The Impact of Bank Competition on the Bank Lending Channel:
Empirical Evidence from the United States
By
Florine de Mol van Otterloo
June 2017
University of Groningen Faculty of Economics and Business
Master Thesis International Economics and Business Supervisor Dr. A.C. (Andreas) Steiner
Co-supervisor Prof. dr. J. (Jakob) de Haan
__________________________________________________________________________________
Abstract: This paper analyses the influence of bank competition on the transmission of monetary policy through the bank lending channel in the United States. The banks’ return on assets ratio and net interest margin are used to proxy the level of competition over the period 2003-2016, thereby assessing the impact of the recent financial crisis. Using both dynamic and static models yield the robust result that the impact of monetary policy on banks’ loans reduces as market power increases. This finding is only valid in the post-crisis period, as there is no evidence of the existence of the bank lending channel in the years prior to 2008. Furthermore, this paper confirms that different competition indicators yield different results about competitive behaviour within countries over time.
Keywords: Bank lending channel; Bank competition; Monetary transmission; Financial crisis ___________________________________________________________________________
Table of Content
1. Introduction………. 1.
2. Literature review………. 2.
2.1 Bank competition……….. 2.
2.1.1 Measuring bank competition……….. 4.
2.2 The bank lending channel……….. 5.
2.3 Bank competition and the bank lending channel……… 7.
2.4 Evaluation of the US banking system………. 8.
3. Methodology……… …… 10.
3.1 Empirical model………. …… 10.
3.2 Discussion of variables………...…… 12.
3.3 Estimation methods………... 14.
3.3.1 Generalized Method of Moments………...…………. 14.
3.3.2 Fixed Effects model………. 15.
4. Data……….. 16.
4.1 Descriptive statistics………... 17.
4.2 Panel data assumptions………... 19.
5. Results………... 20.
5.1 Main results system GMM………... 20.
5.1.1 Complete period……… 20.
5.1.2 Pre- and post-crisis period……….... 22.
5.1.3 Robustness checks system GMM………. 23.
5.2 Main results fixed effects model………. 28.
5.2.1 Complete period……… 28.
5.2.2 Pre- and post-crisis period……… 29.
5.2.3 Robustness checks fixed effects estimations……… 30.
5.3 Alternative robustness checks………. 31.
6. Conclusion……… 35.
6.1 Limitations and future research……….. 35.
References……… 36.
Appendices……….. 39.
Appendix I: Definition and sources of variables………. 39.
Appendix II: Panel data assumption tests……….... 40.
Appendix III: Averages main competition indicators………. 42.
Appendix IV: Robustness checks system GMM………. 43.
Appendix V: Robustness checks fixed effects estimator………. 46.
1
1. Introduction
Banks play a crucial role in the economy, as they are a major source of credit to households, firms and governments. In the aftermath of the global financial crisis one of the key issues in order to recover is the availability of bank credit to firms and households. By providing the necessary funds financial intermediaries are able to finance new investment projects and provide opportunities for sustainable and stable growth (Leroy, 2014). Although the nature of the financial system is currently more market-based than bank-based in the United States (US), the total credit banks extend to the private non-financial sector still amounts over 50% of GDP (BIS, 2017). Therefore, banks’ adjustment of their lending and pricing of loans in reaction to changes in monetary policy rates by the central bank is an important channel through which monetary policy affects the real economy (ECB, 2008). In this respect, renewed interest arose lately in studying the role of bank credit in the monetary policy transmission mechanism; the so-called bank lending channel.
Furthermore, in the United States, regulations have been implemented with the objective to stimulate competition in the banking sector and enhance financial integration (Degryse et al., 2014). Over the past years much research has been done in the field of bank competition, often as a determinant of another factor such as stability (Fiordelisie & Mare, 2014; Zigraiova & Havranek, 2016). Theory explains how the degree of competition in the banking sector is meant to increase efficiency, enhance the quality of financial products and stimulate innovation. More importantly, bank competition leads to economic growth as it facilitates better access to financial services for firms and households (Claessens, 2009). Increased bank competition however also has its disadvantages. It may encourage banks to engage in more risky activities and thereby undermine the stability in the financial sector (Sun, 2011). This implies the need for competition policy measures to consider broader aspects of bank competition (Claessens, 2009). Nonetheless bank runs could happen to monopolist banks too; therefore competition is not a necessary condition of fragility (Sun, 2011).
Changes in regulation as well as events in global financial markets have influenced the degree of competition in the United States banking landscape. For example, the introduction of the Riegle-Neal Act in 1994 removed many of the restrictions imposed on banks to open bank branches across state lines (Bikker & Spierdijk, 2008). More recently, the global financial crisis of 2008 induced the introduction of policy measures in order to enhance financial stability whilst also exposing the need for a change in macro prudential regulation in order to mitigate systemic risk (Claessens, 2009). In response to the crisis, the Dodd–Frank Wall Street Reform and Consumer Protection Act was signed into federal law by the Obama administration in 2010. Amongst many other financial reforms, it imposes more stringent prudential standards and establishes a number of new regulatory agencies. These regulations have most likely influenced bank competition in the United States. In the present-day the presidential election of Donald Trump will also have its reflection on the US banking sector, as Mr Trump is set to dismantle parts of the Dodd-Frank financial regulatory framework (Financial Times, 2017). However these events are too recent at the time of writing to draw firm conclusions.
2
abetting or hampering the transmission of monetary policy decisions (Fungáčová et al., 2014). There is no consensus on the influence of financial intermediaries on monetary transmission (Leroy, 2014). From an empirical point of view, some research link bank competition with the transmission of monetary policy through the bank lending channel (Adams & Amel, 2005; Olivero et al., 2011; Leroy, 2014). However the amount of literature present is limited and the results are inconclusive. The aim in this study is to empirically assess how bank competition affects the bank lending channel in the United States, with the use of a panel dataset covering the period before and after the 2008 financial crisis. It is analysed whether the bank lending channel is shaped by market power as well as different bank characteristics (liquidity, size and capitalization). By estimating both a dynamic and static model, it is found that competition in the US banking sector strengthens the transmission of monetary policy through the bank lending channel, though only in the post-crisis period. The contribution of this paper to the existing literature is threefold. First, the measurement of competition is subject to debate in the current literature. Hence whilst previous research typically uses the H-statistic or Lerner index to measure competition, this paper focuses on different measures of bank competition as proposed by Carbó et al. (2009). Second, this paper provides new evidence of the effects of the global financial crisis on bank competition and the bank lending channel. The use of bank-level data observed over the period 2003-2016 allows testing for bank-specific characteristics and gives us the desired insight on the pre- and post-crisis period. Lastly, research in this field of study focussing on the United States is barely available and out dated. Consequently there is no consensus on the current impact of bank competition on the bank lending channel in the United States.
The remainder of this paper is structured as follows. Section 2 presents the literature overview which summarizes previous empirical studies and describes the research hypothesis. Section 3 provides the methodology used, and afterwards the data is elaborated upon in Section 4. Section 5 presents the findings of the regression; and Section 6 concludes.
2. Literature review
2.1 Bank competition
The competitive conditions of global banking sectors have been a popular subject of academic analysis together with their specific market structure, i.e. perfect competition, oligopoly, monopolistic competition, and monopoly (Apergis et al., 2016). The 2008 financial crisis reignited the interest of policy makers and academics in bank competition and the role of governments in competition policies. Subsequently the crisis, and the policy responses following the crisis, may have altered the competitive behaviour of financial intermediaries in developed economies. The majority of empirical studies show how monopolistic competition dominates the various global banking systems (Bikker & Haaf, 2002a; Claessens & Laeven,
2004; Apergis et al., 2016).The US banking sector is no exception in this respect.
country-
3
specific characteristics are taken into consideration and support that greater foreign bank presence and fewer activity restrictions can make for more competitive banking systems. In earlier research comprising the United States, Bikker and Haaf (2002a) consider 23 countries over the period 1991-1997 by also using the H-statistic. Again the American banking industry was found to be subject to monopolistic competition. Additionally they document competition to be stronger among large banks, which typically operate in an international environment. Smaller banks that predominantly operate in local markets are found to be less competitive. In 2008, Bikker and Spierdijk conduct the first worldwide investigation of the developments in banking competition, by covering 101 countries and spanning the years 1986-2004. Again the Panzar and Rosse approach was used to measure banks’ market power over time. The competitive climate in the US was found to be subject to large fluctuations, with a break detected in 2001 that indicated a period of lower competition. The authors attribute the decline in competition to the process of consolidation, which generally creates larger banks with more market power. Another explanation for the decline in competitiveness is the shift over time from traditional intermediation services to more complex and sophisticated banking products. However they too conclude that monopolistic competition prevailed in the US over the period observed.
Most recently, Sun (2011) applies the popular H-statistic and finds that bank competition declines in the US following the crisis. However this evidence must be viewed as preliminary only, as the time period observed ranges from 1995 to 2009. Similarly, research has shown how competition in the European Union banking sector has a small but significant decline following the adoption of the euro currency and the on-going financial crisis (Apergis et al., 2016). Apergis et al. use the same methodology to assess banks’ competitiveness. When comparing European to US banks, De Bandt and Davis (2000) find European banks to be less competitive than United States banks. Interestingly, different studies describe how the US appears to have slightly less competitive banks than Europe (Bikker & Haaf, 2002a; Claessens & Laeven, 2004), despite using the same methodology (H-statistic) over roughly the same time period. These contradictory findings can be attributed to the fact that the latter studies use significantly broader research samples with more observations. Instead De Bandt and Davis (2000) use a small sample of observations obtained from three countries (Germany, Italy and France) to draw conclusions on European bank competitiveness. This may lead to an underestimation of the effective changes in competition over time.
4
2.1.1 Measuring bank competition
The measurement of bank competition is currently subject of debate. In the empirical banking literature various indicators are used to measure the degree of bank competition, which yield different results about competitive behaviour across and within countries over time (Carbó et al., 2009). The existing techniques used to address the level of bank competition can be divided into structural and non-structural approaches (Bikker & Haaf, 2002b). The structural approach aims to measure bank competition by examining the market structure with concentration ratios and the Hirschman–Herfindahl index (HHI) (Yang & Shao, 2016). Particularly the C3 and C5 ratios –the concentration ratio for the three and five largest banks- are used widely. These measures rely on the traditional Structure-Conduct-Performance (SCP) paradigm, under which concentration is negatively related to the level of competition (Yang & Shao, 2016). However concentration ratios do not necessarily imply a lower level of bank competition (Khan et al., 2016), and numeral studies indicate that concentration is an unreliable measure of bank competition as the market power measured is often much higher or lower than the market structure would imply (Bikker et al., 2012; Claessens & Leaven, 2004).
Under the non-structural approach of assessing the level of bank competition from the New Empirical Industrial Organization (NEIO) literature, the Panzar and Rosse model –the H-statistic– and the Lerner-index are the most popular indicators. These measures are developed from static theory of the firm models under equilibrium conditions and typically use some form of price mark-up over a competitive benchmark (Carbó et al., 2009). Numerous studies use the H-statistic and find banks operating under monopolistic competition in virtually every country observed. The H-statistic examines the extent to which a change in factor input prices such as wage and funding is reflected on revenues earned by banks. Because structural changes made post-crisis may distort the long-run market equilibrium required for the validity of the H-statistic, post-crisis estimates are said to provide preliminary evidence only (Sun, 2011). Since the H-statistic captures the aggregate level of bank competition in a country, the advantage of the Lerner index is that it captures bank-level estimates of competition. This is an important difference as the banking markets may be local in nature, making it difficult to measure competition at the national level (Fungáčová et al., 2014). The Lerner index is a direct measure of competition as it captures the degree to which a firm can increase its marginal price above the marginal cost (Yang & Shao, 2016). Other indicators that are widely used in order to infer competitive behaviour are the net interest margin (NIM), return on assets (ROA), and the Boone indicator. The underlying idea of the Boone indicator is that competition enhances the performance of efficient banks, and weakens the performance of inefficient banks. The disadvantage of this measurement is that it assumes banks pass on at least part of their efficiency gains to their customers (van Leuvensteijn et al., 2007). The NIM and ROA measure overall performance of individual banks, making them more aggregate competition indicators (Carbó et al., 2009). The net interest margin looks at the bank-specific net interest return, whilst the return on assets is more comprehensive and includes off-balance sheet returns and non-interest costs. Table 1 summarizes the different indicators of banking competition.
5
income from non-traditional services significantly affect the non-structural measures of competition, as do differences in business cycles and inflation. These country-specific differences need to be considered and adjusted for when determining the longer-term bank pricing power in cross-country comparisons of banking market competition. Overall, they conclude that the best within country measure is the net interest margin when the focus is on traditional banking (loan and deposit services), and the Lerner index and return on assets are best to be used when examining competition in broader banking activity. Also, these indicators are influenced by country-specific factors, the national law and taxation system, and bank specific characteristics such as risk preference and leverage (Claessens, 2009). Therefore following Carbó et al. (2009), this study uses the NIM and ROA in order to gauge the level of competitiveness of US banks. To the best of knowledge, no other study has focussed on these within country competition indicators so far in the literature of the bank lending channel. The Lerner index and Boone indicator will be employed to verify the robustness of the results.
Table 1 Main empirical indicators for bank competition
Concentration Ratio’s C3/C5 The concentration ratios for the three and five largest banks Hirschman–Herfindahl index The sum of the squares of the market shares of banks Lerner Index The mark-up of price (average revenue) over marginal costs,
calculated as Lerneri = (Pricei - MCi)/Pricei
H – statistic Based on a reduced-form revenue equation, the H-static measures the elasticity of total revenues with respect to changes in factor input prices.
Boone indicator It captures the reallocation of market share form inefficient to efficient firms. A profitability equation measures the level of competitiveness: ln(πi) = α + βln(ci) + εi
Net interest margin The ratio of bank net interest margin to total earning assets Return on assets The ratio of bank net income to the value of total assets Sources: Carbó et al., 2009; Apergis et al., 2016; Khan et al., 2016
2.2 The bank lending channel
6
determine the amount of deposits through the reserve requirement. The latter view argues that monetary policy shocks change the relative yields of deposits (and other assets), and thereby influence the amount of deposits households are willing to hold. Therefore a tightening (loosening) of monetary policy leads to a reduction (expansion) in banks’ loan supply (Juurikkala et al., 2011). A key assumption in this model is that banks cannot easily replace the lost deposits with other sources of funds, such as certificates of deposits (CD's), or issue new equity (Bernanke & Gertler, 1995). If banks would be able to compensate for the contraction of loanable funds without additional costs the bank lending channel would shut down (Fungáčová et al., 2014).
In literature the existence and the functioning of the bank lending channel is heavily debated. Quite recently, Disyatat (2011) questions the relevance of the bank lending channel as he argues how the emphasis on policy-induced changes in deposits is misplaced. Following Bernanke (2007), Disyatat (2011) proposes an alternative mechanism for the bank lending channel that imposes the impact of monetary policy on banks’ external finance premium as determined by their balance sheet strength. The theory behind this entails that contractionary monetary policy will affect the bank’s balance sheet negatively not only in terms of leverage and asset quality, but also in perceptions of risk. This will increase the bank its external finance premium, which increases their cost of fund and in turn is passed on to the lender (Yang & Shao, 2016). Romer et al. (1990) even rejected the existence of the bank lending channel as they consider a bank’s ability to diversify its funding sources. A monetary shock can be cushioned since banks are able to increase the share of financing through financial markets, in particular by issuing CDs. In contrast, Kashyap and Stein (2000) are the first to advance the analysis by focussing US bank-level panel data. They find strong evidence on the existence of the bank lending channel in the period 1976 to 1993. Kishan and Opiela (2000) too analyse the US over a similar period and find small, less-capitalized banks especially affected by monetary policy. Moreover, Peek and Rosengren (2014) consider many empirical studies –including the literature previously discussed- and point out that banks in the United States indeed face difficulties in finding alternative funding sources when faced with deteriorating loanable funds, and so may contract their lending. Additionally they show that in the US, many non-financial firms are bank-dependent as they lack access to external finance from other sources.
7
unable to find alternative sources of funding (e.g. due to its size or capital position) or due to having insufficient liquidity buffers (ECB, 2008).
Beyond size, liquidity and capital, there are other factors discussed in literature that might affect the transmission of monetary policy. The bank lending channel is said to have changed as a result of deregulation and financial innovation (Bernanke & Gertler, 1995). New factors include changes in banks’ business models and market funding patterns, banks’ short-term funding and securitization activities and share of non-interest income activities (Altunbas et al. 2009; Gambacorta & Marques-Ibanez, 2011). Finally, the risk factor is introduced in studying the effects of the monetary policy transmission mechanism (Leroy, 2014). However these factors are beyond the scope of this paper.
Additionally the bank lending channel has been subject of current literature in the light of the recent financial crisis, indicating the importance of bank-specific characteristics in the provision of credit post-crisis (Gambacorta & Marques-Ibanez, 2011). According to Leroy (2014), the strengthening of the monetary policy transmission mechanism can be attributed to the implementation of unconventional monetary policies as well as the downward moves in interest rates. However, the bank lending channel is said to be particularly important for the Eurozone, as the financing of corporations by banks loans is clearly more extensive in Europe than in the US (Fungáčová et al., 2014). Therefore the current effectiveness of the bank lending channel in the United States is questionable.
2.3 Bank competition and the bank lending channel
The speed with which the interest rate of banks is adjusted to changes in the central bank interest rate depends inter alia on the degree of competition among banks (ECB, 2008). Several empirical papers investigate the relationship between bank competition and the transmission of monetary policy through the bank lending channel. The first to do so are Adams and Amel (2005) by using concentration indicators –more specifically the Herfindahl index- to measure the degree of bank competition, and analyse its effect on the supply of lending to small businesses with the use of aggregate data from the United States over the period 1996-2002. They conclude that the impact of monetary policy on the supply of loans is weaker in concentrated markets. However one year earlier, Claessens and Laeven (2004) found no correlation between the degree of competitiveness and concentration using the same competition indicator. Hence Adams and Amel’s analyses are controversial (Gunji et al., 2009).
8
model as main market power measure. In a more recent study, Yang and Shao (2016) focus on micro-data from the Chinese banking system over the period 2003-2014 and find similar results: increased competition in China’s banking sector reduces the effectiveness of monetary policy transmission through the bank lending channel. In this particular case the Lerner index was used as indicator of the degree of competition.
Nonetheless, Leroy (2014) and Fungáčová et al. (2014) find contradicting results when examining countries in the Eurozone throughout approximately the same period (1999-2011). Bank-level data from twelve euro area countries is analysed with the Lerner index as a proxy for bank competition. Both studies have similar results: increased competition enhances the effectiveness of monetary policy. In another recent study on the ASEAN countries over the period 1999-2014, Khan et al. (2016) use both structural and non-structural measures to proxy the degree of competition (CR5, HHI, and the Lerner index), and find that increasing bank competition strengthens the transmission of monetary policy. However the results differ when using the Boone indicator; increased competition weakens the monetary policy transmission. These findings confirm the main finding of Carbó et al. (2009) who suggest that the observed level of market power may differ when using different indicators for bank competition. Therefore cross-country comparisons of banking competition may lack consistency and may be unreliable as presently conducted.
With regard to the US, a thorough research of the relevant literature yielded only one related article. Brissimis et al. (2012) observe the US and euro area banking market over the period 1997-2010, whilst using the Boone and Lerner index to assess the level of bank competition. They find that bank competition strengthens the bank lending channel of monetary transmission for the euro area and the US. This paper extends their work, however it differs from Brissimis and colleagues in many ways. They include the euro area in their sample, and contrast in model specification as they also consider the risk taking channel in addition to the bank lending channel. As such, their aim is not to provide detailed knowledge on the relationship between market power and the bank lending channel in the US, but to conduct a much broader analysis. Also the time period observed in this paper is more recent in order to expose the long-term effect of the financial crisis, and lastly, the set of competition measures is broader as bank-level as well as country-level indices are used in this paper.
2.4 Evaluation of the US banking system
9
States, banks extend around 30% of the total credit that goes to the non-financial sector (Dembiermont et al., 2013).
Regulatory changes removed the restrictions – which were in place for the better part of the twentieth century- that affected branching, product, and price competition. As mentioned before, the 1994 Riegle-Neal Act endorsed bank competitiveness by allowing banks to operate branches across state lines. Another important change was the Gramm-Leach-Bliley Act set in force in 1999. By doing so, it effectively repealed the Glass–Steagall Act of 1933, which prevented competition between commercial banks and non-depository institutions. These regulatory changes allowed US banking organizations to engage in underwriting and other dealing activities, plus banks moved quickly to expand across state borders. These laws have thereby improved competition and contributed to international integration together with the entry of new types of competitors using the Internet (Bikker & Spierdijk, 2008). Following the changes in regulation, the number of banks declined tremendously via mergers and acquisitions and bank failures. The increase in competitive conditions drove the least efficient banks out of the market, although this was partly offset by the establishment of many new branches (DeYoung, 2014). The number of US commercial banks has diminished from approximately 14,000 in the 1980s to about 5,000 today (FRED, 2017). Not only changes in regulation affected the US banking structure, as innovation and technological change during the 1980s and 1990s allowed US commercial banks to remain competitive in the provision of credit to large businesses. Nowadays, the growth of online banking has most likely caused the halt in increases in US bank branches (DeYoung, 2014).
At the European level, the financial crisis that started in 2007 has also altered the degree of competition in the banking sector as it led to changes in prudential regulations, including the introduction of Basel III (Leroy, 2014). Basel III greatly increases the amount of common equity capital that the largest banks are obligated to hold, and the US regulators decided in 2012 to apply it to all US banks and financial institutions with over $500 million in assets (Berger et al., 2014). Furthermore, the Dodd-Frank Act was introduced as a response of the national government following the crisis. The Act extended the Federal Reserve’s supervisory powers substantially, although these are still shared with other federal and state agencies (DeYoung, 2014). The US Congress has established the objectives for monetary policy in the Federal Reserve Act, consisting of three main objectives: maximum employment, stable prices, and moderate long-term interest rates (Federal Reserve, 2017). As the crisis built, the Federal Reserve reduced the interest rate to nearly zero with the aim of generating monetary stimulus in order to boost the real economy. However hitting the zero lower bound for the short-term interest rate effectively means the Fed’s hand are tied, and makes them rely on unconventional monetary policy actions to stimulate the economy. And as the US financial system is said to rely heavily on market-based funding, the effects of the interest rate adjustments on the real economy, and more specifically, on the annual growth of loans could be relatively small.
10
3. Methodology
3.1 Empirical model
To examine the role of bank competition on the monetary policy transmission, this paper follows the empirical model of Erhmann et al. (2003), which has often been used by previous studies on the bank lending channel. Recently, Leroy (2014) used a similar model to study the effect of bank competition on the bank lending channel in the Eurozone. The model of Ehrmann and his colleagues (2003) is based on a simplified theoretic framework from Bernanke and Blinder (1988). It assumes that demand for deposits D equals money supply M, and that both depend on the monetary policy rate (mp).
M = D = −ψ(mp) + χ (1)
where χ represents all factors that influence money deposits other than the monetary policy stance. Banks face a loan demand depending on real GDP (y), price level (p), and the interest rate on loans (r).
Ld = φ1y + φ2p – φ3r (2)
The supply of loans of banks depends on the amount of loanable funds (deposits or money) available, the interest rate on loans and the monetary policy rate. Monetary policy enters the loan supply function here in two ways. The first link is the opportunity costs for banks when using the interbank market for funding loans. Secondly the amount of loanable funds available depends negatively on the policy interest rate (Ehrmann et al., 2003; Fungáčová et al., 2014).
Ls = φ4D(mp) + φ5r – φ6mp (3)
Furthermore the assumption is that not all banks are equally dependent on deposits. The impact of a deposit change is expected to be lower for banks with higher size, liquidity and capitalization characteristics (Χi). In this paper the bank-specific variable bank competition
(COMP) is added as bank characteristic in order to assess whether bank competition influences
the bank lending channel.
φ4 = µ0 – µ1Χi (4)
Assuming the loan market clears and taking into account the above equations, loan supply is modelled as:
L = ay + bp + c0MP + c1ΧiMP + dΧi + constant (5)
Loan supply depends on the economic activity (y), the price level (p), the monetary policy rate
(MP), the bank-specific characteristics (Χi), and the interaction term of the latter two (ΧiMP).
11 ∆ ln !"#$ !,!= !!+ !∆ ln !"#$ !,!!!+ ω!∆MP!!! ! !!! + !!!,!!!+ θ!!!,!!! ! !!! ∗ ∆MP!!! + !!! !!! ! !!! + !!"
where i indexes banks, and t represents the time respectively. The dependent variable ∆ ln !"#$ !,! stands for the change in loans (in natural logarithm) by bank i at time t. The one-year lag of loan growth is added to the equation to capture the persistence of the dependent
variable (Yang & Shao, 2016). ∆MP denotes the change in monetary policy; !!" is a vector of
bank-specific characteristics (competition, size, liquidity, and capitalization), whilst !!
represents the macroeconomic control variables GDP growth and inflation. In this empirical model, the existence of the bank lending channel should be reflected by a significant
coefficient of the monetary policy variable (ωj), as well as the coefficients of the interaction
terms (θj) (Fungáčová et al., Leroy, 2014). Moreover !!" is the random error term, and the
model further includes a bank-specific fixed effect variable !!. Bank competition is indicated
by the variable COMP, and the interaction variable ∆MP * COMP captures the marginal impact
of bank competition on the effects of a monetary policy shock on credit growth. The other three bank-specific characteristics are factors that may influence a bank’s access to and premium on external finance. Following Ehrmann et al. (2003), bank characteristics are defined as: Size!" = !"#!"− 1 !! ! !"#!" (7) Liquidity!"= !!" !!" – 1 ! 1 !! !!" !!" ! ! 8 Capitalization!" = !!" !!" – 1 ! 1 !! !!" !!" ! ! 9 Size is calculated by taking the natural logarithm of total assets. Liquidity is the ratio of liquid
assets to total assets, with the definition of liquid assets defined by Orbis Bank Focus1. Lastly
capitalization is the bank’s own equity to total assets ratio. The three bank characteristics are normalized with respect to their average across all the banks in the sample. To eliminate the unwanted trend in size, the size variable is not only normalized with respect to the mean of the whole sample period, but also with respect to the mean of each single time period (Ehrmann et al., 2003). According to Altunbas and his colleagues (2009), the normalization allows interpreting the coefficient of monetary policy (ωj) as “the average monetary policy
effect on lending for an average bank”. Furthermore the use of bank characteristics (size, liquidity, and capitalization) is accompanied by potential endogeneity problems (Arena et al., 2007). For example, a bank may become larger precisely because of rapid loan (and deposit) growth. Or a financially constraint bank may choose to have higher capital or liquidity ratios,
1 The Orbis Bank Focus defines the following to be liquid assets: cash and balances within central banks,
12
which depletes the usefulness of these indicators as measures for financial constraints. To
control this problem, the model is estimated using lagged values of bank characteristics2.
The approach followed in model (6) assumes that the macroeconomic variables capture the relevant time effect (Ehrmann et al., 2003). Therefore one control that is not adopted is a variable representing time dummies, as similar results can be obtained using macroeconomic variables (Gambacorta & Marques-Ibanez, 2011). To verify this assumption, the macroeconomic variables are substituted with time fixed effects as a robustness check. Lastly, this analysis distinguishes between the pre- and post-crisis time period, from 2003-2007 and 2008-2016 respectively, in order to grant better insight on the effects of the global financial crisis on the relationship between bank competition and the bank lending channel.
3.2 Discussion of variables
As the bank lending channel focuses on the transmission of monetary policy through the lending activity of banks, the dependent variable loan growth captures the annual change in
natural logarithm of a bank’s gross loans3. The four bank characteristics size, liquidity,
capitalization and competition can influence loan supply directly, and indirectly via the bank lending channel. Ehrmann et al. (2003) explains how the distributional effects of monetary policy should be reflected in a significant coefficient of the interaction terms of bank
characteristics with the monetary policy variable (θj).
In order to examine the role of bank competition on the effectiveness of monetary policy, ROA and NIM are used to indicate the degree of competitiveness in the banking sector. The return on assets profitability measure does not merely take traditional loan and security asset holdings into account, but considers all different sources of bank income. Net interest margin reflects the difference of interest income generated by banks and the interest paid to their lenders, relative to the amount of their earning assets. Higher returns on these indicators means a bank has more market power and hence reflects less competitive conditions. The main variable of interest in this paper is the interaction term of bank competition and monetary policy. Expected is that bank competition can either strengthen or weaken the effects of monetary policy transmission via the bank lending channel, and therefore the coefficient of this interaction term can either be positive or negative. A significant and positive (negative) sign of this coefficient indicates that as market power increases, the relationship between the monetary policy stance and bank lending activity weakens (strengthens). Thus, a positive sign means that banks with higher market power are less sensitive to changes in the monetary policy rate, as banks still experience positive loan growth. In this case market power would reduce the effectiveness of monetary policy.
The other bank characteristics incorporated in our analysis (size, liquidity, capital) are believed to also influence transmission process of monetary policy via the bank lending channel, and thereby the loan supply (Leroy, 2014). As previously discussed in the literature
2 Additional computations revealed that endogeneity is indeed a large problem in the empirical model.
Computations of the regression model with current and lagged values of bank characteristics (i.e. Xi,t and Xi,t-1),
leads to values significantly larger than 1 for the coefficients of Xi,t and Xi,t-1. When dropping the current
bank-specific variables (Xi,t) the coefficients in the model lose their significance to a great extent, and the values of the
coefficients of Xi,t-1drop below 1.
3 The loans variable consist of all loans to customers: mortgage loans, consumers loans, corporate loans and other
13
review, smaller, less liquid or less capitalized banks are expected to react more strongly to a monetary policy shock. As such, the coefficients of the interaction terms are expected to be positive and significant. The rationale behind this is that banks that are larger in size, more liquid and better capitalized are less sensitive to changes in monetary policy rate as they still experience a positive loan growth. This would imply that an increase of these bank characteristics would reduce the effectiveness of monetary policy.
The macroeconomic control variables, GDP growth and inflation, are added to the equation in order to control demand effects and capture business cycle fluctuations. By introducing these variables the supply-side bank lending channel is distinguished from the demand-side interest rate channel (Leroy, 2014; Yang & Shao, 2016). The relationship between GDP growth and the dependent variable is expected to be positive, as the growth rate of bank loans is larger in a growing economy where the demand for loans is rising (Olivero et al., 2011). Following Khan et al. (2016) the expected affect of inflation on the loan supply is expected to be either positive or negative. Furthermore, the Federal Funds rate is used as indicator for monetary policy. Bernanke and Blinder (1992) show how the funds rate is a good indicator of monetary policy actions, as it sensitively records shocks to the supply of banks reserves. Hence this measure is most often used in previous literature (e.g. Kashyap & Stein, 2000). Higher values of the federal funds rate are associated with tighter monetary policy. Annual averages of monthly rates are used in the analysis. By averaging, there is accounted for whether changes in the federal funds rate occurred relatively early or relatively late in the year of observation (Adams & Amel, 2005). The expected relationship between monetary policy and loan growth variable is negative, such that in increase in interest rate reduces the bank’s lending activity. Therefore the coefficient of monetary policy (ω) is expected to be negative.
14
therefore nominal interest rates cannot fall below zero (Bullard, 2012). In their study on the macroeconomic impact of monetary policy at the zero lower bound Wu and Xia (2016) calculate the shadow interest rate. Figure 1 illustrates the model created by Wu & Xia (2016), which shows the negative US federal funds rate would there be no zero lower bound.
Figure 1: The Wu-Xia Shadow Interest Rate (in %) Source: Board of Governors of the Federal Reserve System (2017) and Wu & Xia (2016)
3.3 Estimation methods
3.3.1 Generalized Method of Moments
When estimating a dynamic model such as equation (6), the lagged dependent variable is correlated with the unobserved panel data effect and therefore an ordinary least squared (OLS) regression would be biased and inconsistent (Stata, 2017). Therefore the empirical model in equation (6) is estimated using Generalized Method of Moments (GMM) designed by Arellano and Bover (1995) and further developed by Blundell and Bond (1998). The GMM method controls for unobservable heterogeneity resulting from using bank level data. It is best to be used for panel data with small time periods and large cross-sections; dynamic dependent variable (depends on its own past values); independent variables that are not strictly exogenous (they are correlated with past and possibly current realisations of the error term); and there are fixed individual effects and heteroskedasticity and autocorrelation within individuals but not across them (Roodman, 2009a). The Arellano and Bond (1991) model starts transforming the regressors by first-differencing the model, and is called difference GMM. The augmented model, system GMM, combines the standard set of first-differenced equations with a suitable lagged level as instruments, with an additional set of equation in levels with lagged first differences as instruments. The model in equation (6) is estimated by using system GMM. In this methodology an increasing number of instruments is not uncommon. Roodman (2009b) explains how too many instruments can lead to the overfitting of the endogenous variables, imprecise estimates of the optimal weighting matrix, downward bias of the standard errors in two-step standard errors, and a weak Hansen test of instrument
-4.00 -3.00 -2.00 -1.00 0.00 1.00 2.00 3.00 4.00 5.00 6.00 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Wu-Xia Shadow Federal Funds rate (in %)
15
validity. It is unclear when precisely the instrument count is too large, but a rule of thumb is
to keep the number of instruments less than the number of groups4.
In line with previous research, the monetary policy and macroeconomic variables are considered to be strictly exogenous, and the bank characteristics and their interactions endogenous (Jimborean, 2009; Cantero-Saiz et al., 2014). Moreover second-lag instruments
are used for the endogenous variables5, and the exogenous variables are instrumented by
themselves. The GMM method is said to ensure efficiency and consistency in the presence of lagged variables and fixed effects, provided that the instruments chosen take into account the serial correlation properties of the model and that they are valid (Ehrmann et al., 2003; Altunbas et al. 2009). The latter is tested for by using the Hansen over-identification test which is commonly relied upon to check instrument validity (Roodman, 2009b). The Hansen test has the null hypothesis of the instruments being exogenous, therefore the higher the P-value the better. Also the Arellano and Bond autocorrelation test is carried out in order to ensure there is no serial correlation of order two in the equation in first difference (Leroy, 2014). In all regressions the two-step procedure is used, which uses the residuals from the one-step variant estimates and is asymptotically more efficient. Because the standard errors in the two-step estimation tend to be downward biased because of the large number of instruments, the Windmeijer (2005) finite-sample correction is applied to obtain corrected standard errors.
3.3.2 Fixed Effects model
In literature on the bank lending channel the dynamic model in equation (6) is usually run using the System GMM method. Nevertheless it is also opted that a dynamic model is not as relevant in the context of annual data as with monthly or quarterly data (Fungácová et al., 2016). The underlying idea is that there is no economic rationale as for why last year’s lending growth would influence current lending growth. Indeed, the results show that the lagged dependent variable is often insignificant, thus the use of a dynamic model is not completely justified. Because of these findings and as a cogent economic rationale for including last year’s loan growth as a regressor is lacking, equation (10) is also estimated following Fungácová et al. (2014) on annual data for the euro area. Equation (10) is estimated in a standard fixed-effects panel regression model.
∆ ln !"#$ !,! = !!+ ω!∆MP!+ !!!!,!!!+ θ!!!,!!!∗ MP! + !!! !+ !!" (10) The bank characteristics size, liquidity, capitalization and competition are lagged one period to ease endogeneity concerns.
4 Therefore in this paper in all the regressions run with system GMM, the number of instruments is kept below
the number of groups.
5 To avoid over-identification problems, occasionally deeper lags have been used as instruments for some
16
4. Data
In order to analyse the behaviour regarding the loan process in the United States following a monetary policy change, micro-level data is collected from the Orbis Bank Focus database by Bureau van Dijk. This paper uses bank balance sheet and income statement data on commercial banks operating in the US banking sector. Commercial banks are traditionally active in a combination of retail banking, wholesale banking and private banking by extending loans to individuals, SMEs and large corporates. The starting date of the empirical data in this analysis was dictated by data availability. Other types of bank specialization such as savings banks, investment banks, and cooperative banks are excluded from the sample to ensure comparability, and also because they differ in capital structures, business scopes and regulation environment (Yang & Shao, 2016). As Orbis Bank Focus offers annual datasets
consolidated balance sheet data is used6. Using bank-level data has the advantage that it does
not suffer from identification problems, which is the case according to Kashyap and Stein (2000) in studies based on aggregate data. This is due to the fact that monetary policy changes have a more pronounced effect on the credit supply for some banks than for others.
As there are over 5,000 commercial banks in the US, the sample does not include every commercial bank in the US economy. Only banks with data available for the entire period are selected. However the number of time series observations differs for some variables, leading to an unbalanced panel. Furthermore following Leroy (2014), only banks with total assets over one billion dollars during the study period are included in the sample. These larger banks are the focus in the analysis because they are said to have the largest impact on the competitive efficiency of the US banking sector (Bolt & Humphrey, 2012). The data is cleaned following criteria proposed by Arena, Reinhart and Vázquez (2007). This implies that outliers were identified and removed from the sample through the application of several filters. Specifically, bank were removed in the following cases: 1) banks with a growth rate of assets exceeding 200%; 2) banks with growth rate of loans and/or deposits exceeding 300%; 3) banks where the value of loans represented more than 100 times the value of deposits. The final sample includes 1062 observations covering the period 2003-2016 and a total of 177 banks. The sample represents 73% of the total US commercial banks in terms of total assets.
The data on the federal funds rate is retrieved from FRED, Federal Reserve Bank of St. Louis. The shadow interest rates are based on the model of Wu and Xia (2016), presented by the Board of Governors of the Federal Reserve System. The annual data on GDP growth and inflation are abstracted from the Global Development Indicator Database of the World Bank. The Lerner index and the Boone indicator are retrieved from the Global Financial Development Database of the World Bank. Table A1 in Appendix I presents the sources of the data obtained.
6 It is important to know whether a bank is a subsidiary of another bank when assessing a bank’s financial
17
4.1 Descriptive statistics
In Table 2 the descriptive statistics of the complete dataset are depicted, together with the statistics of the dataset when separated into the pre- and post-crisis time period. Due to unavailability of data, the loan growth variable together with the Boone indicator, Lerner index, and shadow interest rate have different sample sizes.
Average values of the different market power measurements provide a brief overview of the competitiveness of the banking system in the US. The aggregate (country-level) measures of US bank competition suggest the degree of market power increased following the crisis. The average value of the Boone indicator from 2003 to 2016 is low (-0.053), which indicates high level of competition. However, this value seems to increase post-crisis, which is an indication of rising market power/decreasing competition. The Lerner index suggests the same as the mean value rises slightly, from 0.274 to 0.302. These findings are contrasted in terms of bank specific measurements of market power. Both the ROA and NIM decrease – although slightly– over the sample period, which suggests that on average the degree of banking competition has increased following the crisis. The ROA shows a decrease of 0.5% (from 1.3% to 0.8%), and the net interest margin decreases 0.4% (from 3.9% to 3.5%). This reflects the finding of Carbó and his colleagues (2009) that different measurements of competition yield different results. The difference can also follow from the fact that these are country-level and bank-level estimates. It can be seen that commercial banks experienced increased competition, whilst overall bank competition decreased in the US banking sector.
Other important variables in the analysis include the loan growth, monetary policy indicators, and other bank characteristics. The average of the change in loans (in natural logarithm) over the entire period is 0.08. Here the financial turmoil of 2008 leaves its mark, as the mean loan growth falls from 0.12 to 0.06 in the post-crisis period. Furthermore the monetary policy indicators have large variations across time periods. The federal funds rate with an average of 2.89% before the crisis falls to nearly zero (0.36%) in the light of the economy-stimulating policies of the Federal Reserve. The average changes in federal funds rate are therefore negative in the period 2008-2016, as the government is lowering the interest rate to nearly zero. In this respect it might be interesting to look at the shadow interest rate, which is 2.94% before the crisis (quite similar to the federal funds rate), but decreases substantially to −0.86% in the post-crisis period. Mean changes in shadow interest rate in the post-crisis period are therefore larger as it reflects the potential interest rate were it allowed to cross the zero lower bound. Finally the statistics of the bank characteristics size, liquidity and capitalization show signs of changes in regulation after the banking crisis. The average values of liquidity and capitalization both increase in the time period following the crisis, indicating stricter policies on equity and liquid asset requirements. Average bank size increases too but only very slightly, which can be contributed to the process of consolidation.
18
Table 2 Descriptive statistics
Obs. Mean St. Dev. Min. Max. Obs. Mean St. Dev. Min. Max. Obs. Mean St. Dev. Min. Max.
Whole sample Pre-crisis (2003-2007) Post-crisis (2008-2016)
Δln(Loans) 2,478 0.079 0.146 -0.598 1.206 885 0.117 0.141 -0.575 1.013 1,593 0.057 0.145 -0.598 1.206
Net interest margin 2,655 0.037 0.012 0.001 0.132 1,062 0.039 0.011 0.001 0.132 1,593 0.035 0.012 0.001 0.129
ROA 2,655 0.010 0.011 -0.148 0.096 1,062 0.013 0.007 -0.017 0.072 1,593 0.008 0.013 -0.148 0.096 Size 2,655 22.64 1.507 20.73 28.36 1,062 22.33 1.437 20.73 27.91 1,593 22.84 1.512 20.76 28.36 Liquidity 2,655 0.076 0.088 0.000 0.778 1,062 0.067 0.075 0.000 0.604 1,593 0.082 0.095 0.001 0.778 Capitalization 2,655 0.108 0.036 0.041 0.388 1,062 0.100 0.038 0.041 0.388 1,593 0.113 0.033 0.042 0.321 Inflation 2,655 0.021 0.012 -0.004 0.038 1,062 0.027 0.006 0.016 0.034 1,593 0.016 0.012 -0.004 0.038 GDP growth 2,655 0.018 0.015 -0.028 0.038 1,062 0.027 0.007 0.018 0.038 1,593 0.013 0.017 -0.028 0.026 Boone indicator 2,301 -0.053 0.015 -0.083 -0.038 1,062 -0.066 0.011 -0.083 -0.055 1,239 -0.042 0.007 -0.059 -0.038 Lerner index 2,301 0.289 0.044 0.201 0.337 1,062 0.274 0.041 0.201 0.324 1,239 0.302 0.042 0.205 0.337
ΔFederal funds rate 2,655 -0.233 1.258 -3.090 1.860 1,062 0.187 1.390 -2.220 1.860 1,593 -0.513 1.075 -3.090 0.260
ΔShadow interest rate 2,478 -0.364 1.379 -3.040 1.940 1,062 0.182 1.373 -2.130 1.940 1,416 -0.774 1.234 -3.040 1.470
19
4.2 Panel data assumptions
Certain assumptions are analysed and tested on the data previously described. The data is examined for multicollinearity, normality, heteroskedasticity, and serial autocorrelation. Apart from testing for potential problems in the panel data, tests are conducted for every regression in order to determine which regression model is most appropriate. Tables A2 to A6 in Appendix II present the results.
The Breusch-Pagan test helps decide between a pooled ordinary least squares (OLS) regression and the random effects model. The null hypothesis that variances across entities are zero is rejected in every regression. Therefore it may be concluded that there are random effects present. Next, the Hausman test compares the coefficients of the random effects model to those of the fixed effects model. All P-values are zero, suggesting that the coefficients of both models are not equal and therefore the fixed effects model will be used for all regression estimations.
Furthermore, the pair-wise correlation coefficients between variables are obtained to identify potential problems in multicollinearity. The correlation matrix in Table A4 in Appendix II indicates that the possibility of collinearity between the variables is generally small. It is however problematic that the correlation coefficients between GDP growth and the monetary policy measures exceed the 0.5 threshold. This is not surprising since Paligorova and Santos (2017) explain how correlation exists between monetary policy measures and the macroeconomic environment; monetary policy easing is said to coincide with adverse economic conditions. As the inclusion of these variables is vital to our estimations, the effect of the inclusion of both variables is tested with the variance inflation factor (VIF) test. This test indicates that there is no sign of excessive multicollinearity
between these variables as the threshold of 10 is not exceeded7. Furthermore the correlation
coefficients between measures of competition (Boone and Lerner) and inflation are less than the -0.5 threshold. As the VIF test results show signs of excessive multicollinearity, these variables will not be included together in the model8. Other correlation coefficients higher than 0.5 are variables that will not enter the model together, therefore this is of no concern.
Moreover a modified Wald statistic is calculated for group wise heteroskedasticity in the residuals of the fixed-effects panel models (Baum, 2001). As can be seen from the results, the null hypothesis of homoscedasticity is rejected at the 1% significance level in all regressions. Additionally, the Wooldridge test for autocorrelation is performed. Again the null hypothesis of no serial correlation is rejected in all regression estimations. Therefore cluster robust standard errors are used in every regression of the fixed effects models. By doing so we guard against possible misspecification due to heteroskedasticity, and control for within individual correlation (Hill et al., 2011). Lastly, the Jarque-Bera test shows that even
20 after the elimination of outliers the regressions show a leptokurtic distribution. However this
is not uncommon in the literature of financial studies9.
5. Results
This section elaborates on the main empirical results of the analysis of model equations (6) and (10), conducted by the use of the two-step system GMM estimator and the fixed-effects model. A distinction is made between the results of the complete dataset and the results of the dataset when divided into the pre- and post-crisis period. Furthermore robustness checks with alternative measures of competition and monetary policy are estimated in order to verify the results obtained. Lastly, alternative robustness checks are presented in order to verify the ability of GDP growth and inflation to capture business cycle dynamics and time fixed effects.
5.1 Main results system GMM
5.1.1 Complete period
Table 3 reports the main result of the empirical model (6) using the complete dataset. To demonstrate the effect of competition four different specifications are tested. Column (1) shows the basic model including bank-specific variables; column (2) displays the effect of the bank lending channel in which the interaction variables of size, liquidity and capitalization are added. Columns (3) and (5) report the effect of market power in the banking system by including the competition measures, but we limit the monetary interaction terms to those tied to ROA and NIM in order to analyse its sign. In columns (4) and (6) the monetary policy interactions for the three bank characteristics are added. Several findings are obtained.
Regarding significant variables in columns (1) and (2), last year’s loan growth has a positive influence on current loan growth, however only in the basic model. This justifies the use of a static model as presented in equation (10) further discussed in section 5.2. The
influence of a monetary policy shock (∆MPt-j) on bank lending has the expected negative
sign, although proves only to be significant in the current period. The finding that an increase in the federal funds rate leads to an immediate reduction of credit supply supports the existence of the bank lending channel. Cantero-Saiz et al. (2014) also find immediate responses of bank lending to monetary policy changes in their research on the bank lending channel in Eurozone countries over the period 1999-2013. Furthermore, in line with our expectations, the interaction term between liquidity and current monetary policy shock is positive and significant at the 10% level, indicating that banks with a higher liquidity ratio are better able to buffer against monetary policy shocks. This finding is confirmed by many previous studies, in particular complementing Kashyap and Stein (2000), who found liquid US commercial banks to absorb monetary shocks by drawing on cash reserves and selling securities. Nevertheless the interaction terms of size and capital with monetary policy are insignificant, which means that these bank-specific characteristics do not influence the way in
21 which bank credit reacts to monetary policy shocks. These results differ from Kishan and Opiela (2000) who obtain significant results for the interaction terms of monetary policy with size and capital in the US from 1980-1995. Beyond the effect on monetary policy, it appears that capitalization can directly affect credit growth which is not uncommon in the literature on the bank lending channel. Leroy (2014) explains how well capitalized banks can easily utilize lending opportunities without experiencing statutory constraints. The advantage of well capitalized banks is that it improves their creditworthiness, allowing them to efficiently collect resources. Size and liquidity do not directly affect credit growth.
Adding market power and its interaction variables to the equation allows us to interpret the effect of market competition on the bank lending channel. The results from columns (3)-(6) are quite similar to those in the models of columns (1) and (2), although the direct negative effect of monetary policy on loan growth has no significance in both the current and lagged period. Also the interaction terms of liquidity and monetary policy become insignificant in both panels. However, the interaction term of size and last year’s monetary policy shock is negative and significant in both panel A and B. This is a counter-intuitive result, as it indicates that smaller banks are better equipped against federal funds rate shocks. Traditional rationing implies that larger banks are better buffered against rising interest rates as they are less financed by deposits and equity (Kashyap & Stein, 1995). The results are not completely at odds with previous literature as similar finding are obtained by Jimborean (2009) who analysed EU banks over the period 1998-2006. A possible explanation could be that small banks tend to have relatively higher liquidity and capitalization ratios than large banks, in order to compensate for their difficulties in financing resulting from higher asymmetric information problems (Jimborean, 2009). This would mean that in times of monetary tightening, small banks could extend more credit without raising new capital because they face fewer problems in the context of binding risk-based capital (Van den Heuvel, 2002), and have more liquid reserves to draw from.
This paper argues that bank competitiveness can influence the monetary transmission mechanism. However the test results indicate no significance of the interaction terms of our competition indicators and monetary policy shock. Only a direct effect on loan growth is found as the coefficient of ROA is highly significant and positive, indicating that banks increase their credit supply when they gain market power. But when using NIM as competition indicator, the coefficient turns negative although only significant at the 10% level. These contradictory findings could be the rationalized by Carbó et al. (2009) who show that inferences about the degree of competition differ substantially using different indicators of bank competition. And indeed, the NIM looks only at interest returns while ROA includes non-interest income (off-balance sheet and fee income) and non-interest costs (operating costs). Therefore it may well be that competition in the traditional deposits and loans market rose (lowering NIM), whilst the expansion of US banks into new business areas and the
reduction of operating costs due to technological innovation have led to a rise in ROA10. In
10 In order to verify this assumption, the averages of NIM and ROA over time are estimated. It can be seen from
22 terms of the direct effect of market power on bank credit in the US, the results are inconclusive and depend on choice of measurement.
Lastly, the control variables are generally significant in the models. GDP growth is significant and positive in accordance with our expectations; economic growth leads to an increase in loans demanded by households and firms, which in turn leads to an increase in credit issued (Leroy, 2014). Inflation too influences the growth of loans positively. This is explained by Zhou (2016) as he rationalizes how inflation mitigates the liquidity risks of investment projects, which promotes economic growth in the United States.
5.1.2 Pre- and post-crisis period
Table 4 reports the results from the period before and after the 2008 financial crisis11. To conserve space the basic model is excluded from the results. Interestingly, the results show how in the pre-crisis period the only significant coefficients are those from the lagged dependent variable. Past values of loan growth are said to have a positive influence on current loan growth, which confirms the use of a dynamic model. Yet size, liquidity, capitalization and competition, together with their interactions, are said to have no significant effect on the growth of loans. Moreover, the control variable current GDP growth also has no significant impact on the dependent variable. The F-tests reported in the table shows the overall significance of the model, with a null hypothesis of every coefficient being simultaneously zero (Hill et al., 2011). It can be seen that the regression models in the pre-crisis period are still viable explanatory models at the 1% significance level, except for the model in column (1) that is significant at the 5% level. This implies that before the 2008 crisis, no support is found for the existence of a bank lending channel in the United States. Consequently, no effect of bank competition (and other bank characteristics) on the bank lending channel can be detected. A possible explanation could be that even though the cost of borrowing increased, banks had enough alternative sources of funding to replace deposits. This would confirm the reasoning of Romer et al. (1990), who emphasized banks’ ability to diversify funding sources. Indeed, the extensive use of securitisation made banks less susceptible to changes in monetary policy changes (Altunbas et al., 2009). Also findings for the US jumbo mortgages market suggest that securitisation could make the bank lending channel less effective pre-crisis (Loutskina and Strahan, 2006).
The US banking landscape seems to have changed when observing the results from the post-crisis period. The lagged dependent variable is again positive, but only significant in panel A. The current monetary policy variable is significant and negative as expected, as an increase (decrease) in interest rates leads to a decrease (increase) in the supply of credit. The rise in interest rate increases the cost of funding for banks, which ultimately leads to a decline in credit supply. In contrast to expectations, the lagged federal funds shock is positive and significant in all post-crisis models, meaning that last year’s federal funds rate increase leads to an increase in this year’s credit supply. A possible explanation might be that although banks face higher cost of funding, they are able to raise the lending rates over a year’s time in order to obtain higher profits. However this finding is very unusual, as no previous literature
11 Additional tests revealed the pair-wise correlation coefficient differs between variables in the pre- and
