The Effect of Bank Capital Ratios on Loan Growth:
Euro Area Evidence
G.J. van der Linde
S1772783
Combined MSc thesis for the MSc Finance and MSc Economics
University of Groningen
Supervisor: Prof. Dr. J.M. Berk
June 26th, 2015
Abstract
This study uses publicly available bank level data on the Euro area to model the impact of bank capital ratios on bank lending growth. A specification based on the model of Kashyap and Stein (1995, 2000) is used to separate credit demand and supply factors. The average impact of a 1% increase in bank capital ratios on loan growth ranges from 0.316% to 0.562% on average (depending on the definition of the capital ratio). The impact of capitalization on loan growth is greater for large banks than for small banks. Also, the effect of bank capital ratios is larger for low-capitalized banks than well-low-capitalized banks, implying that the relation is nonlinear. Furthermore, the relation intensified in recent crisis years. The results are robust to various loan categories (i.e., mortgage loans and corporate and commercial loans). However, the results are sensitive to alternative model specifications.
2
1. Introduction
The recent financial turmoil and new regulatory requirements imposed by the Basel III accords have led to renewed interest in the effect of bank capital ratios on lending. In the beginning of the 1990’s research focuses on the role of bank capitalization in the credit crunch during the early 1990’s in the US. Bernanke and Lown (1991) developed a model to estimate the impact of bank equity capital over assets on loan growth and find a positive effect. Recently, Berrospide and Edge (2010) further develop the methodology of Bernanke and Lown (1991) and find a positive effect of bank capital ratios in the US twenty years later. However, the results are more modest than in Bernanke and Lown (1991). Among others, Gambacorta and Marques-Ibanez (2011), Carlson et al. (2013) and Brei et al. (2013) deal with the recent financial crisis and whether this changes the effect of bank capital ratios on lending growth.
This paper examines how bank capital ratios affect loan growth in the Euro area. When markets are imperfect, asymmetric information problems imply an external finance premium exists on risky bank liabilities. Bank capital ratios then serve as a signal to investors. Low-capitalized banks (banks with low capital ratios) face a higher risk premium on their liabilities. A higher premium makes it more expensive for banks to issue additional liabilities to expand their assets (i.e. loans). Friedman (1991) states that capital may constrain banks in offering new loans if a minimum capital ratio requirement is introduced for banks, as is the case for the Basel accords. The market itself may also require banks to hold a certain level of capital as a disciplining device (Duprey and Lé, 2014). If a bank cannot access new capital, and is limited by its scarce capital, it is left with the option of liquidating loans (particularly risky ones) and extending less new credit. Bank dependent firms may be withheld from investing.
3
What is the effect of banks’ capital ratios on bank lending in the Euro area?
1.1 Findings
In line with previous literature I find evidence of a positive and significant relation between bank capital ratios and loan growth. Using a baseline regression and Euro area data between 2000-2013 the average impact of a 1% increase in bank capital ratios on loan growth ranges from 0.316% to 0.562% for the average bank (depending on the definition of the capital ratio). The capital ratio with the highest loss-absorbing quality has the largest impact on loan growth. The size of the estimated impact differs widely across previous studies. I argue that this could be caused by differences in the average size of banks in the sample of these studies. This research finds that the impact of capitalization is greater for large banks than for small banks. Large banks manage their capital structure more actively and operate closer to the required minimum capital ratio. This implies that the impact of bank capital ratios for low-capitalized banks is larger than for well-capitalized banks. I find further evidence that supports this statement. This study finds that bank capital ratios were more important from 2008-2013 than during the years 2000-2007. The findings in this study are robust to different loan categories (i.e., mortgage loans and corporate and commercial loans). As a robustness check I use two alternative specifications of the model. I find evidence that the model might not entirely control for all demand and supply factors. Also, the finding that the capital ratio with the highest loss-absorbing quality has the largest impact on loan growth is not robust to the alternative specifications.
1.2 Content
The next section explains the theoretical reasoning behind the effect of capital ratios on lending. Thereafter, prior literature on the effect of bank capital ratios on lending growth is dealt with. The third section discusses the methodology used in this study. The fourth section is describes the data used in this study. The first part of this section deals with the sources of the data and the construction of the data set. Thereafter, descriptive statistics on the data are provided. The fifth section discusses the results of the model. The last section of this paper concludes. Furthermore, I touch upon fields for further research.
2. Literature review
2.1 Why bank capital matters
4 (Bernanke and Gertler, 1995; Mishkin, 1996). Bank capital helps mitigate asymmetric information problems (Van den Heuvel, 2002). The literature on the transmission of monetary policy into the real economy helps to get an intuitive feel on why bank capital is important. Therefore, this section discusses three monetary transmission channels that arise because of asymmetric information problems in financial markets: the balance sheet channel, the bank lending channel and the bank capital channel.
The balance sheet channel operates through the net worth of firms. When monetary policy is contractionary, rising interest rates are often accompanied by decreasing cash flows and asset prices (Bernanke and Gertler, 1995). This implies that lenders have less collateral underlying their loans. Moral hazard and adverse selection problems become more severe and therefore lenders require a larger compensation for higher costs regarding monitoring and evaluation. Owners of the borrowing firm with more capital over total assets have better incentives to ensure good financial outcomes (Bernanke, 2007). The external finance premium is inversely related to the financial position of borrowers and thus the premium increases if bank capital ratios worsen. The level of bank-dependence in the economy determines the impact of the balance sheet channel on the real economy (Mishkin, 1996). The balance sheet channel applies to all firms (financial and non-financial), the following two channels are specific to banks.
5 and Morgan, 2000). A cutback in the supply of bank loans transfers to the real economy through a decrease in investments by bank-dependent borrowers (Mishkin, 1996).
The bank capital channel depends on three assumptions. First, markets for bank equity are imperfect due to agency costs and tax disadvantages (Myers and Maljuf, 1984; Cornett and Tehranian, 1994; Calomiris and Hubbard, 1995; Stein, 1998). Second, banks face a maturity mismatch between the assets and liabilities on their balance sheet. Banks attract funding with short maturities to create loans with long maturities and this exposes them to interest rate risk (Freixas and Rochet, 1997). Third, banks are subject to regulatory capital requirements (Van den Heuvel, 2001). Following a rise in the policy rate (contractionary monetary policy) a smaller amount of loans can be renegotiated than deposits, because loans typically have a longer maturity than deposits. Banks face a reduction in profits and thus capital. If equity is below a certain threshold and issuance of new equity is costly, banks might consider to reduce lending. Van den Heuvel (2001) finds that for banks to cut lending, bank capital does not necessarily has to be down to the regulatory minimum. Low-capitalized banks might reject profitable lending opportunities now to prevent the chance of capital inadequacy in the future.
The above-mentioned theories indicate that bank capital is a relevant balance-sheet item that mitigates adverse selection and moral hazard problems and thus affects the supply of bank credit. The real economy is affected when firms have no or limited access to financial markets and therefore have to forego profitable investment possibilities (Mishkin, 1996).
2.2 Prior literature
Bernanke and Lown (1991) investigate whether the US recession in the early 1990’s was partly due to a credit crunch. Using bank level data, they use a model that regresses loan growth on previous year equity capital ratios. They control for economic activity by including the growth rate of employment. Their study finds that the amount of bank capital in the US state New Jersey in 1989 had a significantly positive effect on bank lending growth over 1990-1991. Hancock et al. (1995) use a VAR (vector auto regression) approach to assess the impact of a capital shock on various categories of banks’ assets and liabilities. They find that following a shock, banks’ holdings of securities, loans, capital itself and other liabilities do not adjust at the same speed. Bank capital and securities revert to mean within a year after the capital shock took place. The adjustment of bank loans and liabilities takes two to three years. Furthermore, they find that bank’s responses to capital shocks are larger for undercapitalized and small banks than well-capitalized and large banks1. Peek and Rosengren (1997) use the Japanese financial crisis of the late 1980’s and early 1990’s to construct a natural experiment. Japanese banks had a large financial presence in the US at the time of the crisis. This enables Peek and
1
6 Rosengren (1997) to identify a shock to credit supply of Japanese banks’ subsidiaries in the US while loan demand stays unaffected. They find that a one percentage point decrease in a bank’s risk-based capital ratio led to a four percentage point decrease in commercial loans relative to total assets.
More recently, Berrospide and Edge (2010) further develop the models by Bernanke and Lown (1991) and Hancock and Wilcox (1993, 1994) by using additional control measures for supply and demand factors. They study the effect of bank capital ratios on lending growth in the US over the period 1992-2009 using bank level data. They construct a model that separates demand from supply factors, by using bank specific and macroeconomic control variables. They apply both actual and excess capital ratios. Excess capital is defined as actual capital minus an estimated target of capital. Estimation of their model shows modest effects of bank capital (actual and target ratios) on lending growth relative to the results from the 1990’s. Carlson et al. (2013) use an alternative strategy to deal with the problem of separating demand from supply effects. They construct sets of matching banks. Matching occurs on geographical location and other bank specific factors. They then compare differences in bank loan growth between the matched banks to differences in capital ratios. Carlson et al. (2013) argue that this approach controls for local bank credit demand and other environmental influences. They find that banks with higher capital ratios have higher loan growth in the following year. The growth rate differs between types of loans (for example, commercial real estate loans and commercial and industrial loans). This effect is found to be strong during the recent financial crisis, but insignificant in the years before the crisis. They also find that the elasticity of bank lending with respect to bank capital ratios is higher if banks are low-capitalized. This indicates that the relation between bank capital and loan growth is nonlinear.
7 announced Basel III requirements Francis and Osborne (2012) specifically focus on regulatory capital ratio requirements and how these affect bank behavior. They use a database that includes individual capital requirements set by the supervisor for all UK banks. These requirements greatly influence target capital ratios of banks. This study finds that positive shocks to excess capital ratios (defined as the actual capital ratio over the regulatory requirement) have a positive effect on loan growth.
3. Methodology
3.1 Separating credit supply and demand
Separating the supply and demand for credit is one of the most challenging aspects in research on lending growth. Changes in the economic environment that affect the supply for loans also influence the demand for loans. Previous literature uses several methods to separate demand from supply. The first approach uses a natural experiment in which an exogenous shock impacts only bank capital but not the demand for bank lending. This method often utilizes the cross-country nature of banking. Multinational banks are used in assessing the impact of a capital shock in one country on the bank’s lending behavior in another country (where the demand for credit stays unaffected). For example, Peek and Rosengren (1997, 2000), and Mora and Logan (2012) use this methodology. A second approach is to use a model wherein economic conditions are modelled explicitly. Besides capital ratios, proxies for other supply and demand factors that influence bank lending are included. This method is used in this paper, and has also been applied by, among others, Hancock and Wilcox (1993), Kashyap and Stein (1995, 2000) and Berrospide and Edge (2010). Recently, Carlson et al. (2013) developed another approach to distinguish supply from demand. They match sets of banks based on geographic areas and other business characteristics and argue that this controls for local loan demand and other environmental factors.
3.2 Baseline specification
The model is based on the work by Kashyap and Stein (1995, 2000). They developed an approach that estimates loan growth as a function of demand and supply side factors, where bank capital is one of the supply side factors. This methodology is used extensively in the monetary transmission channel literature. The model takes the following form:
, , , , , (1)
Where, , is the annual growth rate of loans for bank i (i=1,2,…,N) at time t (t=1,2,…, T). The
8 variables, , includes the bank specific capital ratio and is the variable of interest. , is a
dummy variable related to changing accounting standards.
A potential identification problem in testing the effect of capital ratios on bank loan growth is that loan growth could also affect bank capital ratios. Therefore, the right hand side variables in (1) are specified in lags to overcome endogeneity problems (Kashyap and Stein, 1995, 2000; Ehrmann et al., 2003). , is the dependent variable and defined as the annual growth rate of loans, because the
stock of loans exhibits stationarity problems (further discussed in Section 4.5). This section will now first deal with the differently defined capital ratios. Thereafter, the macroeconomic and bank specific control variables are discussed.
Three definitions of bank capital ratios ( , ) are considered in this research (see Table A1
in Appendix A for an overview of the definition of all variables). The impact of these capital ratios could differ, because their nominator and denominator are defined differently. This study includes the tangible common equity (TCE) ratio, the tier 1 (T1) capital ratio and the total capital (TC) ratio. The different definitions of capital are mostly concerned with their loss-absorbing quality. Tangible common equity is the most narrow definition of capital and is assumed to be most loss-absorbing (BCBS, 2010). It forms a buffer that serves as the first-loss position in case of bank failure. This ratio gained increasing attention since the start of the financial crisis in 2008 and the announcement of the Basel III accords. Tier 1 capital and total capital (tier 1 capital plus other capital) have less loss-absorbing features (for a more detailed definition of the capital ratios see Table A1 in Appendix A). Furthermore, the numerator of capital ratios is either defined as risk-weighted assets or unweighted assets. Using a risk-weighted measure does not only measure the impact of capital adequacy on loan growth. It gives guidance on the effect of the composition of a bank’s portfolio on lending growth. Off-balance sheet items are also included in the risk-weighted measure. All else the same, when capital is risk-weighted banks with relatively risky portfolios have a larger denominator and thus a lower capital ratio than banks with less risky portfolios (Choudhry, 2012). Both the T1 and TC ratio are risk-weighted. The TCE ratio is defined as tangible common equity over tangible assets (total assets minus intangible assets). For interpretation reasons the capital ratio is normalized to zero by subtracting the cross-sectional mean:
, ,
,
1 ,
,
Where, the summation is over all observations at time t and ,
, is either defined as the TCE,
9 that bank capital helps overcome asymmetric information problems and thus increases loan growth, I expect 0.
The macro variables are the annual growth rate of GDP ( ), the inflation rate ( ) defined as the annual growth rate of the Harmonized Index of Consumer Prices (HICP) and the overnight interbank borrowing rate ( ) measured as the change in EONIA. and are specified in growth rates and in first-differences to overcome stationarity problems (see Section 4.5). The growth rate of GDP ( ) and inflation ( ) account for the demand for loans and the change in the interbank borrowing rate ( ) accounts for monetary policy2. The coefficient can be interpreted as the effect of the macroeconomic variables on bank loan growth of an average bank (ceteris paribus).
, is a vector that includes bank specific control variables. Liquidity ( , ) is measured as
the holding of highly liquid assets over total assets. It captures the effect of banks using their stock of liquid securities to adjust loan growth. The ratio of impairments to total assets ( , ) is included as a
proxy for bank specific risk. The bank specific variables are normalized to zero by subtracting the cross-sectional mean: , , , 1 , , , , , 1 , ,
Where, the summation is over all observations at time t. This allows for the interpretation of the coefficient as the average effect of bank specific variables on the loan growth of an average bank (ceteris paribus).
During the data period many banks switched from a local Generally Accepted Accounting Principles (GAAP) accounting regime to International Financial Reporting Standards (IFRS). A bank specific dummy variable ( , ) accounts for this regime switch. This dummy takes the value one in
the first year a bank uses IFRS (2005 for most banks) and zero otherwise.
2
10
3.3 Nonlinearity
Evidence in the literature suggests that the relation between bank capital ratios and loan growth is nonlinear. In a situation where minimum capital requirements (as required either by the regulator or through market discipline3) are in place, the marginal impact of a change in a bank’s capital ratio might be larger if the ratio is close to its minimum. This implies that a unit increase to the bank’s capital ratio for a well-capitalized bank has a smaller effect on loan growth than the same increase for a poorly capitalized bank. Following Carlson et al. (2013), dummy variables are created that assign a bank in a specific year to be low, medium or well-capitalized. The dummy variable “low” ( , ) takes the value one if the bank’s capital ratio is below the 25th percentile of the annual
distribution of capital ratios and zero otherwise. The dummy variable “medium” ( , ) equals one if
bank capital is between the 25th and 75th percentile. “Well” ( , ) equals one if bank capital is above the 75th percentile. The lagged dummy variables are interacted with , . Both the lagged dummy
variables and the interaction terms are included in the model. The specification then becomes:
, , , , , ,
, , , ,
(2)
In accordance with the hypothesis that the impact of an increase in bank capital ratios is smaller for
well-capitalized banks than low-capitalized banks, I expect .
3.4 Robustness
The models specified above assume that the relevant time effects are captured by the macroeconomic variables. In order to prove that this assumption holds; (1) is estimated with a set of time dummies ( ) instead of the macroeconomic variables. The regressions then becomes:
, , , , , (3)
As further dealt with in Section 5.1, model (1) exhibits serial correlation in the error terms. Autocorrelation could imply misspecification of the model. Brooks (2008) and Verbeek (2012) argue that a dynamic specification (including a lagged dependent variable) might better fit the data. This approach has frequently been used in previous literature on the topic of bank capital ratios and loan growth (i.a., Ehrmann et al., 2003; Berrospide and Edge, 2010; Brei et al., 2013). Therefore, as a robustness check, a lagged dependent variable ( , ) is added to the right hand side of (1):
3
11
, ∅ , , , , , (4)
4. Data
4.1 Bankscope and limitations
Yearly bank balance sheet data is obtained via Bankscope. This research covers 2000-2013, because Bankscope’s data goes back to 1999 and at the time of writing the data on 2014 was still incomplete. All countries currently in the Euro area are included. See Table B1 in Appendix B for all countries included and the number of banks and observations per country.
Bankscope offers the main advantage that it is publicly available. However, it also suffers from some drawbacks. The coverage of the Bankscope database is incomplete and grows over the sample period. In 2000 a little over 200 banks are included in the sample, while by 2013 this number increased to over 350. According to ECB data the number of financial institutions has actually been decreasing since 2000 (ECB, 2014). This might lead to a biased view. However, creating a balanced panel would seriously reduce the number of observations. The rise in the number of banks will be mainly due to missing observations on smaller banks, because financial reporting increases with company size. The sample is thus biased towards larger banks, even more so for the first few years of this study4. Another drawback of the Bankscope database (also identified by Ehrmann et al., 2003) is that quarterly data is of limited availability. Most studies in this field of research use quarterly data. However, Gambacorta (2005) uses a large data set on Italian banks to test for the effect of capital ratios on growth in bank lending. He compares quarterly results to yearly results and finds no significant differences.
4.2 Construction of the data set
In accordance with the literature the sample consists of consolidated bank data only (Berrospide and Edge, 2010; Brei et al., 2013), because many decisions regarding bank capital are made on the bank holding level rather than at the subsidiary level.
Previous literature takes on a variety of ways to deal with mergers and acquisitions (M&A). This phenomenon is of particular interest when estimating loan growth. Consider for example a bank acquiring another bank of equal size. Ceteris paribus the acquirer will have a loan growth rate of 100% in the year the acquisition takes place. M&A events may therefore seriously affect the outcomes. The most commonly used approach to account for M&A is to assume that the merger takes place at the beginning of the sample period and create one proforma bank, consisting of both the acquirer and the
4
12 target (Brei et al., 2013). However, I reason that this approach might not be the best reflection of the truth. The pre-merger banks might have had different strategies regarding capitalization prior to the merger. Therefore, I take on a different approach also used by Francis and Osborne (2012) and Kashyap and Stein (2000). All observations with over 20% of loan growth are checked for M&A activity using the Zephyr database. The observations for the year wherein a merger or acquisition took place are removed from the sample (84 observations).
After the correction for M&A events, outliers are further dealt with by removing all observations with loan growth or contraction of over 50% (347 observations). I argue that such a large change in a bank’s loan base is most likely due to some external shock. Furthermore, banks with a negative capital ratio or a capital ratio of over 50% are removed (74 observations). Banks with capital ratios outside these boundaries are assumed to engage in other activities than normal banking activities that cause this deviation in capital ratios. This approach for removing outliers is also used by Francis and Osborne (2012) 5.
4.3 Descriptive statistics bank data
The final sample consists of 4,527 yearly observations on 549 independent banks over 14 years. The average number of observations per bank is 8.25. Figure 1 shows the distribution of categories of banks that appear in the sample (categories as defined by Bankscope). The largest category is commercial banks, followed by cooperative banks. Furthermore, bank holding companies, investment banks, savings banks and real estate and mortgage banks are included. The balance sheet characteristics of these banking categories are quite diverse, in particular the differences in size are large. For example, the number of bank holding companies included is relatively small, yet their share in aggregate total assets is quite large. Since this study takes on an aggregate approach and focuses on the (commercial) banking sector as a whole, all banks with commercial activities are pooled together.
Figure 2 shows the yearly median of all bank level variables against time. The structural break at the beginning of the crisis (the fall of Lehman Brothers in September 2008) clearly appears in all panels. To accentuate the differences in the data before and after the crisis, Table 1 shows descriptive statistics over both the entire sample and the subsamples (2000-2007 versus 2008-2013). Panel A of Figure 2 shows that the rise in total assets came to a halt in 2008 (at approximately EUR 10.500 billion). Panel B shows the growth in banks loans. Table 1 shows that the median annual loan growth before 2008 was 10.722%, decreasing to 2.229% during 2008-2013. Panels C, D and E show the capital ratios included in this research. The TCE ratio fluctuates between 5.845% and 6.810%. The T1
5
13 and TC ratio remain relatively stable before the crisis, but increase steeply from 2008 onwards. Given that the TCE ratio does not show the same pattern, banks responded to the crisis increasing lower quality capital or shifting their asset portfolio from risky assets to less risky assets. Panel F shows that liquidity was relatively stable up to 2006, thereafter decreasing down to 12.887%-14.694%. The impairments ratio increased in 2008 and 2009, remaining at a higher level thereafter (as shown in Panel G). Column D of Table 1 shows if the mean of the subsamples are significantly different from each other (using a T test). The variable means of the subsamples differ significantly from each other at the 1% level, except for total assets and TCE ratio (which are significant at the 5% level). This so-called structural break is further dealt with in Section 5.5.
Figure 1: Number of banks included in the sample (per year)
0 100 200 300 400 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
Commercial banks Cooperative banks
Bank holding companies Savings banks
14
Figure 2: Bank specific variables (yearly median)
A. Total assets B. Loan growth
C. TCE ratio D. T1 ratio
15 Table 1: Descriptive statistics bank specific variables
A. 2000-2013 B. 2000-2007 C. 2008-2013 D. T testa
Variable Obs. Mean Median Std. dev.
Min Max Obs. Mean Median Std. dev.
Min Max Obs. Mean Median Std. dev.
Min Max Test-statistic Total assets (in bln EUR) 4,527 75.640 9.971 215.000 0.006 2,200.000 2,188 68.11 8.50 183.00 0.01 1,930.00 2,339 82.69 11.50 241.00 0.04 2,200.00 -2.483** Loan growth 4,527 6.377 5.729 14.223 -49.210 50.000 2,188 10.722 9.977 14.342 -48.553 50.000 2,339 2.229 2.401 12.795 -49.210 49.924 21.367*** TCE ratio 4,527 7.471 6.340 5.104 0.010 50.000 2,188 7.288 6.185 4.893 0.290 48.600 2,339 7.640 6.500 5.295 0.010 50.000 -2.315** T1 ratio 2,536 11.333 9.900 6.554 0.090 48.281 1,139 10.011 8.700 4.860 0.120 47.100 1,397 11.927 10.770 5.476 0.090 48.281 -5.776*** TC ratio 2,875 14.162 12.460 10.226 0.012 48.980 1,307 13.095 11.600 5.056 0.120 46.470 1,568 14.369 13.300 5.077 0.012 48.980 -5.193*** Liquidity 4,527 20.351 15.818 16.019 0.009 93.115 2,188 22.572 18.844 16.406 0.009 93.115 2,339 18.212 13.607 15.339 0.035 91.218 9.426*** Impair-ments ratio 4,527 0.508 0.263 1.135 -2.587 29.685 2,188 0.323 0.223 0.570 -2.587 10.180 2,339 0.680 0.319 1.459 -2.276 29.685 -10.731*** Notes:
Summary statistics of the unbalanced sample (2000-2013) after correcting for M&A events and removing outliers. The columns are divided in the entire sample (A), the pre-crisis period (B) and the crisis period (C). Column D shows the results of the T test.
a
Null hypothesis of the two-tailed T test: The sample mean over 2000-2007 is equal to the sample mean over 2008-2013. ***Significance at the 1% level
16
4.4 Descriptive statistics macroeconomic data
Figure 3 shows the macroeconomic variables against time. All variables are collected for the Euro area as a whole. The nominal annual growth rate of GDP is obtained from Eurostat. The inflation rate is measured as the annual growth rate of the Harmonized Index of Consumer Prices (HICP), and obtained from Eurostat. The interbank lending rate is defined as the change in the Euro Overnight Index Average (EONIA) and is obtained via Bloomberg. The red line marks the start of the crisis (with the fall of Lehman Brothers in September 2008). Especially the fall in the GDP growth rate in 2008 and 2009 is remarkable. Also, the very low level of the EONIA rate since 2009 is exceptional.
Figure 3: Macroeconomic variables against time
Notes:
The left axis shows the inflation (growth rate of the HICP index) and the growth rate of GDP. The right axis shows the level of the EONIA rate.
4.5 Stationarity
In (1), ,
,
and are specified in growth rates and is specified as the change in the interbank lending rate (as is standard in the literature, see i.a., Gambacorta and Mistrulli, 2004; Brei et al., 2013), because these variables have nonstationary properties if included in levels. Stationarity avoids spurious correlation problems. For the variables to be (weak) stationary, the means,0 1 2 3 4 5 EONIA (in %) -4 -2 0 2 4
Growth rate (in %)
1998 2000 2002 2004 2006 2008 2010 2012 2014
year
Inflation GDP growth rate
17 variances and covariances cannot depend on time. Nonstationarity is possibly caused by a unit root. Table C1 in Appendix C shows the results of the augmented Dickey-Fuller (ADF) unit root tests for the macroeconomic (time series) variables. The null hypothesis of a unit root for the change in the interbank rate ( ) and GDP growth rate ( ) are rejected at the 5% and 10% level, respectively. The null hypothesis of a unit root for inflation ( ) is rejected at the 1% significance level. However, the time series results might be biased due to the small number of observations (T=14). Multiple approaches are used to test for the stationarity of the bank specific (panel) variables. Levin, Lin and Chu (2001) propose a test assuming that all cross-sectional units have the same autoregressive parameter. On the other hand, the Im, Pesaran and Shin test allows the autoregressive parameter to vary across the cross-sectional units. This test is based on the cross-sectional average of the individual ADF test statistics. Maddala and Wu (1999) and Choi (2001) propose Fisher-type tests that also combine the individual cross-sectional unit root tests. An advantage of these Fisher-type tests over other panel unit root tests is that they do not (formally) require a balanced panel (Baltagi, 2005). The results are in Table C2 in Appendix C. Using the above described tests the null hypothesis of a unit root is rejected for all panel variables at the 1% significance level, except for the TC ratio. The Levin, Lin and Chu test cannot reject the null hypothesis of a unit root for the TC ratio. However, since the other tests do reject the null hypothesis I conclude that the TC ratio is most likely to be stationary.
5. Results
5.1 Estimation techniques
The OLS estimator assumes that the error terms of observations are uncorrelated. However, including the same individual (bank) repeatedly causes the error terms of an individual to be correlated over time. The fixed effects OLS estimator is therefore used for (1)-(3). The fixed effects model is chosen over the random effects model, because the sample of banks is not a random draw from the industry. Furthermore, the fixed effects model is preferred when the independent variables are correlated with the individual specific error terms. The Hausman test is used to test the random effects versus the fixed effects model. The results for the model in (1) are shown in Appendix D. The Hausman test rejects the null hypothesis of no correlation between the independent variables and bank specific error term. This supports the use of the fixed effects model (see Section E.1 in Appendix E for a basic description of the fixed effects OLS model).
18 correlation is rejected, given all three definitions of the capital ratio. Given the presence of serial correlation, the fixed effects estimator remains consistent. However, it does invalidate the standard errors. Furthermore, the presence of heteroskedasticity also implies that standard errors are invalid. Stock and Watson (2003, pp. 128-129) argue: “At a general level, economic theory rarely gives any reason to believe that the errors are homoskedastic. It is therefore prudent to assume that errors might be heteroskedastic unless if you have compelling reasons to believe otherwise6.” The presence of serial correlation and heteroskedasticity requires appropriately adjusted standard errors. Verbeek (2012) proposes using a variant of Newey-West (1987) standard errors, also named cluster-robust standard errors (where each bank i forms a cluster). This approach allows for serial correlation and heteroskedasticity in fixed effects OLS estimation. As mentioned in Section 3.4, autocorrelation could be due to misspecification of the model. As a robustness check, Section 5.8 will deal with a dynamic specification of the model.
5.2 Baseline results
The results of the fixed effects OLS estimation with cluster-robust standard errors of the baseline regression (1) are shown in Table 2. The first two columns show the results for the model where , is defined as the TCE ratio. The third and fourth columns of Table 2 show the results
with the T1 ratio and the last two columns show the results of model (1) with the TC ratio included.
The coefficients of , are positive and significant (at the 1% level) for the TCE and T1
ratio. The coefficient of the TC ratio is positive and significant at the 5% level. This supports the hypothesis that capitalization increases lending growth. As expected, the TCE ratio has the largest impact on loan growth. The average effect of a 1% increase in the TCE ratio is a 0.562% increase in the loan growth of the average bank. For the T1 and TC ratio the average impact is 0.500% and 0.316%, respectively.
The results are generally in line with previous literature. Prior literature also finds a significantly positive impact of capitalization on loan growth. However, the size of the estimated coefficients fluctuates across studies. Although the interpretation of their results is not entirely comparable, Berrospide and Edge (2010) find that a 1% increase in a bank’s capital ratio increases loan growth by 0.580%-1.008% (range is based on the definition of the capital ratio). Their estimates are thus higher than the results in this paper. Berrospide and Edge (2010) also find that the TCE ratio has the largest impact on lending growth (compared to broader measures of capital ratios). Bernanke and Lown (1991) find that a 1% increase in equity capital increases loan growth in the range of 2% to 2.5%. Carlson et al. (2013) find evidence of a smaller impact of bank capital ratios on loan growth.
6
19 They find that the impact of a 1% increase in a bank’s capital ratio on loan growth is between 0.05%-0.20% (again depending on the definition of the capital ratio).
There are a few characteristics of these studies that could explain the different results. Where this study is on the Euro area the above-mentioned papers use US data. However, this does not explain the differences among the papers using US data. Different banks with different characteristics (especially bank size) appear in the samples of these studies, this could account for some of the variation within the results. Berrospide and Edge (2010) include the 165 largest bank holding companies of the US. Bernanke and Lown (1991) include about 100 (small) local banks from the US state New Jersey. Carlson et al. (2013) also use small local banks. This indicates that bank size might be a relevant indicator for the impact of bank capital on lending growth. Therefore, size is dealt with in Section 5.3. Furthermore, the modelling techniques differ across studies. The specification used in this study is closest to the model of Berrospide and Edge (2010). Bernanke and Lown (1991) include the growth rate of employment to control for economic activity and do not include any bank specific control variables. The estimated impact of the capital ratio might thus capture other effects. As mentioned in the literature review, Carlson et al. (2013) use an entirely different methodology to account for the demand for credit.
The coefficient of the growth rate of GDP ( ) is significantly positive (at the 1% level). Increasing GDP is a sign of increasing economic activity; this boosts the demand for loans. In accordance with previous literature, the coefficient of the inflation rate ( ) is negative and significant at the 1% level (i.a., Ehrmann et al., 2003; Berrospide and Edge, 2010). The coefficient of the change in the interbank rate ( ) is negative, but its significance depends on the definition of the capital ratio. The coefficient is significant at the 1% level if the TCE or TC ratio is included and at the 5% level if the T1 ratio is included. A higher interbank borrowing rate decreases the demand for loans.
The coefficient of liquidity ( , ) is significantly positive for the TCE ratio (at the 5%
level). The coefficient for , is still positive but insignificant if the TC ratio is used as a proxy
for the capital ratio. If the T1 ratio is used however, the parameter estimate for , has the opposite sign (and is insignificant). The argument that banks adjust their lending using liquid securities, as first noticed by Kashyap and Stein (1995, 2000), is thus not fully supported. The coefficient of impairments to total assets ( , ) is significantly negative regardless of the definition of the capital ratio. , is significant at the 5% level if the TCE ratio is included and at
20 As seen from the results in Table 2, the TCE ratio has the largest and most significant impact on loan growth. This could be due to the fact that the TCE ratio is the strictest definition on capital and is most loss-absorbing. Also, the risk-weighted capital ratios are under-represented in the Bankscope database and the results could suffer from a small amount of observations after further cutting up the sample. Therefore, I will continue with a focus on the TCE ratio. Also, among others, Gambacorta and Marques-Ibanez (2013) argue that core capital is the most important determinant of loan growth. This supports the focus on the TCE ratio in further parts of this study.
Table 2: Results of the baseline regression over the entire sample
Variables TCE T1 TC
Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error
Macroeconomic controls
2.322*** 0.188 2.690*** 0.246 2.939*** 0.247
-4.424*** 0.325 -5.393*** 0.410 -5.478*** 0.409 -1.125*** 0.341 -1.045** 0.486 -1.424*** 0.478 Bank specific characteristics
, 0.562*** 0.175 0.500*** 0.171 0.316** 0.135 , 0.098** 0.040 -0.026 0.058 0.042 0.050 , -1.893** 0.942 -3.249*** 0.567 -3.385*** 0.468 Other controls , 8.093** 3.402 3.117 2.425 3.291 2.388 Observations 3813 1999 2261 R2 (within) 0.105 0.157 0.150 Notes:
This table reports the results of the baseline regression (1) over the entire sample period (2000-2013). The parameters are estimated using the OLS model with individual fixed effects. Cluster-robust standard errors are reported. The dependent variable is the growth rate of loans ( ,). Explanatory variables are (all in lags): GDP growth rate, inflation, change in
the interbank rate, capital ratio, liquidity and the impairments ratio. Furthermore, a dummy that controls for a change in accounting standards is included (not in lags).
***Significance at the 1% level **Significance at the 5% level *Significance at the 10% level
5.3 Bank size
21 To account for the effect of bank size on the results of model (1), two groups (subsamples) are created. The first group consists of all small banks, defined as the banks with total assets below the 50th percentile of total assets. The second group consists of all banks with total assets above the 50th percentile. Since the median bank size in the sample is growing over the years (see Panel A in Figure 2), the 50th percentile is determined annually. A bank could be included in the group of small banks in one year, while being included in the group of large banks in the following year. The results of running regression (1) separately for the small and large group of banks are in Table 3. For sake of brevity, only the variables of interest are included (the control variables have the same signs as in Table 2). The table shows that the effect of bank capital ratios on loan growth is greater for large banks than for small banks. The average impact of a 1% increase in the TCE ratio on loan growth is 1.054% for the average large bank (significant at the 1% level) versus 0.425% for small banks (significant at the 5% level). A Chow test is performed to see whether the regression coefficients in (1) differ significantly for the subsamples. The null hypothesis of no structural change (the regression coefficients of the subsamples are equal) is rejected at the 1% significance level. Carlson et al. (2013) find the same result. They estimate that the impact of a 1% increase in the capital ratios on lending growth is between 0.086%-0.267% for smaller banks in their sample. For the larger banks in their sample this estimate increases to 0.513%-0.623%.
Table 3: Results of the baseline regression with bank size subsamples Variables TCE (small banks) TCE (large banks)
Coefficient Std. Error Coefficient Std. Error
, 0.425** 0.209 1.054*** 0.176
Observations 1770 1960
R2 (within) 0.075 0.158
Chow test (F-statistic) 4.559*** Notes:
This table reports the results of the baseline regression (1) over the entire sample period (2000-2013). The parameters are estimated using the OLS model with individual fixed effects. Cluster-robust standard errors are reported. The sample is split in large and small banks (at the yearly median of total assets). Only the variables of interest are shown. The dependent variable is the growth rate of loans ( ,). Explanatory variables are (all in lags): GDP growth rate, inflation, change in
the interbank rate, capital ratio, liquidity and the impairments ratio. Furthermore, a dummy that controls for a change in accounting standards is included (not in lags).
***Significance at the 1% level **Significance at the 5% level *Significance at the 10% level
22 closer to the requirement minimum), because they have the resources to manage bank capital more precisely. Gambacorta and Mistrulli (2004) argue that capital ratios are lower for banks that have easier access to equity markets. This is generally true for larger banks. Also, large banks might have different incentives in managing their capital structure, because there are more likely to get rescued in case of failure (they are too big to fail). The European Banking Authority (EBA) also finds that large multinational banks have lower capital ratios than smaller banks (EBA, 2014). This implies that the impact of bank capitalization is nonlinear, because the impact of an increasing bank capital ratio on loan growth is greater if a bank is close to the required minimum. I deal with nonlinearity in the next section (Section 5.4).
5.4 Nonlinearity
23
Table 4: Results of nonlinearity
Variables TCE Coefficient Std. Error , ∗ , 1.663* 0.863 , ∗ , 0.307 0.312 , ∗ , 0.019 0.223 Observations 3813 R2 (within) 0.107 Notes:
This table reports the results of regression (2) over the entire sample period (2000-2013). The parameters are estimated using the OLS model with individual fixed effects. Cluster-robust standard errors are reported. Only the variables of interest are shown. The dependent variable is the growth rate of loans ( ,). Explanatory variables are (all in lags):
GDP growth rate, inflation, change in the interbank rate, capitalization dummies, capitalization-capital ratio interaction terms, liquidity and the impairments ratio. Furthermore, a dummy that controls for a change in accounting standards is included (not in lags).
***Significance at the 1% level **Significance at the 5% level *Significance at the 10% level
5.5 Pre-crisis and crisis period
Based upon previous literature the relation between bank capital ratios and lending growth might be different during crisis years. Gambacorta and Marques-Ibanez (2011) are interested in the impact of monetary policy on bank lending and whether this relation changes during the crisis (their data set runs from 1999 to 2009). They find structural changes in the behavior of banks during crisis years. Berger and Bouwman (2009) find that banks with higher capital ratios are more likely to survive in crisis years. Furthermore, they find that during the crisis, well-capitalized banks are able to increase their market share (measured by total assets) against low-capitalized banks. Carlson et al. (2013) find that the recent financial crisis enlarges the effect of bank capital ratios on loan growth.
24 that capitalization has a larger impact on loan growth during crisis years. They find that the impact of a 1% increase in the capital ratio on loan growth is 0.707% in pre-crisis years and 0.899% in crisis years. Carlson et al. (2013) find that the effect of bank capital ratios on loan growth is practically zero and highly insignificant during 2001-2007. From 2008 to 2011 however, they find that the impact of bank capital ratios on loan growth is between 0.114%-0.276%.
Table 5: Results of the baseline regression with pre-crisis and crisis subsamples
Variables TCE (2000-2007) TCE (2008-2013)
Coefficient Std. Error Coefficient Std. Error
Macroeconomic controls
0.222 0.483 1.178*** 0.312
-6.910*** 3.687 -2.595*** 0.433
0.612 0.562 -2.412*** 0.693
Bank specific characteristics
, 0.272 0.282 0.481* 0.291 , 0.081 0.055 0.190** 0.084 , -0.264 1.638 -0.563 0.561 Other controls , 4.033 2.983 0.360 1.343 Observations 1711 1810 R2 (within) 0.037 0.053
Chow test (F-statistic) 42.341*** Notes:
This table reports the results of baseline regression (1) over subsets of the sample period (2000-2007 and 2008-2013). The parameters are estimated using the OLS model with individual fixed effects. Cluster-robust standard errors are reported. The dependent variable is the growth rate of loans ( ,). Explanatory variables are (all in lags): GDP growth rate, inflation,
change in the interbank rate, capital ratio, liquidity and the impairments ratio. Furthermore, a dummy that controls for a change in accounting standards is included (not in lags).
***Significance at the 1% level **Significance at the 5% level *Significance at the 10% level
25 Liquidity ( , ) is only significant at the 5% level in crisis years and the impairment ratio
( , ) is insignificant for both subsamples. However, the size of the coefficients is smaller in pre-crisis years for both variables. Note that the coefficient of the IFRS dummy ( , ) is much lower
for the crisis period. However, this is probably due to the fact that most banks switched from accounting regime in 2005. Note that the R2 in Table 5 is equal to 0.037 in the pre-crisis period and 0.053 in the crisis period. In Table 2 the R2 of the model including the TCE ratio is 0.107. The model does a better job in describing the data of the total sample than the subsamples.
5.6 Loan categories
26
Table 6: Results of the baseline regression on residential mortgage and corporate and commercial loans
Variables TCE (residential mortgage) TCE (corporate and commercial)
Coefficient Std. Error Coefficient Std. Error
Macroeconomic controls 1.818*** 0.471 1.505*** 0.390 -3.194*** 0.795 -4.996*** 0.822 -1.847** 0.915 0.771 0.718 Bank specific characteristics , 0.478 0.609 0.769*** 0.254 , -0.042 0.147 0.064 0.103 , -2.901** 1.263 -1.066 1.033 Other controls , 14.445** 2.204 8.439*** 1.604 Observations 506 910 R2 (within) 0.104 0.101 Notes:
This table reports the results of regression (1) over the entire sample period (2000-2013). The parameters are estimated using the OLS model with individual fixed effects. Cluster-robust standard errors are reported. The dependent variable of the second and third column is the growth rate of residential mortgage loans. The dependent variable of the fourth and fifth column is the growth rate of corporate and commercial loans. Explanatory variables are (all in lags): GDP growth rate, inflation, change in the interbank rate, capital ratio, liquidity and the impairments ratio. Furthermore, a dummy that controls for a change in accounting standards is included (not in lags).
***Significance at the 1% level **Significance at the 5% level *Significance at the 10% level
5.7 Time dummies
27 with macroeconomic variables. This is quite remarkable, because this would imply that the model in (3) better fits the data than (1). Possibly, regression (1) does not adequately capture all relevant demand and supply factors.
Table 7: Results of the specification with time dummies
Variables TCE T1 TC
Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Bank specific characteristics
, 0.388*** 0.145 0.427** 0.169 0.260** 0.112 , 0.079** 0.039 -0.010 0.061 0.023 0.053 , -2.083** 0.920 -3.102*** 0.466 -3.642*** 0.447 Other controls , 4.552 3.324 -0.038 2.813 -0.489 2.876 Observations 3813 1999 2261 R2 (within) 0.212 0.291 0.303 Notes:
This table reports the results of regression (3) over the entire sample period (2000-2013). The parameters are estimated using the OLS model with individual and time fixed effects. Cluster-robust standard errors are reported. The dependent variable is the growth rate of loans ( ,). Explanatory variables are (all in lags): capital ratio, liquidity and the impairments ratio. Furthermore, time dummies are included and a dummy that controls for a change in accounting standards is included (dummy variables are not in lags).
***Significance at the 1% level **Significance at the 5% level *Significance at the 10% level
5.8 Dynamic model
The model in (1) is subject to serial autocorrelation in the error terms. Up to now cluster-robust standard errors are reported, to allow for serial correlation in the error term. However, autocorrelation may result from a misspecified model. Brooks (2008) argues that if error terms are autocorrelated, a dynamic specification (including a lagged dependent variable) might better fit the data. This approach has also been used in previous literature (i.a., Ehrmann et al., 2003; Berrospide and Edge, 2010; Brei et al., 2013).
28 The results of (4), using system GMM, are in Table 8. The estimates for , are still
highly significant and positive, but differ in size from the results in Table 2. The coefficients are again lower than in Table 2. Especially, the estimated impact of a 1% increase of the TCE ratio on average loan growth is 0.277% for the average bank in the dynamic model, versus 0.562% in the baseline regression. Remarkably, the coefficients of the T1 and TC ratio are larger than the coefficient of the TCE ratio. This provides evidence against the argument that the impact of the TCE ratio on loan growth should be largest, because of its loss-absorbing quality. However, this finding could be biased, because the data set includes fewer observations on T1 and TC ratios than TCE ratios.
Table 8: Results of the dynamic model
Variables TCE T1 TC
Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error
0.280*** 0.049 0.334*** 0.058 0.363*** 0.054
Macroeconomic controls
0.946*** 0.214 0.864*** 0.288 0.914*** 0.288
-3.479*** 0.361 -3.900*** 0.493 -3.934*** 0.494
0.019 0.424 0.475 0.584 0.426 0.588
Bank specific characteristics
, 0.277*** 0.065 0.419*** 0.085 0.394*** 0.102 , 0.046** 0.023 -0.016 0.031 0.006 0.031 , -0.325 0.293 -0.194 0.436 -0.356 0.403 Other controls , 4.249* 2.564 2.384 2.472 2.300 2.490 Observations 3813 1999 2261 AR(1) (p-value) 0.000 0.000 0.000 AR(2) (p-value) 0.374 0.105 0.090
Hansen J test (p-value) 0.135 0.109 0.103
Notes:
This table reports the results of the dynamic model over the entire sample period (2000-2013). The parameters are estimated using the system GMM model with individual fixed effects. The dependent variable is the growth rate of loans ( ,).
Explanatory variables are (all in lags): loan growth rate, capital ratio, liquidity and the impairments ratio. Furthermore, a dummy that controls for a change in accounting standards is included (not in lags).
29 The validity of the system GMM model depends on some additional assumptions7. The system GMM approach requires that the model cannot exhibit second order autocorrelation (AR(2)). Arellano and Bond (1991) propose a test for autocorrelation of the residuals. By definition, the null hypothesis for no first-order autocorrelation is rejected. The null hypotheses of no second order autocorrelation is rejected at the 10% level for the TC ratio but is not rejected for the TCE and T1 ratio. To ensure the validity of the system GMM estimator the instruments need to be valid. The Hansen J test is used to test for overidentifying restrictions. Rejection of the null hypothesis indicates one or more instruments are invalid. The p-values of 0.103-0.135 imply that the instruments are valid.
6. Conclusion
This study uses data from 2000 up to 2013 to assess the impact of bank capital ratios on Euro area banks. In accordance with the literature my research finds that this relation is significantly positive. This finding supports that bank capital mitigates adverse selection and moral hazard problems and thus positively affects the supply of bank credit. This research includes multiple definitions of capital ratios, which differ in their loss-absorbing quality. I find that the effect of the TCE ratio is largest and most significant, however this finding is not robust to alternative specifications of the model. Furthermore, the effect of an increase in the capital ratio is greater for large banks than for small banks. I argue that this is due to the fact that larger banks manage their capital structure more actively than small banks, and operate closer to the minimum required capital ratio. This would imply that the impact of bank capital ratios on loan growth is greater for low-capitalized banks than well-capitalized banks. Low-well-capitalized banks operate closer to the minimum required capital ratio. At this boundary the impact of a 1% increase in the capital ratio is larger than for banks further away from the minimum requirement. This finding suggests that the relation between capitalization and loan growth is nonlinear. This research shows that the impact of bank capital ratios on loan growth intensified in crisis years. The difficulties in valuing bank assets become larger in financial crises and this increases asymmetric information problems. Furthermore, the impact of bank capital ratios on corporate and commercial loans is larger than total loans and residential mortgage loans. This finding could be due to the fact that consumers have limited alternative financing possibilities when it comes to mortgage loans. Companies (especially large companies) have easier access to other sources of finance. As a robustness check I replace the macroeconomic control variables with a set of time dummies. The results have the expected signs and are equally significant as compared to the baseline model. However, the coefficients for the capital ratios are smaller. This could imply that the macroeconomic variables do not entirely capture the demand for loans. Also, the T1 ratio has a larger average impact on loan growth than the TCE ratio. As an additional robustness check, I estimate a dynamic version of
7
To ensure validity, lags 2 through 5 ( , , . . , , ) are used as instruments to the lagged dependent
30 the baseline model by including a lagged dependent variable on the right hand side. The coefficients are smaller than in the baseline regression. Also, the average impact of the T1 and TC ratio on loan growth is larger than the impact of the TCE ratio.
6.1 Discussion and propositions for further research
As mentioned in Section 4.2, the Bankscope database suffers from several drawbacks. The data I use is unbalanced, because balancing the panel would lead to the loss of a great amount of observations. It would therefore be meaningful to complement this research with studies using data from other sources. Furthermore, the methodology used in this research remains sensitive to modeling assumptions. Additional research on the Euro area using other methods to separate demand and supply effects is needed. However, these other methodologies often require highly detailed data that is not publically available. Further research on the endogeneity problem is needed. The argument that this problem is overcome by taking lagged values of the variables is not fully convincing. The positive relation between lagged capital ratios and loan growth may also come from banks increasing their capital ratio in advance if they expect increasing loan volumes. Berrospide and Edge (2010), for example, argue that this problem can be overcome using a measure of excess capital (defined as actual capital minus an estimated target amount of capital).
6.2 Acknowledgements
For helpful comments and feedback I would like to thank Prof. Dr. J.M. Berk and N. Knop.
References
Albertazzi, U., Marchetti, D., 2010. Credit supply, flight to quality and evergreening: An analysis of bank-firm relationships after Lehman. Working Paper No. 756, Bank of Italy.
Anderson, T. W., Hsiao, C., 1981. Estimation of dynamic models with error components. Journal of the American Statistical Association, 76, 598-606.
Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and and aplication to employment equations. Review of Economic Studies, 58, 277-297. Arellano, M., & Bover, O., 1995. Another look at the instrumental-variable estimation of
error-component models. Journal of Econometrics, 58, 5-28.
Baltagi, B., 2005. Econometric Analysis of panel data. John Wiley & Sons Ltd., Chichester
BCBS., 2010. Basel III: A global regulatory framework for more resilient banks and banking systems. Berger, A., Bouwman, C., 2009. Bank capital, survival, and performance around financial crises.
Wharton Financial Institutions Center Working Paper, 09-24.
31 Bernanke, B., 2007. “The financial accelerator and the credit channel.” Remarks - credit channel of
monetary policy in the twenty-first century. Board of Governors of the U.S. Federal Reserve System.
Bernanke, B., Gertler, M., 1995. Inside the black box: The credit channel of monetary policy. Journal of Economic Perspectives, 9(4), 27-48.
Bernanke, B., Lown, C., 1991. The credit crunch. Brookings Papers on Economic Activity, 2, 205-247.
Berrospide, J., Edge, R., 2010. The effects of bank capital on lending: what do we know, and what does it mean? International Journal of Central Banking, 6(4), 5-54.
Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87, 115-143.
Brei, M., Gambacorta, L., & Peter, G., 2013. Rescue packages and bank lending. Journal of Banking & Finance, 37, 490-505.
Brooks, C., 2008. Introductory econometrics for finance. Cambridge University Press, Cambridge. Calomiris, C., Hubbard, G., 1995. Internal finance and investment: evidence from the undistributed
profit tax of 1936-1937. Journal of Business, 68, 443-482.
Carlson, M., Shan, H., Warusawitharana, M., 2013. Capital ratios and bank lending: a matched bank approach. Journal of Financial Intermediation, 22, 663-687.
Choi, I., 2001. Unit root tests for panel data. Journal of International Money and Finance, 20, 249-272. Choudhry, M., 2012. The principles of banking. John Wiley & Sons Ltd., Singapore.
Cornett, M., Tehranian, H., 1994. An examination of voluntary versus involuntary security issuances by commercial banks: The impact of capital regulations on common stock returns. Journal of Financial Economics, 35, 99-122.
Demirguc-Kunt, A., Detragiache, E., & Merrouche, O., 2010. Bank capital: Lessons from the financial crisis. World Bank Policy Research Working paper 5473.
Duprey, T., Lé, M., 2014. Bank capital adjustment process and aggregate lending. Banque de France Working Paper No. 499.
EBA, 2014. Basel III monitoring exercise (September 2014). ECB, 2014. Banking structures report (October 2014).
Ehrmann, M., Gambacorta, L., Pagés, J., Sevestre, P., 2003. Financial systems and the role of banks in monetary policy transmission in the Euro area. ECB Working paper No. 105.
European Commission (Memo), 2014. Banking union: restoring financial stability in the Eurozone. Retrieved from: http://ec.europa.eu/finance/general-policy/docs/banking-union/banking-union-memo_en.pdf
Francis, W., Osborne, M., 2012. Capital requirements and bank behavior in the UK: Are there lessons for international capital standards? Journal of Banking & Finance, 36, 803-816.
Freixas, X., Rochet, J., 1997. Microeconomics of banking. MIT Press, Cambridge.
Friedman, B., 1991. Comments and discussion. Brookings Papers on Economic Activity, 2, 240-244. Gambacorta, L., 2005. Inside the bank lending channel. European Economic Review, 49, 1737-1759. Gambacorta, L., Marques-Ibanez, D., 2011. Bank lending channel: lessons from the crisis. Economic
Policy (April), 135-182.
32 Hancock, D., Wilcox, J., 1993. Has there been a 'capital crunch' in banking? The effects on bank
lending of real estate market conditions and bank capital shortfalls. Journal of Housing Economics, 3, 31-50.
Hancock, D., Wilcox, J., 1994. Bank capital and the credit crunch: the roles of risk-weighted and unweighted capital regulations. Journal of the American Real Estate and Urban Economics Association, 22, 59-94.
Hancock, D., Laing, A., Wilcox, J., 1995. Bank capital shocks: Dynamic effects on securities, loans, and capital. Journal of Banking & Finance, 19, 661-677.
Jayaratne, J., Morgan, D., 2000. Capital markets frictions and deposit constraints at banks. Journal of Money, Credit and Banking, 32, 74-92.
Kashyap, A., Stein, J., 1995. The impact of monetary policy on bank balance sheets. Carnegie-Rochester Conference Series on Public Policy, 42(1), 151-195.
Kashyap, A., Stein, J., 2000. What do a million observations on banks say about the transmission of monetary policy? American Economic Review, 90(3), 407-428.
Kishan, R., Opiela, T., 2000. Bank size, bank capital and the bank lending channel. Journal of Money, Credit and Banking, 32, 121-141.
Levin, A., Lin, C.-F., Chu., C.-S., 2002. Unit root tests in panel data: Asymptotic and finite-sample properties. Journal of Econometrics, 108, 1–24.
Maddala, G., Wu, S., 1999. A comparative study of unit root tests with panel data and a new simple test. Oxford Bulletin of Economics and Statistics, 61, 631-652.
Mishkin, F., 1996. The channels of monetary transmission: Lessons for Monetary Policy. NBER Working Papers, 5464.
Mora, N., Logan, A., 2012. Shocks to bank capital: evidence from UK banks at home and away. Applied Economics, 44, 1103-1119.
Myers, S., Majluf, N., 1984. Corporate finance and investment decisions when firms have information that investors do not have. Journal of Financial Economics, 13, 187-221.
Newey, W., West, K., 1987. A simple positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 69, 1519-1554.
Peek, J., Rosengren, E., 1995. The capital crunch: neither a borrower nor a lender be. Journal of Money, Credit and Banking, 27, 625-638.
Peek, J., Rosengren, E., 1997. The international transmission of financial shocks: The case of Japan. American Economic Review, 87, 495-505.
Peek, J., Rosengren, E., 2000. Collateral damage: effects of the Japanese bank crisis on real activity in the United States. American Economic Review, 90, 30-45.
Rice, T., Rose, J., 2010. When good investments go bad: the contraction in community bank lending after the 2008 GSE takeovers. Mimeo.
Romer, C., Romer, D., 1990. New evidence on the monetary transmission mechanism. Brookings Papers on Economic Activity, 1, 149-213.
Roodman, D., 2009. A note on the theme of too many instruments. Oxford Bulletin of Economics and Statistics, 71, 135-158.
33 Stein, J., 1998. An adverse-selection model of bank asset and liability management with implications
for the transmission of monetary policy. RAND Journal of Economics, 29, 466-486. Stock, J., Watson, M., 2003. Introduction to Econometrics. Addison-Wesley, Boston.
Van den Heuvel, S., 2001. The bank capital channel of monetary policy. Mimeo. University of Pennsylvania.
Van den Heuvel, S., 2002. Does bank capital matter for monetary transmission? FRBNY Economic Policy Review, 259-265.
34
Appendices
Appendix A
Table A1: Variables definition and source
Variable Notation Definition Source
Bank specific
Loan growth , Annual growth rate of net loans Bankscope
Tangible common equity (TCE) ratio
, Tangible common equity (common equity and retained earnings) divided
by tangible assets.
Bankscope
Tier 1 (T1) capital ratio
, Tier 1 capital (consists of common equity, retained earnings and
perpetuals) over risk-weighted assets (as defined by the regulator)
Bankscope
Total capital (TC) ratio
, Includes all capital (tier 1 plus subordinated debt instruments with
maturity >5 years) divided by risk-weighted assets (as defined by the regulator)
Bankscope
Liquidity , High quality liquid securities over total assets Bankscope
Impairments ratio
, Net impairment charge in relation to the bank's loans and advances over
total assets
Bankscope
Total assets - Total tangible and intangible assets Bankscope Macroeconomic
GDP growth rate , Annual growth rate in GDP at market prices based on local currency Eurostat
Inflation , The annual growth rate of the Harmonized Index of Consumer Prices
(HICP)
Eurostat
Change in interbank lending rate
35
Appendix B
Table B1: Number of banks and observations per country Country Number of banks Number of observations
36
Appendix C
Table C1: Unit root tests macroeconomic variables
Variable Unit root test Test-statistic Probability GDP growth rate Augmented Dickey-Fuller -2.77 0.087
Inflation Augmented Dickey-Fuller -4.30 0.007
Interbank rate (change)
Augmented Dickey-Fuller -3.59 0.022
Notes:
Null hypothesis: Variable has a unit root.
Table C2: Unit root tests bank specific variables
Variable LLC* IPS** ADF (Fisher)*** PP (Fisher)***
Test-statistic Prob. Test-statistic Prob. Test-statistic Prob. Test-statistic Prob. Loan growth -43.100 0.000 -16.199 0.000 1504.187 0.000 1727.827 0.000 TCE ratio -33.232 0.000 -8.706 0.000 1156.439 0.000 1271.722 0.000 T1 ratio -11.551 0.000 -7.311 0.000 687.539 0.000 814.900 0.000 TC ratio 3.646 1.000 -8.218 0.000 827.622 0.000 877.874 0.000 Liquidity -32.880 0.000 -10.967 0.000 1229.752 0.000 1212.794 0.000 Impair-ments ratio -79.500 0.000 -10.912 0.000 1152.368 0.000 1150.432 0.000 Notes:
Null hypothesis: Variable has a unit root.
*Levin, Lin and Chu (LLC) t* (test assumes a common unit root process) **Im, Pesaran and Shin (IPS) W-Stat (test assumes individual unit root process)
37
Appendix D
Table D1: Hausman test
Test-statistic (Chi-square) P-value
Hausman test 48.921 0.0000
Notes:
Null hypothesis: Independent variables and the bank specific error terms are uncorrelated.
Table D2: Woolridge test for autocorrelation
Variables TCE Tier 1 TC
Test- statistic P-value Test- statistic P-value Test- statistic P-value Woolridge test 33.238 0.000 41.192 0.000 54.386 0.000 Notes:
