University of Amsterdam
Msc Finance
Master Thesis Banking and Regulation
Bank Risk and Competition: An international investigation
based on interest rate environment
Supervisor: Stefan Arping
RUOWEN ZHOU
11281774
Statement of Originality
This document is written by Student RUOWEN ZHOU who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
Acknowledgements
First of all, I would like to extend my special thanks to my supervisor, Prof. Stefan Arping, and my wife, Dr.Wenjun Yu. When I was confused about my research proposal, Prof. Arping provided me with suggestions and relevant literatures which helped to determine my research topic. Due to family reasons, I cannot completely devote myself to study. The whole process of Master study is definitely not easy for me. My wife always encouraged me even though I was pessimistic sometimes. Without your help and encouragement, I can never stick to it and finish this Master thesis.
At the same time, I would also like to thank Dr. Zipeng Zhang, who helped me with Excel commands so that I could improve my efficiency in Data mining and processing. In addition, thank my entire fellow members in thesis seminar finance, they provided me with lots of ideas and feedbacks. Thank all the other people who helped me during the process of study.
Table of Content
Abstract ... 1 Introduction： ... 2 Literature review: ... 4 Methodology ... 7 Data ... 12 Empirical Results ... 27 Robustness Check ... 31 Conclusion ... 34 Reference ... 361
Abstract
There are basically two views on the relationship between bank competition and bank risk. One view stands for the competition-fragility relationship, which states that more bank competition leads to more bank risks. The other view stands for the
competition-stability relationship, which is that more bank competition leads to less bank risks. This paper studies the hot-debated topic on the relationship between bank competition and bank risk, but taking a new perspective, which is to analyze this old topic by taking the effects of interest rate environment into consideration in an international environment. The purpose of the study is to test whether there is a general relationship between competition and overall risks for most large banks in the world, and whether such relationship is affected by the level of interest rate
environment.
This study uses Z score to measure bank’s overall default risk, Lerner Index to measure bank’s competition. Global banks with asset value larger than 50 billion US dollars are selected in the sample sets, this in total account for 464 banks in the world. The research period is from 2012-2016.
The finding suggests that, in high income countries the relationship between bank competition and bank risk is affected by the level of interest rate environment. When interest rate environment reaches a high level, or when it reaches a certain threshold, more competition leads to less risk. When the interest rate environment is below the threshold, then more competition leads to more risks. Asset value has significant effects on bank risk. Banks with greater asset value tend to be more risky. In addition, banks from countries with high government debt ratio (% of GDP) are less risky.
For banks from relatively low income countries, interest rate environment does not have significant effects on the relationship between competition and risk. More competition leads to more risk. Therefore, in low income countries, the findings only support the competition-fragility view.
2
Introduction：
The theory of bank competition and bank risk attracts much scholarly attention in recent years. One primary view on this topic holds that bank competition results in more bank risk as banks usually choose more risky portfolios in the face of fierce competition (Keeley, 1990; Besanko and Thakor, 1993). However, this view is challenged by a contrary argument that more bank competition reduces bank risk. Boyd and De Nicoló (2005) argue that banks become more risky when their markets are more concentrated as banks can use their increasing market power to charge higher interest rates to their loan customers and make them optimally choose higher risk projects. This view notices that banks to some extent do have the ability to charge different lending interest rates but ignores that such interest rates are also constrained by monetary policy and interest rate environment.
The two strands of literature both fail to emphasize the impact of interest rate environment on the relation between bank competition and risk. Interest rate policy is one of the primary instruments of monetary policy. Under easy-monetary policy, central bank reduces interest rate directly, which strengthens banks’ lending capacity. On the contrary, contractionary monetary policy leads to lower banks’ lending
capacity. In addition, the interest rate over the whole maturity spectrum, the exchange rate and many other assets prices could also be affected when central bank sets the short term interest rates. Therefore, interest rate environment not only constraints bank activity exerting impact on the structure of asset portfolio, but also affects the evaluation of bank risk. In this sense, the interest rate environment should be taken into account when looking at the relation between bank competition and bank risk. In addition, most previous studies on bank competition and risk exclusively focus on a regional dimension; few studies analyze this topic in an international environment.
In view of that no systematic study has been taken to analyze the impact of interest rate environment on the relation between bank competition and risk in an international environment, this thesis explores the question: how to understand the
3
relation between bank competition and risk for most global large banks by looking at the interest rate environment. It primarily uses Lerner Index to measure bank
competition; Z- score to test bank risk; and yearly real interest rate to evaluate interest rate environment. The thesis studies the impact of interest rate environment on the relation between bank competition and risk by adding the interaction term of bank competition and interest rate environment in the regression analysis. It consists of six sections. Section 1 generally introduces the focus of this thesis and primary research question. Section 2 states the existing literature review on the topic of bank
competition and risk, so as to explain the contribution of this research to the existing debates on bank competition and risk. Section 3 explains the relevant data and empirical model; section 4 explains the main results and its economic meaning. Section 5 gives a robustness check. The main concluding results, discussion and deficiency of this paper are given in the section 6.
4
Literature review
There is one basic idea in bank competition and risk that more bank competition encourages banks to conduct more risk-taking activities, which again leads to more overall bank risk and the instability of financial industry. For instance, Bushman, Hendricks and Williams (2015) argue that greater competition increases both individual bank risk and a bank’s contribution to system-wide risk. Keeley (1990) holds that more bank competition causes bank to take on more risk in the view of the increases in asset risk and reductions in capital. Similarly, by analyzing the relation between franchise value and risk taking from 1986 to 1994, Demsetz, Saidenberg and Strahan (1996) argue that banks with high franchise value take on less portfolio risk. A bank with a high franchise value can indicate the bank has strong market power, thus a low competitive environment. Jimenez and his fellows (2007) also find the consistent result that loan market power and bank risk have a negative relationship by conducting the empirical research of Spanish banking system.
On the other hand, there is another view challenges the above idea by contending that increasing bank competition results in less bank risk. An empirical substantiation of this view comes from the study of Schaeck, Wolfe and Cihak. They make the first cross-country investigation of the impact of competitive bank conduct on banking fragility and conclude that more competitive banking system both reduces the probability of systemic banking failure and increases the survival time of banking system to crisis (2006). This “competition-stability” view is also supported by Boyd and De Nicolo (2005), who assume that there is a positive correlation between bank concentration and the risk of bank failure by considering that loan customers find it harder to repay loans when banks charge higher interest rates. For the same token, a reduction in loan rates caused by more bank competition reduces the loans’
probability of default that coincides with bank’s probability of failure. But Boyd and De Nicolo’s argument also encounter some challenges. Martinez-Miera and Repullo (2010) point out that Boyd and De Nicolo’s finding does not necessarily obtain in view of the imperfect correlation of loan defaults as more bank competition reduces interest payments from performing loans, which in turn offers a buffer to cover loan
5
losses. But they also remind that there is a margin effect which brings bank more revenue from the non-defaulted borrowers that pay higher interest rates. Taking this effect into account, Martinez-Miera and Repullo therefore conclude that the final effect of bank competition on the bank risk is ambiguous; they instead propose a U-shaped relationship between bank competition and the risk of bank failure, which means that the probability of bank failure first declines but then increases beyond a certain point as the number of banks increases.
Martinez-Miera and Repullo’s study indicates the third explanation of the relation between bank competition and risk in addition to the previous two conflicting views. Similarly, based on the review of the “competition-fragility” view and the
“competition-stability” view, Berger, Klapper and Turk-Ariss (2008) argue that the two different views are not necessarily conflict each other considering the impact of competition and market power on stability in banking. By regressing measures of loan risk, bank equity-capital on some measures of market powers and the indicators of business environment, their findings both support the “competition-fragility” view in the sense that banks with more market power could also have less overall risk and the “competition-stability” view in the sense that market power increases loan portfolio risk and this risk could also be partly offset by higher equity capital ratios.
The existing banking studies also present different measures of bank competition and risk. Bank concentration is considered as proxy for bank competition. Boyd and De Nicolo’s analysis of bank competition is primarily relied on the “concentration” measure. However, some scholars also doubt that concentration is not an appropriate measure to evaluate the degree of bank competition. For example, Berger, Demirguc-Kunt and others (2004) argue that the indicators such as the Herfindahl-Hirschman index or n-firm concentration ratio are ambiguous to measure bank concentration and emphasize the difference between competition and concentration. In the same vein, Bikker (2004) criticizes that concentration measures like HHI and the k bank concentration ratio are more likely to overstate the degree of competition in small countries and hence increase misleading inferences and measurement mistakes. Different from Body and Nocolo, Stijn Claessens and Luc Laeven apply another method to measure bank competition. They focus on market contestability and find that banking system with greater foreign bank entry and fewer restrictions appears
6
more competitive (Claessens, Laeven, 2004). On the other hand, Gabriel Jiménez, Jose A. Lopez and Jesús Saurina (2007) use both the standard market concentration and the Lerner index to evaluate bank competition. When using the concentration measure, they do not find evidence to support the argument that bank competition does have effects on bank risk. But when using Lerner index, they find a positive relationship between bank competition and bank risk. Such comparison itself cannot prove Lerner index is better than the concentration measure, but it to some extent supports the finding that increased competition leads to more bank risk. A deficiency of their research is that they only use non-performing commercial loans as the measure of bank risk which, however, cannot represent the systemic risk level for a bank’s assets. Berger, Klapper and Turk-Ariss (2008) emphasize the importance of studying the impact of market power on bank risk and stress that the loan risk and bank risk should be separately analyzed to test the “competition-stability” view. Therefore, they use the Lerner index to measure bank competition, and apply HHI to check for the robustness of the findings. In addition, Schaeck and others (2006) use the Panzar and Rosse (1987) H- Statistic to describe the competitive behavior of banks. They argue that the H-Statistic is a measure of direct competitive conduct since it also studies the competitive behavior of other market participants.
Comparing the different measures presented above, this thesis takes the same measure as Schaeck’s, which is Panzar- Rosse H- statistic to analyze bank
competition. Furthermore, no prior study to our knowledge has considered the potential effect of interest rate environment on the relation between bank competition and bank risk. This paper pays primary attention to the impact of interest rate
7
Methodology
This paper uses Z score to measure overall bank risk. Z score gives an inverse measure of a bank’s default probability and it is calculated as: = / . Zit
is the Z score for entity i in year t. ROAit and Eit/TAit are the return on assets and
equity to total assets for entity i in year t respectively. σROA is the standard deviation of return on assets for entity i over the entire research period. For each bank in our sample, Z score measures their buffers against the volatility of their returns, which means it measures if a bank has sufficient capitalization and returns to buffer against unexpected volatility of returns. Therefore Z score is a good proxy for each bank’s probability of bankruptcy, which again is a good measure for bank’s overall risk. Since Z score is an inverse measure of a bank’s default probability, either a higher ROA or E/TA will lead to a higher Z score and a lower probability of bank default risk. While a higher standard deviation of ROA will leads to a lower Z score and a higher probability of bank default risk.
With respect to the measurement of bank competition, except those concentration measures such as Herfindahl index, we prefer both Panzar-Rosse and Lerner Index. In order to calculate the H-statistic we need to firstly run the regression:
( ) = + + +
( ) + , where the terms in the parentheses
are the common measures of bank’s input prices. Then the Panzar-Rosse H-statistic is calculated by adding the three coefficients of the input prices: Panzar-Rosse H-statistic = + + . The H-H-statistic ranges from 0 to 1. The higher the index is, the more competitive the banking system is, and thus the H-statistic is approaching to 1 under perfect competition. However, our analysis is based on Panel-Data, the Panzar-Rosse approach is not so appropriate because we cannot get a H-statistic for each bank in each year within the research period. The H statistic is more relevant for studies of regional banking industry, for example, to compare the competition
8
situation of the entire banking industry among different regions. Therefore, in this study we use Lerner index as the measure of competition environment.
The Lerner index is calculated as: = . In economics, P is the market price of products sold by a firm. MC is the marginal cost for the firm to sell an additional product. Normally, Lerner index ranges between 0 and1. A larger index indicates that the firm has strong ability to earn a higher markup by selling its
products, a strong market power and a lower competition situation in that case. On the contrary, a lower index implies that the firm has lower market power in charging a higher markup, and thus a more serious competition environment. Therefore, Lerner index gives a direct measure of market power of a business entity in the economic environment where it locates. But it could also be seen as an inverse measure of the competition situation the entity encounters. A larger Lerner index indicates a moderate competition situation while a smaller Lerner index indicates a serious competition situation. Since we are using Lerner index for banking industry, the variables in the model need to be adjusted. The market price of the product P is approximated by Revenue/Total Assets, which is the bank’s average output price. A problem arises when determining the marginal cost (MC) in Lerner model. In most cases, marginal cost data is not available for economic entities. In order to get the marginal cost, one has to firstly determine the function between total cost and production, and then derive the derivation of the cost function to calculate the marginal cost. However, for most economic entities, even though they are of the same industry, their business activities can significantly vary from each other. Taking this fact into consideration, it is not accurate to assume only one cost function for all entities of interest. Unless we construct a cost function for each entity, but it is very time-consuming and not realistic if we have many research entities. For banking industry, it also has the same problem because our banking sample covers banks from different regions throughout the world. For banks, the marginal cost actually can be regarded as the marginal cost of fund or simply the interest rate, which is the interest paid by banks for raising additional fund that is used in the bank’s business activities. In empirical studies, short term average variable cost can be used as a proxy for marginal cost. In our case we use the ratio of total interest on customer deposits to total customer deposits (Total interest on customer deposit/Total customer deposits) as a proxy for the marginal cost
9
for banks. By doing so, we can avoid the problem of mis-estimation for banks’ marginal costs because we cannot just use one simple cost function to represent the entire samples.
However, whether the ratio of Total interest on customer deposit to total customer deposits is a good substitute for marginal cost for banks is uncertain, or at least it is not exactly the original “Lerner index”. Under this consideration, we have to choose other measurements to test the findings from Z score and Lerner index. Practically, due to the unavailability of marginal costs, there are some different measurements which can be regarded as substitutes for Lerner index. For example, some of them can be excess return on sales, profit rate, and the ratio of total market value to total assets value, which is so-called Tobin’s Q.
The excess return on sales is calculated as: economic profit/total revenue. Total revenue can be easily collected from a firm’s annual report or some other databases. But it is difficult to derive a firm’s economic profit. In addition to explicit cost, we also need to take implicit cost into consideration when we calculate economic profit. However, the difficulty is that it is sometimes impossible to get implicit costs accurately for a firm because relevant data are normally unavailable in accounting standard reports. Although this measure is favored in theory, but hardly used in practice. The second measure emphasizes the profit rate, which is the rate of return after adjustment for competitive rate of return. The profit rate equals the accounting rate of return subtract the competitive rate of return, and a positive number of the result indicates that the firm has a positive profit rate. The larger it is, the stronger market power the firm has, and a moderate competition the firm faces. However, this good measure is not appropriate for this study due to the fact that it is difficult to reach a solution to generate a competitive rate of return for banks around the world. In addition, since banks’ business portfolios can be diversified, it could be misleading to select just one competitive industry rate of return as a benchmark for the whole banking industry. The last substitute for Lerner index is Tobin’s Q, which is the market value of the firm divided by total assets of the firm. According to the assumption of the theory of Tobin’s Q, under perfect competition situation, for an industry, if the Q is greater than 1, then more investors will enter into this industry because their investment will generate a higher market value than it actually costs.
10
Therefore, we can use Tobin’s Q as an additional measure for bank’s market power. In this case, Tobin’s Q is calculated as; Q = Total market capitalization/Total Assets. Banks with larger Q ratios imply that they have stronger market power and a less serious competition situation. Some scholars have also used Tobin’ Q to measure bank’s competition. For example, (Keely, 1990) uses Tobin’s q to measure bank’s market power and finds that banks with strong market power have less default risk.
Thus Tobin’s Q is selected as the additional measure of risk in this paper. The purpose of adding Tobin’s Q is to use it as the measure of bank competition to investigate the relationship between bank risk and bank competition on the one hand, and to do the robustness check on the other hand. By replacing Lerner index by Tobin’s Q, we test whether the results are consistent with the main findings by using Z score and Lerner index.
The interest rate environment is determined by the short-term basic rate set up by the central bank in each country. Interest rate plays an important role in monetary policy. As a main channel through monetary policy transmission process, interest rate determines the money demand in the economy to a large extent. The short term basic interest rate set up by the central bank will affect the interest rates of the entire maturity spectrum, assets prices as well as exchange rate. In this paper we use short term (yearly) real interest rate of each country as a proxy for interest rate
environment.
In this paper we aim to investigate the relationship between general overall bank’s risk and bank’s competition by taking the interest rate environment into consideration in a global background. Our sample consisted of banks from the five continents in the word. Our hypothesis is: there exists a general relationship between bank’s overall risk and bank competition for all the major banks in the world and this relationship is affected by interest rate circumstance each bank faces in its region. In order to test these hypotheses, we assume the following models:
11 = + + ∗( ) + (1) = + ^2 + ∗( ) + (2) = + ( ) + ∗( ) + (3) = + + ^2 + ( ∗ ) + ∗( ) + (4) = + + ( ∗ ) + ∗( ) + (5)
12
Data
We derive our datasets primarily from Orbis Bank Focus and World Bank
Database. Due to the problem of data availability, other databases such as “Relbanks”, “tradingeconomics”, and “Ycharts’ are also used to improve our datasets. In some cases, the same variable derived from different databases may have small deviations, which could be resulted from different sources of original data, or different data processing and calculation methods. In this case we choose the database which provides the most complete datasets.
Since this is a study for banks throughout the world, the sample selection criteria is simply the global large banks. Banks with an asset value larger than US $50 billion for any year from 2012-2016 are selected. An asset value of larger than US $50 billion is generally considered as large banks in the world. Large banks have more business activities than small banks and their business portfolios are more diversified as well. Normally speaking, small banks’ profits primarily depend on traditional interest-income. While for large banks, non-interest income, such as consultancy income, income from investment activities and various service fees, accounts for a larger fraction in profit. Non-interest income as a fraction of profit is a key predictor for large bank’s profit, but is insignificant for small banks (Fayman, 2009). The businesses and services for large banks are global. In their own region they might have strong market power and do not face competition pressure. But under the
international environment, they can face more serious competition together with other global major banks. Some scholars’ researches have also confirmed the fact that large banks face stronger competition than small banks. For example, Bikker and Haaf (2000) conclude that large banks have stronger competition than small banks because large banks operate more in international market, while small banks are more active in local markets (Bikker and Haaf, 2000). In a similar vein, Yildirim (2003) finds out that larger banks encounter a more competitive environment in transition countries (Yildirim, 2003). When it comes to the research period of this study, we set the time span from 2012 to 2016 by considering that most data prior 2012 are unfortunately
13
not available in Orbis Bank Focus. The research sample consists of total 464 banks in the world.
In order to precisely conduct this research in an international environment, we need to understand how these banks are located, and the competition environment in banking industry for different global groups and classifications.
We firstly investigate the geographic allocation of these banks. Our sample consists of 464 banks, of which 167 are from Asia, 184 are from Europe, 72 are from North America, 25 are from South America, 12 are from Oceania and 4 are from Africa. Europe has the largest amount of banks. On the one hand, it consists of many highly-developed countries; on the other hand, the European Union is still the largest economic unit in the world (World Bank). Asia ranks the second not only because it consists of countries with highly-developed financial industry such as Hong Kong (PRC), Japan and Singapore; but also because the countries there are developing at very high growth rate, whose financial sectors grow rapidly such as China and India. It is worth mentioning that 61 out of 72 banks from North America are located in the United States, which is the largest number among the 49 countries in the sample. Therefore, there is no doubt that the US has the most developed banking industry in the world.
Next we investigate the allocation by the income level of countries each bank locates. This kind of income level classification is from World Bank Country and Leading Groups by World Bank Database. They use the GNI per capita level as the criteria to adjust such classifications. According to World Bank’s methods, high income countries have a GNI/capita larger than $12236, upper-middle income countries have a GNI/capita between $3956 and $12235, lower-middle income countries have a GNI/capita between $1006 and $3955, and low income level countries have a GNI/capita lower than $1005. Among the 464 banks in our sample, 348 banks are from high income countries, 101 banks are from upper-middle income countries, and 15 banks are from lower-middle income countries. It is not surprising that most banks in the sample are from high income countries, and no bank is from low income countries since large banks are located in relatively highly-developed economies. Correspondingly, in order to achieve better economic performance,
high now econ the bank and bank bank Am indu inco H st Tab Th cal hly-develop wadays bank nomy. The previo locations of ks are most emerging m king industr king industr merica, Ocea ustry from d ome level re tatistics. ble1: Panzar-R is table shows t lculated as the s ped econom k has perfor ous analysis f banks are tly located i markets. We ry competit ry for geogr ania and Afr
different inc egion and lo
Rosse H Statist
the regression r sum of the 3 coe
ies must sup rmed as the s based on th highly relat n regions w e now apply tion in the w raphic regio rica. Table 2 come levels ower-middle tic Calculation results of Panza efficients, and. pport highly most impor he locations ted to region with high inc
y the Panzar world. Table ons of Asia, 2 presents th , namely, hi e income le n by Region ar-Rosse H Stati . The H statistic y-developed rtant financi s of global l nal econom come, highl r-Rosse app e 1 shows th Europe, No he H statisti igh income evel region.
istic for differen cs is summarize d banks in l ial intermed large banks my and incom ly-develope proach to ob he H statisti orth Americ ic calculatio region, upp Table 3 sum nt regions. ). Th ed in Table 3. ight of that diary in the suggests th me. Major ed economie bserve the c calculatio ca, South on for banki per-middle mmarizes th he H statistic is 14 hat es on of ing hese then
Tab Th Th ble2: Panzar-R is table shows t e H statistic is t Rosse H statist the regression r then calculated ic Calculation results of Panza as the sum of t by Income Le ar-Rosse H Stati the 3 coefficien evel
istic for differen ts. The H statis nt Income level tics is summari l countries. ). ized in Table 3. 15
16
Variables WORLD ASIA EUROPE NORTH
AMERICA SOUTH AMERICA OCEANIA AFRICA -0.0667 0.0755 -0.078 0.241 0.129 -0.358 0.759 0.278 -0.521 0.584 0.354 -0.0567 0.434 0.139 ( ) 0.372 0.746 0.0551 0.466 0.667 -1.723 0.722 Panzar-Rosse H Statistic 0.58 0.3 0.5611 1.06 0.74 -1.64 1.62
Variables WORLD HIGH
INCOME UPPER MIDDLE LOWER MIDDLE -0.0667 -0.0497 -0.115 0.723 0.278 0.371 -0.118 0.111 ( ) 0.372 0.358 0.216 1.058 Panzar-Rosse H Statistic 0.58 0.6793 -0.017 1.892
Table 3: Panzar-Rosse H Statistic by Region and Income Level
17
Current Research with Panzar-Rosse Approach
The Panzar-Rosse H statistic generally ranges from 0 to 1. The larger the H statistic is, the stronger the competition in the market is, and less effect of entry barriers produces. Theoretically, under perfect competition environment, H statistic is approaching 1. A H statistic equals or smaller than 0 indicates perfect monopoly or a Short-term oligopoly market, demand can exceed supply in such market. A H statistic between 0 and 1 indicates a monopolistic competition market, where new entries are free.
There are many scholars using Panzar-Rosse approach to study the competition of banking industry. As early as 1982, Shaffer uses H statistic to test the monopolistic situation for banking industry in New York. His finding suggests that local banking industries are mainly monopolistic competition. In the long run, neither perfect competition nor complete monopoly applies for most local banking industries (Shaffer, 1982). Shaffer’s finding indicates H mostly should range between 0 and 1, neither approaching 0 nor approaching 1. Following Shaffer, other scholars also use Panzar-Rosse method and get the same conclusion. For example, Nathan and Neave (1989) calculate the H statistic for Canada’s banking industry for the period between 1982 and 1984. They find that the H statistics are significantly different from 0 and 1, which rejects the hypothesis that Canada’s local banking industry is either perfect competitive or perfect monopolistic (Nathan and Neave, 1989). Bikker and
Groeneveld (1998) study the competitive structure in banking industry of European Union countries. They claim that each country’s H values are significantly different from 0 and 1, which again supports the view that the competitive structure of banking industry is monopolistic competition (Bikker and Groeneveld, 1998).
Most studies in banking industry competition generally hold one primary argument that competitive monopoly dominants in the competitive structure in developed economies. But the exceptions also exist when considering specific characteristics of banks and regions. Bikker and Haaf (2000) divide banks into different size groups by asset value and apply the H statistic method. They find out that for small size banks or local banking industries in Australia and Greece, perfect monopoly hypothesis cannot be excluded, and for some regions with various banking
18
sizes, perfect competition cannot be excluded (Bikker and Haaf, 2000). Molyneux, Lloyd-Williams and Thornton (1994) study the banking industry of European Community countries between 1986 and 1989. They find that only Italy banking industry is monopolistic market, while for banks from France, Germany, UK and Spain, their banking markets are monopolistic competition (Molyneux,
Lloyd-Williams, Thornton, 1994). Vesala (1995) studies the banking industry competition in Finland from 1985 to 1992. His research shows that except for the two years of 1989 and 1990, Finnish banking market is monopolistic competition for the rest years when H statistics increased dramatically.
Molyneux, Thornton and Llyod-Williams (1996) conduct a research on Japanese commercial banking market competition. They conclude that Japanese commercial banks earn revenues under monopoly or short term oligopoly because of the entry barriers from local institutions (Thornton and Llyod-Williams, 1996).
The previous researches show that in recent years, many scholars have used Panzar- Rosse to analyze bank competition with the focus on regional and developed economics. These researches present a similar finding that bank competition structure in most cases is monopolistic competition. But Monlyneux and his colleagues show another different finding that the Japanese bank industry is monopolistic partly because of its institutional characteristics.
Panzar-Rosse Approach in This Paper
The H statistics for our sample banks are presented in Table 3. The H for all banks (WORLD) is 0.58, which is the monopolistic competition. This implies that the majority of banks in our sample operate in the monopolistic competition structure, and this is consistent with the findings of previous studies on banking industry competition structure. Asia has a H value of 0.3, which is lower than the world level 0.58. This implies that Asia has stronger monopolistic power than world average level. Europe and South America have H statistics of 0.5611 and 0.74 respectively, which all belong to monopolistic competition. However, for North America, Oceania and Africa, their H statistics are not usual. North America has a value of 1.06, which means that there is a perfect competitive structure in North America. Africa even has a
19
H of 1.62, which is even larger than North America. Oceania has a H statistic of -1.64, which implies a perfect monopolistic competitive environment.
The H statistics calculated from different income level classifications show that banks from high income countries have a H value of 0.6793, which is also consistent with previous research findings that in industrial economies, banking industry normally has monopolistic competition. Banks from upper-middle income countries has a H value of -0.017, while banks from lower-middle income countries have a H value of 1.892.
Does Oceania have a significant perfect monopoly in banking industry because it has a H statistic of -1.64? Do Africa and lower-middle countries have a perfect competitive structure because of their H statistics are significantly larger than 1? We cannot conclude such findings before checking potential problems which might cause a biased result. Our raw data for Panzar-Rosse H calculations are all from Orbis Bank Focus, thus there should not be data mistakes from different data collection or data processing methods. The potential problem could be that the numbers of observations for these groups are limited. Among the total 464 banks in our sample, only 12 banks are from Oceania, 4 are from Africa and 15 banks are from lower-middle income countries. Such limited number of observations could lead to biased results. In addition, these banks also come from different countries. Without controlling for country specific characteristics, we cannot calculate accurate H statistics for these groups only by analyzing limited number of banks. So we believe the H statistics for Oceania, Africa and Lower-middle income groups cannot represent the real
competitive situation in the period from 2012-2016. Further studies need to be conducted in order to answer the question; and a simple solution we choose is to increase the number of banks for each group.
Exclude the results from Oceania, Africa and lower-middle group, our results in table 3 show that in relatively less developed economies, the level of competition in banking industry is less than that from relatively highly developed economies: for example, Asia has a H statistic lower than those of Europe and North America. But there is an exception that South America’s H statistic is larger than Europe’s, which indicates South America’s banking industry is more competitive than Europe.
Upper-20
middle countries’ group has a H statistic lower than High income level group. Such finding is similar to the results of previous studies.
The current situation of banking industry competition in the global environment tells us that significant differences exist among different banking industries in the world no matter by geographic locations or income level. Therefore, in order to conduct our research in an international environment, region-specific economic characteristics must be carefully taken into consideration.
Control Variables:
Our control variables include Bank’s Asset Value, country’s GDP level, GDP/capita, Inflation Rate, Government DEBT(% GDP), Regional Institutional environment.
The asset value of each bank in our sample is an important control variable because it is the criteria to adjust bank size. As we have already concluded, large banks normally have stronger competition than relatively smaller banks.
The GDP measures the total value of production in goods and services in a region for a certain period, thus it is a key indicator to see the overall performance for a region’s economy. Although GDP cannot measure the average economic performance, a country with high GDP level can also have very active financial markets. A good example is China, a country with large GDP but uneven economic development over different regions. China’s financial industry has been developing at a very fast speed. Among the top 10 banks in the world, China has 4 according to the 2016 ranking. This is not a particular case, dual economy usually exists in most less developed economies, while the financial industry is very active, for example Brazil and India.
From the Pazar-Rosse H statistics by income levels, we can see that groups with different income levels have significant differences in H statistics, which indicate bank’s competition structure also depends on the income level of the country where bank locates. We will use GDP/Capita as a measure for income.
21
will increase the opportunity cost to hold money, which will decrease the level of money demand. When people are not confident in future inflation, they cannot make their investment decisions precisely. Therefore, inflation rate is important when studying issues of bank risk.
Banks are an integral part of the financial system of the countries in which they are located. That’s why the risk indicators of country’s financial systems and the characteristics of country’s financial systems regarding risks and competition have to be added into investigation. The main risks to the financial system of counties are derived from the level of country’s debt. Economic crisis reduces economic activity and income in one country, and increases problems of debt service. That’s why country indicators about debt in GDP will be considered in our model.
Except for these economic data, we also need to consider the importance of a country’s institutional environment. Institutional economics is always important when studying a country’s economic performance. The freedom of institutions directly affects new entry to the industry. Institutions can shape the performance of economic behavior. A good institutional environment will boost economic development; otherwise it will affect economic behavior in a negative way. In highly developed industrial economies, institutional circumstances are usually better than less developed economies. This can be seen by various institutional indicators. For example, World Bank’s database of political institutions, World bank’s country policy and institutional assessment (CPIA), and some economic freedom measurement by Fraser Institute, Heritage Foundation, and Polcon database. The CPIA offers specific indicators to measure the institutions in financial markets and could be regarded as one of the best ways to evaluate institutional environment. However, it is very pity that the relevant data concerning most developed countries are not available somehow in CPIA, we eventually use the index of economic freedom by Heritage Foundation. This index is a comprehensive measurement since it contains measure of Property Rights, Judicial Effectiveness, Government Integrity, Tax Burden, Government Spending, Fiscal Health, Business Freedom, Labor Freedom, Monetary Freedom, Trade Freedom, Investment Freedom and Financial Freedom. The index ranges between 0 and 100. A higher level indicates a better institutional environment.
22
Descriptive Statistics
When collecting all the raw data for the period 2012-2016 from relevant databases, we then summarize these data to investigate the basic characteristics. Before we conduct the regression analysis, we need to observe these data to see if further data processing is needed. Table 4 shows the descriptive statistics for the raw data. Table 5 shows the descriptive statistics for data ready for regression.
Variable Obs Mean Std.Dev. Min Max
TotalAsset 2,180 2.740e+08 4.620e+08 243260 3.470e+09 NetIncome 2,155 1.547e+06 4.136e+06 -2.530e+07 4.870e+07 Equity_Assets 2,180 7.523 5.668 -4.307 90.42 Revenue 2,155 6.813e+06 1.220e+07 -981634 1.040e+08 CustomerDepositt 2,053 1.490e+08 2.850e+08 1000 2.560e+09 DepositInterests 1,729 2.209e+06 4.724e+06 -6806 4.890e+07
GDPgroRate 2,320 2.425 2.970 -21.54 25.50 GDPperCapita 2,320 34061 35959 1638 631007 Inflation 2,320 3.799 22.70 -6.650 422.2 GDP 2,320 4419 5590 42.80 19473 GOVDebt 2,315 84.66 60.01 1.600 250.4 RealRate 1,973 3.589 5.169 -17.37 43.27 Institution 2,320 68.19 10.25 33.70 90.10
StaffExp 1,914 1.943e+06 3.534e+06 441.4 2.450e+07 OtherAdminExp 1,642 1.032e+06 2.011e+06 0 1.910e+07 OtherOperExp 2,116 1.387e+06 2.666e+06 -285900 2.150e+07
Table 4 presents all data that will be used in this study. Further explanations will be given for some variables’ labels and units. “Equity_Assets” stands for equity to total assets ratio. “GOVDebt” stands for the amount of government debt as % of GDP. “RealRate” is the real interest rate for each country in each year. “Institution” is the index of economic freedom for each country in each year. “StaffExp”,
23
“OtherAdminExp” and “OtherOperExp” are staff expense, other administration expense and other operating expense respectively. “TotalAsset”, “NetIncome’, “Revenue”, “CustomerDeposit”,“DepositInterests”, “StaffExp”, “OtherAdminExp” and “OtherOperExp” are in thousands US $. “GDP’ is in billion US $. “GDPgroRate”, “Inflation” and “GOVDebt” are in %.
Due to the problem of data availability, all variables do not have the same number of observations. Therefore, this panel data is unbalanced.
Variable Obs Mean Std.Dev. Min Max
Zscore 2,120 85.70 145.3 -5.258 2215 Lerner 1,713 -0.866 31.75 -1240 115.2 TobinQ 826 0.0915 0.103 0 0.893 GDPgroRate 2,320 2.425 2.970 -21.54 25.50 Inflation 2,320 3.799 22.70 -6.650 422.2 GDPperCapita 2,320 34061 35959 1638 631007 GOVDebt 2,315 84.66 60.01 1.600 250.4 RealRate 1,973 3.589 5.169 -17.37 43.27 Institution 2,320 68.19 10.25 33.70 90.10 GDP 2,320 4419 5590 42.80 19473
TotalAsset 2,180 2.740e+08 4.620e+08 243260 3.470e+09
We use the raw data to calculate the main variables in this research, namely Z score, Lerner Index and Tobin’s Q ratio. The rest of data remain as control variables. From the summary statistics above, we can see that for all the three variables, the standard deviations are larger than the means. This indicates there might exist extreme values in them. This can also be proved by the minimum and maximum observations within the three variables. For Z score, although there is no specified range in which the Z value should lie, a maximum of 2215 and a minimum of -5 is unusual. Lerner Index normally ranges between 0 and 1. The Lerner index calculated from our bank sample has a minimum of -1240 and a maximum of 115.2, which is very weird. Tobin’s Q in our sample has a range of [0 0.893], which is theoretically reasonable.
24
In order to precisely evaluate the extreme values, we need to see how these variables are distributed. Figure 1, Figure 2 and Figure 3 show the distribution graphs for Z score, Lerner index and Tobin’s Q respectively.
Figure 1: Distribution of Z score
25
The three figures show that all the three variables have extreme values. For Z score, most of the observations are around a level of 100, an extreme value such as 2000 can lead to biased results. For Lerner Index, they are almost all within a reasonable range, only a few observations have extreme values such -1200. For Tobin’s Q, they all are in a theoretically reasonable range, but it also has the problem of extreme values because a large majority of observations are below 0.2, thus a value of 0.8 will be an extreme value.
According to the descriptive statistics analysis, we find out some variables have extreme values which will lead to biased results. Therefore, we will “winsorize” these variables before the empirical analysis so as to reach an accurate result.
Hausman Test
In order to determine whether to use fixed effects model or random effects model, we firstly need to take a Hausman Test. Table 6 shows the results. P value is 0.0000, which indicates that the null hypothesis is rejected, and in that case we need to use fixed effects model.
Tablee 6: Hausman Test Results
27
Empirical results
Table 7 presents the regression analysis results of our five models. Model 1 only consists of Lerner Index as the main interested independent variable. It is statistically significant at 1% level. It has a positive coefficient of 15.18, which implies that banks with strong market power tend to have less default risk. In other words, Banks facing stronger competition circumstances tend to have higher risks. This result supports the competition- fragility view from previous researches. Real interest rate does not have a significant effect on risks. Both GDP growth rate and GDP level have negative coefficient and are significant at 5% and 10% respectively. This suggests that a country’s general economic performance has negative effects on its national banking industry. Either a higher GDP or GDP/Capita will result in a lower z score, which implies a higher risk for banking industry. The reason can be that banks will be more active and international when their regional economic performance is promising. Banks will have more diversified business portfolios which eventually increase their risk exposures. Inflation rate has a positive coefficient and significant at 5%
significance level. So an increasing inflation rate will increase Z score and reduce the risks for banks. As we have talked before, on the one hand, inflation will increase the opportunity cost of holding money. On the other hand, inflation can affect decisions on investment. Therefore, a high inflation rate not only tends to increase the amount of deposits banks hold, but also decrease the money demand for investment activities. In both cases, bank will have more buffer against potential risks, which then decrease the overall risk a bank faces. Government debt ratio has a positive coefficient and is significant at 1% level. This implies that a higher government debt level tends to decrease bank risk. Governments create government debts by issuing national bonds and other kinds of securities. Most countries in our sample have fiscal sovereignty. They create government debts in most cases not aiming at borrowing money and raising funds, but keeping it as reserves in each country’s central bank. Normally, this action belongs to monetary policy which aims to decrease the excess revenues of other commercial banks in the country. The purpose of such policy is to maintain regional financial stability. On the one hand, it constraints banks from taking excess
28
business activities, and decreases banks’ risks in some way; on the other hand, banks will always have sufficient reserves in Central Bank, in case of unexpected risks, they come to Central Bank for buffers. At the same time, banks are under strict
supervision. Therefore, a high government debt ratio can indicate a good standard of regional monetary policy. So a higher government debt ratio in our models leads to a higher Z score and decreases the risk in banking industry. Institutional environment does not have significant effects on Z score. Bank assets have a negative coefficient and are statistically significant at 5% level. This means large banks tend to be more risky than relatively smaller banks. Just like we have discussed, large banks are more diversified in business portfolios, non-interest income is a significant determinant for their profits (Fayman, 2009). Large banks are also more active in international market, which bring them more risks (Bikker and Haaf, 2000). Therefore, larger assets result in more risks.
Model 2 chooses the square of Lerner Index as the main interested independent variable. This model aims to test if the relationship between bank risk and bank competition is not linear, but a quadratic relationship. The empirical results of model 2 are highly similar to that of model 1. To be specific, the nature of coefficients for each variable and the statistical significances are very close. The Lerner square has a positive coefficient of 14.58, which is also significant at 1% level. This implies that the effects of bank competition on bank risk exists a U shape relationship.
Model 3 aims to test the effects of interest rate level on bank risk. It uses real interest rate as the main independent variable. However, it does not have significant effects on Z score. In this model, GDP per capita is significant at 1% level, but it only has a few effects on Z score since the coefficient is only 0.000658, which is very negligible. GDP also has a 1% significance level, the coefficient has a value of -0.00183, which supports the view that a higher GDP level tend to increase bank risk. Total assets again have a negative value and significant at 1% level, the result supports the finding of model 1.
In model 4 we add all Lerner index, the square of Lerner index and the
production of Lerner index and real interest rate into the model. We aim to observe the effects and significance of the three main independent variables on Z score. Lerner
29
Index still has a significant positive effect on Z score, which is the same as the previous three models. The production of Lerner index and real interest rate has a negative coefficient and is significant at 1% level. The square of Lerner Index lost its significance after adding Lerner Index and the interaction term. The results from model 4 suggest that Lerner index has a positive effect on Z score, but this
relationship is affected by real interest rate level. After adding the interaction term, those GDP variables do not have strong effects on Z score anymore. Government debt ratio’s effect is still significant.
In order to prove our main findings in model 4, in model 5 we drop the square of Lerner index which is not significant in model 4 to test if we can get the same
findings. The coefficient on Lerner index becomes larger (24.35) and still significant at 1% level. The production of Lerner index and real interest rate is significant at 5% level. In addition, the effects of Government debt ratio and bank’s total assets still remain. Therefore, model 5 proves the main findings from model 4.
Comparing the results of the five regression models, the results suggest that Lerner index does have a positive effect on Z score, which means that banks with strong market power tend to have less default risks. A less competition circumstance decreases bank’s overall risk, while a strong competition increases risk. This finding supports the competition-fragility view from previous debates. But the relationship between bank competition and bank risk is affected by real interest rate level. In model 5, the effect of Lerner index on Z score then becomes: + *Real Rate, where
is the coefficient on Lerner index, and is the coefficient on the interaction term. Therefore, in model 5 the effects of Lerner index then actually become: 24.35-2.063*Real Rate. Theoretically, when Real interest rate in model 5 is larger than 11.8%, then Lerner index will have a negative effect on Z score, which implies that more competition leads to less bank default risk. When real interest rate is lower than 11.8%, then the effects of Lerner index on Z score become positive again, this
demonstrates the competition-fragility view. Among the five models, government debt ratio always has a significant stable positive coefficient, and total bank asset has a negative effect on Z score.
30
(1) (2) (3) (4) (5)
VARIABLES Zscore Zscore Zscore Zscore Zscore
Lerner 15.18*** 19.82*** 24.35*** (3.142) (9.107) (4.793) Lerner2 14.58*** 5.024 (3.027) (7.474) Lr_Interest -2.005*** -2.063** (0.840) (0.802) RealRate 0.163 0.110 -0.0218 1.234** 1.150** (0.247) (0.245) (0.203) (0.519) (0.507) GDPgroRate -0.675** -0.626** -0.0461 -0.582* -0.400 (0.308) (0.305) (0.263) (0.310) (0.309) Inflation 0.678** 0.646** 0.257 0.515 0.536* (0.309) (0.307) (0.281) (0.315) (0.310) GDPperCapita 0.000305 0.000279 0.000658*** 0.000270 -2.09e-05 (0.000216) (0.000217) (0.000173) (0.000217) (0.000179) GOVDebt 0.292*** 0.298*** 0.227*** 0.235*** 0.235*** (0.0832) (0.0831) (0.0670) (0.0863) (0.0524) Institution 0.00409 -0.115 0.354 -0.286 0.00958 (0.351) (0.353) (0.300) (0.368) (0.301) GDP -0.00138* -0.00144* -0.00183*** -0.00120 0.000252 (0.000792) (0.000790) (0.000695) (0.000793) (0.000533)
TotalAsset -1.15e-08** -1.10e-08* -2.03e-08*** -1.08e-08** -1.04e-08** (5.61e-09) (5.62e-09) (4.96e-09) (5.60e-09) (4.69e-09)
Constant 51.31** 62.47** 27.55 71.43*** 48.93** (25.66) (25.85) (21.84) (26.91) (20.20)
Observations 1,012 1,012 1,329 1,012 1,012
R-squared 0.080 0.080 0.052 0.088 0.08
Number of ID 273 273 346 273 273
Robust Standard errors in parentheses *** p<0.01 ** p<0.05 *p<0.1 Table 7: Empirical Results
This Table presents the regression results for the 5 regressions. Lerner2 stands for the square of Lerner Index, Lr_Interest stands for Lerner Index times Real Interest Rate, which is our proxy for Interest rate environment. GOVDebt is government debt as % of GDP.
31
Robustness check
Robustness Check usually comes out from two ways, either check the model or check the data. As for the model way, we have already done a Hausman Test before determining our model for panel data, eventually we use the fixed effects model. In addition we have taken heteroscedasticity and serial correlation into consideration when running the regressions. Therefore, we will do the Robustness check based on data.
We do it by running additional three regressions of model 5 to check the robustness of our previous findings. In model 6, we replace Lerner index by Tobin’s Q, which is a substitute for Lerner Index. Model 7 and 8 use Lerner Index as competition measurement, but the samples are divided into High Income and Upper-Middle Income groups based on the classifications of the country each bank locates. By looking at the distribution of banks in our sample, among the 464 banks, 348 are from high income countries, 101 are from upper-middle income countries, and 15 are from lower-middle income countries. We ignore the banks from lower-middle
countries in the robustness check due to their insufficient number in our test. Table 8 presents the results.
32
(6) (7) (8)
VARIABLES Zscore Zscore Zscore
Lerner 26.09*** 25.95*** (6.497) (9.696) Lr_Interest -2.572** -2.712 (1.083) (1.807) TobinQ 57.90*** (20.70) TobinQ_Interest -1.353** (2.146) GDPgroRate -0.796** -0.356 -0.841 (0.357) (0.415) (0.533) Inflation 0.296 1.182** -0.774* (0.366) (0.473) (0.419) GDPperCapita 0.000894*** 0.000595** 0.00134 (0.000261) (0.000255) (0.00148) GOVDebt 0.415*** 0.251** -0.392* (0.106) (0.105) (0.237) RealRate -0.882* 1.696** 1.310 (0.495) (0.677) (0.824) Institution 0.939** -1.073* 1.040* (0.431) (0.571) (0.549) GDP -0.00291** -0.00206* -0.000632 (0.00135) (0.00110) (0.00141)
TotalAsset -1.02e-08* -2.18e-08*** 1.57e-08* (7.95e-09) (6.78e-09) (1.03e-08)
Constant -29.01 118.8** 11.28 (31.09) (46.94) (29.84)
Observations 525 733 260
R-squared 0.187 0.136 0.129
Number of ID 138 196 71
Group WORLD HIGH INCOME UPPER-MIDDLE INCOME
Robust Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
In model 6, Tobin’s Q has a coefficient of 57.9, and it is significant at 1% level. The interaction term has a value of -1.353, which is significant at 5% level. In addition, the coefficients on Government debt ratio and bank total assets have the same properties as in model 5. Therefore, model 6’s results support our findings.
Model 7 uses Lerner index again to test the results for banks from high income countries. Lerner index and the interaction term have coefficients of 26.09 and -2.572 respectively, they are all statistically significant. The results for government debt ratio
33
and bank total assets are the same. Therefore, model 7 also supports our primary findings.
Model 8 tests the results for banks from upper-middle income countries. Lerner index has a significant coefficient of 25.95, the interaction term has a value of -2.712, but not statistically significant anymore. Therefore, we have to reject part of our previous conclusions. And the new findings are: In relatively lower income countries, interest rate environment does not have significant effects on the relationship between bank competition and bank risk. Instead, banks in such countries tend to be more risky if they have a serious competition environment in the banking industry. While for banks in high income countries, the relationship between bank competition and bank risk is affected by local interest rate environment.
34
Conclusion
In this paper, we study the relationship between bank competition and bank risk by taking the effects of local interest rate environment into consideration in an international environment. We aim to find out a general relationship between bank competition and bank risk for banks in the world. We use Z score to measure bank’s overall default risk, Lerner index to measure bank’s competition level, and yearly real interest rate to measure interest rate environment. Our research period is from 2012 to 2016. Banks with an asset value larger than 50 billion US dollars in any one of the five years are selected. According to such criteria of selection, we finally choose a total of 464 banks throughout the world.
Our empirical results indicate that for banks from high income countries, the relationship between bank competition and bank risk is affected by local interest rate environment. When the local interest rate environment reaches a high level, or a threshold, more competition will lead to lower risks in that country. When the interest rate environment has a low level, or below the threshold, more competition will lead to higher risks. In addition, we find out that banks with high asset value tend to be more risky. This means large banks are more risky than small banks. On the one hand, large banks are more diversified in business portfolios, non-interest income is a significant determinant for their profits. On the other hand, large banks are also more active in international market, which brings them more risks. This finding is
consistent with previous researches. (Fayman, 2009) and (Bikker and Haaf, 2000). Therefore, larger assets bring more risks. Another important finding is that a higher government debt ratio tends to decrease risks in banking industry. This can refer to the general principle of monetary policy. One government’s debt ratio has close relation with the regional monetary policy. By creating government debts, government not only tries to control the revenues of its commercial banks and the sufficient reserves in its central bank in a reasonable range, but also put its banks under strict
supervision,. By doing so, one government attempts to sustain the regional financial stability. Our finding shows that bank risk reduces with the governmental debt level raises one the one hand, and that higher governmental debt ratio contributes to a more
35
reasonable regional monetary policy on the other.
As for banks from relatively low income countries (In our case, the upper-middle income countries), interest rate environment does not have significant effects on the relationship between bank competition and bank risk. In low income countries, strong competition will lead to more risks for banks. Therefore, only competition-fragility view is demonstrated in this case. We therefore cannot confirm the effects of other economic variables on risk. In this case the further studies need to be conducted for relatively low income countries.
For those high income countries, when interest rate environment reaches a high level, the competition-stability view is hardly verified in practice. In that case the interest rate environment must reach a very high threshold. But it is worth to mention that one country could be in high interest rate environment but its interest rate
environment does not necessarily reach the threshold because a high interest rate environment means that the interest rate level is higher than other historical levels. For instance, from our empirical results, the yearly real interest rate must reach a threshold of 11.8% so that the competition-stability view can be held theoretically. But such high interest rate in most industrial economies is rare to see. Therefore, we conclude that, practically speaking, the competition-fragility view stands for most banks in the world.
The experience from this thesis indicates that it is definitely not easy to study the relationship between competition and risk for most global banks. First of all, regional characteristics are hard to control. Secondly, banks from different regions have different characteristics, which can affect the relation between competition and risk. Based on the results of robustness check, when banks are divided into different groups by income level, the R squared has improved significantly. Therefore, findings from regional studies by focusing on a relatively small group of banks should have a strong internal validity. But external validity should not be considered due to the fact that so many different characteristics involved in different sample groups. Under this
consideration, the debate in the relationship between bank competition and bank risk should not exist. Either different sample groups or different research approach can lead to different results.
36
Reference
Belaisch, A.J. (2003). Do Brazilian banks compete? International monetary fund, Working Paper, 3-113.
Berger, Allen N., Klapper, Leora F., & Turk-Ariss, Rima. (2009). Bank competition and financial stability. Financ Serv Res, 35, 99–118.
Berger, AN., Demirguc-Kunt A., Levine, R., & Haubrich, J. (2004). Bank
concentration and competition: An evolution in the making. Money Credit Bank, 36, 433–451.
Bikker, J.A. (2004). Competition and efficiency in a unified European banking market. UK, Edward Elgar publishing.
Bikker, J.A., & Haaf, K. (2000). Competition, concentration and their relationship: An empirical analysis of the banking industry. Financial Structure, Bank Behavior and Monetary Policy in the EMU Conference, October 5-6, Groningen.
Bikker, Jacob A., & Haaf, Katharina. (2002). Competition, concentration and their relationship: An empirical analysis of the banking industry. Journal of Banking & Finance, 26 (11), 2191- 2214.
Boyd, J.H., & G. De Nicoló. (2005). The theory of bank risk-taking and competition revisited. Journal of Finance, 60, 1329-343.
Bushman, Robert M., Hendricks, Bradley E., & Williams, Christopher D. (2015). Bank competition: measurement, decision-making, and risk-taking. Journal of Accounting Research, Forthcoming. Available at SSRN:
https://ssrn.com/abstract=2460371 or http://dx.doi.org/10.2139/ssrn.2460371 Claessens, Stijn., & Laeven, Luc. (2004). What drives bank competition? Some
37
international evidence. Journal of Money, Credit and Banking, 36, 563-583. Demsetz R, Saidenberg MR., & Strahan PE. (1996). Banks with something to lose: the disciplinary role of franchise value. Fed Reserve Bank N Y Econ Policy Rev, 2(2), 1–14.
Egli, Dominik., & Rime, Bertrand. (1999). The UBS- SBC merger and competition in the Swiss retail banking sector. Working paper, Study Center Gerzensee.
Fayman, Alex. (2009). Large banks versus small banks: How different are they? Journal of Economic and Social Policy, 13 (1), Article 6.
http://epubs.scu.edu.au/jesp/vol13/iss1/6.
Jiménez, Gabriel Jose A., & Lopez, Jesús Saurina. (2007). How does competition impact bank risk-taking? Federal Reserve Bank of San Francisco. Working Paper. Keeley, Michael C. (1990). Deposit insurance, risk and market power in banking. The American Economic Review, 80 (5), 1183-1200.
Keeley, Michael C. (1990). Deposit insurance, risk, and market power in banking. The American Economic Review, 80 (5), 1183-1200.
Martinez-Miera, David., & Repullo, Rafael. (2010). Does competition reduce the risk of bank failure? The Review of Financial Studies, 23 (10), 3638-3664.
Molyneux, Phil., Lioyd- Williams, D, M., & Thornton, John. (1994). Competitive conditions in European banking. Journal of Banking & Finance, 18 (3), 445- 459. Molyneux, Philip., Thornton, John., & Liyod- Williams, D Michael. (1996). Competition and market contestability in Japanese commercial banking. Journal of Economics and Business, 48 (1), 33- 45.
38
system: Empirical results. Canadian Journal of Economics, 22, 576-594.
Schaeck, K., Cihak, M., & Wolfe, S. (2006). Are more competitive banking systems more stable? Unpublished Working Paper No. 143, International Monetary Fund, Washington, D.C.
Shaffer, S. (1982). A non-structural test for competition in financial markets. In Bank structure and competition, Conference Proceedings (pp. 225- 243). Federal Reserve Bank of Chicago.
Vesala, Jukka. (1995). Testing for competition in banking: Behavioral evidence from Finland. Scientific Monographs.
Yildirim, H. Semih. (2003). Competition and contestability in central and Eastern European banking markets. Doctoral dissertation. University of Tennessee. http://trace.tennessee.edu/utk_graddiss/2351
