Venture capital financing and the financial distress risk of portfolio firms: How independent and bank-affiliated investors differ

This literature has investigated the synergies, the risks, and the conflicts of interest when lending and the underwriting of public securities are combined (see Puri 1996 and Yasuda 2005 for a review). These papers generally show that when commercial banks make loans to firms and also underwrite securities, the pricing of the underwritten securities is higher than investment-house-underwritten securities, thus confirming the certification role of banks.

See Andrieu (2013) for a recent literature review on BVC.

Elston and Audretsch (2011) suggested that the rigidities and the constraints in providing liquidity that characterize the German finance system engender a lack of entrepreneurial activity in high-tech industries. Similarly, Rogers (2012) showed that the market structure in the banking sector affects the creation of new firms at the state level in the USA. States with higher competition on the banking market show less entrepreneurial activity, while a higher number of bank branches per capita and of small banks positively affect entrepreneurship.

See, e.g., Revest and Sapio (2012) for a review of the peculiarities of the European venture capital sector and of VC effects on the growth and the visibility of the funded companies.

For information on the specific skills employed by bankers to grant loans, see Best and Zhang (1993).

According to Gompers (1996), the reason why some private equity firms have incentives to “grandstand” by taking their invested firms public sooner than that which would otherwise be optimal for the invested firm is that they are seeking to build a reputation (a track record of successful exits) to raise capital from external investors. This phenomenon is mostly relevant for IVCs.

A minimum 6-year window was considered from the investment year to observe the effect of VC investment (until the end of 2010, when available). As a consequence, it was necessary to exclude the financing rounds after 2004.

Since accounting data were not available for the German firms, they were excluded from the sample used in the study.

The VICO dataset has been built thanks to the joint effort of nine universities throughout Europe (École des Mines de Paris, Politecnico di Milano, Libera Università Carlo Cattaneo, Research Institute of the Finnish Economy, Centre for European Economic Research (ZEW), Universidad Complutense de Madrid, University College London, Vlerick Leuven Management School, and the University of Gent) with the support of the 7th European Framework Program. For more details on the procedures used in the data gathering process and on all the variables included in the dataset, see Bertoni and Martí (2011).

The data on VC funds did not include any information on contracts or covenants (a limitation that is common to most of the literature on VC). IVC and BVC investors have different backgrounds in terms of competences, monitoring styles, risk aversion, and investment behavior. Consequently, invested firms in BVC portfolios have different risk profiles (e.g., ex ante default probability and ex post moral hazard) compared with IVC-backed firms. All else being equal, these differences are likely to affect the type of contracts designed by BVC and IVC investors. Differences in the type of contracts between IVCs and BVCs could potentially explain a different impact of VC investors on invested firms.

The matching procedure ensures that the treated units and selected controls are comparable in terms of some covariates X. A matched sample is considered good if the distance between the values of the covariates X for the matched units is minimized. Thus, the selection of a matching method involves the definition of the "distance" between two covariate vectors, x _l and x ₂ (i.e., the values of the covariates X for treated and control units, respectively). Given the sample covariance matrix of the matching variables X from the full set of units, say S, and the distance between covariates x ₁ and x ₂, say (x ₁ − x ₂), the Mahalanobis distance is defined as: M(x ₁, x ₂) = (x ₁ − x ₂)^T S ⁻¹(x ₁ − x ₂). A matching algorithm assigns controls to the treated units using that distance. The treated units are randomly ordered, and the first is paired with the nearest of the m controls, the second is paired with the nearest of the remaining m − 1 controls, and so on (Rubin 1973). If more than one control is needed for each treated unit, once each treated unit has one control, and the first treated unit is assigned the nearest of the remaining m − n available controls.

As a robustness check, we also performed alternative matching procedures based on the same propensity scores, including one-to-one matching and caliper matching within a maximum distance of 25 % propensity score standard deviation of the controls to BVC- and IVC-backed firms (Caliendo and Kopeinig 2008). However, no significant improvement in the matching efficiency or different results were obtained.

In terms of age, we find a difference (albeit only weakly significant) between BVC-backed firms and IVC-backed firms (χ2[3] = 6.86, significant at the 10 % confidence level). BVC-backed firms are somewhat younger than the matched IVC-backed firms, with the foundation dates being less concentrated in the years 1990–1994. In the same light, we find a small and weekly significant difference between IVC-backed firms and the matched IVC control group (χ2[3] = 8.16, significant at the 5 % confidence level), indicating that IVC-backed firms are younger than the matched non-IVC-backed firms, with the foundation dates being more concentrated in the years 1984–1989. In order to control for these differences, we included firm’s age as a control in the model specification. In terms of country of establishment, we find a difference between BVC-backed firms and the matched IVC control group (χ2[5] = 18.96, significant at the 1 % confidence level).

The Centre for Business Research at the University of Cambridge has constructed a creditor protection index of 10 variables (that measures the levels of the restrictions on debtors’ activity, creditors’ contract rights, and creditors’ rights in corporate bankruptcy proceedings). This index outlines the differences that exist in Europe in banking legal rules. These differences claim for the homogeneous distribution of BVC-backed firms and control group firms in the sample in order to exclude that external factors may bias our results.

We acknowledge the two authors for having provided us with the model.

Earlier empirical analysis of corporate bankruptcy prediction based on financial ratios dates back to Beaver’s univariate model (1966). Since then, a plethora of multivariate methods have been developed by researchers (see Altman and Saunders 1998 for a review): discriminant analysis, linear probability models, logit and probit regression analysis, or, more recently, recursive partitioning algorithms, multicriteria decision aid methods, expert systems, and neural networks. These models, which have been validated on training samples, are widely accepted because of their relatively high discriminatory power (see Ughetto and Vezzulli 2011 for a classification of these studies).

Out of the 979 firms that compose the matched sample, only 47 are listed (up to 2010), 16 of which are BVC-backed firms.

Given the longitudinal nature of the dataset, the matching procedure matched a BVC-backed (IVC-backed) firm in the year of the investment with a BVC (IVC) control group (non-invested) firm in a precise year of its life, let us say t. The post-investment period for the matched control group firms starts in the year following t, i.e., t + 1, until the end of the observation period.