Venture capital investor type and the growth mode of new technology-based firms

Note that limited partners have restricted access to deal-level information because any involvement in investment management activity might jeopardize their limited liability status (see, again, Sahlman 1990; Gompers and Lerner 1996).

Note, however, that this view is not unanimously shared in the growth literature. For instance, Davidsson et al. (2009) show empirically that growth in sales before profits is often quite erratic, and star companies are those that are able to combine growth and profitability (i.e., they enjoy profitable growth).

Internet firms best epitomize this situation: “… revenues are the lifeblood of these companies and their stocks because earnings often are nonexistent. Investors want growth at Internet firms. When they can’t look at earnings, they look for go–go revenue growth” (see The Wall Street Journal, Feb. 7, 2000). “This conventional wisdom is supported by empirical findings in Davis (2002), who indicates that revenues are value-relevant for Internet firms” (Bartov et al. 2002, p. 327).

For instance, Clarysse et al. (2011) provide evidence, based on case studies of high-tech start-ups, of exceptionally high sales growth combined with hardly any growth in employment in environments characterized by high stability (i.e., tight appropriability) and low complexity (i.e., easy access to complementary assets). Conversely, if the environment is stable and complex, NTBFs are inclined to enter the “market for ideas” rather than compete in the product market, and exhibit high growth in employment with almost no growth in sales.

The proposition that asset specificity is a fundamental driver of vertical integration was originally proposed by Williamson (1971).

Most empirical studies that have compared IVC and CVC investments have considered, as a measure of outcome, the likelihood of a successful exit (e.g., undergoing an IPO or being acquired), its proceeds (e.g., IPO valuation or acquisition premium), and firm performance (see, e.g., Gompers and Lerner 2000; Maula and Murray 2001; Gompers 2002; Bottazzi et al. 2008; Ivanov and Xie 2010; Chemmanur et al. 2010).

Out of the 550 NTBFs that responded to the questionnaire, we have sufficient information to estimate the econometric models illustrated in Sect. 5 for a sample of 538 firms. We further restricted the sample to 531 firms by eliminating firms invested in by captive investors other than CVC investors (e.g., bank-affiliated VC investors and government-owned VC investors) and investments syndicated by IVC and CVC investors. Both the analysis of a broader set of captive investors and the study of the effect of syndicated investments are extremely interesting research topics. However, the limited number of observations in our sample forced us to focus only on “pure” IVC and CVC investments. In this respect; note also that results obtained including syndicated deals in both the IVC- and CVC- backed groups are virtually identical to those reported in this work.

In several cases, telephone or face-to-face interviews were made with firms’ owner-managers to obtain missing information and ensure reliability of data, including those that refer to the identity of VC investors.

For instance, CVC investments represent a small share (~6%) of the 32,364 VC investments considered by Gompers and Lerner (2000). In the sample of 1,510 VC-backed IPO firms analyzed by Ivanov and Xie (2010), only 219 are backed by one or more CVC investors. Among the 11,556 VC investors considered by Chemmanur et al. (2010), there are 926 CVC investors. They were involved in only 7,180 of the 140,915 VC rounds analyzed in their study. In the sample of 750 VC investors considered by Bottazzi et al. (2008), CVC accounts for 8%; CVC investors were involved in 8.8% of the 1,643 investments under scrutiny.

Apparently, these latter results diverge from those presented in Table 1. The reason is to be traced to the ceteris paribus nature of the multivariate econometric analysis performed herein.

Note that these variables do not switch back to 0 when VC investors exit. The reason is that firms that obtained VC are inherently different from those that never did. For instance, VC financing is assumed to signal the quality of a firm to uninformed third parties, making it easier for the firm to obtain access to additional resources. This effect is likely to persist even after the exit of the VC investor.

Note that here we investigate the impact of IVC and CVC on the growth rate of NTBFs and not their impact on the level of the dependent variables (sales and employees). If, in principle, the two issues are different and result in different conclusions (see on this point the discussion in Vanacker et al. 2011), they largely coincide in our specific case remarkably because 60% of sample firms are observed from their inception. We thank an anonymous referee for having raised the point.

Unfortunately, data on the amount of VC financing received by firms are not available. These data are generally regarded as confidential by owner-managers and could not be obtained from public sources (financial accounts) because the instruments used by VC investors differ (e.g., convertible bonds, straight equity). We also do not know the timing of following rounds of VC financing. We acknowledge this as a limitation of the present study.

Including time dummies in the models results in very erratic estimates for the parameter of LnAge (possibly because of the multi-collinearity between time dummies and LnAge in first-differenced equations). However, estimates of other parameters, most notably those corresponding to DIVC, DCVC, DIVC2 and DCVC2, are largely unaffected by the inclusion of time dummies, such that our core results are robust to this different specification.

In this respect, note that we also tested alternative definitions (results are omitted here for brevity): the more the threshold defining the short-term effect is postponed, the more results converge to the “static” version of Eq. 1.

The actual number of observations included in the estimates is slightly lower due to our model specification, estimation strategy, and the presence of (few) missing data. As a matter of fact, most estimates are made on 3,045 observations for 531 firms in the employment equation and 2,998 observations for 526 firms in the sales equation.

If unobservables also influence the ability of firms to attract VC investors, a spurious correlation between VC investments and growth follows because of unobserved heterogeneity. An opposite bias is also possible if NTBFs with superior growth prospects self-select out of the VC market. In a thin VC market, finding a suitable offer from a VC investor might be difficult; with owner-manager time being the scarcest NTBF resource, the opportunity costs of search for a VC investor are clearly higher the better the prospects of the firm.

Note that we also check if the use of all available moment conditions and the consequent large number of instruments can result in significant finite-sample biases causing potential distortions as suggested by Bond (2002). To deal with this problem but also to consider possible measurement errors (see again Bond 2002), we re-estimated the model with a reduced instruments set using moment conditions in the interval between t – 3 (t – 2) and t – 6 (t − 5) for instruments in levels (differences). The results (available upon request to the authors) are very similar to those exposed in the Sect. 6.

The results obtained from the semi-parametric Cox survival model are exposed in the first column of Table 3.

The functional form of this added regressor is \( {{\uplambda}}_{\text{it}} = \varphi \left[ {\Upphi^{ - 1} \left( {F_{i} \left( t \right)} \right)} \right] \cdot \left( {1 - F_{i} \left( t \right)} \right)^{ - 1} \), where F _i (t) is the cumulative hazard function for VC backing of firm i at time t, φ is the standard normal density function, and Φ⁻¹ is the inverse of the standard normal distribution function.

The coefficient for the inverse Mills ratio control for sorting [IMR(VC)] in the WG estimates reported in column HSS is not significant in either the employment or sales equations. In contrast to the results for US firms, in previous studies (e.g., Sørensen 2007; Chemmanur et al. 2008), there is no evidence in our data that VC investors pick companies with the best future growth prospects. As a corollary, we are confident that selection of firms with future high-growth prospects is not the main reason for the positive relationship between IVC and CVC investments and firm growth highlighted by our estimates. For a more thorough discussion of this issue, see Colombo and Grilli (2010). For similar evidence on the absence of a positive sorting effect for European VC-backed firms, see Engel (2002) and Bottazzi et al. (2008).

We followed as closely as possible the above-mentioned test procedure given our data constraints. In particular, because we estimated a single probit equation for firm exit, the only major simplification with respect to Wooldridge’s original framework and its subsequent extension is the assumption that the impact of each determinant of firm exit does not vary over time. The same procedure has also been employed in Colombo et al. (2009) and Bertoni et al. (2011a).

The inverse Mills ratio in the Semykina and Wooldridge (2010) test is treated as an exogenous covariate (we thank Anastasia Semykina for personal communication confirming this as the correct procedure).

For simplicity, we only report standard GMM-SYS estimates [corresponding to GMM-SYS (1) in Table 5]. The results of the estimates obtained using the other estimators are close to those presented here and are available from the authors upon request.

Because the ratio is extremely leptokurtic, we winsorized the variable at the 1% threshold. Similar results are obtained using 2 and 5% thresholds.