1
The ability of monetary policy to influence domestic bank
lending to households and firms
Master Thesis – International Economics and Business 2017-18
University of Groningen – Faculty of Economics and BusinessAuthor: Ramon van de Wetering – S3260119 Contact: M.S.van.de.Wetering@student.rug.nl
Assessor: dr. A.C. Steiner Co-assessor: prof. dr. D.J. Bezemer
Date: June 19, 2018
ABSTRACT
Nowadays economies are often globally integrated and therefore the ability of domestic policymakers to influence domestic credit and bank lending flows is interesting to analyze. This paper investigates the ability of domestic monetary policy to control and influence these flows in the light of a globally integrated economy. The results suggest that domestic policymakers can still influence domestic credit and bank lending developments with the conventional tools at hand, although external factors –such as the FED funds rate, VIX, banking crises and financial integration– play a significant role as well. Key words: credit; bank lending; household and firm borrowing.
ACKNOWLEDGEMENTS
TABLE OF CONTENTS
1. INTRODUCTION 3
2. LITERATURE REVIEW 5
2.1 Domestic monetary policy in a global economy 6
2.1.1 Monetary policy transmission channels 7
2.1.2 Empirical research on monetary policy through time 8
2.1.3 The tools of monetary policy 10
2.2 Other driving factors of credit and bank lending 13
2.2.1 Internal demand factors 13
2.2.2 Internal supply factors 13
2.2.3 External supply factors 14
3. DATA AND METHODOLOGY 16
3.1 Sample specification 16
3.1.1 Data 16
3.1.2 Data trends 17
3.1.3 Sample 19
3.2 Methodology 19
3.3 Specification of the models 21
3.4 Limitations of the specification 23
4. EMPIRICAL RESULTS 24
4.1 Monetary policy tools 24
4.2 Other determinants of credit and bank lending 26
4.3 Robustness checks 30
4.3.1 Dynamic GMM estimation 30
4.3.2 High-income vs lower-income countries 32
4.3.3 The 2008 financial crisis 33
4.3.4 Excluding outliers 35
5. CONCLUSION 36
REFERENCES 38
APPENDICES 44
Appendix 1 – Measurement of the variables and collection process 44
Appendix 2 – Distribution of bank lending to households and firms 47
Appendix 3 – Descriptive statistics and correlations 49
Appendix 4 – Sample countries 51
Appendix 5 – Data collection process for exchange market intervention 51
Appendix 6 – Model specification tests 52
1. INTRODUCTION
Recent studies have shown an ongoing debate about the importance of different forms of credit that matter for real economic outcomes (Beck et al., 2012; Bezemer et al., 2017). Bank loans are an important source of finance for both households and firms, and empirical evidence has shown the benefits that come along with the use of this form of credit. However, excessive credit growth, partially created by banks, can be dangerous as we have recently experienced during the 2008/9 financial crisis. Analyzing how credit fluctuates, how it is controlled for, and to whom it flows may provide useful information for studying and estimating economic activities and monetary changes.
Among economists there is consensus that credit plays an important role in an economy and empirical research is increasingly investigating what the determinants of credit growth are (Arcand, Berkes & Panizza, 2015; Brown, Martinsson & Petersen, 2017). Empirical cross-country studies made use of aggregate credit measures and analyzed how they influence real economic factors. Arcand et al. (2015) found an inverted-U relationship between credit growth and GDP growth. He argues that there is a threshold level in terms of credit to the private sector as a share of GDP: below the threshold, more credit to firms increases economic growth because its stimulates investments, however, beyond the threshold additional credit is negatively related to growth. Brown et al. (2017) argue that well-developed credit markets foster growth for countries that rely on external finance for capital accumulation. They argue that credit markets play an important role in accompanying economic growth. Credit expansion also has a darker side in the sense that it increases macroeconomic volatility and possibly leads to financial crises and banking crises (Kaminsky & Reinhart, 1999; Schularick & Taylor, 2012; Elekdag & Han, 2015). Rey (2015) found that credit inflows increase the sensitivity to global capital flows, asset prices and credit growth in general, for which she refers to as the ‘global financial cycle’. As a result, such countries see their odds of experiencing asset price bubbles and excess credit creation increasing, and those two factors “are among the best predictors of financial crises” (Rey, 2015, p. 313). In addition, Arcand et al. (2015) found that a too large credit market increases economic volatility and the probability of economic crises. Furthermore, Borio and Disyatat (2011) found a positive feedback loop between more credit and risk spreads, creating a credit boom which will be followed by a bust, harming both the domestic and global economy.
This paper uses quantitative research methods to document (i) cross-country differences in domestic bank lending to households and firms, (ii) reveals how monetary policy tries to control this, and (iii) displays how internal demand, internal supply factors, and external supply factors determine the volume of bank lending. The results are based on a sample of 58 countries covering a time period from 1996-2014. First, I examine the relation between aggregate credit growth and the influence of monetary policy. After that, I specify my research towards aggregate bank lending and the ability of domestic monetary policy to influence this. Thereafter, I will further specify the determinants of domestic bank lending to households and firms and focus on how monetary policy tries to control this. Finally, the paper documents whether the relationship between domestic bank lending and its determinants are homogeneous between countries and over time. The paper is intended to answer the following research questions:
(1) Can domestic monetary policy control/influence domestic bank lending to households and firms?
(2) What determines the observed values of domestic bank lending to households and
firms?
Research on this topic is interesting because many governments use monetary policy frameworks (policy by the central bank to manage the money supply and interest rate to influence inflation, consumption, growth and liquidity) that either stimulate investments by providing credit in order to generate economic growth, or policies that aim to prevent the creation of asset bubbles and macroeconomic volatility that comes along with excessive credit growth (Docco et al., 2011). The usefulness of such policies depends on whether they incorporate tools that determine bank lending. Hence, one needs to know what factors determine bank lending (which refers to credit provided by the domestic banking sector) in order to form the right policies. However, because many economies nowadays are globally integrated, it might be the case that the determinants of bank lending are decided outside the scope of the domestic policies. Therefore, this paper analyzes the strength of domestic monetary policy in relation to bank lending.
cross-border bank lending flows in the last decade. However, domestic bank lending has not been analyzed greatly and therefore this paper adds insights into this area. The fourth contribution of this paper is that it decomposes credit growth into domestic bank lending, which makes this research more specific than those focusing on total credit. Lastly, this paper contributes to the lack of empirical evidence on economic theory about the relation between the cost of borrowing and the demand for it. The remainder of the paper is organized as follows. Section 2 describes what former research has found on the influence of monetary policy on credit and bank lending, and analyzes other determinants of these flows. Section 3 describes the data and methodology as well as the econometric model. The results of the analysis will be discussed in section 4. Finally, section 5 provides the conclusion of this research and describes what future research should examine.
2. LITERATURE REVIEW
Before analyzing how theory and empirical research explain the role of monetary policy in determining bank lending, I first define the main concepts. Credit can be defined as borrowed financial resources provided by financial institutions through loans, purchase of securities and trade credits, under an agreement or promise to repay at a later stage (World Bank, 2018; Business Dictionary, 2018). The main focus of this paper is on domestic bank lending, which refers to credit provided by the domestic banking sector and it is often defined as loans to non-banking agents which are used as a form of financing. Hence, bank lending can be defined as the action of borrowing financial resources from banks under an agreement to pay back later, which is in line with the definition of bank lending by Pham (2014) and the Business Dictionary (2018). Credit differs from bank lending in the sense that bank lending only refers to credit provided by (domestic) banks. Credit provided by domestic banks is only part of the total provision of domestic credit and therefore domestic bank lending is only part of total domestic credit in an economy (see section 3.1.2 for the difference in economic terms). Lastly, monetary policy needs to be defined for which this paper uses the definition of The Economic Times: “Monetary policy is the macroeconomic policy laid down by the central bank. It involves management of money supply and interest rate … used … to achieve macroeconomic objectives like inflation, consumption, growth and liquidity” (The Economic Times, 2018).
lending to households and firms is acknowledged by Beck et al. (2012) and Bezemer et al. (2017). According to Beck et al. (2012) –who examined the composition of bank lending and its effects on the real economy– it does matter to whom credit is distributed; firm credit is positively related to economic growth, but household credit is not. Furthermore, the author found that income inequality is reduced by an expansion of firm credit, but increases with additional household credit. Bezemer et al. (2017) found that bank credit has increasingly shifted towards households, and this ‘debt shift’ has consequences for growth, income inequality and macroeconomic stability. Deregulation in financial markets is found to be an important source for this debt shift. A disadvantage of large credit growth is that it may increase macroeconomic volatility (Kaminsky & Rienhart, 1999), and excessive credit growth goes hand in hand with a higher possibility of facing financial and banking crises (Elekdag & Han, 2015). Moreover, the volatility and probability of crises increases when credit markets become too large (Arcand et al., 2015). Rey (2015) argues that the sensitivity in global capital flows, asset prices and credit growth in general are amplified by credit growth. Eventually such countries face a higher probability of facing asset price bubbles and excessive credit creation. Those two factors are among the best predictors of financial crises.
Since bank lending flows both have advantageous and disadvantageous impacts on the real economy, the ability to control bank lending is an interesting topic to analyze. The responsibility of controlling these flows lies predominantly in the hands of central bankers who via monetary policy exercise influence on domestic credit developments. Yet, since many economies today are globally integrated, it might be the case that there are external fundamentals which are more important in determining the demand and supply for bank lending relative to fundamentals within the framework of domestic monetary policy. The ability of monetary policy to encourage, discourage and control bank lending is the primary focus of this paper.
2.1 Domestic monetary policy in a global economy
are provided with three conventional tools to achieve monetary policy objectives: (1) setting short-term interest rates, (2) managing domestic money supply, and (3) imposing reserve requirements.
Regarding the potential advantages and disadvantages that credit has on the real economy, these flows have to be closely monitored and controlled for. One of the objectives of monetary policy is to maintain a stable economy. Therefore, it is of great importance to implement monetary policy frameworks that either stimulate investments by providing additional credit which stimulates economic growth, or policies that aim to prevent the creation of asset bubbles due to excessive credit growth (Docco et al., 2011). The usefulness of such policy frameworks depends on whether they incorporate tools that have an impact on the amount of bank lending. Hence, one needs to know what factors determine bank lending before being able to form the right policies. However, because many economies today are globally integrated, it might be the case that the determinants of bank lending are decided outside the scope of the domestic conventional policy tools. Therefore, this paper analyzes the strength of domestic monetary policy in relation to bank lending.
2.1.1 Monetary policy transmission channels
This paper uses the work of Mishkin (1996) to explain how monetary policy is transmitted into the economy. I look at the scenario of expansionary monetary policy (situation in which a central bank uses its tools to stimulate the economy) throughout this discussion to compare different channels. First, the traditional interest rate channel of monetary policy transmission can be explained by the Keynesian IS-LM model. With expansionary policy we get:
! ↑ → %& ↓ → ( ↑ → ) ↑, (1)
expansionary monetary policy (M) causes a fall in the real interest rate (ir), lowering the costs of capital
which therefore stimulates investments (I), which in turn increases aggregate demand and output (Y). This channel operates for both businesses and households. Because monetary policy affects the real interest rate, it can still stimulate the economy when the nominal interest rate reaches zero. To see this, assume that the nominal interest rate is zero, with expansionary policy we get:
! ↑ → +, ↑ → -,↑ → %
& ↓ → ( ↑ → ) ↑, (2)
expansionary policy raises the expected price level (Pe) and therefore expected inflation (πe), which in
turns lowers the real interest rate (ir) even when the nominal rate is zero. Furthermore, it still stimulates
the economy as described above by stimulating spending and investment.
The second channel mentioned by Mishkin (1996) is the exchange rate channel. This channel includes the interest rate channel as well, and can be written as:
a fall in the real interest rate causes domestic currency to become less attractive relative to foreign currencies. Hence, the domestic currency depreciates (E), making domestic goods cheaper than foreign goods, thereby increasing net exports (NX) and output (Y).
Next are the so-called credit channels which arise due to information asymmetries in credit markets. There are two channels: (1) the bank lending channel and (2) the balance-sheet channel. The bank lending channel is based on the thought that banks are important because they (partly) solve information asymmetry. Expansionary monetary policy works as follows:
! ↑ → 1234 56789%:9 ↑ → 1234 ;8239 ↑ → ( ↑ → ) ↑, (4)
expansionary policy (M) increases bank reserves and deposits, thereby increasing bank loans causing investment (I) and consumption (Y) to increase as well.
Turning to the balance-sheet channel, Mishkin (1996) explains that a lower net worth of businesses causes adverse selection and moral hazard problems in lending to such firms. Borrowers have less collateral to offer which increases the risk for banks and therefore banks decrease lending, which in turn affects investment and consumption. The balance-sheet channel is also applicable to households. Using the balance-sheet channel to explain expansionary monetary policy gives:
! ↑ → +, ↑ → 25<6=96 96;6>:%83 ↓ → ?8=2; ℎ2A2=5 ↓ → 1234 ;635%3B ↑ → ( ↑ → ) ↑, (5)
the expected price level (Pe) rises and so does a firms’ net worth which in turn lowers adverse selection
and moral hazard problem, leading to higher investment (I) and aggregate demand (Y). Also, expansionary policy lowers the nominal interest rate (i) which improves the firms’ balance sheet as it raises cash flow, which reduces moral hazard and adverse selection problems. This can be written as: ! ↑ → % ↓ → >29ℎ C;8D ↑ → 25<6=96 96;6>:%83 ↓ → ?8=2; ℎ2A2=5 ↓ → 1234 ;635%3B ↑ → ( ↑
→ ) ↑ (6)
The credit rationing theory is related to the mechanism above, where borrowers are denied loans even though they are willing to pay higher interest rates. This holds true because those with the most risky investment projects are the ones willing to pay higher interest rates in the first place. Hence, higher interest rates increase adverse selection and lower rates reduce it. If monetary policy lowers the interest rate, the share of non-risky borrowers in total borrowers increases, fostering both investment and output. To conclude, Mishkin (1996) argues that the credit channel theories of monetary policy transmission gained importance at the expense of traditional interest rate channels.
2.1.2 Empirical research on monetary policy through time
controlling the short-term interest rate. However, in the post-crisis period, the ability of central banks to implement monetary policy via the interest rate eroded because the short-term interest rates reached the zero lower bound (ZLB) (Hormann & Schabert, 2015). In order to stimulate the economy, central banks kept the interest rates near the ZLB and simultaneously switched to other types of instruments like quantitative easing and increasing central bank reserves. The instrumental tool of monetary policy to change the level of international reserves is acknowledged by Steiner (2017), who found that central banks use this tool in order to pursue independent monetary policy. Intervening in the exchange market provides central bankers the opportunity to play around with the level of international reserves.
Nowadays many economies are globally integrated and fluctuations in the supply of credit and bank loans might therefore be the result of external shocks rather than domestic ones. This would imply that domestic monetary policy possibly loses its ability to control credit and bank lending using standard conventional tools like the interest rate, money supply and reserve requirements. The impact of external factors on monetary policy is examined by Devereux and Yetman (2014) who argue that monetary policy is possibly constrained by global shocks that drive down interest rates to the ZLB. However, not all countries can pursue their own independent domestic monetary policy strategies. For example, within monetary unions like EMU (Economic and Monetary Union of the European Union) individual countries do not have the ability to follow an independent strategy. The international macroeconomy is often linked to the impossible trinity framework, also known as ‘the trilemma’, of open capital accounts, fixed exchange rates, and monetary policy independence (Schoenmaker, 2011). This theory proposes that a country can only obtain two out of the three objectives in the trilemma. For example, with free capital flows it is only possible to have independent monetary policy by having floating exchange rates. Rey (2015) examined the effects of the global financial cycle on monetary policy independence. The author found that monetary policy of ‘the center’ –i.e. the U.S.– shapes the global financial cycle via leverage and the pro-cyclicality of cross-border credit flows. Rey (2015) argues in favor of a dilemma instead of a trilemma, where countries can choose either between monetary policy independence or free capital mobility. Hence, if countries want to liberalize their capital accounts, they have to give up monetary policy independence. On the other hand, Steiner (2017) argues that the trilemma can be relaxed by using an adequate monetary reserve policy. This allows central banks to have monetary policy independence and fixed exchange rate, even when there is free capital mobility. This is because foreign exchange interventions act as a substitute for capital controls. Yet, analyzing the existence of either a policy trilemma or dilemma is outside the scope of this paper.
growth in those countries. On the other hand, Pham (2014) reveals that for a sample of developed and emerging economies, both internal and external factors still play a significant role in determining lending behavior of banks. To sum up, empirical studies reveal ambiguous results about the influence of internal and external factors in determining credit growth and bank lending, and therefore an answer to the question whether domestic monetary policy is able to control credit and bank lending remains inconclusive. This paper aims to find a more conclusive answer to this question.
2.1.3 The tools of monetary policy
How monetary policy should operate in practice is under constant debate as there seems to be no simple framework that fits all (Rangarajan, 1997). But in general, central bankers have three conventional tools: (1) setting the short-term interest rate, (2) imposing reserve requirements, and (3) managing the domestic money supply. A more recent tool used among central banks is intervening in exchange markets by changing the level of international reserves (Steiner, 2017). Those tools will be discussed below. Hypotheses are provided after discussing each tool. The question whether domestic monetary policy is able to control credit growth and bank lending in a globally integrated economy, can be answered by the (in)significance of the coefficients for the interest rate, the money supply, reserve requirements and exchange market intervention, while simultaneously adding variables into the models that account for external factors. If the monetary policy tools are significant, then monetary policy has control over credit growth and bank lending in the domestic economy.
Setting the short-term interest rate
Furthermore, the author argues that nominal interest rates can only be brought down by keeping inflation low and stable, so that inflationary expectations and uncertainty disappear.
To conclude, it is expected that higher costs of borrowing from banks are associated with lower demand for bank loans. This leads to the following hypothesis:
Hypothesis 1: The demand for bank lending is negatively related to the short-term interest rate.
Imposing reserve requirements
Another common tool that central banks use is imposing reserve requirements. The invention of fractional reserve banking implies that commercial banks are only required to hold a certain amount of reserves as a fraction of their liabilities. The minimum amount of reserves that banks are required to hold –also known as reserve requirements– are indicated by the reserve ratio. For banks it becomes more difficult to provide additional loans when the reserve ratio goes up, because they need more reserves to back it up. Hence, reserve requirements and the supply of credit are inversely related to each other. Gersbach and Rochet (2017) empirically studied the relation between capital regulation and credit fluctuations and found that restrictions such as reserve requirements are effectively used as a tool to stabilize the credit cycle. Reserve requirements should go up when credit grows too fast, and vice versa. Furthermore, the authors argue that the reason for using counter-cyclical reserve requirements on banks is that the government aims to stabilize the credit cycle. Moreover, Gray (2011) shows that reserve requirements are used in order to have monetary control and liquidity management.
Both theoretical explanations and empirical research predict a negative relationship between reserve requirements and bank lending. This leads to the following hypothesis:
Hypothesis 2: The supply bank lending is negatively related to reserve requirements.
Managing the domestic money supply
Both theoretical explanations and empirical research predict a positive relationship between the money supply and bank lending. This leads to the following hypothesis:
Hypothesis 3: The supply of bank lending is positively related to the money supply.
Exchange market intervention
Hormann and Schabert (2015) stressed the importance of central banks using reserves as a new tool to pursue monetary policy, due to the fact that the interest rates reached the ZLB. One theoretical view on international reserves is that they are accumulated because of mercantilist motives –i.e. accumulating as much wealth as possible (Aizenman & Lee, 2007). Central bankers change the level of international reserves in order to keep an undervalued exchange rate which is beneficial for trade. Furthermore, Steiner (2017) shows that changes in international reserves are a way to relax the trilemma constraint. Hence, pursuing independent monetary policy is still possible with pegged exchange rates and free capital mobility. The author argues that capital controls and changes in reserves act as substitutes, which both allow central bankers to manage net capital inflows. Alberola, Erce and Serena (2016) examined a country’s resident investors and suggest that international reserves are accumulated to prevent and mitigate (at least partly) domestic capital flight, revealing precautionary motives to accumulate international reserves. Many central banks have accumulated international reserves in order to prevent excessive misalignment of the exchange rate and to build up buffers against sudden stops. The empirical analysis of Alberola et al. (2016) reveals that larger stocks of international reserves are linked to higher gross capital inflows but lower gross capital outflows, hence positive net capital inflows. This is because international reserves give residents additional incentives to invest savings domestically rather than abroad, and to extract capital invested abroad which in turn mitigates the lack of foreign finance. Following the argument of Pham (2014) who found that net capital inflows are a source of extending credit for banks, one would expect a positive relation between accumulating reserves and domestic loan creation. On the other hand, if the central bank accumulates reserves but wants to keep the domestic monetary base constant, it can offset the effects of foreign exchange intervention by simultaneously converting domestic bonds for foreign bonds – known as ‘sterilization’ (Steiner, 2014). According to Frankel and Okongwu (1996), successful sterilization programs reduce domestic credit, because central banks increase the market interest rates and reserve requirements.
To sum up, the impact of exchange market intervention on bank lending and credit creation might be ambiguous and therefore remains to be tested in this paper. Since both a negative and positive impact can be expected, this paper tests the following hypothesis:
2.2 Other driving factors of credit and bank lending
The volumes of credit and bank lending are determined by supply and demand factors, but supply and demand do not necessarily have to be in equilibrium. Dong et al. (2016) found that the supply and demand for credit are not always aligned which is reflected by the spread between the lending rate and deposit rate. The authors argue that both would be equal if demand and supply of credit were in equilibrium. Furthermore, Swain (2007) examined the ability of households to fully utilize their demand for credit and found that this is budget and time constrained. Moreover, information asymmetries in credit markets are found to be a severe problem for borrowers that lack collateral and good credit histories, and as a result some potential borrowers are excluded from the credit market (Bernanke & Gertler, 1995; De Janvry et al., 2010). However, although acknowledging their presence, this paper does not aim to assess the impacts of disequilibrium in the credit and bank lending market. In this paper, next to monetary policy tools discussed above, the determinants of supply and demand for credit and bank lending are categorized into: (1) internal demand factors, (2) internal supply factors, and (3) external supply factors.
2.2.1 Internal demand factors
A person’s income, and more specific GDP per capita might affect credit as it takes into account both the health of an economy and its demand for credit (Frankel & Romer, 1999). The demand for credit is typically found to be positively related to GDP per capita. Takats (2010) found that the underlying mechanism works through aggregate demand. Higher income corresponds to higher domestic demand for goods and services. In turn, firms respond to this by investing more, which increases their demand for credit. Moreover, GDP growth is also used as an indicator for economic growth and looking at the effect of economic growth on the demand for loans, Calza, Gartner and Sousa (2003) found a positive relationship between both variables. Sustainable economic growth allows economic agents to go more indebted in order to finance higher consumption and investment demands. However, because many studies have shown a positive relation between credit growth and economic growth (Bagehot, 1873; Schumpeter, 1911; Arcand et al., 2015) reverse causality cannot be excluded.
2.2.2 Internal supply factors
stock market and therefore the stock market can act as a substitute for credit, which implies a negative relation between the two. Demirgüç-Kunt and Maksimovic (1996) studied the relation between stock market development and financing choices of firms and found that initial improvements in a developing stock market result in a higher debt-equity ratio for firms. This implies that banks initially are able to increase their credit supply. However, in stock markets that are already developed, equity acts as a substitute for credit when the stock market further develops.
Schumpeter (1912) stressed the importance of the banking sector in stimulating economic growth by providing credit. The role of banks is to stimulate innovation and growth by detecting and financing productive investment opportunities. Furthermore, Schumpeter distinguishes between credit that finances development, called the primary wave of credit innovations, and credit that finances consumption, speculation, and excess investments, called the secondary wave of credit innovations (Bezemer, 2014). The author argues that the primary wave is associated with positive economic development, whereas the second wave is not. Levine and Zervos (1998) use a measure of bank credit as a share of GDP to account for the development of the banking sector. Following this argument, the amount credit to GDP tells something about the quality of the domestic banking sector itself.
Generally speaking, the performance of the banking system matters for how an economy performs, and on the individual level the performance of banks influences their lending behavior. Honohan (2000) examined banking system failures and found that these failures are a central aspect in many recent financial crises. Furthermore, Aisen and Franken (2010) found that well-performing banks which are able to make (large) profits provide positive and solid signals to the economy, which increases the supply of credit and bank loans because those banks have more resources to set out loans.
Lastly, competition in the banking sector reduces the market power of individual banks, which lowers their mark-up power for determining lending rates. Therefore, it is expected that when competition increases, lending rates will move towards a lower ‘equilibrium’ market price determined by the supply and demand for loans. This theoretical mechanism is supported by empirical evidence from Reed and Laosuthi (2012), who found that banking competition significantly lowers the monopoly power of banks. They also found that low access to finance and high prices for financial products can be the result of low competition in the banking sector. Finally, De Lis, Pagés and Saurina (2001) found that strong competition amplifies the supply of credit and bank loans.
2.2.3 External supply factors
constrained by interest rates abroad, rather credit supply is affected because the profitability of banks is affected by foreign interest rates, which in turn affects their resources to extend credit. On the other hand, Elekdag and Wu (2013) found that the impact of international interest rates on domestic credit growth is only moderate. Yet, Csonto and Ivaschenko (2013) and Tornell and Westerman (2002) argue that a lower Fed Funds rate implies that there is more liquidity, which has a positive influence on lending rates on aggregate and contributes to excessive credit growth. In general, the supply of credit and bank loans seem to increase due to a lower Fed Funds rate.
Rey (2015) analyzed the impacts of the global financial cycle on monetary policy independency, and reports that credit flows are quite volatile and typically move pro-cyclical. One of the key determinants of the global financial cycle is the U.S. monetary policy, for which Rey (2015) found that shocks in the Fed Funds rate are transmitted globally, affecting international credit flows and leverage of banks due to financial spillovers from the hegemony to the rest of the world. Furthermore, the global financial cycle can trigger excess credit growth during good periods, but excess contractions during bad times. In general, Rey’s (2015) most important findings applicable to this study are that: (i) credit flows are intertwined with the global cycle; (ii) credit supply increases if the VIX is low for a relatively long period; and (iii) an increase in the VIX lowers the supply of global domestic credit. The VIX –officially called the Cboe Volatility Index– is a measure for future volatility of prices in the S&P500 Index for the next 12 months (Rey, 2015). When credit is abundant and leverage is high, the spreads on assets are relatively low and so is perceived risk, translated into a lower VIX. A positive feedback cycle is found when there is loose monetary policy leading to a lower VIX, which in turn increases the credit supply and leverage, triggering a further fall in the VIX.
Dell’Ariccia et al. (2008) examined the effects of global banking crises in industrial countries using a difference-in-difference model. They found that global banking crises significantly hurt international bank lending flows, because banks in distress are forced to cut their loan issues. Furthermore, the consequences of not having access to loans due to banking crises struck harder in sectors that rely predominantly on bank credit, whereas sectors that have access to alternative sources of finance are significantly less hurt.
3. DATA AND METHODOLOGY
This section describes the econometric approach in order to analyze the proposed research questions. The works of Beck et al. (2012) and Bezemer et al. (2017) will roughly be followed, whom each in its own part elaborated on how the disaggregation of bank lending matters for the economy.
3.1 Sample specification 3.1.1 Data
This paper uses data from six main sources, namely the Credit Structure Database (CSD), Global Financial Development Database (GFDD), World Development Indicators (WDI), Federal Reserve Economic Data (FRED), Bank for International Settlement (BIS) and International Monetary Fund (IMF). The CSD database is developed by Léon (2018) and reports data on the structure of bank loans by borrower’s type separated into households and firms covering the period 1995-2014 for 143 countries. Only those countries that have a consistent data source are used in this dataset, which contributes to its reliability. The GFDD is comprised by economists from the World Bank and contains annual data from 1960-2015 of financial system characteristics of 203 countries. The WDI database is developed by the World Bank and combines data from multiple sources in order to form a complete database. This database presents the most recent and accurate available global development data. The database of the FRED is developed by the Federal Reserve Bank of St. Louis and consists all sorts of time series data on the international level. The database of BIS contains data about the global financial system, like financial stability, international money spillovers and global liquidity. Lastly, the IMF publishes a range of time series data on lending, exchange rates and other financial indicators. Appendix 1 provides an extensive description of all the variables, including their definitions, unit of measurement and data sources.
In order to construct a dataset that decomposes total bank lending towards households and firms, this paper uses the CSD which contains annual data on the structure of bank loans divided into household- and firm bank lending. Table B in Appendix 2 shows on average how bank lending is distributed among households and firms. The CSD’s credit data is highly correlated (close to one) to that of the GFDD and WDI which contributes to its reliability (Léon, 2018). Yet, a limitation of the CSD database is that it does not account for public borrowing from banks, instead it assumes that total bank lending is only divided among households and firms. The reason not to use other databases for decomposing credit is that either they do not focus on bank lending specifically (i.e. BIS) or they do not provide yearly data (i.e. Beck et al., 2012; Bezemer et al., 2017).
lending can be found in Ireland, whereas the lowest values are found in Indonesia. Furthermore, revealed by the minimums and maximums, there exists significant heterogeneity in the amount of credit and bank lending. The low minimum values of bank lending and bank lending to households and firms are observed in Belarus between 2001-2003. Those values are abnormal outliers when taking the whole sample into account. The highest values for bank lending, bank lending to households, and bank lending to firms are found in Ireland, Cyprus and Iceland respectively. For descriptive statistics of all variables I refer to Appendix 3. This paper accounts for outliers (see section 3.4).
3.1.2 Data trends
Between 1996 and 2014 the share of bank lending in terms of total credit varied from a minimum of 43% to a maximum of 66%, implying that bank lending accounts for about half of the total volume of credit in the sample.1 Figure 1A reveals how bank lending/GDP developed over time. From this graph
it becomes clear that the amount of bank lending has increased strongly between 1996-2009, with a small dip in the 2000s, and flattening from 2010 and onwards. This flattening is mainly due to that households and firms where more reluctant to go more indebted after the crisis, and because banks were either more hesitated to set out loans or because they were more restricted to do so (Ivashina & Scharfstein, 2010). The rapid expansion of bank lending/GDP over time supports the claim that it is interesting to analyze whether monetary policy can control such developments.
How disaggregated bank lending flows to households and firms developed for high- and lower-income countries is revealed in panels B and C of figure 1 respectively. Both upper graphs reveal that the share of bank lending to households has always been higher in high-income countries. Yet, where households in high-income countries gained a larger share over the years, this pattern is much less strong with respect to lower-income countries. The two graphs at the bottom reveal that the shares of household and firm bank lending to GDP are far more similar in high-income countries relative to lower-income countries. Furthermore, the amount of bank lending to GDP diverged away in high- income countries from 2000 and onwards suggesting the creation of an asset bubble. Hereafter the
1 Based on the countries that contain data on both total credit and bank lending. Table 1: Descriptive statistics
Variable Observations Mean Min Max Std. dev.
Total credit/GDP 570 175.22 14.85 422.48 80.92
Bank lending/GDP 952 63.42 0.0008 312.12 50.00
Share of bank lending to households 952 26.70 0.0001 124.20 26.14
amount of bank lending as a share of GDP both for households and firms are significantly higher in high-income countries relative to lower-income countries.
Figure 1: Development of bank lending between 1996-2014. Source: Credit Structure Database
(A) Total bank lending – averages of all sample countries 2
(B) Bank lending to households vs firms (HI) 3 (C) Bank lending to households vs firms (LI) 4
2 All sample countries are listed in table E in Appendix 4.
3 High-income countries: Australia, Austria, Belgium, Canada, Chile, Croatia, Cyprus, Finland, France, Germany,
Greece, Hong Kong, Iceland, Ireland, Israel, Italy, Japan, Korea, Latvia, Lithuania, Malta, Netherlands, Singapore, Slovak Republic, Slovenia, Spain, Switzerland, United Kingdom, United States, Uruguay. Total: 30 countries.
4 Lower-income countries: Albania, Argentina, Armenia, Azerbaijan, Belarus, Botswana, Colombia, Egypt, Georgia,
India, Indonesia, Kenya, Kyrgyz Republic, Macedonia, Malaysia, Mexico, Moldova, Mongolia, Morocco, Pakistan, Peru, Philippines, Romania, Russian Federation, Sri Lanka, Thailand, Tukey, Ukraine. Total: 28 countries.
45% 55% 65% 75% 85% 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Year Bank lending (% GDP) 40% 45% 50% 55% 60% 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Sh are in to ta l b an k le nd in g Year
Share of Households Share of Firms
10 20 30 40 50 60 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 % G DP Year
Share of Households Share of Firms
20% 30% 40% 50% 60% 70% 80% 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Sh are in to ta l b an k le nd in g Year
Share of Households Share of Firms
0 5 10 15 20 25 30 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 % G DP Year
3.1.3 Sample
The countries included in the sample are shown in the table of Appendix 4. The total dataset contains 58 countries of which 31 countries are European, 8 are Northern & Southern American, 5 are African and Middle Eastern and 14 are Asian & Pacific. The sample is biased towards European countries since they form the lion’s share of the sample countries, which is due to their rich data availability relative to other countries. Although this dataset is unbalanced, it has a significant overlap between 1996-2014, and therefore the final sample is a panel dataset with time series data covering this period.
3.2 Methodology
Equations 7-10 form the benchmark models in this paper. First to be analyzed are the impacts of monetary policy tools on total credit growth (equation 7). Thereafter the analysis is specified to the impacts of monetary policy on aggregate bank lending (equation 8). Finally, this paper examines the decomposition of bank lending to households and firms (equations 9 and 10).
EFGH = JG+ LMN(OGH + LP!NGH+ LQORGH+ LS.!(GH+ LT0GH+ UH+ VGH (7)
Z[GH= JG+ LMN(OGH + LP!NGH+ LQORGH+ LS.!(GH+ LT0GH+ UH+ VGH (8)
Z[]]GH = JG+ LMN(OGH + LP!NGH+ LQORGH+ LS.!(GH + LT0GH+ UH+ VGH (9)
Z[_GH= JG+ LMN(OGH + LP!NGH+ LQORGH+ LS.!(GH+ LT0GH+ UH+ VGH (10)
TC in equation 7 refers to total credit/GDP, BL in equation 8 stands for bank lending/GDP, BLHH in equation 9 indicates the part of bank lending towards households, and BLF in equation 10 denotes the part of bank lending that flows towards firms. JG is the unknown intercept for each country
implying that there is exists heterogeneity across individual countries, UH depicts period dummies in order to account for time fixed effects, VGH is the error term, and ‘i’ denotes the country, ‘t’ is time.
measured by a bank’s regulatory capital to risk-weighted assets, which indicates the ratio of total regulatory capital to its assets held, weighted according to the risks of those assets. Data is obtained from the IMF database. The final monetary policy tool is exchange market intervention for which data is gathered from IMF. It is measured as the percentage change in reserve assets from year to year.
Xit refers to the other determinants of bank lending based on internal demand factors, internal
supply factors and external supply factors. The first internal demand factor that will be analyzed is income growth (IG). Income growth is measured by the annual Gross National Income (GNI) growth rate. GNI is also known as Gross National Product (GNP) and captures the sum of value added by all resident produces, plus product taxes, plus net earnings of primary income from abroad. According to Pham (2014), this variable validly captures a country’s income growth. Data is obtained from WDI.
The internal supply factors consist of the efficiency of financial markets (EFM), performance of banks (PB), health of banks (HB) and competition in the banking sector (COMP). The efficiency of financial markets can be measured by the cost-to-income ratio, which is defined as operating expenditures of banks as a share of the sum of net-interest revenue and other operating incomes (Beck et al., 2000). Data for the efficiency and size of financial markets is obtained from GFDD. Following Pham’s (2014) arguments, the performance of banks can be measured by their return on equity (ROE). In the GFDD database, which is the source for this data in this paper, ROE indicates a bank’s after-tax income to its yearly averaged equity stock. The health of the banking sector can be measured by the Bank z-score, which captures the probability of default of a country’s banking system, where individual bank risks are accounted for and weighted by their total assets (Beck et al., 2000). The data is obtained from GFDD. Lastly, the competition in the banking sector can be measured by several indicators, like bank concentration ratio and the Boone indicator. This paper uses the bank concentration ratio because it is a proxy for (the lack of) competition within the banking sector (Boyd & De Nicoló, 2005). Data for this variable is obtained from GFDD.
bank rescue was at least 2 percent of GDP; or non-performing loans reached at least 10 percent of bank assets” (Dell’Ariccia et al., 2008, p. 6). The data for banking crises are obtained from GFDD. A limitation of this dataset is that it only includes data until 2011, hence I assume that there were no banking crises between 2012-2015. The final variable that is added to the model is financial integration. Being financially integrated is a situation in which financial markets are globally linked together (Pham, 2014). This paper uses the Chinn-Ito index –also known as the KAOPEN index– which is a well-known measure of financial integration that measures a country’s degree of capital account openness (Chinn & Ito, 2006).
3.3 Specification of the models
This paper performs regression on panel data to discover the relationship between monetary policy and bank lending. Several tests are conducted to decide whether a pooled model, fixed effects model or random effects model should be used. The pooled regression model assumes that there are no cross-country differences, and that all countries interact with the variables in the same way (Carter Hill et al., 2012). Furthermore, it assumes that the error term has a zero mean and zero variance, and that the errors are uncorrelated over time, across individuals and across the independent variables. However, the sample in this study most likely has cross-country differences in its data and therefore a pooled OLS regression seems inappropriate. Whether a pooled OLS regression is appropriate can be tested with the Breusch-Pagan Lagrange Multiplier test, which tests for the presence of random effects (Torres-Reyna, 2007). The null hypothesis states that the variances of all country-specific components of the error term are zero, implying that a pooled OLS regression can be used. The alternative hypothesis states that the variances are greater than zero and therefore argues in the favor of random effects. The results of this test reject the null hypothesis, and therefore random effects should be used (see table F in Appendix 6).
Both random effects and fixed effects are consistent under the condition that there is no correlation between the error term and the explanatory variables. If there is correlation, the random effects model becomes inconsistent but fixed effects remain consistent. Hence, in this case the Hausman test assumes that there are differences in the coefficients for the random and fixed effects model. If the test rejects the null hypothesis, fixed effects should be used. If the test fails to reject the null hypothesis, random effects should be used because this leads to more precise estimates of the coefficients (Carter Hill et al., 2012). The results of these tests reveal that fixed effects should be used (see table G in Appendix 6). Another test proposed by Torres-Reyna (2007) is to check for time fixed effects when using fixed effects models. Time fixed effects control for time trends which would otherwise be seized by the residuals. The results of this test show that time fixed effects are needed (see table H in Appendix 6). The Shapiro-Wilk test is used to test whether the residuals are normally distributed. The null hypothesis of this test argues that residuals are normally distributed, while the alternative hypothesis states that residuals are not-normally distributed (Hanusz et al., 2014). The results of this test reject the null hypothesis and therefore the residuals are not normally distributed (see table I in Appendix 6). Furthermore, figures I-VIII in Appendix 6 graphically reveal the distributions of the main variables in this paper. The figures reveal that the variables are positively skewed, implying that the mass of the distribution is concentrated in the left side of the figure with a long tail to the right. On the other hand, the rule of thumb –known as the central limit theorem– states that if the sample size is sufficiently large, then the least squares estimators have a distribution that approximates the normal distribution (Carter Hill et al., 2012). There exists no consensus on the question of what sufficiently large is, but sample sizes of 30 and 50 are mentioned by the authors. The sample size in this paper is larger and based on this rule of thumb I assume that the non-normal distributions do not cause too big of a bias.
One of the assumption of a multiple regression model is that that the variances are constant, i.e. the errors are homoscedastic. A Modified Wald test5 is used in order to test this assumption, where
the null hypothesis states that the errors are homoscedastic, and the alternative hypothesis argues that the errors are heteroscedastic which violates the assumption. The results show that heteroscedasticity is present (see table J in Appendix 6). A Wooldridge test is used to test for serial correlation, where the null hypothesis states that there is no serial correlation (Wooldridge, 2010). Standard errors and variances are estimated to be smaller than they really are when serial correlation is present, and therefore influences the efficiency of the estimates. The results indicate that serial correlation is present in all models (see table K in Appendix 6). Clustered standard errors are used in order to account for both heteroscedasticity and serial correlation.
Lastly, I checked whether the data are stationary. The problem of unit roots is that it leads to spurious regressions implying that estimates and seemingly significant findings are unreliable (Carter
Hill et al. 2012). In order to test for this, I used a Fisher-type stationarity test known as the Augmented Dickey-Fuller test. The results of this test are shown in table L of Appendix 6 and indicate that all variables are stationary.
3.4 Limitations of the specification
The approach is subject to criticism since panel regressions assume that there are homogeneous effects, implying that across countries the dependent variables in this research react in the same way to changes in the independent variables. As mentioned this can be partly dealt with by using fixed effects (Carter Hill et al., 2012). However, by adding time fixed effects in long-run regression estimations may produce multicollinearity implying that explanatory variables are highly correlated. A pairwise correlation matrix is used to check for multicollinearity (see Appendix 3). The highest significant correlation can be found between financial integration and the short-term interest rate that equals 0.35 which makes sense because high interest rates attract foreign capital. Furthermore, a highly significant correlation of -0.32 is found between financial integration and the reserve requirements, which might be due to higher reserve requirements make it harder to financially integrate. The other explanatory variables are not that highly correlated to one another. Overall, all correlation coefficients remain well below the 0.90, 0.70 and 0.70 threshold levels proposed by Green et al. (1988), Lehman et al. (1998), and Dormann et al. (2013) respectively, and therefore multicollinearity is not present.
A limitation of using panel regressions of the nature described in this paper is that they cannot make strong causal statements due to endogeneity. Endogeneity refers to the problem that causality might run in both ways, from independent variable to dependent variable and vice versa. It is also referred to as that the explanatory variables and the error term are correlated, implying that the explanatory variables are endogenous. This endogeneity may cause biased coefficients. One possibility to encounter the problem of endogeneity is to use an instrumental variable (IV), which is a variable that is exogenous and correlated to the explanatory variables but not to the dependent variable, at least not directly. However, IV’s can be very difficult to find and in fact may not exist at all. Therefore, this paper uses two other methods to account for endogeneity. The first method is to use lagged forms of independent variables. It is unlikely that the impacts of income growth, return on equity, Fed funds rate, financial integration, the short-term interest rate, reserve requirements and money supply can be excluded from reverse causality. To account for this endogeneity problem, these variables are lagged by 1 period. The second method is to use a dynamic Generalized Method of Moments (GMM) model which will be further discussed in section 4.3.1.
these are not presented in this paper). Outliers are included in the baseline regressions, but a robustness check is performed in order to account for the bias caused by outliers.6
4. EMPIRICAL RESULTS
This section describes the results of the analyzes and is set up in 3 subsections. First, the impact of monetary policy tools on credit and bank lending flows are reported in table 2. Thereafter, in table ,3 I focus on the results of other determinants of credit and bank lending, and on what determines its decomposition to households and firms. In section 4.3 some robustness checks are performed.
4.1 Monetary policy tools
Table 2 reports the results of the regression analyzes performed on total credit/GDP, bank lending/GDP, and on the decomposition of bank lending regressed by the tools for monetary policy. Two regressions are performed for each dependent variable, where one does not encounter the problem of endogeneity (denoted by 1), whereas the other one does (denoted by 2). The overall significance of the models is determined by the F-test, for which the results indicate that each model is highly significant.
The first two columns reveal how total credit/GDP is affected. When the analysis is not controlled for endogeneity, I find a significant impact of exchange market intervention at the 1% level. Once controlled for endogeneity I find that exchange market intervention is significant but only at the 10% level. Its negative sign indicates that a 1 unit increase of exchange market intervention decreases total credit/GDP by approximately 0.03% and 0.05% respectively. This finding is in line with the process of sterilization by which the central bank keeps a constant monetary base. The remaining variables are insignificant. Although the results for exchange market intervention are statistically significant, in economic terms they are rather small.
The 3rd and 4th column of regressions results reports the impact of monetary policy on bank
lending flows. The results do not differ a lot whether or not controlled for endogeneity by using lags, except that the coefficient of reserve requirements becomes more significant. I find significant coefficients for reserve requirements and money supply. Furthermore, these findings all confirm their hypotheses. A 1 unit increase of reserve requirements decreases bank lending/GDP by 0.54% and 0.83% respectively. These findings are significant at the 10% and 1% level and confirm the theory proposed by Gersbach and Rochet (2017) who argue that imposing requirements is an effective tool to stabilize the credit cycle. Moreover, a one unit increase in the money supply increases bank
6 Extreme observations of bank credit to GDP, bank lending to households, bank lending to firms, short-term interest
lending/GDP by approximately 0.0004% and 0.0003% respectively. This finding confirms the theory of Bernanke and Blinder (1988) who argue that domestic money supply and a bank’s loan creation are positively related to each other. Although this finding is statistically significant at the 5% level, in economic terms its impact is really small. The remaining tools do not have a significant impact on bank lending/GDP.
Turning to the regressions in columns 5 and 6 where the decomposition of the share bank lending to households is regressed on monetary policy tools, I find significant impacts at the 5% and 10% level of the money supply only. The money supply has a positive impact on the share of bank lending to households which is in line with the hypothesis and confirms the theory discussed in the literature review. However, in economic terms its influence is rather small since a 1 unit increase in the supply of money only increases the share of bank lending to households by approximately 0.0002% and 0.0001% respectively. The remaining tools are insignificant.
The last two columns report the impact of monetary policy tools on the share of bank lending to firms. In the case of not controlling for endogeneity, reserve requirements and money supply have a significant impact at the 5% level. The signs of these coefficients are in line with their hypotheses. A 1 unit increase of reserve requirements decreases the share of bank lending to firms by 0.45%, which is in line with the theory proposed by Gersbach and Rochet’s (2017). Furthermore, a 1 unit increase in the money supply increases the share of bank lending to firms by approximately 0.003% which in economic terms is rather small. Regarding the regression controlled for endogeneity, I find that reserve requirements are statistically significant at the 1% level. This finding is in line with the hypothesis and a 1 unit increase of reserve requirements decreases the share of bank lending to firms by approximately 0.64%. The coefficient of money supply becomes insignificant after controlling for endogeneity. The short-term interest rate and exchange market intervention are insignificant in both models.
4.2 Other determinants of credit and bank lending
There are more factors than just monetary policy tools which determine the amount and allocation of credit and bank lending. This paper controls for additional factors and the results of these regressions are provided in table 3, where each model is highly significant. Each regression reports the impact of monetary policy tools, internal demand & supply factors, and external supply factors on total credit/GDP, bank lending/GDP, and bank lending to households and firms. The regressions account for endogeneity by lagging some explanatory variables by 1 period, as specified in section 3.4.
The impact of the short-term interest rate is only significant (10% level) in the total credit/GDP model and it has a positive sign which contradicts the hypothesis. A possible explanation for this is that a majority of countries already had interest rates close to the ZLB, and therefore a minor increase in the interest rate might still have a positive impact on the demand for credit. I find insignificant coefficients of the short-term interest rate in the remaining models.
The impact of reserve requirements is significant in the model for bank lending/GDP and also for the share of bank lending to firms, both at the 1% level. The findings are in line with their hypotheses stating that reserve requirements and bank lending are inversely related to each other.
Turning to the impact of money supply, I find positive significant coefficients in the models for bank lending/GDP and both for the share of bank lending to households and firms models. These findings are significant at the 5%, 1% and 10% level respectively, and are in line with their hypotheses.
Exchange market intervention forms the last tool of monetary policy for which I do not find any statistical significant effect in neither model.
Controlling for additional variables allows me to check whether or not the findings in table 2 for monetary policy tools are robust. To do so I compare the findings in table 3 with those earlier discussed in table 2. There are no large differences with respect to significance and signs of each coefficient and therefore the findings are quite robust after controlling for other variables. However, one change is that exchange market intervention becomes a statistical insignificant tool to influence total credit/GDP when comparing the results of table 2 and 3. On the other hand, the short-term interest rate is statistically significant in table 3 but not in table 2. With respect to the share of bank lending to household I find that the money supply becomes statistically significant in the results of table 3 but they were not in table 2.
29
Furthermore, the results confirm the substitution effect of other sources of finance instead of credit when financial markets become more efficient, which is revealed by the significant coefficients of the efficiency of financial markets in models 2, 3 and 4 (significant at the 1% and 5% level).
Regarding the impact of a bank’s performance measured by its return on equity, I only find a significant positive correlation at the 5% level to total credit/GDP which counters the hypothesis. When holding equity becomes more profitable for banks, they might use their revenues for investing in additional equity instead of expanding their loan volumes.
With respect to a bank’s health, I find evidence that confirms the hypothesis. The coefficients of a bank’s health are significantly negative in the total credit/GDP and the share of bank lending to firms’ regressions (both at the 5% level). The health of banks is measured by the bank Z-score, which measures the probability of default of a country’s commercial banking system. An increasing Z-score indicates that the score is above the mean, implying that the average probability of default increases. Hence, when banks become less healthy (i.e. the Z-score goes up) their provision of loans and credit declines.
The coefficients of banking competition are insignificant in all models and therefore I do not find evidence that either confirms nor denies the hypothesis.
The external supply factors are highly significant in each model and all signs confirm their hypothesis except for banking crises. The coefficients of banking crises are positive which is evidence that counters its hypothesis. Taking a closer look at how banking crises are measured in the GFDD makes this finding more sensible. Banking crises in this database are measured in such a way that during a crisis it is assumed that credit growth is negative. Furthermore, an intuitive consequence of a crisis in the banking sector is a fall in GDP. Hence, whenever GDP declines more strongly than credit does, this implies that the ratio of total credit/GDP and bank lending/GDP increases. Compared to the other factors, the coefficients of external supply factors are large and therefore in economic terms their influence is huge (example: a 1 unit increase of the Fed Funds rate decreases total credit/GDP by approximately 5.84%).
that for each regression model, at least one tool has a significant impact on credit and bank lending flows. Hence, even though the credit cycle is subject to external shocks, there is still room for domestic monetary policy to maintain a stable economy by controlling the credit cycle via the tools provided within the framework of monetary policy.
In addition, regressions with standardized coefficients are used because they allow to discover the explanatory variables with the greatest impact on the dependent variables (Carter Hill et al., 2012).7
The results are reported in table M of Appendix 7. Coefficients can be interpreted as the effect of a one standard deviation change in the explanatory variable in terms of standard deviations of the dependent variable. In all models I find that the significant standardized coefficients of external factors are larger than the those of the monetary policy tools. Hence, the largest impact comes from external factors rather than from those factors that are controlled by monetary policy. The factors with statistical significant impacts from large to small are banking crises, the VIX, financial integration, Fed Funds rate, reserve requirements, and money supply respectively. All findings are in line with their hypotheses.
4.3 Robustness checks
This section checks the robustness of the main findings on how monetary policy tools affect total credit/GDP, bank lending/GDP, and the shares of bank lending to households and firms. First a dynamic panel Generalized Method of Moments (GMM) method is used to control for endogeneity concerns (Arellano & Bover, 1995). Secondly, I checked whether the results differ per country groups by splitting the sample into high-income and lower-income countries. Finally, the model is re-estimated by distinguishing between the pre- and post-crisis period in order to test whether the results are affected by the 2008 financial crisis.
4.3.1 Dynamic GMM estimation
The benchmark models in this paper are re-estimated using a dynamic GMM approach. This method was first introduced by Arellano and Bond (1991), and later on developed by Arellano and Bover (1995) and Blundell and Bond (1998). Using these methods has increasingly become popular according to Roodman (2009) who argues that these estimators are designed for producing consistent parameters in situations with relative small time periods and many individuals using a time fixed approach and data that consists serial correlation and heteroscedasticity. Furthermore, the explanatory variables in these models are not completely exogenous, implying that they are correlated with past values and the errors term. GMM accounts for this by using lags of the dependent variable and first differences of the independent variables as instruments. Using a dynamic GMM method therefore
allows to account for endogeneity. Including a lagged feature of the dependent variable as an explanatory variable makes sense because the current volume of credit and bank lending is (at least partly) based on its amount one year before. The Sargan test is used to test whether the results validly overidentify the restrictions, which is another way of saying that the IV is valid (Arellano & Bond, 1991). A rejection of the null hypothesis that restrictions are validly overidentified implies that the model has to be reconsidered. The Sargan test in each model does not lead to a rejection of the null hypothesis. Moreover, Arellano and Bond (1991) argue that the models should be tested for serial correlation of the second-order. The null hypothesis of this test indicates that there is no serial correlation. The p-values of these tests are reported in the table and reveal that in each models the null hypothesis cannot be rejected for AR(2). Estimates in table 4 come from the two-step approach which uses the residuals based on the estimates from the one-step approach. Lastly, Roodman (2009) argues that using too many instruments can cause problems for the endogenous variables and possibly leads to a loss of statistical power. The author uses a rule of thumb which states that the number of instruments should be kept lower than the number of observations. In this case, the number of instruments remained well below the number of observations.
The dynamic GMM results are reported in table 4. The Wald test reveals that each model is highly significant. In most cases the monetary policy tools are significantly correlated with the dependent variables. The results in the first row show that initial levels of all dependent variables are highly significant, indicating that historical values affect present values. Furthermore, the findings conflict with those in table 2 where fewer coefficients where found to be significantly correlated. Furthermore, the coefficients of the short-term interest rate are significant but its sign varies in each model. Compared to table 2, using GMM allows to find supportive results for its hypothesis in the case of model 1 and 3, but not in model 2 and 4. With respect to reserve requirements the findings in table 4 are in line with those in table 2, hence those are robust. The same accounts for the money supply except that its significant impact is not found in model 3 of table 4. Lastly, looking at the coefficient of exchange market intervention, table 4 indicates that it is significant in models 1, 2 and 3. Compared to the results in table 2, exchange market intervention was only significant in model 1. Hence, the findings of this variable are not robust. All in all, using a dynamic GMM approach to estimate the models allows me to confirm that only the findings for reserve requirements in table 2 are robust.
4.3.2 High-income vs lower-income countries
