The effect of holding more liquidity on the lending behavior of European Banks

(1)

Master thesis Business Economics (Finance track)

The effect of holding more liquidity on the lending

behavior of European Banks

Name

: Saskia Lodders

Student number

: 10191224

Supervisor

: Razvan Vlahu

(2)

2

Abstract

During the recent financial crisis it became clear that banks held insufficient liquidity buffers to handle the withdraw of short-term debt and the collapse of asset prices. In response to the crisis the Basel Committee introduced Basel III in 2011. Basel III included multiple reforms and one of the reforms was the introduction of liquidity regulation. Due to the introduction of liquidity regulation, critics are afraid that banks will limit the availability of credit. Since liquidity regulation is relatively new, not any empirical research has been done to the effect of holding more liquidity on the outstanding loans of a bank and therefore this thesis examines empirically the following research question: how does the amount of liquidity a bank holds affect the lending behavior of European Banks? Since central banks determine the rules for banks, European banks is a homogeneous group to examine. This research is done by conducting a fixed effects regression for the time period 2005-2014 and the crisis period in Europe, specifically 2008-2012. Moreover, the regression is conducted for small, medium and large banks separately. The main independent variable is the liquidity ratio and the dependent variable used is the log of total loans. Considering the whole period (2005-2014), this research suggests that when a bank holds more liquidity, the effect on total outstanding loans is negative and significant at the 1% level. The effect is somewhat smaller but still negative and significant when only the crisis period (2008-2012) is considered. When small, medium and large banks are examined separately regarding the time period 2005-2014, the effect of holding more liquidity on total outstanding loans is again negative and significant for each group. This effect is the largest for small banks. With respect to the crisis period, the effect for small banks is negative and significant. Again, holding more liquidity negatively affects the total loans outstanding of small banks.

Statement of Originality

This document is written by Student Saskia Lodders who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

(3)

3

Table of contents

1. Introduction 4

2. Literature review 6

2.1 Liquidity creation 6

2.1.1 Theories regarding liquidity creation 6

2.1.2 Empirical findings regarding liquidity creation 8

2.2 How liquidity evolved over time towards the crisis 8

2.2.1 Theories regarding how liquidity can evolve over time 9

2.2.2 Empirical findings regarding how liquidity actually evolved towards the crisis 9

2.3 Liquidity and lending during crises periods 10

2.3.1 Theories regarding liquidity and lending during crises periods 11

2.3.2 Empirical findings regarding liquidity and lending during crises periods 11

2.4 Liquidity regulation 12

2.4.1 Theories regarding liquidity creation 12

2.4.2 Empirical evidence regarding liquidity creation 14

2.5 Summary of the literature 15

3. Hypotheses, Methodology and Data 16

3.1 Hypotheses 16

3.2 Methodology 16

3.3 Data and descriptive statistics 20

3.3.1 Data description 20

3.3.2 Descriptive statistics 21

4. Results 25

4.1 Empirical results 25

4.2 Robustness check 30

5. Discussion and Conclusion 33

5.1 Discussion 33

5.2 Conclusion 33

(4)

4

1. Introduction

During the recent global crisis that started in 2007, many banks relied (partly) on wholesale funding. Wholesale funding is most of the time short-term and needs to be rolled-over on a short-term basis (Huang and Ratnovski, 2011). Banks that relied (partly) on wholesale funding experienced difficulties when funding dried up during this crisis, because banks held insufficient liquidity buffers. In this financial crisis it became clear that Basel II (introduced in 2004 by the Basel Committee) had some shortcomings. One of these shortcomings was the lack of a framework against liquidity risk. Liquidity risk refers to the risk that debt holders (retail depositors and wholesale depositors) do not rollover, and the bank has to liquidate assets to come up with the cash (Perotti, 2010).

The Basel Committee, introduced in 1988, is a global committee that sets standards for prudential regulation. Until the recent crisis, the Basel Committee did not focus on liquidity regulation. The regulators thought liquidity regulation was a concern of the national authorities and before 2007 there had not been an event that showed financial markets had difficulties due to liquidity risk. However, the recent crisis showed what kind of damage could be caused by irresponsible liquidity management. Therefore, the Basel Committee introduced Basel III with a number of reforms in 2011, compared to Basel II (Bonner and Hilbers, 2015). One of the reforms was the introduction of liquidity regulation. The aim of the regulations is to strengthen the resilience of the banks in periods of stress (Basel Committee on Banking Supervision, 2011).

Many critics on Basel III are afraid that the reforms, changes in capital regulation and the introduction of liquidity regulation, will change the lending behavior of banks and limit the availability of credit (Allen et al., 2012). Lending behavior of banks can be understood as follows: loans to corporations and private customers (for example mortgages). These specific loans are part of the illiquid assets on the balance sheet of a bank. It may be the case that when banks need to hold more liquid assets to conform to the liquidity requirements, liquid assets (cash, securities) will replace these specific illiquid assets on the balance sheet of the bank. However, it could also be the case that banks will expand their balance sheet or reduce other assets (for example real estate owned by banks).

There has already been many research conducted on the impact of capital requirements on the lending behavior of banks. However, since the liquidity regulation is relatively new, there is much less research conducted regarding the effects of holding more liquidity on lending behavior of banks.

Until now, not any empirical research have explored the possible relationship between liquidity and lending behavior of banks for the time period 2005-2014 specifically. That is why this thesis examines empirically the following research question: how does the amount of liquidity a bank holds affect the lending behavior of European Banks?

(5)

5 In the time period 2005-2014 the recent crisis is included, but also periods of normal growth are captured. Since Basel III was introduced in 2011, the transition period for banks to conform to the liquidity requirements is captured too.

Moreover, this research is focused on European banks specifically. The Basel Committee provides guidelines for the central banks and the central banks are the entities that determine the rules for banks. European banks could therefore be seen as a homogeneous group to examine.

To answer the research question empirically, two sources will be used. From Datastream all relevant bank characteristics for European listed banks will be obtained and macroeconomic data like GDP growth will be gathered from Quandl.

Based on the existing literature about banks and liquidity, it is expected that changes in the amount of liquidity of banks do not immediately affect its lending behavior. To capture this time element, a fixed effects regression will be conducted. Since the (possible) relationship might differ during the recent crisis compared to the total time period (2005-2014), two different regressions will be performed: one regression including all data and one regression including only the crisis years. Moreover, based on Berger and Bouwman (2009a), the regression will also be conducted for small-, medium- and large banks separately. The findings of this thesis will be relevant for regulators and policymakers, to illustrate if the criticisms on the reforms are grounde

d.

This thesis is structures as follows. The second chapter gives an overview of the existing literature on liquidity and the effect on loans outstanding. The third chapter presents the hypotheses, explains which method was used for analyzing the data and which method was used to obtain the data for answering the research question. Next, chapter four discusses the results of the data analysis. Finally, chapter five discusses this research and concludes and provides suggestions for further research.

(6)

6

2. Literature review

Banks create liquidity for the economy by borrowing short-term en lending long-term to investors. This maturity transformation is risky for banks and to reduce this risk they hold liquid assets and issue equity (Berger and Bouwman, 2009a).

Since this thesis investigates the effect of holding more liquidity on the lending behavior of European banks, an overview of the existing relevant literature regarding liquidity, liquidity creation and the lending behavior of banks is given in this chapter. First the general relevant literature regarding liquidity creation is reviewed. Since liquidity creation changes over time, the relevant literature with respect to how liquidity creation changes over time is considered next. Thirdly, the lending behavior of banks and amount of liquidity a bank holds during crises periods is discussed. Finally, the literature on the possible impact of the recently introduced liquidity regulation is summarized. This chapter ends with a summary and the contribution of this research to the existing literature.

2.1 Liquidity creation

A representative bank can hold two types of assets: safe, liquid assets and risky, illiquid assets. The liability side of a bank can consist of short-term debt, long-term debt and equity (Eisenbach et al., 2014).

Such a representative bank performs a valuable maturity transformation to the economy. It provides illiquid loans to investors on the asset side of their balance sheet, while it provides liquidity on demand to short-term creditors on the liability side of the balance sheet (i.e., liquidity creation) (Diamond and Rajan, 2001).

The following two subsections will give an overview of the theories and empirical findings regarding liquidity creation.

2.1.1 Theories regarding liquidity creation

Due to the structure of a bank and the maturity transformation it performs, a bank is subject to different kind of risks. One of those risks is liquidity risk. In the existing literature, two types of liquidity risk are distinguished: market liquidity risk and funding liquidity risk. If market liquidity is high, a bank can raise money easily by selling assets. When it is low, assets have to be sold at depressing prices. Funding liquidity is measured by the ease a bank can attract funding and this can be influenced by three sources: changes of margins, impossibility to roll-over short term debt or withdrawal of funding (Bonner et al., 2015). When short-term creditors do not rollover their debt, it might be the case that banks are forced to

(7)

7 sell their assets at a lower prices to provide liquidity to the depositors that run. Therefore, funding liquidity is closely related to market liquidity (Bonner et al., 2015).

Brunnermeier and Pedersen (2009) developed a theoretical model that links funding liquidity to market liquidity by focusing on changing margins. They state that traders provide market liquidity, but this depends on their availability of funding. When traders buy assets, they often borrow the money needed by financiers and use the asset as collateral. However, traders cannot borrow against the whole price of the asset and part of the asset has to be financed with capital. The difference between the price of the asset and collateral value is called the margin. When markets are liquid, financiers provide traders with favorable margin requirements (funding liquidity increases), which makes markets liquid. However, when markets are illiquid, financiers demand higher margins (funding liquidity decreases) which restricts traders from providing market liquidity. This further increases margins demanded by financiers, which decreases funding liquidity and market liquidity even more. This process is called the margin spiral. Brunnermeier and Pedersen (2009) state that when traders have enough capital to have no funding risk, market liquidity will be the highest and will be insensitive to changes in margins. Besides from the margin spiral, Brunnermeier and Pedersen (2009) identified a loss spiral: when market liquidity decreases, speculators might lose on existing positions, which induces speculators to sell more assets and that will cause a further price drop. The margin spiral and loss spiral reinforce each other and this results in a greater total effect compared to the sum of each liquidity spiral separately.

On the one hand the illiquidity of assets provides the rationale for the existence of banks, but on the other hand it explains why banks are vulnerable for runs of depositors (Diamond and Dybvig, 1983). Diamond and Rajan (2001) state that in a perfect world this financial fragility is a desirable characteristic of a bank to perform liquidity creation, because in a perfect world with no uncertainty about asset values, banks have specific skills to obtain knowledge about borrowers and to collect the loans they have outstanding. Since it is undesirable for banks to have a run of their short-term creditors (because then they will have no earnings), banks will commit to collecting loans with their specific skills.

However, when there is uncertainty about asset values, bankers have to trade off liquidity creation against the cost of bank runs in deciding their capital structure and how much liquid assets they will hold (Diamond and Rajan, 2001). When asset values decline during a crisis, a bank that holds more capital is better able to absorb loan losses during a crisis (Diamond and Rajan, 2002). Moreover, in this situation a safer, liquid asset portfolio makes the bank less vulnerable to bank runs, because then short-term creditors are more certain they will get their deposits/investments back (Eisenbach et al., 2014).

(8)

8

2.1.2 Empirical findings regarding liquidity creation

By introducing a comprehensive measure of bank liquidity creation, Berger and Bouwman (2009a) explore liquidity creation by banks. They use approximately all banks in the U.S. for the time period 1993-2003. The liquidity creation measure is constructed in three steps. Firstly, they classified all balance sheet items of a banks as liquid, semi-liquid or illiquid. Secondly, they assigned weights to the activities from step 1. Thirdly, a liquidity creation measure is obtained by combining step 1 and step 2 in different ways. The preferred measure is measured by loans classified based on category and off balance sheet items are included too, the “cat fat” measure.

They find that liquidity creation had nearly doubled in the specific time period based on the “cat fat” measure. In 2003, the banking sector created on average $4.56 of liquidity per $1 of capital. Moreover, Berger and Bouwman (2009a) state that liquidity creation differs for some important bank characteristics. They find that large banks (assets > $3 bln) create more liquidity compared to medium banks (assets $1 bln - $3 bln) and small banks (assets < $1 bln). Moreover, liquidity creation and value (measured by market-to-book) are positively correlated. Lastly, the effect of capital on bank liquidity creation is significantly positive for large banks, but significantly negative for small banks (Berger and Bouwman, 2009a).

Jiménez et al. (2012) analyze the effects of monetary policy and economic activity on the accepting of loans by a bank. They use bank data from Spain for the time period 2002-2008. When the amount of liquidity is taken into account separately, Jiménez et al. (2012) find that banks that have more liquidity have a lower probability of granting loans to new borrowers. Moreover, they find that when economic activity decreases (i.e. lower GDP growth) and/or under tighter monetary policy (i.e. short-term interest rates increases), credit availability decreases. However, when banks have higher liquidity (liquid assets/total assets) or capital compared to their competitors, the negative effect of higher interest rates or lower GDP growth is statistically lower. Therefore, under tighter monetary conditions or decreasing economic growth, lower bank capital or bank liquidity negatively affects the availability of credit and might cause a credit crunch.

2.2 How liquidity evolved over time towards the crisis

Until recently, bank regulators did not focus explicitly on liquidity risk. However, the recent crisis showed the influence of liquidity risk on the economy, in which liquidity risk increased significantly towards the recent crisis (Bonner et al., 2015). The following two subsections describe the theories and empirical findings regarding how liquidity evolved over time towards the crisis.

(9)

9

2.2.1 Theories regarding how liquidity can evolve over time

Banks’ short-term debt can partly consist of wholesale funding. Banks increasingly used wholesale funds prior to the crisis. Wholesale funds are on a short-term rollover basis (Huang and Ratnovski, 2011). Due to wholesale funding, banks have the possibility to exploit profitable investment opportunities without being constrained to the deposit supply. Moreover, wholesale financiers shall monitor banks, because they are not insured as depositors are. Wholesale funding also lowers banks’ interest expenses. However, when banks rely too much on wholesale funding, they could experience difficulties when this funding dries up in economic downturns. Huang and Ratnovski (2011) provide a theoretical model based on insured depositors and uninsured wholesale funding. It starts with the fact that in reality wholesale financiers have a seniority claim on liquidation values, since depositors are passive investors. In the years prior to the crisis, there were noisy, costless, public signals to the wholesale financiers about the values of banks’ assets, like house prices or mortgage backed securities. They find that wholesale financiers increasingly choose to rely on the public noisy signal instead of private costly monitoring when their seniority increases. When they increasingly rely on the public noisy signal, a signal that could be wrong, inefficient liquidations will be higher (when wholesale funds are not rolled-over). This is what happened before the crisis, banks were increasingly relying on wholesale funding and eventually this kind of funding dried up at the start of the crisis. Therefore, a bank should consider the private benefits from lower interest expenses against the higher risk of inefficient liquidations (Huang and Ratnovski, 2011).

Regarding the asset side of the balance sheet of banks, Acharya et al. (2010) provide a theoretical model of banks choice of liquidity that is driven by strategic considerations. Since in economic upturns the pledgeability of risky, illiquid assets is high, banks tend to hold less liquid assets. However, when pledgeability of risky assets is low, like in economic downturns, banks may even hold more liquidity than that is socially optimal to be able to buy assets that are sold at fire sale prices. Concluding, the choice of bank liquidity is countercyclical: too low in economic upturns but too high during crisis periods.

2.2.2. Empirical findings regarding how liquidity actually evolved towards the crisis

Demirüc-Kunt and Huizinga (2011) consider the effect of specific short-term funding strategies of banks and their activities on risk and return. Risk is measured by the Z-score. Their sample consists of 1334 banks in 101 countries for the time period 1995-2007. Non-deposit funding has risen for investment banks, nonbank credit institutions and other financial institutions in the sample period, where commercial banks decreased the non-deposit funding share in the period before the crisis. Moreover, developed countries relied more on non-deposit funding compared to developing countries. They find empirical evidence that

(10)

10 banking strategies relying mainly on non-deposit funding are very risky. This is in line with the theoretical model of Huang and Ratnovski (2011). However, at low levels of wholesale funding there might be risk diversification benefits, but this result is not conclusive (Demirüc-Kunt and Huizinga, 2011).

Acharya et al. (2010) examine the amount of liquidity a bank holds and find that cash holdings of US banks for the time period 1980-2007 decreased from 10% of total assets to approximately 3% of total assets. Since the start of the crisis (September 2008), cash holdings increased again. These findings confirm their conclusions from the theoretical model, described in section 2.2.1.. Moreover, Aspachs et al. (2005) confirm these findings with empirical evidence. They use banking data of UK banks for the time period 1985-2003. Liquidity is measured by constructing a liquidity ratio as follows: 𝐿𝑖𝑞𝑢𝑖𝑑 𝑎𝑠𝑠𝑒𝑡𝑠

𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑝𝑜𝑠𝑖𝑡𝑠. In line with

Acharya et al. (2010), they find that in periods of stronger growth, measured by GDP growth, UK banks hold smaller amounts of liquidity. A 1% increase in GDP lowers liquidity by about 2% (significant at the 1% level). Based on their findings, Aspachs et al. (2005) conclude that UK banks tend to pursue a countercyclical liquidity policy.

Bonner and Hilbers (2015) also find that liquidity decreased towards the crisis, but they link this decrease to regulating capital. They suggest that regulating capital (started in 1988) is negatively correlated with the amount of liquidity banks hold and therefore capital and liquidity should be considered jointly in regulation.

Lastly, Berger and Bouwman (2009b) examine bank liquidity creation around five financial crises. They use the measure for liquidity creation that was also used in Berger and Bouwman (2009a). They show that for the time period 1984-2008, actual liquidity creation increased. They also investigate abnormal liquidity creation (relative to a time trend) and find that abnormal liquidity creation was high for the period 2005 till 2007. Berger and Bouwman (2009b) did find this abnormal liquidity creation in the periods before all banking crises investigated and suggest that while financial fragility might be needed to create liquidity (Diamond and Rajan, 2001), too much liquidity creation might lead to financial fragility.

2.3 Liquidity and lending during crises periods

When a crisis hits, banks have to manage the liquidity shocks that occur. The next two subsections describe the theories and empirical evidence regarding how a bank reacts on liquidity shocks during crises with respect to their liquidity holdings and lending behavior.

(11)

11

2.3.1 Theories regarding liquidity and lending during crises periods

Diamond and Rajan (2005) provide a theoretical model of how bank failures can exacerbate aggregate liquidity shortages. When some loans are paying with a delay but eventually will pay, a bank could finance this with new depositors. However, banks can experience difficulties when too many borrowers are paying their loans with a delay or even default on their loans. If depositors’ losses are unavoidable, they shall demand their money. This demand can cause a bank failure and since banks are using a common market for liquidity, one bank failure can lead to a contagion of failures. A bank failure and the subsequently contagion could be dampened if banks holds more liquid assets.

Moreover, when banks need to sell their assets at fire sale prices to prevent such a bank failure as described by Diamond et al. (2005), Cifuentes et al. (2005) state that liquidity buffers might prevent systemic stress (contagion) that follows from the need of selling assets at fire sale prices. They explore liquidity risk in a model where banks are interconnected. They argue that liquidity buffers internalize the costs of selling assets in an economic downturn. When banks increase their liquidity buffers, confidence is restored and fire sales can be avoided, because creditors believe that banks have enough liquidity to pay off their creditors. Therefore, liquidity buffers could prevent systemic crises.

2.3.2 Empirical findings regarding liquidity and lending during crises periods

Using the liquidity creation measure of Berger and Bouwman (2009a), Berger and Bouwman (2009b) find that after a build-up of abnormal liquidity creation before the recent crisis, liquidity creation decreased substantially when the crisis hit. They also investigate the different components of liquidity creation. Semi-liquid assets (especially mortgages) and illiquid assets (especially and business lending) decreased slightly during the recent crisis, where loan commitments and other off-balance sheet guarantees decreased significantly. Furthermore, since the start of the crisis, liquid assets have shown an upward trend. This is in line with banks building up liquidity in investment securities (Berger and Bouwman, 2009b).

Cornett et al. (2011) discuss how U.S. banks managed the liquidity shocks during the recent financial crisis (2007-2009) and how this depends on their liquidity risk. Liquidity risk is defined across four dimensions: compositions of asset portfolio (liquid versus illiquid), core deposits as a fraction of financial structure, capital as a fraction of financial structure and funding liquidity exposure due to loan commitments. Cornet et al. (2011) report that banks that were more exposed to liquidity risk increased their amount of liquid assets and this reduced their capacity for new loans. Moreover, banks that relied more on stable funding sources (i.e. core deposits and capital) had the possibility to continue lending compared to other banks.

(12)

12

2.4 Liquidity regulation

As a response to the recent crisis, the Basel Committee introduced liquidity regulation for banks in 2011. The following two subsections describe the liquidity regulation included in Basel III and the (possible) effects of liquidity regulation on the behavior of banks.

2.4.1 Theories regarding liquidity regulation

Since 1978, regulators have realized that capital requirements are needed to regulate bank capital. In 1993, the Basel Committee (committee of banking supervisory authorities) implemented capital requirements for 12 participating countries, which became the Basel I accord. However, there were numerous shortcomings, like too rude risk classes and too little attention for interest-rate risk (Boot et al., 2016). As a response to the numerous shortcomings, the Basel II Capital Accord was introduced in 2004.

The recent crisis showed that due to poor liquidity management and insufficient liquidity buffers institutions experienced severe problems. Till then, regulators mainly focused on capital regulation but did not have much attention for liquidity regulation. Regulators thought that liquidity issues were a matter for national authorities and it did not have to be harmonized, like capital regulation. Moreover, the financial markets did not experience difficulties due to liquidity before this recent crisis. However, the recent crisis gave supervisory momentum and made liquidity harmonization more likely, it highlighted substantial shortcomings in the existing regulation of Basel II (Bonner and Hilbers, 2015). For example, countercyclical buffers were not included. Moreover, it became clear that current regulation did not deal with the maturity mismatch on the balance sheets of banks. Basel III, introduced in 2011, tries to tackle the shortcomings of Basel II, by introducing two liquidity measures: Liquidity Coverage Ratio (LCR) and the Net Stable Funding Ratio (NSFR) (Boot et al., 2016). The LCR requires banks to have enough liquid assets for the possible net cash outflows in the next 30 days under a stress scenario:

𝐿𝐶𝑅 = _{𝑁𝑒𝑡 𝐶𝑎𝑠ℎ 𝑜𝑢𝑡𝑓𝑙𝑜𝑤𝑠 𝑖𝑛 𝑡ℎ𝑒 𝑛𝑒𝑥𝑡 30 𝑑𝑎𝑦𝑠}𝑆𝑡𝑜𝑐𝑘 𝑜𝑓 𝐻𝑖𝑔ℎ 𝑄𝑢𝑎𝑙𝑖𝑡𝑦 𝐿𝑖𝑞𝑢𝑖𝑑 𝐴𝑠𝑠𝑒𝑡𝑠 .

The NSFR requires a minimum amount of stable funding relative to the liquidity of the assets of a bank (Basel Committee, 2011):

𝑁𝑆𝐹𝑅 = 𝐴𝑣𝑎𝑖𝑙𝑎𝑏𝑙𝑒 𝑆𝑡𝑎𝑏𝑙𝑒 𝐹𝑢𝑛𝑑𝑖𝑛𝑔

𝑅𝑒𝑞𝑢𝑖𝑟𝑒𝑑 𝑆𝑡𝑎𝑏𝑙𝑒 𝐹𝑢𝑛𝑑𝑖𝑛𝑔.

Even though the LCR has already been realized since 1 January 2015, the NSFR has not been approved yet. It is expected that the NSFR will be realized from 1 January 2018 on (Basel Committee, 2011).

Allen et al. (2012) discuss the economic impact of the Basel III reforms on banking regulation. They state that the material risks will be that the supply of credit to the economy will be disrupted by the implementation of the new regulations on capital and liquidity. They claim that the problem is not the higher

(13)

13 capital and liquidity requirements per se, because these requirements have little effect on the fundamental cost of banking activities. However, when there are difficulties within the adaptation and implementation of the new rules in the financial industry, then the reform may reduce economic activity and lending. Allen et al. (2012) even state that if the efficiency improvements from higher capital and liquidity requirements are high enough, these requirements will eventually reduce the cost of intermediation and increase the level of output (instead of a decrease).

In 2010, The Macroeconomic Assessment Group (MAG) of the Bank of International Settlements modeled the macroeconomic impact of the introduction of the liquidity requirements on the lending spreads. Their model predicts that a 25% increase in liquid asset holdings will increase the lending spreads by approximately 15 basis points and decrease GDP by a maximum of 13%. On average, they suggest that the liquidity requirements might have an impact on bank lending in the implementation period of the liquidity requirements. However, in line with Allen et al. (2012), as long as the implementation period is sufficiently long and other conditions in the financial economy are supportive, the effect of the liquidity requirements should not be large (MAG, 2010).

In addition to MAG (2010), Angelini et al. (2011) conduct counterfactual experiments to assess the long-term impact of the new liquidity requirements on economic fluctuations. The focus of their experiments is on the costs of regulation, not the benefits. They use the NSFR as relevant liquidity requirement. Meeting the NSFR is modeled as an increase of 50% in the ratio between liquid assets and total assets and find that meeting the NSFR causes a median 0.08 percent decline in the level of steady state (long-run) output. Angelini et al. (2011) also report that capital and liquidity requirements tend to reduce output volatility. However, it should be noted that these results are surrounded by considerably uncertainty.

King (2010) also discusses the introduction of liquidity requirements to investigate what the effect will be on bank lending spreads. In contrast with MAG (2010), which focuses on the transition period, King (2010) focuses on a steady-state period. Moreover, he mentions that this study should be seen as a starting point to understand the behavioral response of banks to the new liquidity regulations. King (2012) states that when banks meet the NSFR, this should cause a fall in banks’ risk weighted assets, which is included in the capital requirements. When this synergy is taken into account, King (2010) states that meeting the NSFR would increase the lending spreads by 12 basis points, instead of 15 basis points as predicted by MAG (2010).

(14)

14

2.4.2 Empirical findings regarding liquidity regulation

Without the existence of regulatory liquidity buffers, the amount of liquidity a bank holds is influenced by a number of factors, for example profits or the capital ratio of a bank. Bonner et al. (2015) investigate whether the liquidity requirements of Basel III neutralize the incentives to hold liquid assets. They state that without regulation, liquidity buffers are determined by bank-specific and country-specific factors. However, when liquidity regulation is in place, only disclosure requirements and the concentration of the banking sector remain important.

Moreover, they find suggestive evidence that during stress, banks subject to liquidity regulation reduced their lending volumes and have higher liquidity buffers. During stress it is more difficult for banks to obtain funding and since they have to hold higher liquidity buffers, this might be the reason for the negative relationship between liquidity regulation and lending volumes during stress. However, Bonner et al. (2015) state that in normal times, banks subject to liquidity regulation have higher lending volumes. Since higher liquidity buffers function as insurance against liquidity shocks, banks may feel more comfortable by issuing credit (Bonner et al., 2015).

Since the liquidity requirements are relatively new and only the LCR is already implemented since January 2015, not much empirical research has been done to the effect of the liquidity requirements of Basel III on the liquidity buffers of banks. Since 2003, Dutch banks are subject to liquidity regulations that resemble Basel III. Therefore, De Haan and Van Den End (2013) investigate Dutch banks’ actual liquidity management under the liquidity requirements that has been operational since 2003 for the time period 2003-2010. Overall, they find that banks hold an amount of liquid assets as a buffer against liquid liabilities. When banks know they will have net cash inflows next month, they hold fewer liquid assets, but they do not completely decrease their liquid assets by the amount of net cash inflows. Since they do not reduce their buffer by the same amount as the cash net inflows, they conclude that Dutch banks pursue prudent risk management (De Haan and Van Den End, 2013).

(15)

15

2.4 Summary of the literature

The previous subsections gave an overview of the existing literature regarding liquidity and lending behavior of banks. It started with explaining liquidity creation and that it is influenced by a number of bank specific characteristics, for example size or the profitability of a bank. Moreover, macroeconomic conditions as well have an effect on liquidity creation (Berger and Bouwman, 2009a).

Next it showed that liquidity creation increased substantially before the recent crisis and when the crisis hit it became clear that banks held insufficient liquidity buffers to handle the withdraw of short-term debt and the collapse of asset prices. In response to this event, the Basel Committee introduced two liquidity requirements: the Liquidity Coverage Ratio and the Net Stable Funding Ratio (Bonner and Hilbers, 2015). However, critics are afraid that the obligatory liquidity requirements affect the lending behavior of banks (Allen et al., 2012).

Before the recent crisis there was not much attention for liquidity regulation and how much liquidity a bank holds (Bonner and Hilbers, 2015). Therefore there has not been many empirical research done to the effect of holding more liquidity on the lending behavior of banks. Jiménez et al. (2012) found that Spanish banks with more liquidity have a lower probability of accepting loans. However, when the economy worsens, higher liquidity positively affects the granting of loans.

Some studies related to this research (mostly studies on liquidity regulation) found that introducing liquidity requirements and forcing banks to hold more liquidity possibly reduce lending and induces banks to increase their lending spreads. However, Bonner et al. (2015) found that banks subject to liquidity regulation have higher lending volumes in normal times. Moreover, evidence regarding the crisis is mixed. Cornet et al. (2011) found that banks that were less exposed to liquidity risk had the possibility to continue lending compared to other banks. However, Bonner et al. (2015) found a contradicting result that during a crisis banks subject to liquidity regulation lower their lending volumes more compared to banks without liquidity regulation.

There is a lack of empirical literature regarding to the effect of holding more liquidity on lending behavior specifically. This research will contribute to the existing literature by considering the (possible) relationship between the amount of liquidity and lending behavior for European banks in particular and for the time period 2005-2014. Moreover, the crisis period is considered separately.

(16)

16

3. Hypotheses, Methodology and Data

3.1 Hypotheses

This thesis investigates the effect of holding more liquidity on the lending behavior of European banks. As described in the literature review, evidence regarding liquidity regulation in particular is mixed. Based on the total existing literature as described in the previous section, the first hypothesis is the following:

H1: Holding more liquidity will negatively affect the lending behavior of banks.

This hypothesis will be tested on European Banks in the time period 2005-2014.

Based on the literature review it is expected that during a crisis period, banks that have higher liquidity buffers are less exposed to liquidity risk compared to banks having lower liquidity buffers. Therefore, considering the crisis period specifically, the second hypothesis is the following:

H2: The effect of holding more liquidity will be less negative during crisis periods compared to the total period considered.

3.2 Methodology

The Basel Committee provides guidelines for the central banks. In the end, the central banks are the entities that determine the rules for banks. The banking landscape in Europe will be investigated, because all banks in Europe are subject to the same laws of the European Central Bank (ECB) and can therefore be treated as one group.

Based on Jiménez et al. (2012), the amount of liquidity of a bank might influence the lending behavior of banks in a later period of time. This possible effect does not happen immediately. To capture the time element, panel data will be used (Stock and Watson, 2012).

The relevant time period for this research is 2005-2014. In this time frame periods of “normal” growth are captured, including the crisis period in Europe of 2008-2012. To find out if the effect of the amount of liquidity on outstanding loans is different in a crisis period, two regressions are conducted: First, a regression is done using the whole time period and second, a regression is done for a specific time period, the crisis (2008-2012).

Since it is of interest to measure the effect of the amount of liquidity on total loans outstanding of banks, the variable total loans is the dependent variable. This variable is transformed using logarithms, because the absolute numbers are too big (Stock and Watson, 2012).

(17)

17 In line with the research question is the variable of interest a measure for the amount of liquidity a bank holds. Unfortunately, banks are obliged to publish the LCR in the annual reports from 1 January 2015 on. Therefore, using the LCR as stated by BIS (Basel Committee on Banking Supervision, 2013) will give too few data. Taken into account the LCR and that this research discusses if liquid assets will replace the specific illiquid assets (loans), a proxy for the liquidity ratio is used:

𝐶𝑎𝑠ℎ 𝑎𝑛𝑑 𝑠𝑒𝑐𝑢𝑟𝑖𝑡𝑖𝑒𝑠

𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑝𝑜𝑠𝑖𝑡𝑠 ∗ 100.

Aspachs et al. (2005) and Freedman and Click (2006) also used this proxy to measure the liquidity holdings of a bank. The numerator, cash and securities, can be seen as high quality liquid assets. Deposits are used in calculating the expected net cash outflows in the denominator of the LCR (Basel Committee on Banking Supervision, 2013). Using this liquidity ratio, liquidity and liquidity mismatch of a bank are measured (Aspachs et al., 2005).

Regarding both hypotheses (H1 and H2), it is expected that the coefficient of this variable is negative considering the whole period and the crisis period specifically. When negative, an increase in the liquidity ratio (increase of cash and securities or a decrease in total deposits) with 1, will decrease the total loans outstanding by 100 * 𝛽1 % (𝛽1 in equation (1) below).

Based on the literature review, there are also five control variables collected which may influence the outstanding total loans of a bank. Lending behavior varies across banks, therefore some bank characteristics need to be included, such as the size of a bank or its profitability. Moreover, there are also two country-specific characteristics that may influence the lending behavior of banks (Berger and Bouwman, 2009a) (Jiménez et al., 2012) (Berger and Bouwman, 2013). Following the literature, the control variables included are:

1. Total assets of a bank 2. Return on equity of a bank 3. Capital ratio of a bank 4. Herfindahl index per country 5. GDP growth per country

With respect to the first control variable, Berger and Bouwman (2009a) indicate that size differences between banks, measured by total assets, are substantial and that components of liquidity creation vary greatly by bank size. Therefore total assets are included to control for the size of a bank. Following Berger and Bouwman (2009a) and the ECB (2014), a dummy variable will be created, separating

(18)

18 banks into: small banks, medium banks and large banks. Since the sample consists of European listed banks, the assets groups of the ECB (2014) are used:

1. Small banks: < 100 EUR bln

2. Medium banks: 100 EUR bln – 500 EUR bln 3. Large banks: > 500 EUR bln

Next to the size of a bank could the profitability of a bank affect bank lending (Jiménez et al., 2012). Therefore a profitability measure is included, the return on equity of a bank (control variable 2). It is expected that the relationship between the outstanding loans and profitability of a bank is positive (Jiménez et al., 2012).

The third control variable included is the capital ratio. Berger and Bouwman (2009a) state that banks with higher lagged capital ratios create more liquidity, i.e. have more illiquid assets. Since the capital ratio might influence the lending behavior of banks, this ratio is included as a control variable. It is expected that the coefficient is positive (Berger and Bouwman, 2009a).

Besides the three bank variables that may affect lending (size, profitability and capital), there are two country-specific characteristics too that may affect lending: the competition in a country and the macroeconomic conditions in a country. Jiménez et al. (2012) find that the degree of competition in a country influence the lending behavior of banks. To control for competitiveness, the Herfindahl index per country for credit institutions is included as a lag. An increase in the index indicates an increase in market power / decrease in competition. Following Jiménez et al. (2012), it is expected that the coefficient of this control variable is positive.

Next to the degree of competition, Jiménez et al. (2012) state that macroeconomic conditions have an effect on the total loans outstanding of a bank. To control for the business cycle, GDP growth per country is included. Following the literature, it is expected that the coefficient is positive.

Using the variables mentioned, the main panel data (fixed effects) regression of this research is:

(1) ln (𝑇𝑜𝑡𝑎𝑙 𝑙𝑜𝑎𝑛𝑠)𝑖𝑡 = 𝛽0+ 𝛽1∗

𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑝𝑜𝑠𝑖𝑡𝑠 _𝑖𝑡−1+ 𝛽3∗ 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ𝑗𝑡+ 𝛽4∗ 𝑆𝑚𝑎𝑙𝑙 𝑏𝑎𝑛𝑘𝑠𝑖𝑡−1+ 𝛽5∗

𝑀𝑒𝑑𝑖𝑢𝑚 𝑏𝑎𝑛𝑘𝑠𝑖𝑡−1+ 𝛽6∗ 𝐶𝑎𝑝𝑖𝑡𝑎𝑙 𝑟𝑎𝑡𝑖𝑜𝑖𝑡−1+ 𝛽7𝐻𝐻𝐼𝑗𝑡−1 + 𝛽9𝑅𝑂𝐸𝑖𝑡−1+ 𝑡𝑖𝑚𝑒 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 +

𝑓𝑖𝑟𝑚 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝑒𝑖𝑡

* GDP growth and HHI varies per country, this is indicated with subscript j. The liquidity ratio, dummies for the bank size, capital ratio and ROE varies per bank, indicated with subscript i.

(19)

19 The regressions are estimated with robust standard errors, clustered by banks. The clustered robust standard errors control for possible correlation among the observations of one bank in multiple years and for heteroskedacity (Stock and Watson, 2012).

As can be seen in the main regression (1), firm fixed effects and time fixed effects are included. Firm fixed effects capture all the bank specific characteristics that do not change over time. Time fixed effects capture all variables that change over time, but are the same for every bank. By conducting a panel data regression including firm fixed effects and time fixed effects, a possible endogeneity problem is reduced, because there is more control over possible omitted variables (Stock and Watson, 2012).

Regression (1) will also be conducted for each group of banks separately (small-, medium- and large banks) to see if there are differences between the different groups:

𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑝𝑜𝑠𝑖𝑡𝑠 _𝑖𝑡−1+ 𝛽3∗ 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ𝑗𝑡+ 𝛽4∗ 𝐶𝑎𝑝𝑖𝑡𝑎𝑙 𝑟𝑎𝑡𝑖𝑜𝑖𝑡−1+ 𝛽5𝐻𝐻𝐼𝑗𝑡−1 +

𝛽6𝑅𝑂𝐸𝑖𝑡−1+ 𝑡𝑖𝑚𝑒 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝑓𝑖𝑟𝑚 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝑒𝑖𝑡

Comparing regression (2) to (1), it can be seen that the dummy for size is excluded since this regression is conducted for each groups of banks separately based on size.

In the end, a pooled OLS regression including lags will be performed to check robustness of the main regression (1) (Stock and Watson, 2012):

(3) ln(𝑇𝑜𝑡𝑎𝑙 𝑙𝑜𝑎𝑛𝑠)𝑖𝑡= 𝛽0+ 𝛽1∗𝐶𝑎𝑠ℎ 𝑎𝑛𝑑 𝑠𝑒𝑐𝑢𝑟𝑖𝑡𝑖𝑒𝑠_{𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑝𝑜𝑠𝑖𝑡𝑠}

𝑖𝑡−1+ 𝛽3∗ 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ𝑗+ 𝛽4∗ 𝑆𝑚𝑎𝑙𝑙 𝑏𝑎𝑛𝑘𝑠𝑖𝑡−1+ 𝛽5∗

𝑀𝑒𝑑𝑖𝑢𝑚 𝑏𝑎𝑛𝑘𝑠𝑖𝑡−1+ 𝛽6∗ 𝐶𝑎𝑝𝑖𝑡𝑎𝑙 𝑟𝑎𝑡𝑖𝑜𝑖𝑡−1+ 𝛽7𝐻𝐻𝐼𝑗𝑡−1+ 𝛽9𝑅𝑂𝐸𝑖𝑡−1+ 𝑒𝑖

In a pooled OLS regression as described in (3), the time dimension of panel data is ignored. Moreover, this regression is conducted with clustered standard errors.

(20)

20

3.3 Data and descriptive statistics

3.3.1 Data description

To gather the data required for answering the research question, two databases are used: Datastream and Quandl.

All bank specific variables can be found in the database Datastream. In Datastream, data are available for all listed banks in Europe (list called “Banks Europe”). This list consists of 162 public listed banks. The following variables are extracted and available for 149 banks:

1. Total loans: total loans represents the total loans made by a bank to their customers (for example real estate mortgage loans or commercial and industrial loans).

2. A liquidity ratio: the liquidity ratio is a proxy for liquidity of a bank gathered from Datastream. The liquidity measure is constructed as follows:

𝐶𝑎𝑠ℎ 𝑎𝑛𝑑 𝑠𝑒𝑐𝑢𝑟𝑖𝑡𝑖𝑒𝑠 𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑝𝑜𝑠𝑖𝑡𝑠 ∗ 100

3. Total assets: this item represents the total assets of a bank. It includes for example the total investments, plant and equipment and account receivables.

4. Return on equity: return on equity (ROE) is a profitability measure.

5. Common shareholders’ equity: this item represents the common shareholders’ investment in a bank. It includes for example the common stock value and retained earnings.

6. Herfindahl index: the HHI is measured by the ECB and it represents a Herfindahl index per country for credit institutions.

The variables described above are used to obtain the variables as described in regression (1). First, a capital ratio for each bank is constructed by taking common shareholders’ equity (5) divided by total assets (3).

Since almost all variables are included as a lag, lag variables are generated from the following characteristics: liquidity ratio (1), total assets (3), ROE (4), HHI (6) and capital ratio.

Moreover, three asset groups (small banks, medium banks and large banks) are generated as described in Methodology (3.1) using the lag of variable (3), total assets.

To obtain a balanced regression, the sample is restricted to banks that have non-missing values in total loans (dependent variable). This resulted in a sample of 112 banks. Moreover, total loans is transformed using logarithms.

(21)

21 In the end, the database Quandl is used to obtain a measure for the macroeconomic conditions in a country. Quandl shows GDP growth per country, obtained from the World Bank. It is also called a World Bank World Development Indicator.

3.3.2 Descriptive statistics

Summary statistics for the initial variables used are depicted in Table 1.

Table 1: Summary statistics of initial variables

(1) (2) (3) (4) (5) (6)

VARIABLES N #banks mean sd min max

GDP growth 1,120 112 1.508 3.153 -9.132 10.83

Total loans 1,120 112 1.042E+08 1.851E+08 157,365 1.548E+09

HHI 770 77 0.0708 0.0395 0.0170 0.220

Total assets 1,120 112 2.177E+08 4.411E+08 194,608 2.974E+09

Liquidity ratio 1,094 110 62.23 49.84 13.09 201.2

Capital ratio 1,120 112 0.0744 0.0332 0.0289 0.160

ROE 1,108 112 10.04 9.392 -15.37 25

* The values of total assets and total loans are displayed in thousands of Euros

As can be seen, the HHI of some banks is missing. This is because Datastream did not have the HHI of the following countries: Switzerland, Croatia, Norway, Russia and Turkey. The summary statistics on the Herfindahl index show that on average European Banks are located in a competitive landscape, since on a scale from 0.0 to 1.0, the average HHI is 0.0395.

Moreover, as shown in Table 1, the liquidity ratio differs significantly across European banks. The liquidity ratio in the sample ranges from 13.09 to 201.2 and on average they hold 62.23 times as much cash and securities compared to their deposits.

The summary statistics show that European banks have low levels of capital compared to total assets. This is in line with the existing literature, since banks’ funding consists on average for more than 90% of short-term debt and long-term debt (Perotti, 2010). On average, banks in the sample finance 3.32% of their total assets with capital.

Lastly, GDP growth ranges from -9.13% to 10.83% and the ROE ranges from -15.37% to 25%. The negative values for GDP growth and ROE are mainly reported by banks during the crisis years, 2008-2012.

(22)

22 In Table 2 and 3 the summary statistics for the main regression are depicted, for the total period and the crisis period respectively.

Table 2: Summary statistics of the main regression variables for the period: 2005-2014

(1) (2) (3) (4) (5) (6)

GDP growth 1,120 112 1.508 3.153 -9.132 10.83

Ln(total loans) 1,120 112 17.15 1.705 11.97 21.16

Lag HHI 770 77 0.0696 0.0382 0.0170 0.217

Lag small bank 1,120 112 0.704 0.457 0 1

Lag medium bank 1,120 112 0.160 0.367 0 1

Lag large bank 1,120 112 0.131 0.338 0 1

Lag Liquidity ratio 1,091 110 61.46 49.91 12.53 201.2

Lag Capital ratio 1,114 112 0.0735 0.0337 0.0279 0.162

Lag ROE 1,103 112 10.90 9.027 -13.98 25.27

* The values of ln(total loans) are displayed in thousands of Euros

Table 3: Summary statistic of main regression variables for the crisis period: 2008-2012

(1) (2) (3) (4) (5) (6)

GDP growth 560 112 0.365 3.433 -9.132 9.157

Ln(total loans) 560 112 17.25 1.683 12.56 20.96

Lag HHI 385 77 0.0693 0.0362 0.0180 0.217

Lag small bank 560 112 0.698 0.459 0 1

Lag medium bank 560 112 0.159 0.366 0 1

Lag large bank 560 112 0.143 0.350 0 1

Lag Liq. ratio 550 110 61.56 50.86 12.53 201.2

Lag Cap. Ratio 560 112 0.0729 0.0330 0.0279 0.162

Lag ROE 560 112 9.958 9.128 -13.98 25.27

As already mentioned in the explanation of Table 1, most negative values of GDP growth and ROE are reported by banks during the crisis years. When comparing Table 2 and 3, it can be seen that indeed GDP growth was on average lower during the crisis period compared to the total period (0.365 versus 1.508). This is in line with the existing literature which states that crises are characterized with low or negative GDP growth (Borio, 2014). Furthermore, as depicted in Table 2 and 3, the average profitability of banks, measured by ROE, was smaller in the crisis period compared to the total period. The average ROE was 9.96 and 10.90 respectively.

(23)

23 Moreover, on average European banks had a higher amount of outstanding loans during the crisis compared to the whole period. This contradicts with the literature that states that during the crisis banks contradict lending (Berger and Bouwman, 2009b).

Lastly, banks held on average less capital during the crisis period, but the difference compared to the total period is small: 7.29% compared to 7.35%.

As mentioned in the methodology, regression (1) will also be conducted for each group of banks separately (small, medium, and large banks). These summary statistics can be found in Tables 4, 5 and 6.

Table 4: Summary statistics of the data for all small banks (< 100 EUR bln assets)

(1) (2) (3) (4) (5) (6)

GDP growth 788 68 1.757 3.294 -9.132 10.83

Ln(total loans) 788 68 16.28 1.085 12.04 18.60

Lag HHI 482 39 0.0739 0.0402 0.0170 0.217

Lag Liquidity ratio 776 67 45.21 36.51 12.53 201.2

Lag Capital ratio 788 68 0.0835 0.0332 0.0279 0.162

Lag ROE 778 68 11.72 8.391 -13.98 25.27

Table 5: Summary statistics of the data for all medium banks (100-500 EUR bln assets)

(1) (2) (3) (4) (5) (6)

GDP growth 179 27 0.906 2.955 -9.132 8.535

Ln(total loans) 179 27 18.81 0.409 17.85 20.01

Lag HHI 156 23 0.0720 0.0398 0.0170 0.214

Lag ROE 178 27 10.16 10.33 -13.98 25.27

(24)

24

Table 6: Summary statistics of the data for all large banks (>500 EUR bln assets)

(1) (2) (3) (4) (5) (6)

GDP growth 147 17 0.794 2.311 -5.619 4.174

Ln(total loans) 147 17 19.90 0.524 18.68 21.16

Lag HHI 127 15 0.0494 0.0167 0.0170 0.0950

Lag ROE 147 17 7.472 9.735 -13.98 25.27

Tables 4, 5 and 6 illustrate that most banks in the total sample are small banks (68 banks).

Moreover, the value of the independent variable of interest, the liquidity ratio, is larger for large banks compared to medium and small banks. Large banks hold on average 138.2 times as much cash and securities compared to their deposits, where small banks hold on average 45.21 times as much cash and securities compared to their deposits. A possible explanation might be that the different groups (small, medium and large banks) are based on total asset size, which partly consists of cash & securities, the numerator of the liquidity ratio.

Regarding total loans outstanding of a banks, these Tables show that large banks have on average the largest amount of loans outstanding (EUR 4.39*10^11) compared to medium banks (EUR

1.48*10^11) and small banks (EUR 1.18*10^10). This makes sense, since the different groups (small, medium and large banks) are based on total asset size, which partly consists total loans.

Furthermore, Tables 4, 5 and 6 show that small banks hold on average the highest amount of capital compared to total assets in this sample. On average, small banks in the sample finance 8.35% of their total assets with capital, where the capital ratio of medium banks and small banks is 5.46% and 4.30% respectively.

(25)

25

4. Results

4.1 Empirical results

In Table 7, the results of the main regression are depicted:

(1) ln (𝑇𝑜𝑡𝑎𝑙 𝑙𝑜𝑎𝑛𝑠)𝑖𝑡 = 𝛽0+ 𝛽1∗𝐶𝑎𝑠ℎ 𝑎𝑛𝑑 𝑠𝑒𝑐𝑢𝑟𝑖𝑡𝑖𝑒𝑠_{𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑝𝑜𝑠𝑖𝑡𝑠}

𝑖𝑡−1+ 𝛽3∗ 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ𝑗𝑡+ 𝛽4∗ 𝑆𝑚𝑎𝑙𝑙 𝑏𝑎𝑛𝑘𝑠𝑖𝑡−1+ 𝛽5∗

𝑀𝑒𝑑𝑖𝑢𝑚 𝑏𝑎𝑛𝑘𝑠𝑖𝑡−1+ 𝛽6∗ 𝐶𝑎𝑝𝑖𝑡𝑎𝑙 𝑟𝑎𝑡𝑖𝑜𝑖𝑡−1+ 𝛽7𝐻𝐻𝐼𝑗𝑡−1 + 𝛽9𝑅𝑂𝐸𝑖𝑡−1+ 𝑡𝑖𝑚𝑒 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 +

𝑓𝑖𝑟𝑚 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝑒𝑖𝑡

Table 7: Results main regression (period 2005-2014), dependent variable: ln(total loans)

(1) (2) (3) (4) (5) (6)

VARIABLES base extra1 extra2 extra3 extra4 extra5

Lag Liq. ratio -0.00302*** -0.00301*** -0.00252*** -0.00232*** -0.00244*** -0.00257***

(-3.778) (-3.786) (-3.400) (-3.465) (-3.651) (-3.782) GDP growth -0.000483 -0.00424 -0.00615 -0.00635 -0.00541 (-0.119) (-1.053) (-1.404) (-1.469) (-1.347) Lag ROE 0.00837*** 0.00602*** 0.00653*** 0.00683*** (5.604) (3.538) (3.754) (4.127) Lag HHI 2.183* 2.132* 1.880* (1.934) (1.900) (1.669)

Lag Cap. ratio -1.708* -1.701*

(-1.862) (-1.967) Lag s. bank -0.308** (-2.535) Lag m. bank -0.105 (-1.106) Constant 17.55*** 17.55*** 17.49*** 17.57*** 17.71*** 17.94*** (283.2) (281.2) (308.7) (183.0) (144.0) (114.2) Observations 1,091 1,091 1,080 754 754 754 R-squared 0.502 0.502 0.527 0.548 0.553 0.565 Number of banks 110 110 110 77 77 77

Year fixed effects Yes Yes Yes Yes Yes Yes

Firm fixed effects Yes Yes Yes Yes Yes Yes

Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1

Firstly, only the lag of the liquidity ratio is regressed on ln(total loans) including the year fixed effects and firm fixed effects. In the base regression (1), the effect of a higher liquidity ratio on the outstanding loans at a bank is negative and statistically significant at the 1% level. In regression (2) till

(26)

26 (6), all control variables are added to the model. As can be seen, the effect of a higher liquidity ratio on total loans becomes smaller but is still negative and statistically significant at the 1% level when all control variables are included (regression (6)). The regressions displayed in Table 7 confirm hypothesis 1: holding more liquidity will negatively affect the lending behavior of banks.

When the liquidity ratio of a bank rises with 1% (the ratio is given in percentages), the expected average change in ln(total loans) of a bank is -0.00257 next year (regression (6), when all control variables are added), holding the other effects constant. It can also be viewed another way to analyze the economic effect on total loans immediately (instead of ln(total loans)): when there is a one unit change in the liquidity ratio, it is expected that on average total loans changes by -0.00257*100=-0.257% next year, holding the other regressors constant (Stock and Watson, 2012). Moreover, the model predicts that a bank holding the average amount of total outstanding loans (EUR 1.042*10^11, since total loans are displayed in thousands of Euros, Table 1) will decrease its total loans to: 1.042*10^11 * (1-0.00257) = EUR 1.039*10^11, when the liquidity ratio rises by 1% and the other regressors are held constant. This is a decrease of EUR 2.678*10^8 in absolute economic terms.

When looking at the profitability ratio, ROE, this coefficient is positive and statistically significant at the 1% level. On average, when the ROE of a bank rises with 1%, total loans outstanding will rise with 0.683% next year, holding the other regressors constant. The effect of an increase in ROE could be applied to a bank holding the average amount of total loans, to measure the effect economically. When a bank that holds the average amount of total loans has an increase in its ROE of 1%, total loans outstanding will increase to 1.042*10^11 * 1.00683 = EUR 1.049*10^11 next year, holding the other regressors constant. This is an absolute increase of EUR 7.117 *10^8. Jiménez et al. (2012) also find a positive relationship between a profitability ratio (ROA) and loans applications granted next year, which increases the total loans outstanding.

The coefficient on the Herfindahl index is positive and statistically significant at the 10% level. On average, there is a positive relationship between a decrease in competition (increase in the HHI) and the bank providing more loans, holding the other regressors constant. This outcome was also predicted by the existing literature of Jiménez et al. (2012) that used Spanish banks, because when there is less competition in a country, a bank has the possibility to increase the availability of credit for their bank specifically.

The effect of a higher capital ratio is negative and statistically significant at the 10% level. This might be due to deleveraging of banks (Drehmann et al., 2012): when banks reduce total assets, the capital ratio becomes larger and total loans might be reduced due to the deleveraging.

In regression (6), the dummy for the large bank is omitted, because of the dummy variable trap (small bank and medium bank are included) (Stock and Watson, 2012). Large banks is therefore the reference category when interpreting these variables. The coefficient for small banks is negative and

(27)

27 statistically significant at the 5% level. This coefficient represents the incremental effect of being in that category, relative to large banks (Stock and Watson, 2012). On average, small banks will have less total loans compared to large banks, holding the other variables constant. This is in line with the summary statistics (Tables 4, 5 and 6) which showed that large banks have the highest amount of loans outstanding. Lastly, the joint F-test was conducted to check if the fixed effects are significantly different from zero. The F-test is significant at the 1% level. Therefore the fixed effects should be included in the regression and this fixed effects panel data regression is the right model to use (Stock and Watson, 2014).

The main regression is also conducted for the crisis period (2008-2012) only, to find out if there is a difference compared to the total period. These results are depicted in Table 8.

Table 8: Results main regression for crisis period (2008-2012), dependent variable: ln(total loans)

(1) (2) (3) (4) (5) (6)

VARIABLES

Lag Liq. ratio -0.00251*** -0.00261*** -0.00242*** -0.00226** -0.00233*** -0.00247***

(-3.020) (-3.018) (-2.709) (-2.508) (-2.662) (-2.886) GDP growth 0.0159*** 0.0130*** 0.00231 0.00197 0.00278 (4.715) (4.058) (0.596) (0.507) (0.721) Lag ROE 0.00464*** 0.00212 0.00285* 0.00259 (3.352) (1.326) (1.770) (1.588) Lag HHI 1.540 1.002 1.087 (0.936) (0.581) (0.652)

Lag Cap. ratio -2.185** -1.870**

(-2.397) (-2.257) Lag s. bank -0.284** (-2.640) Lag m. bank -0.0790*** (-2.802) Constant 17.50*** 17.51*** 17.47*** 17.58*** 17.77*** 17.94*** (320.3) (307.3) (289.8) (127.2) (104.8) (98.10) Observations 550 550 550 385 385 385 R-squared 0.261 0.301 0.333 0.119 0.146 0.166 Number of banks 110 110 110 77 77 77

As can be seen in table 8, during the crisis period the coefficient of the liquidity ratio is still negative and statistically significant at the 1% level when the first lag of the liquidity ratio is regressed on the log of

(28)

28 total loans. Moreover, the coefficient of interest stays negative and statistically significant at the 1% level when the control variables are added in regression (2) till (6). When there is a one unit change in the liquidity ratio of European banks within this period, it is expected that on average total loans changes by -0.00247 * 100 = -0.247% next period, holding the other regressors constant (Stock and Watson, 2012). This effect is somewhat smaller compared to the effect using the whole period (-0.257%). This is in line with the second hypothesis that the effect of holding more liquidity on outstanding loans would be negative, but less negative compared to the whole period (2005-2014).

Regarding the economic importance of the coefficient, a bank that holds the average amount of total loans is considered. When a bank that holds the average amount of total loans increases its liquidity ratio by 1%, its total loans outstanding next year will decrease to: 1.042*10^11 * (1-0.00247) = EUR 1.039*10^11, holding the other regressors constant. This is a decrease of EUR 2.574*10^8 in absolute economic terms compared to EUR 2.678*10^8 including the whole period.

Furthermore, the effect of a higher capital ratio is still negative and statistically significant at the 10% level when conducting the regression for the crisis period. The negative effect might be due to deleveraging of banks, as already mentioned before. However, as shown in Table 8, the coefficient for the capital ratio is larger now, compared to the regression output of the whole period (1.870 compared to -1.701). This could be explained with the trend towards deleveraging among European banks in medium terms since the start of the crisis (Drehmann et al., 2012).

As shown in Table 8(6), the coefficient regarding ROE is not statistically significant anymore when only the crisis period is considered. When considering the coefficient in economic terms, it can be seen that the coefficient is approximately three times as low compared to the total period: 0.00259 compared to 0.00683. A possible explanation relating to the existing literature could be that during a crisis, banks will use less of the increase in their profitability to support lending, since banks are on average less eager to increase lending during a crisis (Berger and Bouwman, 2009b).

Moreover, the coefficients on GDP growth and competition (measured by HHI) are not statistically significant in Table 8(6). This contradicts the existing literature of Jiménez et al. (2012), which reports that the effect of GDP growth and competition should be positive and statistically significant since these are important characteristics that may affect lending. The statistical insignificance of GDP growth and HHI could be due to the relative small sample size of 77 banks compared to 350 banks in the sample of Jiménez et al. (2012).

In regression (6), the dummy for the group of large banks is omitted again because of the dummy variable trap (small bank and medium bank are included). To interpret these variables, large banks is again the reference category (Stock and Watson, 2012). As depicted in Table 8, the coefficients for small banks and medium banks are negative and statistically significant (at the 5% and 1% level). On average, small

(29)

29 banks and medium banks will have less total loans compared to large banks in this specific period, holding other regressors constant. This is in line with the summary statistics (Tables 4, 5 and 6) which showed that large banks have the highest amount of loans outstanding.

The joint significance of the fixed effects was also tested for the crisis period. Again, the fixed effects are jointly statistically significant at the 1% level (Stock and Watson, 2014).

Regression (1) is also conducted for each group of banks separately (small, medium, and large banks), to examine if there are differences between the groups:

𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑝𝑜𝑠𝑖𝑡𝑠 _𝑖𝑡−1+ 𝛽3∗ 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ𝑗𝑡+ 𝛽4∗ 𝐶𝑎𝑝𝑖𝑡𝑎𝑙 𝑟𝑎𝑡𝑖𝑜𝑖𝑡−1+ 𝛽5𝐻𝐻𝐼𝑗𝑡−1 +

𝛽6𝑅𝑂𝐸𝑖𝑡−1+ 𝑡𝑖𝑚𝑒 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝑓𝑖𝑟𝑚 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝑒𝑖𝑡

The regression is done for the two specific time periods: 2005-2014 and 2008-2012. The results can be found in Table 9.

Table 9: Results regressions per bank, for total period and crisis period (2008-2012), dependent variable: ln(total loans)

Large banks Medium banks Small banks

(1) (2) (3) (4) (5) (6)

VARIABLES All years Crisis All years Crisis All years Crisis

Lag Liq. ratio -0.00310** -0.00177 -0.00217** -0.000896 -0.00498*** -0.00509***

(-2.419) (-1.651) (-2.291) (-1.090) (-4.313) (-3.847) GDP growth 0.00939 0.0159 -0.00684 -0.00525 -0.00464 0.00492 (0.577) (0.779) (-0.770) (-0.943) (-1.314) (1.198) Lag ROE 0.00327 -0.000630 0.00667** 0.00472** 0.00764*** 0.00350 (0.848) (-0.135) (2.610) (2.784) (3.638) (1.612) Lag HHI -3.141 -2.625 4.171** 3.813*** 1.043 -0.375 (-0.933) (-0.476) (2.747) (3.216) (0.969) (-0.201)

Lag Cap. ratio -8.762 -9.889 0.0174 -0.572 -1.759* -1.883**

(-1.281) (-1.220) (0.00816) (-0.286) (-1.938) (-2.280) Constant 21.04*** 20.90*** 18.72*** 18.62*** 16.79*** 16.81*** (46.97) (43.60) (103.8) (138.7) (122.2) (88.01) Observations 127 70 155 76 472 239 R-squared 0.462 0.144 0.708 0.349 0.568 0.283 Number of banks 15 15 23 18 39 35