Performance and Risk Profile of Pure-Play Internet Banks:
Evidence from the European Banking Industry
Name: Arjen Smeijers1
Student number: S2396750
Programme (course code): MSc Economics (EBM877A20), MSc Finance (EBM866B20)
Supervisor: dr. M.A. Lamers
Date: 19 January 2018
Abstract:
In recent years, traditional banks have invested heavily in digital financial services. Banks that adopted a click-and-mortar business strategy use a structure of operating branches alongside their digital channels, while pure-play internet banks have implemented a complete branchless strategy. The objective of this paper is to compare the financial performance and risk profile of these two types of banks by studying 91 banks in Austria, Belgium, Germany, Luxembourg, The Netherlands, and Poland over the period 2011-2016. I find that internet banks outperform click-and-mortar banks in terms of financial return, but do not exhibit significantly different risk profiles.
Keywords: Banks, Internet Banking, Performance, Risk JEL classification: G21, O33
1 A.J. Smeijers
2
1. Introduction:
The revolution of Information Technology holds the potential to fundamentally change the banking sector. In recent decades, Information Technology has functioned as a catalyst for generating operating efficiencies and improve customer convenience by offering online banking services. This technological progress has made digital financial services widely available to the general public through the use of personal computers, cell phones, and other mobile devices. However, at the same time, these technological innovations lead to increased pressure for banks from other financial institutions and nonbank competitors that are offering improved payment services and lending platforms. Consequently, the banking industry is ripe for disruption and banks have to innovate in order to differentiate themselves from these outside rivals. This disruption will affect the products offered by banks and the way banks are servicing their customers.
In order to transform into successful financial institutions, banks have invested heavily in financial technology in recent years. During this transformation in the banking sector, some banks have adopted a click-and-mortar business strategy. These banks diversify their distribution channels by offering a network of physical locations alongside their online channels. An alternative strategy is the adoption of a complete online strategy without making use of operating branches. These banks adopt a pure-play internet banking model by providing their product portfolio exclusively over the internet. The strategic core of this pure-play internet banking model is realizing a reduction in overhead costs which may result in more consumer-friendly loan and deposit pricing. However, although this branchless business model is often cited as highly innovative (e.g. Guettler and Hackethal, 2005), the profitability of these pure-play internet banks is questionable (e.g. DeYoung, 2001, 2005; Delgado et al., 2007).
3 contribution to the current literature by analyzing the viability of this evolved online banking model.
Previously, only two studies analyzing the performance of pure-play internet banks use a multi-country sample (Delgado et al., 2007; Arnaboldi and Claeys, 2008). These two studies are focused on the theoretical cost advantage of the pure-play internet model, while paying only little attention to the risks related to this online banking model. This study adds value to the existing body of literature by studying both the performance and the associated risks of internet banks in comparison to click-and-mortar banks in six European countries (Belgium, The Netherlands, Luxembourg, Germany, Austria, and Poland).
The main goal of this study is analyzing the viability of internet banks by performing regression tests based on annual data from 2011 to 2016. In this study, 91 banks are analyzed of which 25 are considered as internet bank. I use several financial performance ratios to distinct the performance of internet banks to their branching counterparts. In total seven performance ratios (e.g. Return on Equity (ROE) and Net Interest Margin) are used in order to study the profitability, pricing and cost structure of internet banks. Also, I use three risk-related ratios (e.g. Nonperforming Loan Ratio and Wholesale Funding over Assets) to analyze the loan quality, funding and, capitalization. We apply two different estimation techniques: Ordinary Least Squares (OLS) and Generalized Least Squares Random Effects (GLS-RE). In general, the estimation results are robust to the estimation technique. On the one hand, the results indicate that internet banks outperform branching banks in terms of financial return. Although internet banks are not able to fully exploit their potential overhead reductions, these banks are more profitable by reducing their labour expenses and paying lower interest rates on deposits. On the other hand, no evidence is provided that internet banks have significantly different risk profiles compared to click-and-mortar banks.
4
2. The Potential impact of Pure-Play Internet Banking on Performance and Risk:
In this section, I position the pure-play internet banking model and discuss its implications vis-à-vis the click-and-mortar business model. A distinction is made between the theoretical impact on lending, the cost structure and funding of pure-play internet banking. This should give us better understanding which aspects related to internet banking can affect the return and risk profile of banks.
Firstly, internet banking has implications on lending that stems from a different customer relationship of pure-play internet banks and click-and-mortar banks. Internet banking builds on a digital relationship with its customer. In contrast, a click-and-mortar business strategy makes use of a network of physical branches that provide personal interaction with its customers. This difference between the business models can impact the business strategy of these banks and the product portfolio in particular. Following DeYoung (2005), the pure-play internet business model is more appropriate for transaction-based loans (e.g. mortgage loans, credit cards and auto loans) where risk management is done through automated credit scoring systems. On the other hand, a click-and-mortar business strategy is more suitable for ‘relationship lending’ that requires in-person interaction between the bank and its clients (e.g. small business loans) where the assessment of risk is done through personal knowledge.
Consequently, a potential competitive advantage of click-and-mortar banks is offering customers the option to conduct transactions over the internet while retaining the clients that prefer personal interaction through their network of offices (DeYoung, 2005). The strategic core of these banks is to route commoditized services through relatively inexpensive digital channels, whereas the network of physical branches is used for more complex and value-enhancing transactions. Hereby, it is in the interest of click-and-mortar banks to efficiently match the transaction in question with the corresponding delivery channel. According to DeYoung (2005), an efficient match between the transaction and the delivery channel may result in an increased profitability for banks offering both channels, because customers are willing to pay a premium for these high-value services.
5 internet banks pay higher wages to their relatively highly educated ICT employees. While this effect on total employment costs is not clear beforehand, the reduction in rental and maintenance costs should give internet banks a theoretical competitive advantage. If these banks are able to convert this lower cost level into lower loan rates, higher deposit rates or lower fee rates, these banks should attract more customers and grow relatively faster (DeYoung, 2001).
Finally, the pure-play internet business model can have implications for the funding of these banks. In the situation where banks are not able to attract sufficient levels of core deposits, they can use wholesale funding. According to Arnold and van Ewijk (2011), successful internet banks do not have to rely on these wholesale sources of funding. They state that internet banks are built on the ability to rapidly collect market shares in mature savings markets. If banks gather these market shares successfully, they can completely fund their operations using core deposits, while not relying on wholesale funding. Related to this, DeYoung (2001) argues that internet banks suffer from a less stable funding base since their deposit customers are more sensitive to changes in the interest rate. The author calls these customers the financially savvy “hit-and-run” clients, who are constantly looking for the best interest rates. These customers are attracted by high short-term teaser rates. However, if these interest rates decline and lag behind market standards it will lead to an outflow of deposits. Therefore, in order to keep these interest rate sensitive customers, a pure-play internet bank needs to keep its interest rates at a high level (Arnold and van Ewijk, 2011).
3. Literature Review:
In this section, the empirical research on internet banking is discussed. The first part focuses on the impact of pure-play internet banking on profitability and risk. The financial literature on pure-play internet banks is relatively small and much attention is given to newly established internet banks (DeYoung, 2001, 2005; Delgado et al., 2007) because of the novelty of the business model. The second part focuses on the impact of internet adoption on the performance and risk profile of banks.
6 employment costs spent on the highly educated workforce. Also, the author provides evidence that internet banks have difficulty generating deposit funding which may be explained by the fact that internet banking was not widely accessible at that time, requiring internet banks to create a network of ATMs to receive deposits.
Since DeYoung (2001) focuses on young banks, the advantages of the business models might occur over time as these internet banks accumulate experience and grow larger. Therefore, in a follow-up paper, DeYoung (2005) studies the scale and experience effects of a dozen US internet banks that started between 1997 and 2001. The paper confirms the low average profit levels of internet banks, mainly because the expected reduction in overhead and other expenses have not materialized. On the one hand, the paper finds evidence that internet banks may achieve more scale effects compared to branching banks, suggesting internet banks have to grow larger in order to stay in business. On the other hand, only little evidence is provided that experience effects accelerate the financial performance of internet banks as these banks age and gain experience.
These results are confirmed by a multi-country study by Delgado et al. (2007). They find that European internet banks underperform newly charted multi-channel banks mainly because of their higher overhead costs resulting from large fixed ICT investments and high marketing expenses. However, this profitability gap decreases as internet banks grow larger because they exhibit technology scale economies. This effect is mainly the result of internet banks being better able to control their operational costs more efficiently as compared to newly chartered branching banks. The results by both DeYoung (2005) and Delgado et al. (2007) indicate that if internet banks are able to grow larger over time, the efficiencies resulting from scale effects might close the probability gap between internet-only and click-and-mortar banks. In terms of risk, both studies do not provide any conclusive evidence that internet banks have a significantly different capitalization or credit risk exposure.
7 deposit-based products, they are not able to gain from more rewarding banking activities. Customers asking for high-value services prefer in-person interaction which hinders the profitability of internet banks.
More recently, a case study on ING Direct, one of the leading pure-play internet banks, is performed by Arnold and van Ewijk (2011). They find that ING Direct exhibit low-cost levels and easy scalability. The ability of this internet bank to easily scale their operations enables it to quickly gather core deposits. However, according to the authors, this may create a challenge for internet banks to increase the lending portfolio at the same pace as their deposits. It might even drive internet banks towards transaction-oriented business, like securities, or result in internet banks relaxing their lending standards. This may create an overexposure in relatively risky markets and will have negative implications for the stability and credit risk of internet banks in a macroeconomic downturn.
Other studies analyze the impact of the adoption of internet banking on banks’ performance by comparing the financial performance between click-and-mortar banks and traditional brick-and-mortar banks. The impact of online banking in Europe is studied by Hernando and Nieto (2007) who examines 72 click-and-mortar banks in Spain. They find that the adoption of the internet as a delivery channel results in a higher profitability and that the provision of online banking services is a complement, rather than a substitute for these physical branches. The increase in profitability is mainly the result of a lower level of overhead expenditures, particularly caused by a cost reduction in staff, marketing, and ICT.
Furthermore, Ciciretti et al. (2009) analyzes the bank return and risk of traditional banks based in Italy during the period 1993-2001. Their findings indicate a strong positive relationship between the adoption of internet activities and bank return. Additionally, they find a marginally significant negative relationship between the offerings of online banking products and bank risk resulting from an increased level of diversification. Finally, DeYoung et al. (2007) examine US community banks and suggest that the internet distribution channel is complementary to a network of locations. The authors find that internet adoption improved the financial performance substantially due to a higher level of revenues from deposit services charges.
8 increased. Therefore, a deeper understanding is needed whether the performance of these pure digital banks has improved. Furthermore, no conclusive evidence is provided that internet banks have distinctly different risk profiles compared to click-and-mortar banks. In this study, I will further address these doubts using evidence from the European banking industry.
4. Data and Methodology:
In order to study the financial performance and risk profile of internet banks, a regression model is estimated which is presented first. This is followed by a discussion of the performances measures and the explanatory variables used in the model. This section concludes with an analysis of the sample data and the representativeness of this sample.
The performance and risk profile of internet-only banks is analyzed using the following equation:
𝑃𝐸𝑅𝐹𝑂𝑅𝑀𝐴𝑁𝐶𝐸𝑖𝑡 = 𝛼 + 𝛽1∗ 𝐼𝑁𝑇𝐸𝑅𝑁𝐸𝑇𝑖 + 𝛽2∗ 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡+ 𝜃1∗ 𝐶𝑂𝑈𝑁𝑇𝑅𝑌𝑖+ 𝜃2 ∗ 𝑌𝐸𝐴𝑅𝑡+ 𝜀𝑖𝑡 (1)
In this regression specification, the subscript i refers to a cross-sectional index of banks and the subscript t refers to time index in years.2 The dependent variable is PERFORMANCE and is estimated separately using ten different measures of a bank’s financial performance. INTERNET is a dummy variable which takes the value of one if bank i is classified as a pure-play internet bank, and zero otherwise. Therefore, the corresponding coefficient β1 is the main
test statistic. A statistically significant value for β1 means a financial performance gap
between internet banks and click-and-mortar banks. I assume the error term εit to be normally
distributed and independent with zero mean.
As stated earlier, the dependent variable PERFORMANCE is estimated separately using seven performance measures and three risk measures. The first performance measures I include are the net interest margin and its two components, the interest income to earning assets ratio and interest expenses to deposits ratio. These measures are included to study if internet banks are able to convert their potential cost reduction into narrower margins, i.e. higher deposit rates and lower loan rates. The ratio of noninterest income to total income is included since we are interested in the extent to which internet banks rely on deposit-based
2 The independent variables have not been lagged which may result in an endogeneity issue. The rationale
9 products or, alternatively, if these banks can benefit from more rewarding fee-based activities. I use the labour expenses to total assets ratio in order to study if internet banks are able to reduce their employment costs since these branchless banks do not need a workforce operating at the offices. The cost-to-income ratio is another cost-related measure I add.3 This ratio reflects the ability of a bank to realize income from its expenditure and is included to study if internet banks are able to reduce their overhead costs. Finally, the return on equity (ROE) is a good profitability indicator as it reflects the ability of a bank to generate return for its equity financing.
In terms of risk measurements, I include the nonperforming loan ratio to study if internet banks have an overexposure in risky loans. The nonperforming loan ratio is an important indicator of a bank’s financial health and creates comparability of the credit quality of loan portfolios among different banks. Also, I include the ratio of wholesale funding to total assets ratio to study to what extent internet banks have to rely on wholesale funding instead of core deposits. It is often argued that core deposits exhibit lower levels of funding liquidity risk since this source of funding is more “sticky” compared to wholesale funding. Finally, the Z-score is considered, which is a measure that relates the equity capital of a bank to the variability in its return. In particular, it measures how much variability in return can be absorbed by a bank’s equity capital before it becomes insolvent. The denominator of this ratio is given by the standard deviation of the Return on Assets (ROA), which is a representation of the variability of returns. The numerator consists of the equity ratio plus the ROA. A lower Z-score indicates an increased level of bank risk.
The performance and risk profile of click-and-mortar banks and internet banks might be - apart from cross-sectional differences in bank characteristics – affected by country-specific economic features. General economic features like market concentration and macroeconomic conditions can influence a bank’s performance and risk profile and therefore a number of control variables are included in the regression model.
Firstly, a set of bank-specific factors are included in the regression that affect the profitability of a bank. The control variable ASSETS reflects the size of banks measured by total assets and is included to control for scale effects, as indicated by DeYoung (2005).4 EQUITY, measured as the ratio of equity to total assets, is included in the model to control for standard bank riskiness. Athanasoglou et al. (2008) find the relationship between equity
3 According to Orbis Bank Focus, the cost-to-income ratio is given by the ratio of total operating expenses to
operating revenues.
4 The natural logarithm of ASSETS is considered as marginal benefits from economies of scale decrease with
10 capital and profitability to be positive as equity capital can act as a safety net in an economic downturn. LOANS, measured by the ratio of total loans to total assets, is another measure of bank risk. By definition, increasing the total loan portfolio will raise the interest income and therefore will affect the profitability positively. For instance, Trujillo-Ponce (2013), and Dietrich and Wanzenried (2011) find a positive relationship between the loan ratio and a bank’s profitability. However, an increase in loans outstanding can lead to higher credit risk, which will have a negative impact on profitability because of the incurrence of loan losses (Bikker and Hu, 2002).MBHC is a dummy variable that takes the value of one if the bank is an affiliate in a multibank holding bank company and zero otherwise. Following DeYoung (2001, 2005, 2007), I include this variable to control for organization structure.
Furthermore, two industry-specific factors are included in the regression specification in order to control for competitive market effects. HERFINDAHL, measured by the Herfindahl-Hirschman Index (HHI)5 and C5RATIO, measured by the market shares of the five largest banks in their industry, are two measures of market concentration. Market concentration is extensively studied as a determinant of bank profitability. Two theoretical models form the basis of this literature. The structure-conduct-performance hypothesis claims that banks operating in highly concentrated industries have higher profitability levels. These banks are able to gain from their market power and therefore charge higher than competitive prices. Based on empirical evidence, Goddard et al. (2004) find evidence for the structure-conduct-performance hypothesis for European banks. Alternatively, the efficient-structure theory claims that more efficient banks generate higher levels of profitability resulting in these banks to generate greater market shares (Berger, 1995).
The appended Table A1 displays the market concentration of the countries studied in the paper. It can be seen that the Netherlands has a highly concentrated banking industry according to both the HHI and the ratio of the five biggest banks in this country (C5 Ratio). It is illustrated that in the Netherlands the five biggest credit institutions hold roughly 85% of the total assets. Belgium is also characterized by a high concentration level in the banking industry. In Poland, the market share of the five largest institutions is about 45% whereas in Austria, Germany and, Luxembourg it is about 33% only.
The last set of explanatory variables are included in order to control for macroeconomic conditions. LTIR refers to the long-term interest rate and controls for growth
5 Herfindahl-Hirschman index is calculated by the sum of squares of market shares of all firms in a single
11 of the cost of funding.6 STIR is the short-term interest rate and its effect on net interest income and therefore profitability is found to be positive (e.g. Demirgüç-Kunt and Huizinga, 1999).7 REALGDP refers to the real growth in domestic product (GDP) and is included to capture the business cycle effect. For instance, Bikker and Hu (2002) find a positive relationship between real GDP growth and bank profitability. They argue that the business cycle affects the loan provision since in prosperous economic times the demand for credit is higher which will, in turn, improve bank profitability.
As indicated earlier, the different regressions are estimated using both OLS and GLS-RE estimation technique on pooled annual bank data. Both time (YEAR) and country (COUNTRY) fixed effects are included, unless including these dummies lead to serious multicollinearity problems. These time fixed effects control for the possible effect of changes in the macroeconomic environment. The country dummies are included to capture the influence of omitted variables that are country-specific and do not vary over time (e.g. country-specific regulatory effects).8
Finally, the GLS-RE estimation technique adds bank-specific random effects to the regression estimation. These random effects are included to account for bank-specific unexplained variation in the dependent variable. Additional to the normal random disturbance term, the random effects approach adds a bank-specific disturbance term to the regression which is identical for each year. As explained by DeYoung (2005), bank-specific fixed effects are not appropriate in this regression specification, since the aspect that is tested is for themselves fixed effects. In particular, the variable INTERNET is time-invariant in our sample and therefore, it would not be possible to test for the effect of coefficient β1 if
bank-specific fixed effects are included in the regression bank-specification.
Our study focuses on the financial performance of banks in the countries Belgium, The Netherlands, Luxembourg, Germany, Austria and, Poland during a six-year time period from 2011 to 2016. By studying multiple countries with different characteristics, the results will be more broadly applicable. These countries differ in terms of the structure and market concentration of their banking industry. Combined with the fact that these countries contain variation in their economic systems, it will create more robust results than studying a single market. Data is obtained from Orbis Bank Focus by Bureau van Dijk, that provides detailed
6
Long-term interest rate is measured by 10-year government bond yield of the particular country.
7
Short-term interest rate is measured by 3-month Euribor, i.e. 3-month money market rate.
8
12 balance sheet information of public and private banks worldwide using comparable standards. Data availability forms a major limitation in studying the performance of internet banks since banks are not required to specify to what extent their revenues, costs, loans and, deposits are the result of internet activities. Another limitation is that Orbis Bank Focus only reports a maximum of six years of bank balance sheet information for a limited amount of banks. Therefore, the panel data set in this study is unbalanced with missing information for a number of banks in the years 2011 and 2012.
I consider unconsolidated statements to distinguish between the performance between different entities within a banking group. Therefore, a distinction can be made between, for example, the performance between the pure-play internet bank ING-Diba (Germany) and the click-and-mortar bank ING Bank NV (The Netherlands), which are both part of the ING Group NV. However, in case unconsolidated statements are not available, I use a single consolidated statement of the banking group. According to Duprey and Lé (2016), to be consistent you have to select either consolidated or unconsolidated financial statements of a particular bank. By analyzing both consolidated and unconsolidated financial statements for a single company, a pure double counting problem will occur. So by either selecting the unconsolidated statement or the consolidated statement, I can prevent this double counting issue.
13 internet bank has to offer a full range of banking services. These services include deposit-taking, offering current and savings accounts and making loans. In total 25 banks meet these three conditions, and therefore, are considered as internet bank in this study.
To identify the effect of being a pure-play internet bank on the performance and risk profile, I include 66 click-and-mortar banks as a benchmark sample. A categorization of the total sample of 91 banks is presented in Table 1. The descriptive statistics are presented in Table 2. A distinction has been made between income statement items and balance sheet items. Table A2 in the appendix gives an overview of the banks considered in this study.
The benchmark sample meets the above-mentioned conditions with the exception that these banks operate through a network of physical bank offices alongside their online channels. During the selection process of benchmark banks, the effect of age and size of internet banks has been considered. Regarding age, internet banks are relatively younger compared to click-and-mortar banks because of the novelty of this pure-play internet banking model. Previous studies on the performance of internet banks are focused on newly chartered banks (DeYoung, 2001, 2005; Delgado et al., 2007). As indicated earlier, these studies did not find evidence for technology-based learning effects, such that internet banks do not have additional learning effects as these banks age. This means that internet banks do not benefit from technology-based learning that accelerates the financial performance of internet banks in their early life. Therefore, I have chosen to combine both newly chartered and established banks into the sample and to put no conditions on the age of banks.
14
9
In order to classify a bank’s ownership, I use the BvD Independence Indicator, which is defined by Orbis Bank Focus as an indicator that characterizes the degree of independence of a company with regard to its shareholders. The BvD independence Indicator is classified as A, B, C and D. A indicates that no shareholder is present with shareholdings more than 25%, B indicates that shareholders do not have more than 50% of direct or total ownership, and at least one shareholder with more than 25% and C and D relate to a company with at least one shareholder with direct or total ownership of more than 50%. I consider a bank as independent if they are classified as A or B.
10
As can be seen in Table 1, some internet banks are identified as ‘Established’ and therefore are operating for at least 20 years. This is due to the fact that some internet banks were previously operating as a click-and-mortar bank and changed their business model to a pure-play internet model. Although it is possible that banks changed from a click-and-mortar business strategy to a pure-play internet banking model in their history, I have made sure that no changes in the business model occurred during the sample period of 2011 – 2016.
Table 1: Sample breakdown categorized per country. Age is measured in years since its establishment in its current
organizational form, where Newly Chartered < 20 years; Established ⩾ 20 years. Size is measured by Total Assets, where Small < €50 billion; Medium = €50-250 billion; Large = €250-500 billion; Very Large > €500 billion.
Source: Orbis Bank Focus.
Austria Belgium Germany Luxembourg Netherlands Poland
15
Table 2: Summary Statistics for all regression variables. All dependent variables are self-explanatory, except for
Z-score. Z-score = ratio of the standard deviation of return on assets (ROA) over equity ratio plus the ROA. Definitions of the control variables: Ln(Assets) = natural logarithm of total assets, MBHC = one if a bank is an affiliate in a multibank holding company, Herfindahl = Herfindahl-Hirschman Index (HHI), C5 ratio = market share of five largest banks in a country, LTIR = long-term interest rate given by 10 years government bond yield, STIR = short-term interest rate given by 3-month money market rate; Real GDP Growth = real GDP growth rate. All dependent variables have been winsorized at the 1 percent and 99 percent levels of the sample distribution prior to taking the summary statistics. Source: Orbis Bank Focus (for bank-specific information), ECB, Statistical Data Warehouse (for market concentration and interest-rate information), and IMF DataMapper (for GDP growth).
Dependent Variable Mean Std. Dev. Minimum Maximum
A. Income Statement Items
Interest Income to Earning Assets 0.0352 0.0195 0.0079 0.1235 Interest Expenses to Deposits 0.0172 0.0122 0.0000 0.0515
Net Interest Margin 0.0199 0.0228 -0.0211 0.1331
Noninterest Income to Total Income 0.3689 0.2524 -0.7131 1.0077 Labor Expenses to Total Assets 0.0066 0.0044 0.0000 0.0260
Return on Equity 0.0595 0.1113 -0.5540 0.4039
Cost to Income Ratio 0.6565 0.2297 0.1934 1.5202
B1. Balance-sheet (asset side) and related items
Cash to Total Assets 0.0402 0.0470 0.0000 0.2709
Securities to Total Assets 0.3624 0.2294 0.0257 0.9373 Loans to Total Assets 0.5033 0.2304 0.0023 0.9653 Nonperforming Loans to Loans 0.0435 0.0844 0.0000 0.9412 Fixed Assets to Total Assets 0.0043 0.0045 0.0000 0.0513 Other Assets to Total Assets 0.0841 0.1150 0.0000 0.6010
B2. Balance-sheet (liability side) and related items
Deposits to Total Assets 0.5660 0.2605 0.0158 0.9773 Other Liabilities to Total Assets 0.0907 0.1220 0.0043 0.7542 Wholesale Funding to Assets 0.2705 0.2320 0.0000 0.9034 Equity to Total Assets 0.0710 0.0344 0.0141 0.2146
16 whether our estimation results hold when internet banks are matched with similar sized click-and-mortar banks.
In the appended Table A3, an overview is presented of the representativeness of the sample in terms of total assets analyzed in this study related to the total assets of the total banking industry. Despite the limited amount of banks analyzed in this study and a lack of available information in the years 2011 and 2012, they represent a large share of the banking industry in the concerned country. On average, the sum of the banks’ total assets included in this study comprises more than 67 percent of the size of the total banking industry. The only exception is Luxembourg where the sample only captures an averaged 40 percent of the banking activity in the country.
The sample includes a mixed group of banks with large banks, medium-sized and, relatively smaller sized banks. Typically, these large banks operate through a network of physical branches and are classified as click-and-mortar banks. Also, a limited number of these banks are independent, whereas the vast majority belongs to a banking group.
5. Results:
This section is divided into two parts. In the first part, I will present indicative results using a univariate analysis. These results give a broad perspective on the performance and riskiness of internet banks. In the second part, I will use a multivariate regression framework to study the actual impact of being an internet bank on performance and the risk profile.
First of all, a univariate analysis is performed of which the results are displayed in Table 3. The items in segment A represent income statement items and provide information about the financial performance of the internet and benchmark banks. In segment B, the balance sheet items are presented. Although not all balance sheet items considered in the univariate analysis are used in the regression framework, they are useful for comparative purposes. A statistically significant difference between the mean of the internet-only bank and the mean of the click-and-mortar bank provides evidence for a difference in performance or risk profile between both types of banks.
17
Table 3: Subsample means and difference-of-means test. Data are an unbalanced panel of 546 annual
observations of 66 click-and-mortar banks and 25 pure-play internet banks between 2011 and 2016. The superscript *,** and *** denote significance at the 1%, 5% and 10%, respectively. All dependent variables have been winsorized at the 1% and 99% levels of the sample distribution prior to taking the means.
Source: Orbis Bank Focus.
(1) Click-and-mortar (2) Pure-play internet (2)-(1) T-test
A. Income Statement Items
Interest Income to Earning Assets 0.0367 0.0307 -*** Interest Expenses to Deposits 0.0183 0.0125 -***
Net Interest Margin 0.0197 0.0209 +
Noninterest Income to Total Income 0.3625 0.3869 + Labor Expenses to Total Assets 0.0067 0.0063 -
ROE 0.0535 0.0758 +**
ROA 0.0047 0.0055 +
Cost-to-Income Ratio 0.6484 0.6775 +
B1. Balance-sheet (asset side) and related items
Total Assets (in million €) 143,000 16,000 -***
Cash to Total Assets 0.0327 0.0630 +***
Securities to Total Assets 0.3740 0.3284 -***
Loans to Total Assets 0.5231 0.4490 -***
Nonperforming Loans to Loans 0.0446 0.0221 -*** Other Assets to Total Assets 0.0656 0.1359 +*** Fixed Assets to Total Assets 0.0051 0.0019 -***
B2. Balance-sheet (liability side) and related items
Deposits to Total Assets 0.5090 0.7193 +***
Other Liabilities to Total Assets 0.0988 0.0688 -**
Wholesale Funding to Assets 0.3195 0.1382 -***
Equity to Total Assets 0.0712 0.0696 -
18 the strategic core of being an internet bank, which is not confirmed in this univariate analysis. This may imply that internet banks are able to reduce their expenses related to maintaining their network of physical branches, but this effect is offset by higher ICT development costs. Also, internet banks are not able to reduce their labour costs. This may provide evidence for a second substitution effect where a reduction in staff working at the branches may be offset by an increase in highly skilled and relatively expensive ICT workers.
Given the result that internet banks do not exhibit different cost levels, it makes sense that these banks are not able to set narrower interest margins because it would deteriorate their returns. Although our results indicate that internet banks are able to charge lower loans rates compared to click-and-mortar banks (3.07% relative to 3.67%), internet banks simultaneously pay lower deposit rates to their depositors (1.25% relative to 1.83%). On the one hand, this relatively low implicit deposit rate may indicate that the depositors of internet banks are less price sensitive than suggested by Arnold and van Ewijk (2011). These authors argued that internet-only banks have to pay high deposit rates in order to prevent a deposit outflow from their interest rate sensitive customers. Our results show that internet banks are largely funded by deposits, which may provide evidence that internet banks do not face difficulty to fund themselves with deposits and that the deposit clientele of internet banks is less interest-rate sensitive as earlier suggested. On the other hand, this relatively low loan rate may indicate that internet-only banks provide less risky loans. The relatively low nonperforming loan ratio of internet banks provides some evidence for this result, as it indicates that internet banks have a higher quality loan portfolio.
19 In Table 3, I also compare the size of internet banks in terms of total assets. These results confirm that internet banks are significantly smaller compared to click-and-mortar banks. Internet banks measure at €16 billion whereas click-and-mortar banks have an average size of €143 billion measured in terms of total assets. On the asset side of the balance sheet, it turns out that internet banks have difficulty in realizing lending activity. Our results indicate that, despite offering lower loan rates, pure-play internet banks generate fewer loans (44.90% relative to 52.31%). However, this result does not lead to internet-only banks investing more in risky securities. The securities to assets ratio of internet banks is about 5 percentage points lower compared to the benchmark banks, which indicates that internet banks do not heavily resort to relatively risky securities in case these banks have an excess of deposit funds. This lower risk profile is confirmed by the level of cash internet banks hold. The results show that internet banks have, on average, almost double the amount of cash relative to click-and-mortar banks (6.30% relative to 3.27%). Also, it is not surprising that click-and-click-and-mortar banks have a higher level of fixed assets compared to their online counterparts since these click-and-mortar banks have a network of physical branches.
Looking at the funding of both types of banks, I find that internet banks rely substantially more on deposits (71.93% relative to 50.90%). This is in line with conventional wisdom that internet banks mainly handle deposit-based products. Also, as indicated by Arnold and van Ewijk (2011), internet-only banks are built on the ability of easily generating core deposits in mature saving markets. Therefore, successful internet banks should be able to operate without making use of on wholesale funding. The sample means in Table 3 provide some evidence for this. Although internet banks still need wholesale funding (approximately 14%), these banks are able to mostly fund themselves using core deposits. This indicates that internet banks face less funding liquidity risk because these internet-only banks are predominantly funded by relatively stable retail deposits. Finally, the results show no statistically significant difference between the leverage and the related Z-score of internet-only and click-and-mortar banks. Decomposing the Z-score, I find that internet banks, on average, exhibit more volatile profits.11 Combined with the notion that both types of banks have almost identical levels of equity cushions (in terms of equity capital plus ROA), it turns out that branching banks have a higher averaged Z-score. However, this Z-score is not significantly different, which provides preliminary evidence that internet banks do not have an increased probability of insolvency.
11 This result is not reported in Table 3, but the average standard deviation of ROA is 0.0048 for internet banks
20 5.1 Regression Results:
In this section, I will estimate equation (1) separately for each of the seven performance measures and three risk-related measures. The main results of the OLS estimation are presented in Table 4a, whereas the results of the GLS-RE are displayed in Table 4b. As stated earlier, β1 is the main test statistic and indicates a possible performance or
risk gap between internet banks and click-and-mortar banks.
In general, the estimation results are robust to the estimation techniques, although the GLS-RE results tend to be larger in absolute magnitude. Also, the signs of β1 are generally in
line with the performance gaps inferred by the difference of means tests in Table 3. However, the regression results show less significant results and are different in terms of magnitude because they are conditioned on a set of control variables.
I first consider the impact of being an internet bank on the income statement items. According to my results, pure-play internet banks outperform click-and-mortar banks in terms of profitability. The estimated effect of being an internet-only bank is an increased ROE of 6.31 percentage points.12 This result is in line with our indicative result inferred from the mean tests. However, this result contrasts studies by DeYoung (2005) and Delgado et al. (2007) who find a relatively low level of profits at pure-play internet banks because of their high levels of overhead costs. However, both studies point towards the scale effects of internet banks, which may be an explanation for this improvement in profitability. In this study, I consider substantially larger internet banks compared to previous studies by DeYoung (2005) and Delgado et al. (2007). It might be possible that internet-only banks have gained from their technology-based scale effects resulting in a higher profitability.
The estimation results indicate that the performance gap is driven by two efficiencies of internet banks. First, internet banks are able to attract deposits while paying lower deposit rates. In particular, internet-only banks pay deposit rates that are 0.62 percentage points lower compared to click-and-mortar banks. This result refutes DeYoung (2001) who argues that internet banks mainly attract relatively price-sensitive depositors that are only interested in benefiting from high deposit rates.13 Also, it contradicts the conventional idea that internet banks have to pay higher deposit rates to bid away customers from click-and-mortar banks. Possibly internet banks have superior ICT systems more appropriate for low value-added
12
Unless stated differently, the estimation results reported in the main text correspond to the GLS-RE estimation results presented in Table 4b.
13 The result that internet banks attract less interest-rate sensitive customers is preliminary. As indicated by
21 transactions. If these systems are highly valued by depositors, it may be possible that these clients are willing to deposit cash at internet banks while receiving lower deposit rates.
Nonetheless, the result that lower interest expenses lead to an improved ROE of internet banks should be interpreted with caution, because these banks do not show a distinctive performance in terms of interest income and interest margin. In particular, while internet banks pay relatively lower deposit rate to their depositors, these banks do not pay significantly different loan rates and, simultaneously, are not able to convert the advantage of paying lower deposit rates into higher interest margins. Therefore, we should not over-estimate the benefit of paying lower deposit rates on the actual profitability of internet-only banks.
Second, evidence shows that the level of labour expenses of internet banks is 0.38 percentage points lower compared to click-and-mortar banks. This indicates that internet-only banks are able to benefit from operating without a network of branches by reducing their employment costs. This result is in contrast to DeYoung (2001) who finds that pure-play internet banks have relatively high labour expenses due to increased wage costs to their (highly skilled) ICT workers. Although it is not possible to test for the average wage of internet bank employees in this sample, the results seem to provide evidence that the effect of a smaller workforce exceeds the effect of increased wages paid to the ICT workers, which make internet banks successful in substituting low with highly skilled workers.
The reduction in labour expenses does not translate into a lower cost-to-income ratio of internet banks, perhaps caused by increased ICT expenditures or higher marketing costs.14 As indicated by DeYoung (2001), internet banks are more likely to encounter difficulties in creating a brand identity since they lack the physical presence that may provide them promotional advantages. The result that internet banks do not exhibit a lower cost-to-income ratio has important implications because it refutes the idea that the strategic value of a pure-play internet bank is derived from lower overhead costs. This finding is consistent with another European study performed by Arnaboldi and Claeys (2008).
Again, Arnaboldi and Claeys (2008) state that noninterest income is an important driver of the viability of the pure-play internet banking model. It is necessary for internet banks to address themselves towards fee-based income, which may be especially relevant in low-interest environments where banks’ interest margins may be compressed. My estimation
14
22 results show that the pure-play internet banks do not have a significantly different level of noninterest income. This indicates that internet-only banks do not face additional difficulties to cross-sell fee-based financial products. This is a surprising result since internet banks make use of a distribution channel based on a digital relationship with their clients. Therefore, internet banks may be less able to benefit from relatively rewarding activities that require personal interaction with their clients (e.g. investment banking). However, this effect may possibly be offset by the agility of internet banks. These digital banks have a continuous information flow about their customers provided by their online channels. This could provide them with the advantage of being better able to rapidly adapt to the changing needs of their clients.
Regarding the risk profile, there is no significant difference between the nonperforming loan ratio of internet-only banks and click-and-mortar banks. Therefore no evidence is provided for the hypothesis that internet banks may invest in relatively more risky assets. In case the lending growth cannot keep up with a strong deposit growth, internet banks may be inclined to relax their lending standards in order to earn a high yield. A relatively high nonperforming loan ratio of internet banks could have been an indication for this mechanism. However, our results cannot confirm this, and therefore, they are line with DeYoung (2005) who also finds no risk gap in terms of the nonperforming loan ratio.
Regarding funding, my results only provide weak evidence that digital banks rely substantially less on wholesale funding compared to click-and-mortar banks. However, although the estimation results provide a negative sign for β1, which is in line with the means
23 Finally, no statistically significant performance gap is found for the Z-score. This indicates that internet-only banks and click-and-mortar banks do not have a different buffer (measured by equity capital plus ROA) to absorb their variability in returns. Therefore, it is concluded that the probability of insolvency is not distinctively different between digital and branching banks. This may have important implications. As indicated by DeYoung (2005), equity capital may be more useful for internet banks to realize their growth potential. My finding that internet banks do not have higher equity buffers can potentially impede this potential of internet banks.
Our regression model shows explanatory power that is in line with previous research on this topic (Delgado et al., 2007; Arnaboldi and Claeys, 2008). Equation (1) is estimated separately using ten different performance and risk measures, and they exhibit an average r-squared of 0.2467. In the same context, the Wald Test is used to test for joint explanatory power of the entire regression model. In all regression estimations, the Chi-squared statistic is significant which provides evidence that the explanatory variables jointly affect the performance variables.
Now we have interpreted the effect of the main test statistic β1, I would like to analyze
24
Table 4a: Panel Estimates. Estimation Technique: Ordinary Least Squares (OLS). INTERNET is a dummy variable which takes the value of one for banks classified as a
pure-play internet bank, and zero otherwise. The superscript *,** and *** denote significance at the 1%, 5% and 10%, respectively. The reported statistic is distributed as an F with 14 numerator degrees of freedom and the number of observations minus 15 denominator degrees of freedom. In the second and third equation, year dummies are excluded to avoid multicollinearity problems. White-corrected standard errors are shown in parentheses.
Performance Risk Variables Interest Income to Earning Assets Interest Expenses to Deposits Net Interest Margin Noninterest Income to Total Income Labour Expense to Total Assets ROE Cost-to-Income Ratio Non-performing Loan Ratio Wholesale Funding to Total Assets Z-score INTERCEPT 0.3091** 0.0328*** 0.0369 -17.22*** 0.0409 -4.1387*** -24.70*** 0.1602 5.3492*** -1051.86 (0.1247) (0.0110) (0.0610) (2.9845) (0.0295) (1.4839) (7.1117) (0.3654) (1.6095) (668.29) INTERNET -0.0022 -0.0055*** 0.0045 -0.0484 -0.0016*** 0.0596*** -0.0524 -0.0003 -0.1530*** -16.38 (0.0031) (0.0020) (0.0037) (0.0383) (0.0005) (0.0171) (0.0325) (0.0063) (0.0323) (15.41) ln(ASSETS) -0.0013** -0.0006 -0.0024** -0.0073 -0.0004** 0.0062 -0.0173** 0.0047*** 0.0215*** -7.6384** (0.0006) (0.0006) (0.0011) (0.0086) (0.0002) (0.0048) (0.0083) (0.0011) (0.0066) (3.5537) EQUITY -0.0470* -0.0830*** -0.0276 1.2415*** 0.0532*** 0.6080** -1.1191** 0.1981*** -0.9358*** 258.43** (0.0245) (0.0207) (0.0408) (0.4103) (0.0078) (0.2399) (0.4325) (0.0515) (0.2907) (112.24) LOANS 0.0338*** 0.0011 0.0334*** -0.2760*** 0.0026*** -0.0248 0.0159 0.0382*** -0.0792* 57.53*** (0.0041) (0.0035) (0.0070) (0.0680) (0.0009) (0.0250) (0.0519) (0.0082) (0.0451) (15.21) MBHC -0.0036 -0.0026 -0.0064** 0.0891*** -0.0013*** -0.0171 0.0438** -0.0023 0.0659*** -25.11*** (0.0022) (0.0016) (0.0033) (0.0261) (0.0004) (0.0114) (0.0206) (0.0036) (0.0185) (8.6574) HERFINDAHL -0.1297 0.0032 0.1385 -3.8864 -0.0642 -2.5477 1.9658 0.0650 -0.8140 -1075.16 (0.4049) (0.1251) (0.5933) (3.2980) (0.0401) (2.2844) (3.0454) (0.4058) (1.8914) (702.11) LTIR 0.7713** -0.138* 0.6312*** -7.4582 -0.0941 3.6264 -5.4206 -0.1581 4.9521 1412.51 (0.3678) (0.0802) (0.2198) (5.1637) (0.0871) (2.8107) (5.0017) (0.7620) (3.6079) (1557.41) STIR -196.22** 0.0649 -0.1240 8225.51*** -19.13 -1662.83*** 4432.52*** -98.77 -3925.70*** -453965* (78.58) (0.3843) (0.8227) (1288.25) (19.17) (593.10) (1207.45) (169.52) (1066.83) (258317) REALGDP -0.1115 -0.1283** 0.0887 0.8727 0.0158 2.6691** -2.665 -0.2001 -2.1029 -69.95 (0.1392) (0.0599) (0.1317) (1.7052) (0.0359) (1.1621) (2.219) (0.2910) (1.2925) (509.86)
Bank Random Effects No No No No No No No No No No
Year dummies Yes No No Yes Yes Yes Yes Yes Yes Yes
Country dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 433 271 269 433 434 439 427 439 439 435
25
Table 4b: Panel Estimates. Estimation Technique: GLS-RE. INTERNET is a dummy variable which takes the value of one for banks classified as a pure-play internet bank,
and zero otherwise. The superscript *,** and *** denote significance at the 1%, 5% and 10%, respectively. The reported statistic is distributed as a chi-square with 19 degrees of freedom. White-corrected standard errors are shown in parentheses.
Performance Risk Variables Interest Income to Earning Assets Interest Expenses to Deposits Net Interest Margin Noninterest Income to Total Income Labour Expense to Total Assets ROE Cost-to-Income Ratio Non-performing Loan Ratio Wholesale Funding to Total Assets Z-score INTERCEPT 0.0291 0.0400* 0.0207 23.16*** 0.0424*** -6.5232*** 17.42*** 0.3111 0.0671 -168.42* (0.1029) (0.0205) (0.0593) (1.1639) (0.0163) (0.8573) (1.3162) (0.2215) (0.4040) (94.55) INTERNET -0.0013 -0.0062* 0.0031 -0.0638 -0.0038*** 0.0631** -0.0593 -0.0000 -0.0363 -2.6786 (0.0070) (0.0034) (0.0077) (0.0723) (0.0015) (0.0266) (0.0671) (0.0142) (0.0831) (27.68) ln(ASSETS) -0.0004 -0.0006 -0.0022 -0.0128 -0.0013*** 0.0061 -0.0257 0.0048 0.0653*** -0.4982 (0.0011) (0.0011) (0.0020) (0.0178) (0.0004) (0.0076) (0.0162) (0.0030) (0.0198) (3.8564) EQUITY 0.0154 -0.0622** -0.0122 1.1546** 0.0178* 0.5870* -1.7236*** 0.1478 -0.5851 514.75*** (0.0319) (0.0305) (0.0739) (0.4495) (0.0091) (0.3456) (0.6034) (0.0929) (0.5140) (72.73) LOANS 0.0277*** -0.0030 0.0350*** -0.2745** 0.0023* 0.0020 0.0264 0.0045 -0.0694 -8.5867 (0.0077) (0.0055) (0.0131) (0.1170) (0.0012) (0.0445) (0.0947) (0.0161) (0.0672) (6.0551) MBHC -0.0042 -0.0027 -0.0068 0.0779 -0.0010 -0.0176 0.0353 -0.0018 0.0506 -23.22 (0.0052) (0.0029) (0.0069) (0.0570) (0.0010) (0.0191) (0.0404) (0.0085) (0.0448) (23.00) HERFINDAHL 0.0595 0.0676 0.1087 -4.1783** -0.0996*** -2.3570 0.8444 0.2724 0.1036 -155.70 (0.3719) (0.1140) (0.6205) (2.0593) (0.0349) (2.2756) (2.3295) (0.2990) (0.8485) (106.68) LTIR 0.6985** 0.0602 0.6146 -4.0608 0.0270 3.6671 -2.5212 0.0660 1.3832 315.37 (0.3030) (0.2570) (0.4167) (3.4825) (0.0424) (2.7946) (4.5651) (0.3834) (1.3087) (319.06) STIR 5.1922 -0.1189 0.2855 8219.09*** 2.7967 -2381.09*** 5986.25*** 131.05 371.61* -78552*** (45.55) (0.6709) (1.0634) (461.26) (5.8016) (357.31) (528.60) (85.90) (196.81) (29620.66) REALGDP -0.1936* -0.0891 -0.1248 0.2217 -0.0012 2.3086** -2.2075 -0.1511 -0.5938 -81.91 (0.1083) (0.0627) (0.1426) (0.9740) (0.0171) (1.1421) (1.6803) (0.1133) (0.4246) (77.69)
Bank Random Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Country dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 433 271 269 433 434 439 427 439 439 435
26 only slightly affects the performance and cost components. Although an increased market concentration is associated with lower labour expenses and lower fee-based income, it has no impact on the profitability of banks. On the one hand, this result suggests that higher market concentration not adequately gives incentives to be cost-efficient, as the cost-to-income ratio is not affected by the Herfindahl index. On the other hand, the result that market concentration has no impact on ROE indicates that banks in highly concentrated markets are not able to earn monopoly rents by setting higher than competitive prices. Sixthly, long-term interest rates (LTIR) have little impact on the performance of banks. Although rising bond rates result in higher interest income, it does not affect profitability or funding. Seventhly, it turns out that increased short-term interest rates (STIR) depress ROE, mainly caused by an increased cost level. Finally, economic growth (REALGDP) drives up ROE, which is consistent with the hypothesis that in economically prosperous times, there is increased demand for loans, which may offset the lower loan rate charged by banks that is suggested by the estimation results.
The bottom line in Table 4b is that internet banks have an improved profitability compared to click-and-mortar banks. On the one hand, internet-only banks are able to reduce interest expenses by paying a lower deposit rate. However, this reduction in interest expenses does not result in a higher interest margin, which may provide evidence that the effect on effect on return may be marginal. On the other hand, internet banks reduce their labour expenses. Although this reduction does not result in a lower cost-to-income ratio of internet banks, the combined effect does result in an ROE improvement of 6.31 percentage points.
5.2 Robustness Checks:
In order to check whether the baseline results presented in Table 4a-b hold, a number of robustness checks are performed. Firstly, we re-estimate equation (1) using a different market concentration ratio. Initially, the Herfindahl-Hirschman index is used to control for the effect of market concentration in the banking sector. The HHI is replaced by the C5 Ratio, measured by market shares of the five largest banks in the industry. Also, I use alternative proxies for bank risk, by replacing equity to assets ratio and loans to assets ratio by the nonperforming loan ratio (NPL) and the liquid assets to total assets ratio (LIQASS).
27 smaller, there are no major differences in the key estimation results that would make any significant changes to the conclusion of the regression results.
Furthermore, as indicated previously, internet banks studied in our sample are, on average, smaller compared to their branching counterparts. In order to compare our internet bank observations with comparable click-and-mortar observations, I estimate a slightly different model using Propensity Score Matching. Propensity Score Matching was introduced by Rosenbaum and Rubin (1983) and is a method of facilitating comparability between different groups. The idea behind Propensity Score Matching is to match an internet bank with a click-and-mortar bank based on the assumption that both types have an equal probability of being classified as internet bank.
The process consists of two phases. Firstly, a propensity score is calculated for both internet banks (pi) and for click-and-mortar banks (pj). E.g. Brookhart et al. (2006) argue that
researchers should include all variables that potentially affect the outcome even if they have no relationship to the assignment of the treatment. In my case, this indicates that all variables relating to the outcome should be included independent if they are related to the assignment of being an internet bank. However, due to the balancing property condition of the propensity score, I only include the bank size into the propensity scores. The propensity scores tell us that smaller banks are indeed more likely to be identified as a pure-play internet bank and therefore a correction has been made for this characteristic. Secondly, a matching technique is executed, such that for each observation of internet banks is matched with an observation of click-and-mortar banks based on the size of the bank. In particular, I use the nearest-neighbour matching where a match between an internet bank observation and a click-and-mortar bank observation is selected that minimizes the distance between both propensity scores:
min | 𝑝𝑖− 𝑝𝑗 | (2)
28
Table 5: Panel Estimates. Estimation Technique: GLS-RE. INTERNET is a dummy variable which takes the value of one for banks classified as a pure-play internet bank, and
zero otherwise. The superscript *,** and *** denote significance at the 1%, 5% and 10%, respectively. The reported statistic is distributed as a chi-square with 19 degrees of freedom. In case the Non-performing Loan Ratio is the dependent variable, the degrees of freedom is 18. White-corrected standard errors are shown in parentheses.
Performance Risk Variables Interest Income to Earning Assets Interest Expenses to Deposits Net Interest Margin Noninterest Income to Total Income Labour Expense to Total Assets ROE Cost-to-Income Ratio Non-performing Loan Ratio Wholesale Funding to Total Assets Z-score INTERCEPT -0.0274 0.0105 0.0480 22.31*** 0.0351** -7.8832*** 19.87*** 0.2115 0.4930 -619.76*** (0.0778) (0.0244) (0.0788) (1.0724) (0.0141) (0.8716) (1.2087) (0.1688) (0.3349) (145.22) INTERNET -0.0028 -0.0057* 0.0048 -0.0509 -0.0044*** 0.0527** -0.0310 -0.0002 -0.0231 -10.61 (0.0074) (0.0034) (0.0080) (0.0786) (0.0016) (0.0254) (0.0649) (0.0142) (0.0779) (28.11) ln(ASSETS) -0.0008 -0.0005 -0.0006 -0.0168 -0.0015*** 0.0034 -0.0176 0.0040 0.0704*** -4.1965 (0.0013) (0.0010) (0.0018) (0.0185) (0.0004) (0.0075) (0.0165) (0.0030) (0.0180) (5.0089) NPL 0.0254 0.0183 0.1102** 0.2219 -0.0024 -0.1242 0.2868 0.1582 -6.4987 (0.0239) (0.0196) (0.0509) (0.2402) (0.0069) (0.2173) (0.1993) (0.0980) (20.10) LIQASS -0.0145** 0.0023 -0.0043 0.0756 -0.0041** 0.0663 -0.1920* -0.0216 0.0664 -4.8848 (0.0057) (0.0087) (0.0099) (0.1190) (0.0012) (0.0445) (0.1048) (0.0157) (0.0500) (11.05) MBHC -0.0034 -0.0021 -0.0046 0.0728 -0.0010 -0.0117 0.0182 -0.0019 0.0467 -21.47 (0.0055) (0.0030) (0.0064) (0.0607) (0.0011) (0.0192) (0.0390) (0.0083) (0.0458) (22.61) C5RATIO -0.0313 0.0427 -0.0766 -1.3242*** -0.0242*** -0.5225 0.1176 0.0831 -0.0401 -93.20** (0.0636) (0.0286) (0.1114) (0.4404) (0.0078) (0.4654) (0.5516) (0.1089) (0.2131) (40.12) LTIR 0.6863* 0.1141 0.5744 -5.8608 -0.0112 2.6935 -2.5243 0.2254 1.2437 295.88 (0.3986) (0.2751) (0.5876) (3.8168) (0.0556) (3.1938) (4.9242) (0.3465) (1.4269) (379.00) STIR -31.68 -0.1920 0.1178 7761.15*** -5.2691 -2969.98*** 7004.96*** 89.86 598.63*** -300804*** (38.51) (0.6855) (1.1810) (405.57) (5.1024) (353.26) (501.82) (70.06) (129.51) (33585.94) REALGDP -0.1689 -0.1054 -0.0600 0.5971 0.0059 2.5658** -2.5450 -0.1825 -0.6097 -16.61 (0.1179) (0.0657) (0.1636) (0.9949) (0.0175) (1.1531) (1.7358) (0.1250) (0.4455) (88.54)
Bank Random Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Country dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 433 271 269 433 434 439 427 439 439 435
29
Table 6: Average effect of Treatment on the Treated (ATT) using the Nearest-Neighbour Matching Method. INTERNET is the treatment variable which takes the value of one for
banks classified as a pure-play internet bank, and zero otherwise. The superscript *,** and *** denote significance at the 1%, 5% and 10%, respectively. Bootstrapped standard errors are shown in parentheses. N (Treatment) refers to the number of observations of internet banks. N (Control) refers to the number of observations of click-and-mortar banks.
30 interpreting these results since this matching technique discards a large number of observations, which leads to reduced power of the model. However, this reduction may be acceptable because the used matching technique ensures that very similar groups in terms of bank size are being compared which may, in turn, temper this power reduction.15
Overall, the robustness checks confirm the main regression results that internet banks pay lower deposit rates and have lower labour expenses. However, the increased profitability of digital banks is not robust using the Propensity Score Matching Method, which weakens the evidence for this outperformance of internet banks.
6. Conclusion:
Internet banking has become increasingly important in the business plans of banks throughout the world. Banks adopting a click-and-mortar business strategy provide a network of offices alongside their digital channels. Other banks have adopted the pure-play internet business model by servicing their clients solely by the use of online channels. Strategic value for these internet banks is created by this lack of physical locations, resulting in potential overhead reductions and more attractive prices for consumers. This study attempts to identify whether internet banks are able to benefit from these theoretical sources of value.
Although the advantages of internet banks to realize lower overhead costs are often stressed in previous literature, the empirical evidence that internet-only banks are able to convert this strategic value into higher profits is rather lacking. In contrast, this study provides some evidence that internet banks outperform branching banks in terms of ROE. On the one hand, this profitability gap seems to stem from the ability of internet banks to pay lower deposit rates. On the other hand, the results show that internet banks are able to reduce their labour expenses.
However, there is little evidence that internet banks fully exploit the features of the low-cost business model, indicated by a cost-to-income ratio that does not substantially differ from click-and-mortar banks. Also, there is no conclusive evidence that internet banks have a significantly different risk profile. The result that many aspects between internet banks and branching banks do not differ might be driven by the fact that click-and-mortar banks already run most of their volumes over the internet. Although there is little information on how click-and-mortar divide their banking operations over their two distribution channels, it is in the
15 According to Snedecor and Cochran (1980), comparing more identical groups based matched characteristics
