1. Introduction
The volume of credit granted by the banking system to Portuguese companies has positively evolved over these last few years (
Banco de Portugal 2022). Concurrently, small- and medium-sized enterprises (SMEs) have played a pivotal role in the Portuguese economy, representing approximately 99.9% of all businesses, generating 63.4% of the wealth produced, and accounting for 77.4% of employment in 2019 (
PORDATA 2021).
In this context and considering that bank financing constitutes the main external funding source for SMEs (
Meslier et al. 2020), assessing how bank concentration and the relative power of banks may condition the operations of Portuguese companies is of utmost importance, namely in terms of firm profitability, cost of debt, and capital structure. Banking markets in Germany, the Netherlands, and Portugal are overall more competitive compared to other EU nations, as observed by
Wang et al. (
2020). Despite the numerous studies conducted, an integrative theory capable of comprehensively explaining the intricate field of corporate finance is yet to be formulated.
Building on the hypothesis of perfect capital markets, numerous works argue that financial decisions exert a major impact on business performance (
Abdullah and Tursoy 2021). However, contemporary studies recognize the existence of several market imperfections, such as bank concentration, the market power of banks, information asymmetry, and conflicts of interest inherent to corporate finance.
Therefore, the analysis should adopt a collaborative and interconnected approach. The predominance of studies focusing on large companies and markets highlights the importance of delving deeper into how these aforementioned issues specifically impact smaller companies.
Much of the current research tends to focus on macroeconomic trends, often overlooking the microeconomic dynamics that can significantly influence firm-level outcomes. However, further studies are required to further understand this relationship.
The research also addresses the need for size specificity in the current literature, particularly in a small market like Portugal. Understanding that the impact of bank power concentration may vary depending on the size of the companies is crucial for small countries. By including companies of different sizes operating within small geographical contexts, this study aims to uncover nuanced patterns and provide valuable insights to be used for both academic and practical applications in corporate finance. Overall, the research is driven by the need to address these critical gaps and provide a more robust understanding of the implications of bank power concentration on firm-level financial dynamics.
The study makes a timely and relevant contribution to the existing literature, by providing empirical evidence on the significance of bank concentration and the relative power of banks—Bank-Related Power—, for each of the different companies, particularly in terms of firm profitability, cost of debt, and capital structure. The work contributes, therefore, to a better understanding of the impact of the banking system on small business activity. Banking relationships affect value creation and provide both current and potential shareholders with analytical tools that enable them to shape future relationships. Considering the sample and the limited number of studies focusing on the Portuguese reality, our research provides results and significant information on the critical role played by banking relationships in company performance. Additionally, the time frame considered allowed us to assess the impact of the subprime crisis (2008–2009) and the sovereign debt crisis (2010–2013) on the dimensions addressed. The innovative nature of this study stems from the introduction of a new variable (Bank-Related Power), its substantial sample size (434.990 companies), and the extended time frame addressed (2006–2018).
Certain limitations regarding access to the database provided by Banco de Portugal emerged during this research, mainly due to the imperative to comply with the General Data Protection Regulation.
The analysis of the results, sorted by business size and year, was conducted using a fixed-effects regression model and the ordinary least squares model (pooled OLS) (
Hedges and Vevea 1998;
Bell et al. 2019). The fixed-effects model with Driscoll–Kraay standard errors is employed when evidence suggests that one of the assumptions of the fixed-effects estimator needs to be addressed (
Topcu and Gulal 2020).
In addition to this introductory chapter, the body of this work comprises five chapters. The second chapter is dedicated to contextualizing the problem under study, identifying the key concepts, and reviewing the studies conducted on this topic. The third chapter encompasses sample selection and the variables under investigation. The fourth covers the methodology used, as well as the resulting outcomes. Finally, the fifth chapter presents the main conclusions, outlines the study’s limitations, and suggests avenues for future research.
3. Data and Methods
The Banco de Portugal cooperated in the collection of information, granting restricted access to two databases, but ensuring anonymity regarding the identification of companies: (i) Credit Responsibilities Center (CRC), which contains information on credit offers from banking institutions operating in Portugal, aggregated by companies, and (ii) Harmonized Panel of the Central Balance Sheet, which encompasses financial information in panel data from the Central Balance Sheet database and harmonized variables (
Bank of Portugal Microdata Research Laboratory (BPLIM) 2019).
The analysis included information from 434,990 Portuguese companies that established relationships with financial institutions operating in Portugal between 2006 and 2018. To ensure information for the entire period under study, a sample comprising 2,669,785 observations was obtained. Over the 13-year time frame under study, two crises unfolded: the subprime crisis (2008 and 2009) and the sovereign debt crisis (2010 to 2013) (
Rua 2017).
Return on assets (ROA), defined as the ratio between operating profits and total assets (
Serrasqueiro and Nunes 2012;
Gun 2020); interest rate (IR), expressed by the quotient between financial expenses and total debt (
Banco de Portugal 2019); and debt-to-assets ratio (DAR), which evaluates the percentage of total assets acquired using borrowed capital (
Schjelderup 2016;
Gupta 2021) were the dependent variables considered.
To assess bank market power, two independent variables were created: bank concentration and bank-related power.
Bank concentration was determined using the principal component analysis (PCA) method using three indicators that, according to
Table 1, exhibited a very high degree of correlation: (i) Max_relationship—greater bank relationship, which represents the percentage of the company’s largest loan with the bank; (ii) Nb_relationship—number of bank relationships, which indicates the number of banks with which the company establishes relationships and (iii) Hhi_relationship—concentration of bank relationships, which measures the concentration of bank relationships using the HHI index to assess the level of bank concentration, and highlights the dimension of the bank within the banking system (
Sulaiman et al. 2019). The PCA method was selected since it is a statistical approach that can be used to analyze data and also to prepare it before conducting subsequent statistical tests (
Kherif and Latypova 2020).
According to
Table 1, all variables are correlated. A negative relationship is observed between greater bank relationships and the number of banks with which the company establishes a relationship. This correlation is justified by the fact that an increase in the number of relationships entails a decrease in credit availability for each single bank. The HHI index shows positive correlations with both greater banking relationships and the number of banks involved.
The bank-related power variable is obtained from Equation (1). This novel variable was developed and introduced in this study and its primary objective is to make a contribution to the existing literature by enabling the assessment of the power wielded by a particular bank compared to the others with which the company has banking relationships. It is constructed from the average bank share, represented by Zm, the average share of each bank in the company’s credit, and Max_relationship, and is expressed as follows:
Equation (1)—Expression of the “Bank-Related Power” variable
Table 2 displays the descriptive statistics of the variables. To ensure security, data protection, and customers’ confidentiality, Banco de Portugal does not provide the minimum and maximum values of the variables, and hence these statistics are not presented.
The table shows that 90% of the observations indicate an interest rate equal to or lower than 14.36%. As for profitability, the values suggest that in 90% of the observations, the ROA is less than 17.61%. Finally, 90% of the observations recorded a debt-to-assets ratio of less than 66.17%. As for bank concentration, 90% of the observations exhibit values lower than or equal to 1.288. The median for all variables stands at values considered reasonable (2.3% for ROA, 4% for the interest rate, and 17% for debt-to-assets ratio.
The average for the bank-related power variable is 9.29, reflecting a 9-fold difference between the power held by the bank with the lowest share and the power of the bank with the highest share. The 90th percentile indicates that 90% of the observations related to the bank-related power variable are less than or equal to 19, which suggests a difference of up to 19 times between the bank with the highest share and the bank with the lowest share. These results suggest the existence of a high concentration of credit.
According to the measures of dispersion used (mean and standard deviation), some of the companies’ datasets analyzed can be considered outliers. They are, nonetheless, regarded as true and real outliers, since the database from which the information was collected was provided by a governmental entity. But considering that we are dealing with more than 2.5 million observations, the extreme values displayed by a few companies do not jeopardize the statistical inference. That is why we used the 1st and 2nd, median, 90th, and 99th percentiles to ensure that the values displayed by most companies can be considered normal, and that there are only a few outliers. However, those outliers cannot be excluded because they are real data and real companies. Furthermore, some authors argue that real outliers should be studied as they are often found to be interesting and influential (
Aguinis et al. 2013).
Gallup (
2020) states that, since panel datasets are typically large, it is quite common to have a few distant outliers, which do not significantly affect the estimates. The author further notes that the researchers must ensure that the undisplayed observations are not significant to the estimated relationship.
According to
Sullivan et al. (
2021), the median resists extreme values, while the mean is sensitive to them. However, the mean can be modified by only a single extreme value, regardless of the sample size. Likewise, the standard deviation and variance are sensitive to extreme values; nevertheless, the interquartile range and the median are more robust to outliers.
The Driscoll–Kraay method estimates the covariance matrix of the errors in a way that accounts for potential heteroscedasticity and serial correlation.
Heteroscedasticity and serial correlation can be linked to outliers, especially in time-series data where observations may exhibit temporal dependencies. Serial correlation implies that there is a systematic relationship between the current observation and past observations. Outliers in time-series data may manifest as unexpected spikes or deviations from the typical pattern, potentially leading to serial correlation. By allowing for temporal correlation in the errors, the Driscoll–Kraay standard errors aim to provide more efficient estimates in the presence of serial correlation.
While not explicitly designed to handle outliers, the efficiency gains from addressing serial correlation may help produce more reliable estimates when outliers are present in time-series data.
To complement the analysis, control variables, such as the age of the company (maturity), and its size (micro-, small-, medium-, and large-sized) were considered—following the recommendations of the European Commission issued on 6 May 2003 (2003/361/CE). Total assets (global investment) were also considered. The transformation of total assets using the logarithm serves to minimize significant differences and smoothen out extremes.
Data analysis was conducted using statistical inference techniques, such as static panel data regression, fixed-effects regression model, random-effects regression model, and pooled OLS regression, as well as regression with Driscoll–Kraay standard errors. These techniques were employed to address the research questions under investigation. The use of multiple regression is grounded in certain assumptions: (i) the error term has a population mean of 0; (ii) the error term has a constant variance, measured by homoscedasticity; (iii) the error terms are uncorrelated with each other, in other words, there are no correlations between errors and explanatory variables (absence of multicollinearity); and (iv) the error term is normally distributed (
Pesaran 2015). Data treatment was carried out with STATA (statistic data analysis). Multiple linear regression with static panel data was the model used. Panel data provide more information, more data variability, less collinearity among variables, more degrees of freedom, and more efficiency (
Wooldridge 2003).
The Hausman test was employed to determine whether a fixed-effects model or a random-effects model is appropriate to analyze the sample. If we cannot reject the null hypothesis (
H0) using the Hausman test, it means that the random-effects model is more appropriate; if the null hypothesis is rejected, the fixed-effects model should be preferred (
Rabe-Hesketh and Skrondal 2008). When the Hausman test does not reject the null hypothesis, providing evidence of the heterogeneity/variability between individuals, the Breusch–Pagan test is used to determine which of the models (random-effects vs pooled OLS) is more suitable for panel data analysis. It also allows the presence of heteroscedasticity, based on the assumption that the error terms follow a normal distribution (
Zivot and Wang 2013).
Table 3 provides a summary of the test results and identifies the most appropriate model to be used in the construction of the regression model for the variables under study. The fixed-effects model will be used for most variables, since
Prob > chi2, with a value close to zero, for a significance level of α = 5%, except for situations marked with an asterisk (Breusch–Pagan test was applied in those cases).
The table includes an “annual review” section where the impact of independent variables on dependent variables across different years is assessed. In addition to the dimensions identified by Banco de Portugal—micro-, small-, medium-, and large-sized companies—two new dimensions were created: (i) micro and small companies (encompassing micro- and small-sized companies) and (ii) SMEs (comprising micro-, small-, and medium-sized enterprises).
The regression model estimation for the fixed-effects specifications is provided by expression (1):
where,
represents bank concentration,
bank-related power and
control variables (age and total assets). Firm size was used as a grouping variable. Age and total assets were not displayed in this study as our main focus was to examine the impact of bank concentration power on each of the companies. Under these circumstances,
represents the unobserved, time-invariant individual characteristics that may affect
, that is, individual heterogeneity is captured by
. The term
represents the error term.
represents each dependent variable: ROA, financing costs, and debt-to-assets ratio.
Estimation with robust standard errors was used since we were unable to achieve heteroscedasticity correction and determine the existence of autocorrelation of errors. In this context, the cluster-robust standard errors model should also be used, according to
Arellano (
1987). To verify whether the formulated assumptions were met, regression with Driscoll–Kraay standard errors was employed for fixed-effects estimation. The error structure of this method is robust to heteroscedasticity, correlation, and cross-sectional dependence (
Topcu and Gulal 2020). In situations where the OLS pooled regression model proves more appropriate for panel data analysis, the specification closely resembles that of the fixed-effects model, according to expression (2). To control the correlation of error
, over time, the coefficients are presumed to be constant across all units and periods, with no expected heterogeneity between units (
Bell et al. 2019).
Figure 1 summarizes the research methodology employed.
5. Conclusions
This research aims to assess the influence of bank concentration and the relative power of banks on firm performance, financing costs, and capital structure. This study covers an extended time frame (spanning the years from 2006 to 2018) characterized by two periods of crisis, the subprime crisis (2008 and 2009) and the sovereign debt crisis (from 2010 to 2013) that conditioned the activity of Portuguese companies.
The data sample, consisting of 2,669,785 observations, was collected from Banco de Portugal and analyzed using statistical inference techniques, such as static panel data methods (fixed-effects model), the pooled OLS model, with robust standard errors, and regression with Driscoll–Kraay standard errors.
The results suggest a positive impact of bank concentration on the business performance of micro-enterprise and SMEs. Since micro-enterprise and SMEs are more informationally opaque and entail greater credit risk, banks operating in concentrated markets have more incentives to invest in collecting customer-specific information. This endeavor will ensure greater assistance, supervision, and control, ultimately contributing to the improvement of business performance.
On the other hand, since companies tend to establish privileged relationships with a limited number of banks, bank concentration does not seem to represent a significant constraint on firm performance for the remaining companies. However, during the years of the financial crisis, it was noted that one of the primary objectives of companies and banks wielding greater relative market power was to maintain relationships to minimize potential negative impacts and enhance business performance, which happened in the post-crisis period (after 2014).
The analysis also reveals that bank concentration increases the debt burden borne by companies and that bank-related power has been statistically insignificant during the period under analysis. Irrespective of the relative influence that banks hold in the financing of Portuguese companies, these still have to face high costs. It is widely accepted that, in bank-concentrated markets, companies often depend on one or two banks, which allows banking institutions to charge higher interest rates. This trend extends into the period of the sovereign debt crisis (2010–2013) and the subsequent years.
Finally, Portuguese companies operating in bank-concentrated markets tend to improve their capital structure as indebtedness decreases (bank concentration variable). Micro-enterprise and SMEs subjected to greater funding restrictions are highly informationally opaque and granted limited guarantees. This often forces them to resort to self-financing and only turn to bank debt as a last resort. Consequently, banks with greater banking power find it more difficult to adapt and offer flexible loans to companies, as SMEs rely less on bank financing, thereby conditioning bank-company relationships. In contrast, when considering bank-related power, large companies have access to higher financing capacity. Perhaps, those with high self-financing capacity are free from the constraints faced by SMEs when they resort to the banking system. When a company establishes relationships with a creditor wielding greater relative power, this relationship can influence its capital structure. These companies maintain their loyalty to a bank and may, consequently, have to bear higher costs due to informational monopolies and increased difficulties in switching banks.
This research aims to contribute to the existing literature by providing empirical evidence of the impact of bank concentration and of the relative power of banks—Bank-Related Power, a new variable introduced in this study—on business performance, financing costs, and debt structure. This sort of information is, unequivocally, of utmost relevance to all stakeholders. On the other hand, given the scarcity of studies focusing on the Portuguese reality, our investigation makes an additional contribution: it provides significant results and relevant information about the pivotal role played by bank relationships in shaping business activity.
5.1. Theoretical Implications
This study offers several theoretical contributions to the existing literature. Firstly, we use a wide range of firm characteristics and introduce a new variable, bank-related power, to assess bank relationships. Secondly, the study contributes to a better understanding of the importance of bank relationships in the performance of smaller companies. Finally, we examine how the subprime crises (2008–2009) and the sovereign crises (2010–2013) influenced banking relationships and business activity.
5.2. Practical Implications
This study makes various contributions to different stakeholders. It matters to Portuguese companies, because it analyses the impact of certain characteristics of the banking system (bank concentration and relative power of banks) on business performance (profitability, cost of debt, and capital structure). It enables managers to improve business policies and understand the impact that their decisions have on business performance. It helps business leaders clarify the importance of the relationship between bank market power and the different dimensions of business activity. It can also be of great significance for both current and potential shareholders, as banking relationships enhance corporate value creation. The financial difficulties faced by certain companies are deeply influenced by the decisions and actions of political decision-makers, business partners, suppliers, customers, and public authorities. Reducing information barriers between SMEs and lenders is another vital area that demands the undivided attention of policymakers, especially if they genuinely want to assist financial institutions in achieving a more effective allocation of their funds. A more complete/deeper understanding of the current context can help improve future circumstances.
Secondly, to the best of our knowledge, this study stands as a pioneer in introducing a novel indicator used for assessing banking relationships in the context of Portuguese companies.
Finally, through this work, scholars, researchers, and professionals will gain a more comprehensive understanding of the current state of the art of a subject that addresses the context of a small economy. Drawing inspiration from this study, they will be capable of introducing new and innovative approaches and contributing to the evolution of an increasingly relevant topic in corporate finance.
5.3. Limitations and Avenues for Future Research
The primary limitation of this study stems from the difficulty in accessing corporate information from the Banco de Portugal, due to the General Data Protection Regime and the confidentiality terms in force. These constraints have also limited the development of a more exhaustive analysis of the data collected.
This study focuses on a period marked by financial instability, as it spans some of the biggest financial crises that ever occurred. Future research studies should consider a similar analysis but expand its scope to include other geographic and temporal contexts, and use other indicators and variables, such as credit contract conditions. Furthermore, future studies should use dynamic panel data models to complement the findings presented in this study.