, we consider different definitions of certain variables to show that the results are not driven by the model specifications. These models use dummy variables for seed, expansion, and late stages. At the time of the first VC investment, there are 69 investments, 82 expansion investments, 32 late investments, and 40 buyouts in the data. The models also use a dummy variable for entrepreneurs with rankings of seven or more (as an alternative to the other models with the ranking variable on the one-to-ten scale) and a second VC dummy for both captive and non-captive VCs. (We suppress the captive bank VC dummy for reasons of collinearity). Because there are 28 exits of foreign investments and nine unexited foreign investments, we also include a variable for the exits in which the investment is initially in a foreign firm. Also, we consider the use of different contract terms, which we include as right-hand-side variables. When we use common equity as a proxy for weak VC control rights, we find that it is associated with a 12.3% greater chance of an IPO and a 30.1% smaller chance of an acquisition. These findings are consistent with the comparison tests in Table 21.3. In a similar specification, we find that drag-along rights are more important in effecting an acquisition than are redemption or antidilution rights. Drag-along rights are associated with a 15.8% reduction in the probability of an IPO and a 31.5% increase in the probability of an acquisition. We also considered other robustness checks, such as Heckman (1976, 1979) sample selection corrections to control for the non-random selection of an exit event versus an ongoing investment in the portfolio. 7 The results are robust. These checks are not reported here but are available on request.