Bank Market Power and Access to Credit: Bank-Firm Level Evidence From the Euro Area

In 2017 SMEs contributed to 56.8% of value added and employed 66.6% of the work force in the EU (European Commission 2017).

Orbis Bank Focus (previously Bankscope) contains information on over 40,000 public and private banks around the world, while Amadeus contains financial information on over 24 million public and private European companies. Both databases are compiled by Bureau Van Dijk, a Moody’s Analytics company.

Amadeus Banker provides reference banks’ names only for a limited number of Euro Area: Austria, Cyprus, France, Germany, Ireland, Latvia, Malta, Netherlands, Portugal, Slovenia and Spain.

To do so we perform a fuzzy merge using bank names and country locations as reported in each database. We use the Stata ado file reclink2 written by Micheal Blasnik which uses a bigram string comparator to calculate the fraction of consecutive character matches between two string variables. To ensure accuracy, we also perform a clerical review of all matches.

In terms of coverage, banks in our sample manage around 720 billion euros as of December 2014, corresponding to 3.7% of total consolidated assets reported by Main Financial Institutions for the countries we consider. Similarly, firms in our sample employed about 437.000 workers, i.e. 0.45% of total employment in these countries (see Table 8).

Admittedly, the estimation of the cost function still requires specifying the geographical scope of the market (Beck et al. 2013). Throughout the analysis we use a Lerner index based on a cost function estimated on a country by country basis, and in Section 4.1 we show that using a Lerner index based on a cost function estimated at the Euro Area level does not alter our results.

The close unconditional correlation between the Lerner index and the H-statistic may call into question our decision to focus on the lerner index as our preferred measure. However, in Table 10 we show that the correlation between these two measures declines substantially once we control for observable bank characteristics, suggesting that the two measures capture slightly different dimensions of banks’ market power.

We also assume that the market power of firms that supply trade credit is industry-specific and hence netted out by industry fixed effects. This assumption reflects the fact that discount terms in trade credit contracts are typically set at the industry level (Petersen and Rajan 1994). Recent research has cast doubts on the reliability of trade credit as proxy for financing constraint by suggesting that trade credit sends a positive signal to banks on the creditworthiness of potential borrowers: rather than a substitute, trade credit would then be a complement to bank lending (Giannetti et al. 2011; Agostino and Trivieri 2014). However, since such signal should be especially valuable for relatively uninformed banks, we regard this specific concern to be less relevant in our analysis since we focus on banks that are already in a lending relationship and therefore unlikely to be uninformed about their corporate borrowers.

Variables are winsorized at 1% and 99% levels to minimize the influence of outliers. The complete list of variables is reported in Table 7 alongside descriptive statistics and data sources.

By contrast, we cannot include firm-year fixed effects to completely absorb firm-level credit demand as in Khwaja and Mian (2008). The reasons is that while we have a small sample of multi-bank firms, our credit variables refer to the total amount of credit received in a given year from all its lenders. As a result, including firm-year fixed effects is unfeasible since this leaves no residual variation in the dependent variable. Analogously, we cannot include bank-year fixed effects, as this effectively compares different firms borrowing from the same bank in the same year: in this case, we have no variation in the dependent variable. The inclusion of time-invariant firm or bank fixed effects is also problematic. First, the inclusion of firm fixed effects exploits within-firm variation to estimate coefficients. While the dependent variables (e.g. firm’s credit outstanding) exhibit substantial time-variation, the dependent variable (the lerner index) is a slow moving variable with very limited time-variation. Specifically, the within variation in the lerner index (standard deviation σ = 0.068) is less than half its between variation (σ = 0.139). Similarly, because of the preponderance of single-bank relationships, including a bank-fixed effect implies too relying on within-bank variation in the lerner index to estimate the coefficient. Results reported in Table 9 confirm this view, with almost all estimates being statistically insignificant at conventional levels when either firm or bank fixed effects are included to our baseline specification. This finding suggests that the variation used to identify the effect of bank market power on credit conditions stems entirely from cross-sectional differences between firms related to banks with different degree of market power, as opposed to deriving from changes in bank market power over time.

We cannot simultaneously include all three variables because of collinearity.

Results are qualitatively unchanged if we take the European commission definition of small and medium-sized enterprises and use a dummy for SMEs that takes values 1 if a firm has (i) 250 or less employees or (ii) operating revenues up to 50 million euros and total assets up to 43 million Euros and 0 otherwise, as in Ryan et al. (2014).

We are grateful to an anonymous referee for suggesting us to investigate this additional hypothesis.

To minimize the incidence of outliers, all variables are winsorized at 1%.