«Tobias Schmidt (Deutsche Bundesbank) Wolfgang Sofka (Department of Organisation and Strategy, CentER, CIR, Tilburg University) Discussion Paper ...»
Organizations can gain legitimacy and hence access to resources through external validation in establishing institutional linkages with established institutions or succeeding in contests (Baum & Oliver, 1991; Rao, 1994). We focus predominantly on a firm’s ability to attract external funds for R&D or, more precisely, government R&D subsidies and venture capital investment. Venture capital investors are known to be highly selective in their investment decisions (Eckhardt et al., 2006). They monitor firms intensely, conclude growth/performance-oriented contracts, facilitate crucial personnel decisions and provide additional services (e.g. access to strategic alliances) (e.g. Gompers & Lerner, 2001b). As a result, the chances that the firm will be successful in the future and generate positive returns should increase. On the basis of this signal, banks should therefore be able to provide additional funds for the firm and its R&D investment. Similarly, many governments provide R&D grants for firms to stimulate R&D investment. Information requirements in applications are extensive and competition for grants is intense (Czarnitzki & Toole, 2007). Successful grant awards are highly selective and can signal the exceptional value of an R&D project (Aerts & Schmidt, 2008; Kleer, 2010). Banks may therefore rely on this external assessment
for overcoming information asymmetries. We propose:
Hypothesis 5: A firm’s R&D investment increases with the degree of market share of its main bank’s corporate client portfolio in its industry and this effect is reinforced by the patent stock of the focal firm, the presence of a venture capital investor or a government R&D subsidy, i.e. there is a positive moderating effect.
Hypothesis 6: A firm’s R&D investment decreases with the degree of specialization of its main bank’s corporate client portfolio in its industry, but this effect is mitigated by the firm’s patent stock or if the focal firm has attracted a venture capital investor or received a government R&D subsidy, i.e. there is a positive moderating effect.
4.1 Data We construct a unique panel dataset for testing the theoretical predictions. Data requirements are extensive because comprehensive information is required for banks and their client portfolio across multiple industries. What is more, the bank information needs to be linked to firm R&D investment. We achieve this by linking multiple databases in Germany.
The crucial starting point is the Mannheim Enterprise Panel (MUP).
This is a firm-level database collected by Creditreform, the leading credit rating agency in Germany. Since 1999, ZEW has been receiving a full copy of Creditreform’s data-warehouse of firm level data and constructing the panel twice a year. Creditreform collects its data on the basis of regional firm registries. The Creditreform data cover nearly the entire population of 3 million German firms with a few exceptions that are not legally required to register with the authorities (e.g. farmers). The Creditreform data are also the German input for the widely used AMADEUS database. Creditreform provides credit information and insurance services based on its data. Hence, the Creditreform data covers information that allows an assessment of a firm's credit worthiness. Most importantly for our study, it contains firms’ bank relations, including the bank that firms consider as their main bank. Data quality can be considered to be high since keeping information on financial solvency and relationships up to dare is a core part of Creditreform’s business model of and firms are not overly concerned about revealing their bank relationships (similar information can be found on a typical invoice). Given the population character of the database, we can calculate the industry composition of each bank’s client portfolio. The bank information is very precise based on the German eight-digit bank code, which allows a precise identification of the banks’ location and type (e.g. private bank vs. savings bank). Based on this information, we can track 2,432 banks. The banking code is mandatory for banks in Germany for obtaining a banking license. Coverage is therefore not limited. It should be acknowledged that the database does not contain information on the extent of each bank’s lending engagement with individual firms. To the best of our knowledge, no such database is publicly available or accessible.
We link this dataset to the “Mannheim Innovation Panel” (MIP) which provides information on firm R&D investment; the dependent variable of our analyses. The dataset is drawn as a representative, stratified random sample based on the German MUP firm population. In contrast to other studies analyzing bank-based financing for innovation we can therefore form perfect matches between the two databases, i.e. we do not have to rely on regional banking indicators (e.g. Benfratello et al., 2008) or statistical matching. The MIP survey is conducted annually by the Centre for European Economic Research (ZEW Mannheim) on behalf of the German Federal Ministry of Education and Research. For our study we use data from the 2002-07 surveys and analyze 7,294 firm observations from 4,363 firms. The panel is unbalanced.
The MIP survey targets R&D decision makers. These can be heads of R&D departments, innovation managers or CEOs, which is most likely the case in smaller firms where no elaborate functional structures exist. Several mechanisms are in place to secure the quality of the survey and its results. All core constructs in the survey follow the OECD’s “Oslo Manual” on measuring innovation inputs, outputs and processes (OECD, 2005). Furthermore, the MIP is the German contribution to the Community Innovation Survey (CIS) of the European Union. CIS methodology and questionnaires have been refined over the years in international application. They are subject to extensive pre-testing and piloting in various countries, industries and firms with regard to interpretability, reliability and validity (Laursen & Salter, 2006). This multinational application of CIS surveys guarantees quality management and assurance. CIS data have been the basis for several recent publications in highly ranked management journals (e.g. Laursen & Salter, 2006; Leiponen & Helfat, 2010).
The merged dataset contains precise identifiers for the European Patent Office statistics as well as the Bureau van Dijk ZEPHYR since it is also the basis for the AMADEUS database.
The former linkage allows us to obtain the number of patents granted to each firm, the second one tracks venture capital investments. The final dataset contains 7,294 observations from 4,363 firms between 2002 and 2007 encompassing firm, innovation and R&D characteristics, bank information, patent activity and venture capital investments.
4.2 Empirical model 4.2.1 Dependent variable Our dependent variable is R&D investment measured as the share of R&D expenditures on total sales. This R&D intensity measure is frequently used to take into account size effects (e.g. Cohen & Levinthal, 1989).
4.2.2 Independent variables We calculate for each firm the composition of the client portfolio of its main bank in its industry. We define the industry for this purpose at the two-digit NACE level for an aggregation that is neither too coarse nor too narrow, e.g. NACE34 covers firms in the automotive sector including OEMs and suppliers but not other transportation equipment. On the basis of this aggregation, a bank portfolio can be described along 84 different industry dimensions. We calculate the market share of a firm’s main bank in an industry as the number of firms in the focal firm’s industry in its client portfolio divided by all firms in this industry in Germany. This follows the basic rationale that all firms in an industry are the total pool of information from which a bank can potentially draw. We measure specialization as the percentage of firms in the focal firm’s industry of the total portfolio of a firm’s main bank portfolio. To account for differences in firm size we include the number of employees as weights in the calculations of the shares. This should be considered as a proxy in the absence of detailed bank lending information for all firms. Individual lending information for each bank and its clients would be preferable. However, after consulting with industry experts, we conclude that banks are highly unlikely to divulge such information owing to concerns about confidentiality and competition. The coefficients of the main bank market share and specialization variables test hypotheses 1 and 2 respectively. We test hypotheses 3 through a multiplicative interaction of both variables, i.e. whether the specialization of a bank’s client portfolio in an industry where it has a large market share has an additional effect.
For the test of hypothesis 4 we construct a variable at the industry level capturing the use of scientific knowledge. Cohen et al. (2002) construct a similar variable based on the Carnegie Mellon innovation survey. We follow their approach and access the equivalent survey for Germany conducted in 2001, one year ahead of our observation period. The particular survey question asks heads of R&D departments about external knowledge sources that were important for their innovation activities during the preceding three years. Question layout and design are well established in strategic management research on firm’s knowledge search (e.g.
Laursen & Salter, 2006; Leiponen & Helfat, 2010). The survey is representative of the German firm population and we obtain projected values for how many firms in an industry used knowledge from universities and/or research institutes. We include the variable as a main effect in the model and split the sample into two sub-samples with above/below average use of scientific knowledge in the industry. Hypothesis 4 would be supported if the effects from a main bank’s market share and specialization are stronger in the above average subsample.
Hypotheses 4 and 6 suggest effects from patenting, government R&D subsidies and venture capital investment. We add the EPO patent stock for each firm to the model (scaled per employee) and a dummy variable for the presence of a venture capital investor. The MIP survey provides information on whether the firm has received a government R&D subsidy at the European, federal or regional level. We include a dummy variable indicating whether this is the case or not.
Our theoretical model predicts several relationships based on the “ceteris paribus” assumption. We add several control variables to ensure unbiased results. At the bank level we control for the type of bank. The structure of the German banking system is often described as the “Three Pillar System“ (see e.g. Krahnen & Schmidt, 2004; Engerer & Schrooten, 2004) and consists of municipality-owned savings banks (in German: “Sparkassen”), cooperative and private banks. All these banks are active as universal banks. However, there is evidence that private banks differ in their decision making from say government-owned banks (Sapienza, 2004; Porta, Lopez-de-Silanes & Shleifer, 2002). Hence, we add a dummy variable indicating whether the firm’s main bank is a savings or cooperative bank. Similarly, banks may differ in their size and regional orientation because investments have been found to be more likely in geographical proximity (Coval & Moskowitz, 1999; Grinblatt & Keloharju, 2001). We add control variables for the total client corporate portfolio of a firm’s main bank (number of firms weighted by employees in logs) as well the share of client firms located in the same agglomeration area.
Firms may work with multiple banks which can in turn alter their relationship with the main bank (Boot & Thakor, 2000; Ongena, Türmer-Alkan & Westernhagen, 2007; Elsas 2005).