NET INTEREST MARGIN OF BELGIAN BANKS

Transcription

1 NET INTEREST MARGIN OF BELGIAN BANKS Aantal woorden/ Word count: Bertrand Le Noir Stamnummer/ Student number : Promotor/ Supervisor: Prof. dr. Rudi Vander Vennet Co-promotor/ Co-supervisor: Thomas Present Masterproef voorgedragen tot het bekomen van de graad van: Master s Dissertation submitted to obtain the degree of: Master of Science in Business Engineering Academiejaar/ Academic year:

2 This page is intentionally left blank.

3 NET INTEREST MARGIN OF BELGIAN BANKS Aantal woorden/ Word count: Bertrand Le Noir Stamnummer/ Student number : Promotor/ Supervisor: Prof. dr. Rudi Vander Vennet Co-promotor/ Co-supervisor: Thomas Present Masterproef voorgedragen tot het bekomen van de graad van: Master s Dissertation submitted to obtain the degree of: Master of Science in Business Engineering Academiejaar/ Academic year:

4 i Permission of use on loan PERMISSION I declare that the content of this Master s Dissertation can be consulted and/or reproduced, provided that the source is referenced. Name student: Signature:

5 ii Foreword I would like to thank professor Rudi Vander Vennet and Thomas Present for the guidance while writing my thesis and providing the interesting topic. Furthermore, I would like to thank my parents for giving me all the opportunities that I could have wished for. I am also very grateful for the support that I received from my girlfriend and my roommates.

6 iii Contents List of Figures List of Tables List of Abbreviations v v vii 1 Introduction Role of banks Banking sector Belgium Overview Literature General Ho and Saunders Other studies The relationship between the net interest margin and the interest rate risk, default risk and off-balance sheet activities Saunders and Schumacher The impact of global competition Factors explaining the interest margin in the principal European banking sectors The determinants of commercial bank interest margins Net interest margin on a macro level The determinants of bank interest margins in the Central and Eastern European countries The determinants of bank margins in European banking Determinants of Interest Margins in Colombia Conclusion Methodology Approach Econometric Pooled ordinary least squares Fixed effects least squares Random effects least squares

7 CONTENTS iv 4 Data Independent variables Cost income ratio Liquidity Size Credit risk Asset structure Risk aversion Diversification Summary Dependent variable Descriptive statistics General information Results Econometric issues Parameter estimates The banks as fixed-effect The years as fixed-effect Robustness Explanation Conclusion General conclusion Practical implications Limitations Future research A Mathematical model equity to assets 40 B Descriptive statistics 41 C Correlation 45 D Banks 47 E Regression results 48

8 v List of Figures 1.1 Balance sheet breakdown of Belgian retail banks (assets and liabilities) based on balance sheet data retrieved from Bankscope Natural logarithm of total assets: large banks versus small banks in Belgium (averages) Evolution of 3 month euribor and the median of the net interest margin of the Belgian retail banks Yield curve basis for measuring three components of net interest margin (Hempel and Simonson, 1999, p. 89) Difference between pooled OLS regression and fixed effects least squares dummy variable (LSDV) model Mean of the net interest margin over the different banks Mean of the net interest margin over the different years Correlation between the different variables. When crossed the relation is not significant on a 20% level Net interest margin (fitted values) in function of natural logarithm of total assets. For each bank separately and for the Belgian retail banks as a whole Evolution of the natural logarithm of size and loan to assets ratio C.1 Variables plotted against Net Interest Margin (1) C.2 Variables plotted against Net Interest Margin (2)

9 vi List of Tables 4.1 Independent variables: overview Model fit performance figures Lagrange multiplier test for random effects Redundant fixed effects test Fixed-effect (banks) LSDV regression with net interest margin as dependent variable Fixed-effect (years) LSDV regression with net interest margin as dependent variable Economic relevance of the net interest margin determinants Economic relevance of the net interest margin determinants B.1 Cost income ratio (in %) B.2 Net loans to deposits and short term funding (in %) B.3 Market Share (in %) B.4 Natural logarithm of amount of total assets B.5 Loan Loss Prov/Net interest revenue (in %) B.6 Loans to assets (in %) B.7 Equity to assets (in %) B.8 Total non-interest income to total income (in %) B.9 Net interest margin (in %) B.10 Spread (in %) B.11 Overview variables C.1 Correlation Matrix E.1 Cross-section fixed effects LSDV regression without net interest margin outliers. 48 E.2 Period fixed effects LSDV regression E.3 Cross-section fixed-effects LSDV regression with net interest margin as dependent variable E.4 Cross-section fixed-effects LSDV regression with spread as dependent variable.. 51

10 vii List of Abbreviations CIR ETA LLP LTA MS NBB NIM NLTST OECD cost income ratio equity to assets loan loss provision to net interest revenue loan to assets market share National Bank of Belgium net interest margin net loans to deposits and short term funding Organisation for Economic Co-operation and Development

11 1 1 Introduction 1.1 Role of banks The primary role of banks is to accept deposits and grant advances. Doing so they serve both borrowers and lenders. As a result of performing this role they receive the title of financial intermediary. These two main activities play an important role in the financial system and economy. A bank is able to minimize the transaction costs and information asymmetries that otherwise would occur. By exercising their role banks perform several transformations which are of vital importance, namely size transformation, risk transformation and maturity transformation. The first transformation has to do with the fact that a lot of short term deposits can be used to serve one big long term loan. The second transformation deals with the fact that banks are able to better cope with loan defaults compared to individuals. The final transformation handles the fact that there is a maturity mismatch between lenders and borrowers. Banks borrow mostly short term but lend out long term.(casu, Girardone, and Molyneux 2006) Banks get their funds from deposits or the money-market. The obtained funds are used to lend out money to individuals or institutions who require these funds at that time. The price paid or received is expressed by the interest rate a bank sets. When performing this role banks are faced with a certain amount of costs and risks. The transformations mentioned before bring along a liquidity risk, credit risk and interest rate risk. These risks and costs need to be compensated and the higher these are the higher the required remuneration will be. The compensation to fulfill their primary role can be expressed as the net interest margin. Retail or personal banking relates to financial services provided to consumers and is usually small scale in nature (Casu, Girardone, and Molyneux 2006). Many banks have a distinct retail banking division. It is common for large banks to have a retail banking department but also other smaller types of banks often perform retail banking activities. Consumers and small businesses are the core customers of retail banks. Besides the deposit and loan activity, retail banks also offer other products. Many banks try to diversify their income sources but there is no consensus about the effect on the profits for banks.

12 Introduction Banking sector Belgium When looking at some figures published by the National Bank of Belgium we notice a shift of activities since the banking crisis. Belgian banks are focusing more on the traditional and domestic business activities. This makes the net interest margin, which is the topic of this paper, more important. Due to this reallocation Belgian banks were able to de-risk and de-leverage their balance sheets. The total balance sheet of 2015 amounted to 970 billion euro. Figure 1.1: Balance sheet breakdown of Belgian retail banks (assets and liabilities) based on balance sheet data retrieved from Bankscope. Figure 1.1 shows the breakdown of assets and liabilities of the Belgian banking sector. It clearly shows the diminution of the overall balance sheet size and the reduction of derivatives, interbank activities and debt securities. This de-levering was mainly done by the bigger Belgian banks, the smaller banks kept growing (figure 1.2). In line with the de-risking and de-levering Belgian banks reduced their foreign activity heavily. Due to their focus on the domestic market, the amount of deposits and loans went slightly up. (Loose, Neuville, Pauwels, and den Bruel 2006) Figure 1.2: Natural logarithm of total assets: large banks versus small banks in Belgium (averages)

13 Introduction 3 In theory the decline in interest rates should undermine the margins of banks. On the one hand the funding costs should decline although the saving deposits are only partially dependent on the money market rate. But on the other hand, long term loans and securities reach their maturities or are refinanced. This lowers the compensation for the earning assets. The decline in interest rates (figure 1.3) had a positive/neutral 1 influence on the net interest margin of the Belgian banks due to the fact that the funding costs went down faster than the return on interest-earning assets. But one needs to keep in mind that this effect is temporarily. Belgian retail banks lowered their interest rates on deposits and are reaching the lower boundaries. Until now they were able to keep the margins on new loans at a reasonable level but if these low interest rates would stay in place the net interest margin will drop according to the National Bank of Belgium. So a profound understanding of the net interest margin of the Belgian retail banks is in place. Certainly because the net interest margin stays the most important source of income for banks. (Hilgers 2015) Figure 1.3: Evolution of 3 month euribor and the median of the net interest margin of the Belgian retail banks 1 This depends on the summary statistic (table: B.9).

14 Introduction Overview The remaining of the paper is divided in five sections. Section 2 will deal with the net interest margin in the literature. The focus will lie on the model of Ho and Saunders (1981). Section 3 will clarify the methodology of our research and the underlying econometric models. Section 4 will disclose all variables which will be used and their link with the literature. Section 5 discusses the results and the interpretation of the coefficients. The final section will present a conclusion with the most relevant results. Section 2: Literature Section 3: Methodology Section 4: Data Section 5: Results Section 6: Conclusion

15 5 2 Literature 2.1 General The net interest margin is composed of three spreads (Hempel and Simonson 1999). Credit spread, interest rate spread and funding spread. In figure 2.1 (Hempel and Simonson 1999) you can clearly see the components. Credit spread compensates for the time that a bank needs to obtain and analyze all the information of the counterpart. They have to check the creditworthiness and have to execute all the administrative tasks. The default risk is carried completely by the bank and is also included in this spread. Due to the fact that there is a mismatch between loans and deposits banks face a certain interest rate spread. If interest rates would rise, the short term funding costs would eventually rise with them and this would cost the bank money. Figure 2.1: Yield curve basis for measuring three components of net interest margin (Hempel and Simonson, 1999, p. 89)

16 Literature 6 There are done a lot of studies on the determinants of the interest margin of banks. These studies focus on different risks and start from different levels. When we look at the multiple levels for which there can be opted we distinguish four levels. Hawtrey and Liang (2008) explain the four levels in their study about the net interest margin in OECD countries. The micro level examines the difference between business divisions of banks. The meta level uses the banking conglomerate as subject to analysis. Next is the macro level, this looks at the country level. And finally we can find the continental approach where economic regions are the subject of the investigation. Several authors mentioned the importance of the net interest margin for the economy. The net interest margin is an important figure when we look at the efficiency of the financial intermediates. It also plays an important role for the growth rate of the economy. The net interest margin influences the return on savings and the return on investments. 2.2 Ho and Saunders The starting point of a lot of researches in the field of bank interest margins is without question written by Ho and Saunders (1981). There were two major models of bank behavior at the publication date of their paper (November 1981). First of all we have the hedging hypothesis. This model assumes that banks make an attempt to match the maturities of both assets and liabilities. In practice this is almost impossible so banks have to cope with a certain refinancing or reinvestment risk. Banks generally have a lot of short term liabilities and a lot of long term assets. The second model is based on the microeconomics of the bank. According to this model banks focus on profit maximizing or maximizing the expected utility of profit. This second model is the focus of a lot of studies. Ho and Saunders saw the bank as a dealer or intermediate that exercises their primary role (accept deposits and grant loans). Due to the fact that loans (outflows) and deposits (inflows) arrive at different moments, the bank demands a positive interest spread or fee to cope with this uncertainty. The decision period of the model is set at one single period. The purpose of their model is to find the factors that maximize the utility of profit by optimizing the spread. The optimal spread (later referred to as pure spread) according to Ho and Saunders is: s = (a + b) = α β Rσ2 l Q (2.1) The first term gives us the spread demanded by a risk neutral bank. R gives us the risk aversion of the bank, σ 2 l gives us the variance of the interest rate on deposits and loans and Q is the bank s transaction size. This function also implies that bank margins will always be positive as long as banks are risk-averse. Ho and Saunders assumed that this pure spread is equal across

17 Literature 7 banks for which the market structure in which they operate in similar and when the risk aversion and interest rate risk is close the equal. They tested this theoretical model with an empirical model. The margin that is caused by the transaction uncertainty is called the pure spread (β 0 ). In order to arrive at the actual interest margin (M) certain imperfections need to be taken into account. The first one is the implicit interest payments (IR). Second, the bank s opportunity cost of holding reserves (OR). Third, the default premiums on outstanding loans (DP ). All the other possible imperfections are included in the error term. M = β 0 + β 1 IR i + β 2 OR i + β 3 DP i + υ i (2.2) Their results state that the interest margin is mainly influenced by the transaction uncertainties and implicit interest payments. They did some additional research on the impact of different balance sheet structures. They divided their sample in large and small banks to cover differences in structure. One of the conclusions was that larger banks have to deal with smaller pure spreads. A possible explanation for his result could be that larger banks seem to operate in a more competitive market framework compared to smaller banks. Smaller banks are also able to exploit their regional monopoly positions. 2.3 Other studies This first important study on the subject of net interest margins had some shortcomings. A lot of studies started from the model of Ho and Saunders (1981) and added some explanatory variables which had a significant influence and were omitted in the initial model. Others tested the initial model of Ho and Saunders. McShane and Sharpe (1985) assessed the relationship between the pure spread and the observed spread. They concluded that the model from Ho and Saunders had some empirical relevance for the Australian banks. Allen (1988) extended the Ho and Saunders model by adding diversification between the loans. Different types of loans were incorporated in the model. The probability of a loan arrival depends now also on the interest rates of other products. λ(loan m) = α m β l b m + β n b n (2.3)

18 Literature The relationship between the net interest margin and the interest rate risk, default risk and off-balance sheet activities Angbazo (1997) investigated the relationship between the net interest margin and the interest rate risk, default risk and off-balance sheet activities. This study extends the Ho and Saunders model by including default risk and the interaction with interest rate risk. Although he extended the Ho and Saunders model this study is not an empirical test of the Ho and Saunders model. The interaction is included because there is strong evidence that the interest margin of large banks is influenced by default risk but not by interest rate risks. The latter conclusion is due to the large amount of off-balance sheet hedging instruments that larger banks have at their disposal. The contrary conclusion can be made for the smaller regional banks. The net interest margin is a function of the desired interest spread (S it ) and bank specific factors (X it ). NIM it = F (S it(.), X it, ɛ it ) (2.4) The desired interest spread compensates for the bank inventory risk caused by the uncertain deposit and loan arrivals. A lot of studies investigate the imperfections that cause a lower or higher net interest margin. The intercept in their equation is often viewed as the pure spread from Ho and Saunders. A two step approach is mostly followed Saunders and Schumacher Saunders and Schumacher (2000) applied a two-step approach to empirically test the model from Ho and Saunders. In the first step they make an attempt to isolate the pure spread. They do this by including certain imperfections such as regulatory forces, credit risk and operational inefficiencies. In the second step they use cross-country and time-varying determinants to explain the variations in the pure spreads. The important assumption that they made is that the pure spread is equal across banks in any country in any given year.

19 Literature The impact of global competition Gischer and Juttner (2003) also started from the model of Ho and Saunders. The focus of their study lies on the impact of global competition. They included two variables to capture other factors than the pure spread that could influence the actual interest margin. First of all managerial efficiency and secondly implicit interest payments. Gischer and Juttner also included risk variables. They stated that riskier banks need to compensate their stockholders with additional return and this additional return involves increasing the margins. Although they mentioned that one has to be cautious when using different risk rankings Factors explaining the interest margin in the principal European banking sectors Maudos and De Guevara (2004) analyzed the factors explaining the interest margin in the principal European banking sectors. They focus on the banks operating costs and the degree of competition (Lerner Index). They extended the model from Ho and Saunders by including operating costs and credit risk in the formula for the pure spread. They opted for a single-stage methodology where all variables, both of the theoretical model of Ho and Saunders and the additional variables of imperfections are included. The reasoning behind this decision is that a two-stage approach requires a time-series that is long enough and their study only focuses on seven years. According to them this relatively shorter time-span makes it impossible to conduct a two-stage approach. The econometric approach followed by Maudos and De Guevara (2004) consists of introducing several fixed effects. They capture the individual, time, country and bank-type effects by introducing dummy variables The determinants of commercial bank interest margins Demirgüç-Kunt and Huizinga (1999) used bank level data from 80 countries to determine the determinants of commercial bank interest margins. They considered two spreads. An ex ante spread and an ex post spread. The difference is that ex ante spread looks at the contractual rates and ex post spread at the actual expenses and revenues. They opted for the ex post spread. They used a variety of determinants such as: bank characteristics, macroeconomic conditions, deposit insurance regulation, overall financial structure and underlying legal and institutional indicators. They included country and year fixed effects in their regression. When determining the factors influencing the net interest margin they also included an interesting dummy variable. This variable indicated whether or not the bank is foreign owned or domestic owned. When a bank has foreign ownership the net interest margin is expected to be higher. This was confirmed by their results.

20 Literature Net interest margin on a macro level Hawtrey and Liang (2008) investigated the net interest margin on a macro level (country level). The reason for examining the net interest margin on a macro level was caused by the assumption that different banks in a country act like one. The assumption is that they more or less mimic each other. The econometric approach which was followed consists of using panel data. In their study the following models are considered: pooled regression model (PRM), fixed effects model (FEM), random effects model (REM) and generalized least squares (GLS). They started with a pooled least squared estimation and added the cross-section and period fixed effects later on. These effects improved the model and were included in the final model. Finally they opted to deploy a model with the GLS estimator because this assumes heteroscedastic and autocorrelated error terms The determinants of bank interest margins in the Central and Eastern European countries Claeys and Vander Vennet (2008) investigated the determinants of bank interest margins in the Central and Eastern European countries. They investigated determinants on three dimensions, time-variation, cross-country variation and bank-level variation. They mention an interesting theory which can influence our study. The relative-market-power hypothesis states that banks with a lot of market power are better suited to reach high interest margins. The results of their study did not confirm this hypothesis for Western Europe. Next to this we also have the efficient-structure (ES) hypothesis. The X-efficiency (X-ES) version implies that banks which are more efficient have lower cost and can offer more competitive interest rates. This efficiency measurement is a function of total costs, total loans, total other earning assets, personnel expenses, price of labor, price of funds and bank capital. This measurement ranks a bank as inefficient when it has higher costs compared to other banks with the same input and output characteristics. The scale-efficiency (S-ES) version states that banks which produce at a more efficient scale will be able to demand smaller margins. This figure is a function of bank capital, total costs and total loans and other earning assets The determinants of bank margins in European banking Valverde and Fernández (2007) investigated the determinants of the bank margins in European banking. Their paper not only focuses on the classic interest margin but assumes a more broader definition of bank margins. Also non-traditional assets are included in their model. They stated that diversification and additional income sources could explain the combination of a decrease in

21 Literature 11 interest margins and an increase of market power 1. According to various models this behavior is in theory not possible 2. A diversified bank is able to undercut a specialized rival s prices and force him out of the market while pertaining viable income sources. Their bank margin is a function of market power, risk variables, bank specific variables (solvency and efficiency), specialization, diversification and macroeconomic and regulatory control variables. To arrive at their final regression equation they toke the first difference of their initial equation. This removes any firm-specific effect. Finally they applied a two-step GMM (Generalized Method of Moments) estimator Determinants of Interest Margins in Colombia Estrada, González, and Hinojosa (2006) analyzed the determinants of the interest margins in Columbia. They based their model on the model of Ho and Saunders and saw the interest margin as a function of bank-specific institutional imperfections and a pure spread. The model extends the model of Ho and Saunders by letting interest rates of competitors influence the probability of loan and deposit arrival. They also included a value for the market power in their formulas for the arrival probability of loans and deposits. A high market power can be caused by reputation, size or other determinants and implies that a bank can charge higher rates. The pure spread is modeled as a function of market power, interest rate volatility and credit risk exposure. Their paper is similar in purpose compared to our paper because only microeconomic determinants are included to capture the differences between the credit institutions. Their model includes a two step estimation approach. This means that the interest margin is first estimated using an unbalanced fixed effects panel data with a time-varying intercept. In the second step, the estimated pure spread (which was the intercept in step 1) is regressed against interest rate volatility and market power. So this paper not only focuses on the determinants of the net interest margin but also on the determinants of the pure spread. The pure spread in their paper is defined as the fraction of the interest margin which is not affected by certain institutional imperfections. They made the assumption that the interest margin of a bank is a function of the pure spread plus additional institutional imperfections. This pure spread is equal across banks when the assumption is made that the banks have a similar market structure, risk aversion and interest rate volatility. In theory the pure spread is calculated as below: s i = N (N 1) 2α β + ρ i Qσ2 I (1 + r) (2.5) 1 This relationship was founded by Maudos and De Guevara (2004) 2 Subsection contains an equation (2.5) that demonstrates the expected behavior.

22 Literature 12 α β gives us the elasticity of supply and demand. A high value for this ratio results in a more inelastic supply and demand function. ρ i gives us the risk aversion of the bank, σl 2 gives us the variance of the interest rate on deposits and loans and Q is the bank s transaction size. N gives the number of institutions. The first term indicates the market power or competitiveness of the banks. The formula states that an increase of market power for the banks as a group results in a higher pure spread. The institutional imperfections that need to be taken into account are the credit risk, opportunity costs, efficiency and other intermediation costs. The assumption made in their paper is that the pure spread varies over time but not across banks. They measured the net interest margin using the following formula: Interest Income P erforming Loans Interest Expense Cost Bearing Liabilities (2.6) The model is tested using following formula: IM it = θ ot + θ 1 CR it + θ 2 OC it + θ 3 Eff it + θ 4 Com it + ɛ it (2.7) All variables were found to be significant. Credit risk, operational costs and reserves had a positive effect on the interest margin. Net commission had a negative effect. 2.4 Conclusion A lot of options are available to conduct an investigation into the determinants of the net interest margin. - There are four different levels on which we can conduct a research. We distinguish the micro, macro, meta and continental level. There is also the possibility to combine several levels. - There are two approaches to determine the factors that influence the net interest margin, a single-stage approach or a two stage approach. - We need to choose the different imperfections and the proxies to express them. There is an extensive amount of variables which are already utilized in past studies. - The econometric approach also needs some attention. When working with pooled data several methodologies are at our disposal 3. Another possibility is to first-difference the variables, this way any individual specific effect is taken care of. 3 Fixed effects model, Random effect model and pooled OLS regression.

23 13 3 Methodology 3.1 Approach We have seen that the spread can be calculated by adding up the margins set on loans and deposits (a + b). The purpose of this paper is to determine the factors influencing this total spread for the Belgian retail banks. Research Question 1. Which determinants influence the net interest margin of the Belgian retail banks and in which direction? We will use two different formulas to represent the net interest margin. One with average earning assets as denominator and one with both average earning assets and average bearing liabilities as denominators. We will discuss these formulas in more detail in section 4.2. The dependent and independent variables are extracted from Bankscope. We extracted all data included in the annual reports of the banks mentioned in appendix D. The data is extracted for the years We opted for these years because previous years would not have given a realistic and representative view on the influencing determinants due to the banking crisis in Europe. The basic model resembles the following formula: k=n s t i = x 0 + α k x t k,i k=1 x k represents the different factors that will be taken into account. n is the amount of variables. t is the time period which varies between 2009 and i indicates the bank. We will examine the net interest margin on a meta level (bank level) in Belgium. The focus lies on banks for which the net interest margin is a very important income source, namely retail banks. Due to the limited amount of banks we opt for a single stage approach where all explanatory variables are investigated in one regression. The focus will lie on pooled panel data regression. More advanced regressions will be performed and compared with the basic model and with each other. This study does not test the model of Ho and Saunders (1981) nor any extended model. The purpose is to get an insight into the different determinants that influence the net interest

24 Methodology 14 margin. A lot of these determinants are already examined in papers that did use the model of Ho and Saunders (1981) but that is not the scope of this paper. To the best of our knowledge this is the first paper examining the determinants of the net interest margin for the Belgian banks. 3.2 Econometric We have three possible regression types available to use for our empirical analysis (Gujarati and Porter 2009). Time series, cross-section and panel data. There are several advantages that result from using panel data (Hsiao 2007): - Better suited to address to complexity of human behavior. - Panel data can ensure that the differences per year and the differences per entity limit possible shortcomings originated from omitting variables. - Most economic relationships are dynamic. Panel data reduces the collinearity between the explanatory variables due to the availability of different entities. - If individuals are assumed to behave similar to certain triggers we can use this behavior to learn more about the behavior of others. This results in more accurate and efficient models. - Minimize the bias. Panel data simplifies computations compared to time-series and crosssections. Ability to perform transformations on the variables. Our panel data could be categorized as a short balanced panel, this means that there are more cross-sections than periods of time (years) and that every year and bank has the same amount of observations. Within panel data we can choose between pooled ordinary least squares (OLS) model, fixed effects least squares dummy variable (LSDV) model and the random effects model (REM). We will make use of each of these modeling techniques. We perform these additional regression models because we assume that there will be a lot of heterogeneity between the different banks and (to a lesser extent) between the different years. This problem is graphically displayed in the graph below with random data. If we would perform a pooled OLS regression we get the black regression line. If instead we would perform a fixed effects least squares dummy variable (LSDV) model with the banks as dummy variables we would get the blue lines, which are visually more correct. We can extent this model to cope with both the heterogeneity between the banks and years. We can test whether the basic pooled OLS model is better than the model with the fixed effects.

25 Methodology 15 Hypothesis 1. There is a significant heterogeneity across the different banks. Hypothesis 2. There is a significant heterogeneity across the different years. Figure 3.1: Difference between pooled OLS regression and fixed effects least squares dummy variable (LSDV) model Pooled ordinary least squares NIM t i = β 0 + β 1 CIR t i + β 2 NLT ST t i + β 3 LLP t i + β 4 LT A t i+ β 5 ET A t i + β 8 size t i + β 8 diversification t i + υ t i (3.1) Due to the fact that our sample is rather small and the advantages that pooled data brings along we will perform a pooled ordinary leased squares regression. By pooling the data we ignore the fact that the data is composed of different banks. As long as the variance of the standard errors is equal (homoscedastic) and as long as the standard errors are not correlated, the ordinary least squares estimator is the most efficient estimator. When using panel data the possibility of encountering cross-sectional heteroscedasticity, correlation among cross sections or serial correlation within and across cross-sections is considerable. We expect that that there will be a lot of heterogeneity (hypothesis 1), this heterogeneity will be captured in the error term and will result in heteroscedasticity. The OLS estimator will not remain optimal if one of the previous mentioned problems would arise. When coping with these problems we can choose between several different solutions. We could use Feasible Generalized Least Squares (GLS), this option was also used by Hawtrey and Liang (2008). We expect that this heterogeneity will need to be taken care of by use of a fixed-effects model or a random-effects model.

26 Methodology Fixed effects least squares NIM t i = β 1 CIR t i + β 2 NLT ST t i + β 3 LLP t i + β 4 LT A t i + β 5 ET A t i+ β 6 size t i + β 7 diversification t i + α 1 D t 1,i α 20 D t 20,i + υ t i (3.2) Due to the possible shortcoming of not taking the different banks into account we make some adjustments compared to the previous model. We can test the bank effect by applying a Lagrange multiplier test. The imported dummy variables take care of the unobserved heterogeneity between the different banks. In order to perform a fixed effects model every bank should be unique in some sort of way and this uniqueness should impact the net interest margin. In order to obtain estimators that are BLUE we need to meet the following two assumption: - The explanatory variables are non-stochastic, independent of the errors. - The random terms are independent, homoscedastic with zero mean Random effects least squares When using the random effects model we capture the ignorance through the error term. This is composed of the individual differences between banks and the idiosyncratic error term. NIM t i = β 1 + β 2 CIR t i + β 3 NLT ST t i + β 4 LLP t i + β 5 LT A t i + β 6 ET A t i+ β 7 size t i + β 8 diversification t i + ω it (3.3) ω it = ɛ i + υ it (3.4) There are some assumptions that we need to make when performing REM. E(ɛ i ) = 0 E(ɛ i ɛ j ) = 0 E(ɛ 2 i ) = σ 2 ɛ E(υ it ) = 0 E(υ t iυ s j ) = 0 E(υ 2 it) = σ 2 υ E(ɛ X t i ) = 0 E(υ t ix t i ) = 0 Within cross-sections correlation of error terms is accepted. An important assumption is that the individual error term needs to be uncorrelated with the explanatory variables. If this would be the case the random effect estimator is inconsistent. To test this assumption we can use the Hausman test. The null-hypothesis is that there is no correlation between the error term and the explanatory variables. In this case there is no difference between the fixed estimator and the random effects estimator. Another important assumption of the REM is that the sample is a random sample of a much bigger population.

27 17 4 Data 4.1 Independent variables Cost income ratio The cost income ratio is an important measure for productivity and efficiency. Productivity can be expressed as the relation between output and input. The term efficiency is harder to explain. A lot of definitions follow the Pareto-Koopmans concept: Full (100%) efficiency is attained for an object [... ] if and only if none of its inputs or outputs can be improved without worsening some of its other inputs or outputs (Cooper, Seiford, and Zhu 2004). A company with a high cost income ratio will be seen as a company with low productivity or efficiency. This ratio is a widely used measurement to compare companies (and banks in our case) because its simplicity and intuitiveness. We expect that the net interest margin will increase when the cost income ratio is high due to the fact that banks need to compensate for the low productivity with a higher margin. This is also the reasoning followed by Gischer and Juttner (2003). Angbazo (1997) and Maudos and De Guevara (2004) follow a different viewpoint. They assume that a bank with a low cost income ratio will have a high-yield and low-cost composition of assets and liabilities due to their high quality administration. Estrada, González, and Hinojosa (2006) concluded that a bank s inefficiency is the most important micro-economic factor that influenced the net interest margin. Higher efficiency resulted in lower net interest margins. They stated that this conclusion was important for policy makers in order to improve market conditions for banks. We can also make a reasoning with the ratio as starting point. Because the net interest margin is an important part of the earnings, a low net interest margin will result in a low denominator and a high cost income ratio. When we take a look at the average cost income ratios between 2009 and 2015 we notice a steady drop since 2011 indicating an increasing productivity across the banks (table B.1). In appendix C we take a look at the relationship between the net interest margin and the cost income ratio, we observe a slightly negative correlation. We calculate the cost income ratio by applying the following formula: CIR = T otal Non Interest Expenses Net Interest Income + T otal Non Interest Operating Income (4.1)

28 Data 18 Due to the slightly negative correlation and the different empirical results we assume that a higher cost income ratio will result in a lower net interest margin. Hypothesis 3. We expect that a higher cost income ratio results in a lower net interest margin Liquidity An important liquidity ratio is the net loans divided by deposits and short term funds. NLT ST = Net Loans Deposits + Short T erm F unding (4.2) From 2009 until 2015 this ratio showed a steady increase (table B.2), probably due to the refocus of the Belgian banks to the domestic market. We expect that this ratio will be positive, a higher ratio will bring along a higher liquidity risk and this risk needs to be compensated. Valverde and Fernández (2007) also used a liquidity ratio (as determinant of the pure spread) and used liquid assets divided by short term lending. They follow the reasoning of Angbazo (1997), more liquid assets will give a higher opportunity cost and this is charged to the customer. This would indicate a negative sign in our case. The lower the ratio, the higher the amount of reserves on the balance sheet and the higher the opportunity costs. We assume that the liquidity risk will be more prominently present than the opportunity costs. In appendix C we notice a positive relationship between the two variables which gives us a first indication of the direction of the relationship. Hypothesis 4. We expect that a higher net loans to deposits and short term funding ratio will result in a higher net interest margin Size Market share is an important measurement in this study because we compare all Belgian retail banks. We focus on one market. We expect that a higher market share will result in a higher net interest margin. With this assumption we follow the relative-market-power hypothesis, which was also used in the previously mentioned paper of Claeys and Vander Vennet (2008). Three explanations are given in Market share - A key to profitability (Buzzell, Gale, and Sultan 1975). According to this paper a higher market share results in (1) economies of scale, (2) market power and (3) a higher quality of management. We will express this variable by dividing total assets of bank i at time t by the total amount of assets for all banks at time t. MarketShare t i = T otal assetst i T otal assets t (4.3) We observe a negative correlation with the net interest margin (appendix C). This would indicate that larger banks operate in a more competitive framework than the smaller more regional banks.

29 Data 19 We will also include a variable to capture the size. A bigger size will subsequently result in a bigger market share. Size = ln(t otal assets) (4.4) We assume that a higher market share will result in more market power and the ability to charge higher net interest margins. The same reasoning is followed for the size of a bank. Due to the high correlation between these two variables they will not be used in the same regression (table C.1). Hypothesis 5. We expect that a larger natural logarithm of total assets will result in a higher net interest margin. Hypothesis 6. We expect that a higher market share will result in a higher net interest margin Credit risk We expect that credit risk will result in a higher net interest margin. Due to the higher risk the bank will demand a higher margin for this exposure. Credit risk is present due to the fact that a bank has a certain amount of non-performing loans and defaults. Credit risk can be expressed in different ways, some example are: loans divided by total assets, loan default divided by total loans (lagged value), capital to assets ratio, etc. The different formulas result from different starting points in the reasoning. When we look at loans divided by total assets, a certain default percentage which is always present will result in a higher risk when this ratio is higher. Capital to assets will give you a proxy on how easy a bank can withstand a certain percentage of defaults. This ratio could also be seen as a proxy for risk aversion. The amount of loan loss provisions shows how much a bank expects to lose due to loan defaults. The higher this figure is, the more expected losses are present. A higher net interest margin could be used to compensate for this risk. Negative loan loss provisions occur when the quality of loans is improving or when there are lower loss rates. We will use the following formula: Loan Loss P rovision N et Interest Revenue (4.5) Hypothesis 7. We expect that a higher loan loss provision to net interest revenue ratio will result in a higher net interest margin.

30 Data Asset structure An interesting ratio included by Claeys and Vander Vennet (2008) is the loans to total assets ratio. T otal loans T otal assets This is calculated by dividing the total amount of loans by assets. Claeys and Vander Vennet (2008) assumed that a higher ratio resulted in a higher interest margin. This assumption was made because they reasoned that a higher ratio will be associated with increased risks and costs. Loans are seen as the most risky and highest yielding earning assets. Their assumption was also confirmed by their results. This measurement was also used by Demirgüç-Kunt and Huizinga (1999). They found a positive relationship with the net interest margin. A different reasoning was followed by Valverde and Fernández (2007). They argued that more loans result in more specialization which in turn leads to higher efficiency. And as mentioned before a higher efficiency results in lower margins. Another aspect that they address is the fact that by the growing number of loans the bank misses other technological learning opportunities related to diversification. assets ratio and the net interest margin. (4.6) Their result indicated a negative relationship between the loan to Loan to assets increased the last years indicating that banks are, as already mentioned, focusing more on their core business (see table B.6 and figure 1.1). This ratio gives us information not only about the credit risk but also about the diversification grade and the asset structure of the bank. Hypothesis 8. We expect that a higher loan to assets ratio will result in a higher net interest margin Risk aversion Wong (1997) found that risk averse banks have a higher optimal interest margin compared to risk neutral banks. Different proxies can be used to express the degree of risk aversion. Examples could be: equity divided by total assets ratio and securities plus other assets divided by volume of loans (Hawtrey and Liang 2008). We will use the following ratio: Equity T otal assets (4.7) A higher ratio will indicate a more risk averse bank and we assume that this will cause a higher net interest margin. The cost that the additional equity requires needs to be compensated by a higher net interest margin. Equity to assets increased the last years steadily (see table B.7). This increase is mainly caused by the implementation of the new basel 3 rules which are more

31 Data 21 strict compared to basel 2. This variable is also included by Saunders and Schumacher (2000) but they used capital instead of equity. Due to the fact that for our sample the total capital to assets ratio is not always at our disposal we opted for equity to assets as proxy. Demirgüç-Kunt and Huizinga (1999) also included this variable but used a lagged value for the total assets. Valverde and Fernández (2007) found that a higher capital assets ratio results in higher interest margins. They stated that banks demand a premium as a result of the solvency regulation on a bank s lending activity. Claeys and Vander Vennet (2008) found a positive relationship between the capital ratio and net interest margin. A higher capital ratio has two advantages according to them. First of all holding additional capital in addition to the amount required results in trust from the depositors. Secondly, higher buffers allow the bank to lend out funds to more varying risk profiles. We are able to support the assumption of an increasing net interest margin when the equity ratio rises with a mathematical model (Vander Vennet 2015). The derivation can be found in appendix A. The final equation shows that an increase in the capital ratio results in an increase of the spread. The required return adjusted for reserves is assumed to always be larger than the return on deposits. k present the reserves deposit ratio and e represents the equity to assets ratio. s e = r E(1 k) r D > 0 (4.8) Hypothesis 9. We expect that a higher equity to total assets ratio will result in a higher net interest margin Diversification Valverde and Fernández (2007) introduced numerous diversification variables. They assumed that a higher diversification grade would result in a higher net interest margin. The reasoning is that the more a bank is focused on their core business the more specialized and efficient this bank is. As seen before, an increase in efficiency should result in lower net interest margins. This assumption was confirmed by their results. Gischer and Juttner (2003) used a fee income variable. They made a different assumption as Valverde and Fernández (2007) and assumed that a higher fee income, in other words a higher diversification grade, leads to a broader range of activities hence a less risky bank. These additional activities will result in less credit risk and more cash flow certainty. The downside are the lower margins on these activities. Demirgüç- Kunt and Huizinga (1999) also included a ratio that expressed the diversification of a bank. They introduced non-interest earning assets to total assets. The sign of this variable in the regression is negative which means that a higher grade of diversification leads to a lower interest margin. We will assume that a higher diversification grade results in a lower net interest margin.

32 Data 22 We include the following variable: T otal non interest income T otal income (4.9) Both terms are net values. The ratio can take value below 0% and above 100%. Because a ratio between 0% and 100% is preferred we transform the variables to values between 0% and 100%. We will test both ratios. This transformed variable will be named DIV1. Hypothesis 10. We expect that a higher total non-interest income to total income will result in a lower net interest income Summary Variable Formula Expected sign CIR NLTST MS T otal non interest expenses Net interest income+total non interest operating income - Net loans Deposits and short term funding + Assets bank i T otal bankassets + Size ln(t otal assets) + LLP LTA ETA DIV Loan loss provision Net interest revenue + Loans T otal assets + Equity T otal assets + T otal non interest income T otal income - Table 4.1: Independent variables: overview

33 Data Dependent variable As mentioned in section 3.1 we will use two different formulas to represent the net interest margin. First of all we have the following formula: Net interest margin = Net interest income Average earning assets (4.10) The last seven years we noticed that the median of this ratio was steadily increasing (see table B.9). Net interest income is calculated by subtracting two terms. The first one (gross interest and dividend income) is calculated as the sum between interest income on loans, other interest income and dividend income. The latter one (total interest expense) is calculated by adding up the interest expenses on customer deposits and other interest expenses. Average earning assets is the average of the earning assets of the previous year and the current year. Average earning assets is a better measurement compared to total earning assets because the net interest income is the result of the entire year and the denominator needs to take this into account. The second formula used is the following one: Spread = Gross interest and dividend income Average earning assets + T otal interest expense Average bearing liabilities (4.11) This interpretation of the net interest margin will we call the spread in the remaining of this paper. This formula corrects for banks which have much less interest bearing liabilities. If the amount of interest bearing liabilities would be very low and the amount paid for these liabilities is very high we would get two very different results for the interest margins. The net interest margin would not be influenced very much compared to the spread which would undergo a downwards move. The average spread fluctuated around 2% the last seven years (see table B.10).

34 Data Descriptive statistics The reader can find the descriptive statistics in appendix B. An overview of the correlation between the variables is included in appendix C. Two graphs are produced to get an initial view on the relationship between the net interest margin and the different years and banks. These graphs give a first insight whether or not to include fixed effects (fixed effects will be discussed in the following two sections). Figure 4.1: Mean of the net interest margin over the different banks. Figure 4.2: Mean of the net interest margin over the different years. We notice that the variability between the banks is much larger than the variability between the different years. This is a first indication that cross-section fixed effects may be needed. Figure 4.3 shows which variables are correlated and the extent to which they are correlated. We notice that size, LTA and NLTST are the most important variables to take into account.

35 Data 25 Figure 4.3: Correlation between the different variables. When crossed the relation is not significant on a 20% level. 4.4 General information Data is extracted from bankscope, a database which gives access to detailed information of banks. The interest of our study lies on the income statement, balance sheet and the already calculated ratios which are at our disposal. In total we have 13 variables (NIM, CIR, ETA, LLP, NLTST, LTA, size, MS, DIV, DIV1, spread, year, bank) and 140 observations. These observations are composed of 7 years and 20 banks. Some of these variables have missing values. We opted to fill these gaps up by using the average figure for that bank (example: bank a has a missing value for CIR, this missing value will be filled up with the average of the other cost income ratios that are at our disposal for bank a), this way we keep a balanced panel.

36 26 5 Results 5.1 Econometric issues We performed a basic pooled ordinary least squares model as starting point. Additionally we performed a fixed effects model with the cross-sections and/or periods as dummies. The comparison statistics are based on the regressions with DIV1 and the natural logarithm of total assets as proxies. We compared the models based on the model fit. We follow the same methodology as Hawtrey and Liang (2008). When adding the dummy variables we notice an improvement of the model fit (table 5.1). The durbin-watson statistic increased from 0.10 to Additionally, the standard error of regression, sum squared residuals, akaika info criterion and schwarz criterion showed smaller values indicating an improvement in model fit. These improvements were most substantial when we compared the pooled OLS model with the model where the cross-sections (banks) acted as fixed elements. The addition of the period fixed effects (years) did not improve the model substantially. D.W. S.E. of regr. S.S. resid. AIC Schwarz crit. Pooled Fixed (banks) Fixed (years) Fixed (two-ways) Table 5.1: Model fit performance figures Besides these model fit performance figures we did additional comparisons to decide whether or not to include the periods as fixed effects. First of all we performed a test on the basic pooled OLS model to check for possible absence of individual and time effects (table 5.2). The null-hypothesis states that there is no absence of effects.

37 Results 27 Breusch-Pagan Honda King-Wu Cross-section Time Note: p<0.1; p<0.05; p<0.01 Table 5.2: Lagrange multiplier test for random effects We can clearly reject the hypothesis that every individual has the same intercept. We cannot reject the null-hypothesis of equal intercepts across different periods. Secondly we calculated the F-statistics and chi-square statistics to determine if the cross-section fixed effect and period fixed effect should be included (table 5.3). Once again there is strong evidence that the cross-sections need to be included as fixed effect and the period effect not. We confirm hypothesis 1 and reject hypothesis 2. Statistic d.f. Probability Cross-section F (19,113) 0.00 Cross-section Chi-square Period F 0.08 (6,126) 0.99 Period Chi-square Table 5.3: Redundant fixed effects test Additionally we performed a random effects model. At first sight the coefficients of the fixed effects model and the random model do not differ much. An important assumption of the REM is that the explanatory variables cannot be correlated with the random effects. We can check this assumption by means of the Hausman test and this gives us an indication whether or not to use the fixed effects model or the random effects model. We retrieved a p-value op which is in our opinion inconclusive. We cannot reject the null-hypothesis of uncorrelated random effects on a 10% significance level but this result is not convincing. Because our sample is not a random sample chosen from a larger population we prefer the fixed effects model. If the true model would be a random effects model, the fixed effects model is still consistent. We cannot make the same conclusion in the other direction. To cope with the heteroscedasticity between the different cross-sections and the correlation of the error-terms withing the cross-sections we re-estimated the model introducing cross-section GLS weights and white period robust estimators. The former is robust against cross-section heteroscedasticity which means that the residual variance can differ between different crosssections and the latter against cross-section heteroscedasticity and within cross-section error correlation.

38 Results 28 Table 5.4: Fixed-effect (banks) LSDV regression with net interest margin as dependent variable Dependent variable: NIM (1) (2) (3) (4) CIR ( ) ( ) ( ) ( ) NLTST ( ) ( ) ( ) ( ) SIZE ( ) ( ) MS ( ) ( ) LLP ( ) ( ) ( ) ( ) LTA ( ) ( ) ( ) ( ) ETA ( ) ( ) ( ) ( ) DIV ( ) ( ) DIV ( ) ( ) Observations R Adjusted R F Statistic Note(1): Method: Panel EGLS (Cross-section weights); Linear estimation after one-step weighting matrix; White period standard errors and covariance (d.f. corrected) Note(2): p<0.1; p<0.05; p<0.01

39 Results 29 Table 5.5: Fixed-effect (years) LSDV regression with net interest margin as dependent variable Dependent variable: NIM (1) (2) (3) (4) CIR ( ) ( ) ( ) ( ) NLTST ( ) ( ) ( ) ( ) SIZE ( ) ( ) MS ( ) ( ) LLP ( ) ( ) ( ) ( ) LTA ( ) ( ) ( ) ( ) ETA ( ) ( ) ( ) ( ) DIV ( ) ( ) DIV ( ) ( ) Observations R Adjusted R F Statistic Note(1): Method: Panel EGLS (period weights); Linear estimation after one-step weighting matrix; White cross-section standard errors and covariance (d.f. corrected) Note(2): p<0.1; p<0.05; p<0.01

40 Results Parameter estimates The banks as fixed-effect Table 5.4 gives an overview of all the regressions that are performed with banks as dummy variables and net interest margin as dependent variable. The differences between the regressions are caused by different proxies for the size variable and the diversification variable. All the regression display a very high R 2 but not all variables are highly significant. All the dummy variables (not mentioned in table 5.4) are highly significant indicating that hypothesis 1 cannot be rejected and is found to be true. When interpreting the coefficients from table 5.4 it is important to keep in mind that we added cross-section fixed effects. Lets clarify with an example: loans to assets has a positive sign which means that if bank x would increase the amount of loans,keeping anything else equal,bank x is expected to receive a higher net interest margin. The cost income ratio always displays a negative sign but is not significant. Subsequently an increase in efficiency will not lead to a lower net interest margin. Net loans to short term funding and deposits shows variable signs and is never significant. Although if we remove the loan to asset ratio,which is positively correlated with the NLTST variable,the net loans to short term funding and deposits receives a positive significant sign. Loan loss provision to net interest revenue shows variable signs and no significance. We cannot confirm hypothesis 3 and 7. We can confirm hypothesis 4 but this variable seems to be too highly correlated with the loan to asset ratio. Our two proxies for the size are both significant on a 10% level and negative. Natural logarithm of total assets shows a negative sign and is found to be significant on a 1% level. We reject hypothesis 5. This implies that there is a trade-off between volume and profit margin. A bank has to choose between growing or retaining a reasonable interest margin. Market share displays a similar trend. The sign is also negative and significant on a 1% level and 5% level. Loan to asset has a positive sign and is significant on different levels. This result confirms hypothesis 8 and follows the reasoning of Claeys and Vander Vennet (2008) which states that loans are the most risky but highest yielding assets. The more loans compared to assets a bank has the higher his net interest margin is expected to be. Equity to total assets is negative and significant for half of the cases. It loses its significance once the market share is introduced as proxy for the size of the bank. We cannot confirm hypothesis 9 based on these results. This means that banks do not have an incentive to increase their equity to total assets ratio. Due to the extremely low funding costs that banks face there is no incentive to be less risky because it is not possible to receive better funding conditions.

41 Results 31 The proxies for the diversification grade all have a negative sign. The diversification variable which underwent winsorization (DIV1) is significant on a 5% level. The other proxy that did not undergo this transformation is not significant on a 10% level. Based on the results obtained for the transformed diversification variable we confirm hypothesis 10. The bank has to choose between a higher net interest margin and profits from varying sources. To examine the economic relevance of the significant variables we carried out a variance decomposition. The economic relevance of a variable is equal to: Coef f icient Standard deviation (5.1) Statistic Economic relevance Absolute value Variance decomposition SIZE % LTA % ETA % DIV % Total % Table 5.6: Economic relevance of the net interest margin determinants This table gives an insight in the real relevance of the variables. We used the average coefficients of the four regression. The standard deviation is the average of the individual standard deviations of each bank. This makes clear that the size of a bank has the biggest impact on the net interest margin followed by the loans to total assets ratio. We also introduced a second dependent variable namely the spread (table E.4). When we use the natural logarithm of total assets as proxy for the size we do not notice big differences. But once that we introduce market share as proxy some signs change. The market share variable has a positive sign and is significant. The equity to total assets ratio also displays a positive sign which it did not showed before. The coefficients (of equity to total assets) are not significant on a 10% level. Especially the fact that market share received a different sign is remarkable.

42 Results 32 Due to the fact that we looked at the within bank relationships we omitted a lot of variability. These regressions may be the most optimal ones when we evaluate them on model fit but they do not tell us much about the effects of the different variables. The most important conclusions drawn from these regressions are the following ones: - There is a lot of heterogeneity between the different Belgian banks caused by unique characteristics. - Within the banks there is a trade-off between growth and interest margin. - The loan to assets hypothesis (8) is confirmed. - There is a trade-off between interest income and non-interest income The years as fixed-effect Table 5.4 resulted in the best model but we cannot make any conclusions about the effects of the variables between the different banks. To cope with this shortcoming we performed an additional model with the different years as fixed effect (table 5.5). A lot of the significance that went to the bank dummies in the previous model goes to the different variables in the new model. The cost income ratio becomes significant and receives a negative sign meaning that the more efficient a bank is the higher the net interest margin is expected to be. This follows the reasoning of Angbazo (1997) and Maudos and De Guevara (2004) which is mentioned in subsection They reason that a bank with a low cost income ratio has a highly qualitative administration which enables them to obtain interest earning assets with a high yield and interest bearing liabilities with a low cost. Net loans to short term funding and deposits also becomes significant and positive implying that the higher the loan to deposit ratio is the higher the net interest margin will be. This result is in line with the expected sign. A higher ratio will bring along more loans (which are as already mentioned the highest yielding earning assets) but also more risk. Equity to total assets turns significant and positive. This means that banks with a higher ratio have advantages when they need to pay funding costs. The fact that this variable becomes significant can be caused by the fact that there is not much variability of this structural ratio within the banks. The remaining variables: size, loan to asset ratio and diversification keep their sign and remain significant. We also performed a variance decomposition for this model. We used equation 5.1 and calculated the standard deviations by taking the average of the standard deviations of the different time periods.

43 Results 33 Statistic Economic relevance Absolute value Variance decomposition LTA % ETA % DIV % SIZE % CIR % NLTST % Total % Table 5.7: Economic relevance of the net interest margin determinants 5.3 Robustness In order to check the robustness of our initial model we made some iterations to our final fixed effects model. First of all we removed some outliers. Our dataset contains three banks that clearly have a much higher net interest margin. These net interest margins lie above 3.9 %. Secondly, we repeatedly removed one variable from our equation (equation 3.2). The result for the first iteration can be found in table E.1. The result remains broadly the same except that the equity to total assets ratio is no longer significant and that net loans to short term funding and deposits is becoming significant. This would imply that when this ratio goes up, the opportunity costs go down and that the customer can benefit from this by means of a lower net interest margin. When looking at the output of our second iteration (table E.3) we notice that all variables (except NLTST) display the same sign and significance. We also performed two iterations on the period fixed effects model. The loan to assets ratio and the net loans to short term funds and deposits are subsequently removed because they were highly correlated with each other (table E.2). 5.4 Explanation The loans to total assets ratio returned the expected sign and gave no surprises. It is generally accepted that loans are the highest yielding assets and this study confirms that. The most important and remarkable determinants are the size of the bank and the diversification grade. We illustrate the relationship between the net interest margin and the size by the following graph:

44 Results 34 Figure 5.1: Net interest margin (fitted values) in function of natural logarithm of total assets. For each bank separately and for the Belgian retail banks as a whole. There are already given possible explanations for this behavior in past researches. Ho and Saunders (1981) stated that larger banks operate in a more competitive framework and that smaller banks are able to exploit regional positions. Research in Belgium indicates that financial stability and delivered services are key components in the perceived trustworthiness of a bank (Ballegeer 2014). These components could give the smaller banks an advantage to the bigger banks. The smaller banks are perceived as financially more stable. The big Belgian banks were the subject of investigation and were more prominently present in the actuality during the banking crisis. All these reasons could give cause to the higher margins that small banks are able to demand to their customers. If we assume that due to these reasons the demand for deposits and loans increased at small banks and that small banks do not strive for an exponential growth. We could assume that small banks lower the probability of arrival by widening their margin. λ l = α l β l a (5.2) λ d = α d β d b (5.3) These two equations also show the trade-off that banks face. Does the bank choose for a higher margin or does he choose for the additional volume? Other explanations can be found in the literature. The information advantage hypothesis states that small banks are better informed and have an advantage when evaluating and monitoring loan quality (McNulty, Akhigbe, and Verbrugge 2001). They also mentioned that small banks are the primary source of small business lending. This could explain the higher margins. Berger, Miller, Petersen, Rajan, and Stein (2005) concluded that smaller banks are more willing to lend out money to firms with no or few financial records. These credits would have a higher yield compared to more established bigger firms.

45 Results 35 Akhigbe and McNulty (2003) mentioned five different theories that could explain why smaller banks perform better than relative bigger banks. The most relevant one is the structure performance hypothesis (Gilbert 1984) which states that the relative smaller banks can charge higher rates on loans and pay lower rates on deposits due to the less competitive environment. The quiet life hypothesis implies that small banks limit their output on loans and cherry pick potential loans. This way they achieve a high return rate on the outstanding loans. We can conclude that there are already a lot of theories that confirm our results. The situation in Belgium may intensify the relationship but we are not able to verify this. Other reasons which could be caused by the specific framework 1 are unclear and need to be examined in future research. We already mentioned that the bigger banks were decreasing their balance sheet size and focusing more on the core businesses. It seems that these actions, based on our results, are the correct efforts to retain a viable net interest margin. Figure 5.2: Evolution of the natural logarithm of size and loan to assets ratio The diversification grade also proved to be an important variable. This variable was significant on the within level and the between level. It seems that banks need to choose between a higher margin and additional profits from other sources. Lepetit, Nys, Rous, and Tarazi (2008) investigated this relationship more in depth. They found a similar relationship, namely higher diversification results in lower interest margins. According to them this was caused by underpricing the loans in order to cross-sell other products. Subsection explained how size, diversification and net interest margin could be related. 1 Country specific reasons or the time frame in which this research was conducted.

46 36 6 Conclusion 6.1 General conclusion We performed two fixed effects models, one with the banks (cross-sections) as fixed effect and one with the years (periods) as fixed effect. The cross-section fixed effects model performed better based on various performance measurements due to the presence of the bank dummy variables. We observed a lot of heterogeneity between the different banks. In order to get a better understanding of the impact of the different variables between the different banks a period fixed effects model was introduced. - The size proxy was significant and negative for the two models resulting in two interesting findings. First of all there is a trade-off between volume and profit. Secondly, small banks are able to exploit their position in the market to obtain better rates. - Loan to assets was significant and resulted in the expected sign. More loans resulted in a higher net interest margin. - An important finding was also the sign of the diversification variable. The variables displayed a negative significant sign which implies a trade-off between interest income and non-interest income. - Equity to total assets did not show a lot of variability within the banks so the influence on the net interest margin within banks was limited. On the between level we observed a positive significant sign which means that an increase in equity results in better funding conditions or the ability to hold a riskier, higher yielding interest earnings assets portfolio. - Cost income ratio presented a negative sign which was also significant. The economic relevance was considerably lower compared to the previous mentioned variables. - The last significant variable is net loans to short term funds and deposits. This variable displayed the same sign as the loans to assets ratio which was positive.

47 Conclusion Practical implications It is important to get an understanding of the drivers of the net interest margin especially when this performance figure is expected to decrease. Our research made clear that Belgian retail banks need to choose between volume and interest margin. Current conditions do not allow to maximize both. Banks will also need to choose between interest income and non-interest income. The results made clear that it is difficult to combine the two income sources. The last years we noticed an improvement of the cost income ratio and an increase in the loan to asset ratio. These two efforts should result in increasing net interest margins and are already excellent responses of the banks. The results also indicated that increasing the amount of equity provides a stronger position on the market. 6.3 Limitations Our research was conducted on the Belgian retail banks which is a rather small population. Due to the limited amount of observations the results need to be used carefully. There is no certainty about the impact of the different variables in a different scope. We chose a specific time-frame, namely the first years after the banking crisis so there is no evidence that the result can be generalized to other years. Important to remember is the fact that a lot of significance in the cross-section fixed effects model went to the dummy variables which could imply that there are a lot of missing explanatory variables or that the banks are very heterogeneous. The credit risk proxy did not resulted in a significant coefficient for the Belgian retail banks so another proxy should be used to investigate the impact of the credit risk before we could make a final conclusion about the impact of this variable. 6.4 Future research It could be interesting for future studies to examine the relationship between a bank s size and the net interest margin more in depth. Future research could investigate the link between size, bank trust and net interest margin in order to get a better understanding of the impact on this important profitability figure in Belgium. In order to validate these results additional research could be done for countries which have similar characteristics as Belgium.

48 38 References Akhigbe, A., and J. E. McNulty The profit efficiency of small US commercial banks. Journal of Banking & Finance 27 (2): Allen, L The determinants of bank interest margins: a note. Journal of Financial and Quantitative analysis 23 (02): Angbazo, L Commercial bank net interest margins, default risk, interest-rate risk, and off-balance sheet banking. Journal of Banking & Finance 21 (1): Christophe Ballegeer 2014, % van de Belgische consumenten heeft vertrouwen in zijn bank van-de-belgische-consumenten-heeft-vertrouwen-in-zijn-bank. Accessed: Berger, A. N., N. H. Miller, M. A. Petersen, R. G. Rajan, and J. C. Stein Does function follow organizational form? Evidence from the lending practices of large and small banks. Journal of Financial economics 76 (2): Buzzell, R. D., B. T. Gale, and R. G. Sultan Market share-a key to profitability. Harvard business review 53 (1): Casu, B., C. Girardone, and P. Molyneux Introduction to banking, Volume 10. Pearson Education. Claeys, S., and R. Vander Vennet Determinants of bank interest margins in Central and Eastern Europe: A comparison with the West. Economic Systems 32 (2): Cooper, W. W., L. M. Seiford, and J. Zhu Data envelopment analysis. In Handbook on data envelopment analysis, Springer. Demirgüç-Kunt, A., and H. Huizinga Determinants of commercial bank interest margins and profitability: some international evidence. The World Bank Economic Review 13 (2): Estrada, D. A., E. G. González, and I. P. O. Hinojosa Determinants of interest margins in Colombia. Citeseer. Gilbert, R. A Bank market structure and competition: a survey. Journal of Money, Credit and Banking 16 (4): Gischer, H., and D. J. Juttner Global competition, fee income and interest rate margins

49 REFERENCES 39 of banks. Kredit und Kapital 36 (3): Gujarati, D. N., and D. Porter Basic Econometrics Mc Graw-Hill International Edition. Hawtrey, K., and H. Liang Bank interest margins in OECD countries. The North American Journal of Economics and Finance 19 (3): Hempel, G. H., and D. G. Simonson Bank management: text and cases. Wiley. Hilgers, J Financial Stability Report National Bank of Belgium. Ho, T. S., and A. Saunders The determinants of bank interest margins: theory and empirical evidence. Journal of Financial and Quantitative analysis 16 (04): Hsiao, C Panel data analysis advantages and challenges. Test 16 (1): Lepetit, L., E. Nys, P. Rous, and A. Tarazi The expansion of services in European banking: Implications for loan pricing and interest margins. Journal of Banking & Finance 32 (11): Loose, K. D., S. Neuville, S. Pauwels, and T. V. den Bruel The cummulative impact of regulation, taxes and low interest rate environment. Maudos, J., and J. F. De Guevara Factors explaining the interest margin in the banking sectors of the European Union. Journal of Banking & Finance 28 (9): McNulty, J. E., A. O. Akhigbe, and J. A. Verbrugge Small bank loan quality in a deregulated environment: the information advantage hypothesis. Journal of Economics and Business 53 (2): McShane, R., and I. Sharpe A time series/cross section analysis of the determinants of Australian trading bank loan/deposit interest margins: Journal of Banking & Finance 9 (1): Saunders, A., and L. Schumacher The determinants of bank interest rate margins: an international study. Journal of international money and finance 19 (6): Valverde, S. C., and F. R. Fernández The determinants of bank margins in European banking. Journal of Banking & Finance 31 (7): Vander Vennet, R. 2015, 12. Basel III Impact assessment. Lecture December Wong, K. P On the determinants of bank interest margins under credit and interest rate risks. Journal of Banking & Finance 21 (2):

50 40 A Mathematical model equity to assets Equation A.1 displays the balance sheet structure. L is loans, R is reserves, D is deposits and E is equity. The capital (equity) to assets ratio is presented as e (E/L). k presents the reserves deposit ratio. L + R = D + E (A.1) L(1 e) = D(1 k) (A.2) Equation A.3 expresses the profit function (π is profit). π = r L L r E E r D D (A.3) π = r L L r E el r D 1 e 1 k L (A.4) When profit is differentiated to the amount of loans L we get the following result: dπ dl = r L r E e r D 1 e 1 k = 0 (A.5) r L (1 k) r E e(1 k) r D (1 e) = 0 (A.6) r L r }{{ D = r } E e(1 k) + kr L r D e (A.7) spread The final equation shows that an increase in the capital ratio result in an increase of the spread. The required return adjusted for reserves is assumed to always be larger than the return on deposits. s e = r E(1 k) r D > 0 (A.8)

51 41 B Descriptive statistics Statistic N Mean St. Dev. Min Median Max CIR CIR CIR CIR CIR CIR CIR Table B.1: Cost income ratio (in %) Statistic N Mean St. Dev. Min Median Max NLTST NLTST NLTST NLTST NLTST NLTST NLTST Table B.2: Net loans to deposits and short term funding (in %)

52 Descriptive statistics 42 Statistic N Mean St. Dev. Min Median Max MS MS MS MS MS MS MS Table B.3: Market Share (in %) Statistic N Mean St. Dev. Min Median Max Size Size Size Size Size Size Size Table B.4: Natural logarithm of amount of total assets Statistic N Mean St. Dev. Min Median Max LLP LLP LLP LLP LLP LLP LLP Table B.5: Loan Loss Prov/Net interest revenue (in %)

53 Descriptive statistics 43 Statistic N Mean St. Dev. Min Median Max LTA LTA LTA LTA LTA LTA LTA Table B.6: Loans to assets (in %) Statistic N Mean St. Dev. Min Median Max ETA ETA ETA ETA ETA ETA ETA Table B.7: Equity to assets (in %) Statistic N Mean St. Dev. Min Median Max DIV DIV DIV DIV DIV DIV DIV Table B.8: Total non-interest income to total income (in %)

54 Descriptive statistics 44 Statistic N Mean St. Dev. Min Median Max NIM NIM NIM NIM NIM NIM NIM Table B.9: Net interest margin (in %) Statistic N Mean St. Dev. Min Median Max spread spread spread spread spread spread spread Table B.10: Spread (in %) Statistic N Mean St. Dev. Min Median Max CIR NLTST MS SIZE LLP LTA ETA DIV DIV NIM SPREAD Table B.11: Overview variables

55 45 C Correlation Figure C.1: Variables plotted against Net Interest Margin (1) Figure C.2: Variables plotted against Net Interest Margin (2)