SOCIAL CAPITAL AND THE EFFECTS ON
BANK RISK-TAKING
Bouwe Jan Hak
University of Groningen
Faculty of Economics and Business
Supervisor Boris van Oostveen
June 2017
ABSTRACT
By using a large cross-cultural dataset and a multi-level modelling way of analysing the data this study addresses, and improves the methodology applied to current empirical work that study the effect social capital has on bank risk-taking behaviour. This study also examines two moderators of the relationships between the different indicators of social capital and risk-taking behaviour. The results show that social capital leads to significantly more risk-taking behaviour for the Legatum Institute indicator of social capital. The strength of a legal system and the notion of a country being developed have no effect on the impact of the indicators of social capital on bank risk-taking behaviour. No definitive conclusion to the question of what the effect of social capital is on bank risk-taking behaviour can be made. Theoretical and practical implications are discussed.
Keywords: Social capital; bank risk-taking
I. INTRODUCTION
Although research suggests a link between social capital and risk-taking behaviour in general, empirical studies show mixed results on this effect. Yet, specific research between social capital and risk-taking behaviour by banks is rare. By using a large cross-cultural dataset and a multi-level modelling way of analysing the data this study contributes to research on the effect of social capital on risk-taking behaviour by banks. Furthermore, by adding a new way of measuring social capital I provide more insight in the relationship between social capital and risk-taking behaviour by banks.
The most recent banking crisis has been attributed to banks taking too much risk. After this banking crisis several significant changes to the regulations were announced and new regulations are still being passed, refined and broadened today by governments to limit and cover the various types of risks banks can take (Demirgüc-Kunt, Detragiache and Merrouche, 2013). Therefore, the risk-taking behaviour by banks remains a heavily discussed topic. The abundancy of rationales for the specific reason why banks engage in risk-taking behaviour and its effects (Stulz, 2014; De Nicolo and Boyd, 2005; Leaven and Levine, 2009) make it a relevant academic and practical important and relevant topic.
None of the new regulations and policies above-mentioned take social capital within a country into consideration. Since the mid 1990’s social capital, defined here as ‘’the networks of relationships among people who live and work in a particular society, enabling that society to function effectively’’, has increasingly become a part of the academic landscape, crossing over to a broad range of disciplines and fields. This definition encompasses the generalized social capital at the country-level, thus not only within a subgroup, such as friends and relatives or within a family, but the social capital shared with a person you have never met, but live in the same country with. For instance, the effects of social capital on economic growth and economic development have been studied thoroughly in the past two decades (Knack and Keefer, 1997; Helliwel and Putnam, 1995; Fukuyama, 1995). The results indicate that the relationship between social capital and risk-taking behaviour has not gone unnoticed by economists as social capital is considered an important determinant of macro-economic performance (Beugelsdijk, de Groot and van Schaik, 2004).
showed a negative effect between social capital, measured as trust of a country and bank risk-taking, meaning that as the amount of social capital in a country increases the risk-taking by banks decreases. The paper uses the results to give policy makers another instrument to use in controlling bank risk-taking and the stability of banks. Moreover, examining the relationship between social capital of a country and risk-taking behaviour by banks should shed more light on a not often researched topic.
In the following I will provide how and why I improve results of Xie et al (2016). First, the results of Xie et al. (2016) are important but at the same time show potential to advance the economic and behavioural research through the concept of social capital. The paper lacks of a reliable and valid instrument for the measurement of social capital. Although social capital is defined in many ways, a common instrument to measure social capital used in the field of economics is trust and civil norms (Martinez, 2012; Xie et al., 2016; Knack and Keefer, 1997; Guiso et al., 2004). As most measures of social capital, this measure is based on survey results rather than the actual outputs. Meaning, rather than the actual outputs of social capital, the outcomes of social capital are measured. For example is the level of trust a banker exhibits in his daily economic behaviour, which is an output measure, result of good law enforcement or of high levels of social capital? To provide more insight in the relationship between social capital of a country and risk-taking behaviour in this thesis I will use two measurements (The Legatum Institute indicator and the SolAbility indicator), for robustness purposes, of social capital with similar methodologies, and that are not influenced by legal or economic incentives and encompasses the definition of social capital used by this paper better. In Section III of this Thesis I will explain and define both social capital indicators.
Second, Xie et al. (2016) collected data for a relatively short time frame, data from the years 2004-2006, with most data coming from just the year 2005. In this thesis, I use more recent data over a longer time frame. The most important benefit of this is the fact that I will be able to observe the workings of social capital and bank risk-taking more accurately. Furthermore it will allow me to distinguish whether the results found by Xie et al. (2016) are a snapshot from a certain time period or an actual long-term phenomena through which I hope to progress the topic. Third, Xie et al. (2016) makes use of a simple OLS regression. In this thesis, with a multi-level that approach that has several advantages over the simple OLS regression used in the Xie
et al. (2016). In sum, my study can be seen as an attempt to estimate the positive and negative
In this study, I do not find conclusive evidence that social capital has a significant effect on risk-taking behaviour. As for one indicator a significant relation is found, showing that more social capital leads to more bank risk-taking and for the other indicator no significant relation is found. In this way, the study shows, that the methodology, the definition and measurement of social capital do have significant effects on the outcome. This indicates the importance of a unified and overarching definition of the term social capital and its measurement. I furthermore find, that the legal system and the notion of a country being developed have no effect on the impact social capital has on bank risk-taking behaviour. My study puts the results of Xie et al. (2016) in a broader perspective, as I do not find conclusive results of the relationship between social capital and bank risk-taking
The paper proceeds as follows. Section 1 discusses the notion of social capital, bank risk-taking, and the effects of social capital on risk-taking behaviour and economic growth and development. This section furthermore describes the hypotheses that will be tested in Section 2. Section 3 describes the methodology. Section 4 presents the data used, and how it is used in this paper. Section 5 shows the results of the mixed model regressions. Section 6 provides a conclusion, assesses limitations on this work and suggestions for future research.
II. LITERATURE REVIEW
In this section I will elaborate on the most recent literature on the topic of social capital and risk-taking behaviour, starting off with literature social capital. Thereafter the external factors influencing risky behaviour for banks and the impact social capital has on bank risk-taking behaviour will be discussed. This is to be followed by the effects the strength of legal system and the state of development have on the influence of social capital on risky behaviour by banks to formulate the hypotheses.
Social Capital
In this section, I elaborate on the term social capital and present an overview of how the term is used in different research traditions in the past and more recent. This will help me to define social capital.
society to function effectively”; this definition reflects the current scope of the social capital concept. There are connections, social ties between people from which people can derive value. This means that social capital is related to the social safety net in the society which allows people to act as they want whilst making sure that others in the society have that same prosperity.
However, I must state that the social capital concept is a complex one to pin down under one universal definition (Farr, 2004). Additionally, Robison, Schmid and Siles (2002) state that most definitions of social capital do not even satisfy the requirement of a definition. As most definitions are not limited to the question; what is social capital? But rather they include answers to questions such as; how can social capital be used? Where does social capital reside? And how can social capital be changed? Interesting questions per se for research, but not essential to defining the concept. The term social capital has been used since the late 1800s, only becoming a household word in the last 25 years.
There are four traditions, according to some others five, of social capital, that can be identified. Each are associated with a grand theorist of sociology: 1) Durkheim, 2) Weber, 3) Marx, 4) Simmel and 5) Bentham (Woolcock, 1998). These founding fathers of sociology looked and defined at capital from the social point of view. The concept of social capital has since the founding traditions changed a lot. For example social capitalist now take and define social capital from more a capital’s point of view (Farr, 2004).
Recently social capital is viewed more as a neutral resource, this view was highlighted in the influential articles by Coleman (1988) and Fukuyama (1995). This can be seen in the definition of social capital of Coleman (1988) he defined it as follows; ‘’a variety of entities with two elements in common: they all consist of some aspect of social structure, and they facilitate certain actions of actors within the structure’’. Societies endowed with generalised trust enjoy a form of social capital. This endowment factor is complementary to traditional factor endowments like labour and capital. Furthermore non-family or generalised trust is of great importance for successful performance in advanced economies (Fukuyama, 1995). This view is not shared by all, as social capital should not be seen as complementary to the traditional endowment factors, as it may be seen as a figurative term for a ‘’prospective and productive fund that is created by shared, public work’’ according to Farr (2004).
networks, trustworthiness and the norms of reciprocity that arise from them. Putnam argues that social capital can be measured by the amount of trust and "reciprocity" in a community or between individuals. Honing in on the connections among individuals and the social network, social capital can also be seen as the location people have in a social structure, from which they can derive an advantage (Burt, 2004) The social structure represented by Burt corresponds to a division of labour from Durkheim ([1893] 1939).
Social capital is probably one of the most successfully introduced ‘new’ concepts in economics in the last decade (Beugelsdijk et al., 2004). However, there is some criticism with how the concept is currently used and measured in the field of economics. According to Fine (2002), the measurements for the traditionally sociological term social capital in the field of economics is disconcerting. These measurements are often indicators of trust, norms and civic cooperation (Guiso et al., 2004; Knack and Keefer, 1997; Xie et al., 2016). The use of these indicators is, however, mostly outcome based rather than the actual output and activity based mechanisms of measurement. This means, that rather than the actual outputs of social capital, the outcomes are measured (see Farr, 2004). An issue with this is that these measures of social capital are contaminated by other factors. Again I use the following example, is the level of trust a banker exhibits in his daily economic behaviour result of good law enforcement or of high levels of social capital? According to Farr ‘’to cleave to measures of trust and bonding, or for researchers to pin their hopes on picnics and organized sports’’ could not be any more wrong when measuring social capital, indicating that the use of indicator as trust for measuring social capital must be done with caution. Related to this, Glaeser, Laibson and Sacerdote (2002) show that economist however do seem to understand the role that repeated social interactions have in solving the free-rider problem and reduce opportunism and acknowledge the importance. They furthermore acknowledge, that due to the lack of a commonly accepted theoretical framework economists use flawed indicators such as trust.
value, but it is at the same time open enough for an economic interpretation, through the fact that it can be used for different purposes, such as moral and material support, work and non- work advice. Furthermore, this definition of social capital can withstand the criticism outed on the usual indicator used in the field of economics, namely an indicator of trust. In this part of section 2 I elaborated on the definition of social capital and how the concept is used in my paper. In the next section I will elaborate on bank risk-taking behaviour.
Bank Risk-taking
In this part of the section 2 I will describe the current views on why banks take risks and under what conditions risk-taking behaviour becomes more prevalent for banks. The goal of this section is to assess the current views on bank risk-taking, and how the regulatory architecture and the political environment affect the financial market. In turn
In short, risk taken by banks can be very profitable, but also very costly, not only for the bank itself but for the society. These profits (Stulz, 2014; De Nicolo and Boyd, 2005; Leaven and Levine, 2009) and costs (Stulz, 2014; Diamon and Dybvig, 1983; Etsy, 1998), have been widely discussed in the field of economics. In the last couple of decades, the costs have been seen more profoundly. For example the collapse of the American subprime mortgage market in 2007, paving the way for more research to be done on the effects of external factors on risk-taking by banks. This and earlier crises since the 1980’s has focused economists’ attention on bank insolvency issues and common factors of bank risk-taking.
Furthermore, an ever-growing body of literature shows the effect of the regulatory system on the risk-taking by banks. Calem and Rob (1999) find, that that as the bank’s capital increase, the risk the bank takes up to a certain level when it will start to make more risk is going down. However, they also find that an increased capital requirement tends to increase the risky behaviour by banks. Demirgüc-Kunt and Detragiache (1998) show, that banking crises tend to emerge more when there are weak macro-economic conditions, such as a low level of GDP. The seminal paper by La Porta et al. (1997) showed, that laws are better enforced in wealthy countries than in the lesser wealthy countries. Furthermore, greater investor protection is associated with larger capital markets, as their quality is enforced more by the workings of the law. Houston et al. (2010) find that stronger creditor rights increase the likelihood of a financial crisis. They find, that stronger creditor rights increase risk-taking by banks, and consequently, increase the likelihood of a financial crisis. Buch and DeLong (2008) suggest, however, that strong bank supervisory can reduce the amount of banking risk in a country.
In this section I found the common factors influenced bank risk-taking and found that social capital was not a factor studied extensively yet. Furthermore I found that regulatory architecture and political environment have an influence on bank risk-taking. In the next part of section 2 I will elaborate on the effects of social capital on bank risk-taking.
The Effect of Social Capital on Risky Behaviour
In this section I will examine the effects of social capital on bank risk-taking, and furthermore analyse the mechanism that lie behind the effects. Social capital emphasizes "specific benefits that flow from the trust, reciprocity, information, and cooperation associated with networks’’ (Thomas, 2015). Therefore, it makes sense to assume that social capital has an influence on the behaviour of banks in their primary functions. The primary function of banks is to accept deposits, and grant advances, also known as providing loans. These primary functions are transactions that depend on many factors such as the legal environment, but also factors such as trust and knowledge of information, factors that are incorporated in social capital.
less often dishonest and would show less risky behaviour. The rationale behind this is, that in areas with higher levels of social capital, the costs of breaching a contract or of being dishonest is higher. The social and reputational cost of not keeping a promise is higher in a country with higher levels of social capital, because in high social capital countries operations rely more on social interactions. In addition, in countries with higher levels of social capital, CEOs and managers of firms and banks learn to keep their promises better because establishing moral norms of keeping promises represent a larger coefficient in their utility function. Guiso et al. (2004) strengthens this statement by stating ‘’since financial contracts are trust intensive contracts par excellence’’, social capital should have major effects on the development of financial markets. Financing in this sense is an exchange of a sum of money today for a promise to return more money in the future. Whether such an exchange will take place, does not solely depend upon the legal enforceability of contracts, but also the extent the financier trusts the borrower, or in other words the amount of social capital there is
Knack and Keefer (1997) claim and find that trust and civic cooperation are associated with stronger economic performance. Boulila et al. (2008) shows robust evidence that social capital is related with economic growth and that the same level of social capital exerts indirect effect on GDP per income growth through the development of institutions. The increase in economic growth rates caused by the increase in social capital as shown by Boulila et al. (2008) and Knack and Keefer (1997), causes banks to reduce their risky behaviour. This works in two ways: the interest rates are not required to be that high, as banks already have an increased profitability, which makes products less risky to borrowers, and in turn reduces the risk banks bear. The banks become less eager to seek the low probability, high return outcome. As the banks profit increases, they intentionally seek less risk (Boyd and De Nicolo, 2004). Secondly, higher levels of social capital are associated with less selfish and more group oriented behaviour, which should result in more products that are less risky, as banks and borrowers have trust in each other (Goette, Huffman and Meier, 2006). Meaning, that the social aspect of borrowing motivates efficient behaviour even when ordinary incentives fail, through which the bank’s profitability should increase, and reduce the chance of default, thus reducing the risk for the bank.
associated with less selfish and more group oriented behaviour. Meaning that in areas with higher levels of social capital, inter-bank and bank–firm cooperation is expected to be more efficient, and increasing the profitability of banks and reducing the probability of bank insolvency.
Li et al. (2014) find that social capital increases the amount of information shared, meaning, that there is less information asymmetry per definition, as there is more information on the market and thus available to banks. Furthermore, both the content and quality of the shared information is improved which further helps the problem of information asymmetry. This was earlier accentuated by Lin (1999), who states that social ties, which are increased when social capital increases, provide useful information about opportunities and choices in imperfect markets, which in turn helps banks to make the correct decisions. Houston et al. (2010) support this claim, and find that the sharing of information increases bank profitability, contributes to economic growth, and reduces bank risk-taking. This illustrates, that higher social capital decreases the amount of risk taken by banks, as there is less information asymmetry, which per definition decreases the amount of risk taken by banks, as they are more certain of the creditworthiness of borrowers.
In line with the previous line of reasoning Xie et al. (2016) show through an OLS regression that banks do in fact take less risk in countries where higher levels of social capital are found. The paper uses a previously slated trust and civil norms measurement to measure social capital. Additionally, the paper uses a relatively small sample of 2657 banks in 53 countries, and focusses on the 2004-2006 period. The paper focusses on the period by taking averages from this period and values and pool them together.
From these perspectives, bank risk should be lower in countries with higher levels of social capital. Accordingly, I formulate the following hypothesis:
Hypothesis 1: There will be a negative relationship between social capital of a country and risk-taking behaviour by banks, such that in countries with higher levels of social capital, the risk-taking by banks will be low.
In the following I will elaborate the effect of two moderators on this relationship.
Social capital on Strength of Law
(2016) show, that the regulatory system impacts the effect that social capital has on bank risk-taking. It shows that the impacts of trust on bank risk-taking are stronger in countries with stronger legal systems, as these countries are more capable of regulating the behaviour of their citizens and firms. The paper reasons, that a good working legal system guarantees the contract enforcement, and makes it costly to violate the contract. Social capital should work in a similar fashion, as it is not the regulatory costs that increase, but rather the social and reputational costs. Thus, constructing a better institutional environment, and building a society with a high level of social capital, and a well-developed political/legal system should help avoid bank crises. This line of reasoning are based on Knack and Keefer (1997), who find that trust and norms of civic cooperation are stronger in countries with formal institutions that effectively protect property and contract rights, and in countries that are less polarized along lines of class or ethnicity. These results are furthermore confirmed by Zak and Knack (2001) by extending the analysis, and by Calderon et al. (2001). Which, in turn are roughly confirmed by Beugelsdijk
et al. (2004).
enforcement is weaker, suggesting that trust matters more where the legal system is less efficient.
This leads me to formulate the following hypothesis as follows:
Hypothesis 2: A strong legal system strengthens the negative relationship between social
capital and bank risk-taking.
Social capital and Development
This leads me to formulate following hypothesis:
Hypothesis 3: Developed countries strengthen the negative relationship between social capital
and risk-taking activity by banks
III. METHODOLOGY
In this paper, I use multi-level analysis as the main methodology. Data of the bank is nested in countries, meaning that the data is not independent of each other. In addition, time effects are nested in banks. The non-independence of the data refers to the degree to which responses from banks in a certain year are influenced by, depend on, the country they are in. This also indicates, that data of banks are influenced by, depend on, the time period the data was produced. There are logical reasons to believe, that the variables at the individual level are clustered at a higher level, despite this variable not referencing to the cluster at a higher level. This means, that, because the banks are from the same country, they share more characteristics in terms of national culture, governance structure, legal systems and economic systems than banks from different countries. A measure to calculate the interdependency is the Interclass correlations 1 (ICC1, Bliese, 2001). ICC1 confirms, that the data is not independent of each other: ICC1 = .20 and .19 for the level 3 model. The level 2 ICC1= .97, a nonzero ICC1 value indicates, that the group membership is related to lower-level observations. For this reason, I use a multi-level model analysis to test my hypothesis (Raudenbush, 1989; Raudenbush and Bryk, 2002; Snijders and Bosker, 1999). Hierarchical, or nested, data structures are common throughout many areas of research, for example employees are nested in teams, and organizations, and students are nested in classes.
The multi-level model analysis in this study estimates the between and within effects of the country, bank and time effects in the same model. The data has two nested levels of clustering, creating a three-level model once all observations are considered. In other words: by means of multi-level modelling, I correct for the variance on the country, bank level and time period, and therefore estimate the effects in a better way compared to ordinary regression models. Fitting a three-level model requires me to specify two random-effects equations: one for level two: bank level and one for level three: country level.
The representation of the mixed model above looks formally like:
Where subscripts 𝑖 is an index for firm, 𝑡 is a time index and 𝑗 is an index for country. 𝛽0 is a constant, 𝛽1𝑆𝑜𝑐𝑖𝑎𝑙𝑠𝑐𝑜𝑟𝑒𝑖𝑗𝑡 refers to the key variable of interest; social capital score. 𝛽2𝐶𝑜𝑢𝑛𝑡𝑟𝑦𝑖𝑡+ 𝛽3𝐹𝑖𝑟𝑚𝑖𝑗 refers to data nested in the second level that includes a vector of bank-level control variables, bank size, too-big-to-fail and the rate of loan loss reserves (LLR), and data nested in the third level includes a vector of country-level control variables, inflation, log GDP per capita and the legal system (official supervisory power, control of corruption and regulatory quality). 𝑣𝑡 is a time fixed effect, 𝜀𝑖𝑡 is the random disturbance term.
Model 1 is the regression model between social capital indicators and the bank risk-taking with a time-fixed effect and no control factors included. In the next couple of models the control variables are included in the model. The issue of heteroskedasticity and autocorrelation is accounted for by using robust standard errors. There are still issues with robust standard errors; White/Eicker standard errors are consistent under heteroskedasticity, but are biased in small samples even under homoskedasticity, and inconsistent under serial correlation or clustering. Newey-West standard errors correct for autocorrelation, but only up to a certain lag (Hayes and Cai, 2007). Any of these standard errors are not robust: outliers going to infinity unsettles them. Despite these issues with standard robust errors not using them still is a worse outcome than using a robust standard error that does come with some issues.
To deal with the issue of omitted variables and missing data, I will be using the hierarchical nature of multilevel data to obtain estimators, which are robust to presence of omitted variables. The estimators are adjusted for time-fixed effects, as time-fixed effects are useful in dealing with omitted variables. Furthermore, multilevel data, as in my case, contains rich information to deal with omitted variables and missing data.
highly unlikely, and disregards many other factors, that then will come into play. Besides that the knowledge about social capital and risk-taking is not widespread enough or researched enough for it be realistic that any shareholder if it would already know this information would act upon it.
Equation 2 to test hypotheses 2 reads as follows:
𝑍𝑠𝑐𝑜𝑟𝑒𝑖𝑡 = 𝛽0+ 𝛽1𝑆𝑜𝑐𝑖𝑎𝑙𝑠𝑐𝑜𝑟𝑒𝑖𝑡+ 𝛽2𝐶𝑜𝑢𝑛𝑡𝑟𝑦𝑗𝑡+ 𝛽3𝐹𝑖𝑟𝑚𝑖𝑡+ 𝐷1∗ 𝛽𝑖 + 𝑣𝑡+ 𝜀𝑖𝑡 Where all variables read as mentioned earlier. The new term 𝐷1∗ 𝛽𝑖 can be read as the interaction variable between the indicator for social capital and a dummy indicating whether the strength of law indicators are below or above average.
Equation 3 used to test hypotheses 3 reads as follows:
𝑍𝑠𝑐𝑜𝑟𝑒𝑖𝑡 = 𝛽0+ 𝛽1𝑆𝑜𝑐𝑖𝑎𝑙𝑠𝑐𝑜𝑟𝑒𝑖𝑡+ 𝛽2𝐶𝑜𝑢𝑛𝑡𝑟𝑦𝑗𝑡+ 𝛽3𝐹𝑖𝑟𝑚𝑖𝑡+ 𝐷2∗ 𝛽𝑖+ 𝑣𝑡+ 𝜀𝑖𝑡 Where again all variables read as above, with the inclusion of the new variable 𝐷2∗ 𝛽𝑖 that indicates an interaction variable between the indicator of social capital and a dummy variable indicating whether a country is developed or not. This will indicate whether the development of a country in the sample has an influence on the effect that social capital has on the bank risk-taking.
All models are estimated for both indicators of social capital thus in fact doubling the number of models and estimations rendered. Furthermore, for all three equations and all individual models a sub-analysis will be done. Subsample robustness tests are conducted to prevent some countries, or country in this case to have an overly strong influence on the results. In this particular case that means excluding the USA, as it has an extraordinary number of banks and makes up most of the sample (≈81%). This means that the data will split for each equation and re-estimated after removing the data for US banks.
IV. DATA
The initial dataset consists of 11,066 commercial banks from 148 countries covering the years 2012 to 2016. The Bank level accounting data is obtained from the Orbis Bank Focus
Table I Definition of variables.
Definition Sources
Bank-level data
Z-score Equals log of (ROA + CAR)/σ(ROA), where ROA = π/A is return on assets and CAR = E/A is capital-asset ratio
Orbis Bank Focus
Bank size Natural logarithm of total assets, in millions US dollars
Orbis Bank Focus LLR (%) Loan loss reserves divided by gross
loans
Orbis Bank Focus
Too-big-to-fail A dummy variable that takes a value of one if the bank’s share in the country’s total deposits exceeds 10%
Orbis Bank Focus
Country-level data
Social Capital 1 The Legatum Institute indicator of social capital in a country
Legatum Institute Social Capital 2 The SolAbility indicator of social
capital in a country
SolAbility Sustainable Intelligence Company
Log GDP per capita Log real GDP per capita, in US dollars Orbis Bank Focus Inflation (%) Percentage inflation rate, GDP deflator Orbis Bank Focus AverageCC The indicator measures the extent to
which public power is exercised for private gain, including both petty and grand forms of corruption, as well as “capture” of the state by elites and private interests.
Worldwide Governance Indicators
AverageRL The indicator measures the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence.
Worldwide Governance Indicators
AverageRR The indicator measures the ability of the government to formulate and implement sound policies and regulations that permit and promote market competition and private sector development.
Worldwide Governance Indicators
Strength of Law Indicator of strength of law in a country, average of three law sub indicators (corruption control, regulatory quality and rule of law). A dummy variable indicating whether a countries’ strength of law is above average.
Worldwide Governance Indicators
Developed A dummy variable indicating whether a country is developed. 1 is equal to developed
World Bank
Inter Developed Interaction term of Developed and Social Indicator 1 & 2
These two firms have social capital data for most countries in the world. The Legatum indicator contains 149 countries, and has data covering the years 2007-2016. The SolAbility indicator contains 179 counties and has data starting in 2009, however in 2013 a drastic change in the methodology was put in place making the data before 2013 inaccurate and unusable. The definitions of variables used in this study are summed in Table I.
Measuring Bank Risk-taking
The method of measuring the bank risk-taking, I use in this paper, is the Z-score. The Z-score is a widely-used risk measure in the banking literature (Leavin and Levine, 2009; Houston et
al., 2010; Pathan, 2009), attributed to Boyd and Graham (1986), Hannan and Hanweck (1988)
and Boyd, Graham and Hewitt (1993). It is a useful assessment of bank risk and the overall financial stability as it measures the stability of a bank and the probability of insolvency, and can be used for cross-sectional studies as in time-varying panel studies applicable to all listed and unlisted financial institutions (Leppit and Strobel, 2013). There is a lack of consensus on what is the best way of constructing the Z-score. Leppit and Strobel (2013) propose the use of an alternative time-varying Z-score; one that uses mean and standard deviation estimates of the return on assets, which are calculated over the full sample, and combines these values of the capital-asset ratio. This method shows a low level of intertemporal volatility on bank level, making it very straight forward to implement. The Z-score calculation looks as following:
𝑍 − 𝑠𝑐𝑜𝑟𝑒 =𝐶𝐴𝑅𝑡+𝜇𝑅𝑂𝐴,𝑡 𝜎𝑅𝑂𝐴,𝑡 .
Where CAR is the capital asset ratio and the ROA the return on assets. ROA and CAR are calculated as the means for the 2012-2016 period, σ(ROA) is the standard deviation of ROA over the same period, and μ(ROA) is the mean ROA during that particular year. Banks insolvency occurs when losses exceed equity thus if the profits are normally distributed the inverse of the probability of insolvency should equal the calculation above. The higher the Z-score the more the bank can be considered stable. The Z-Z-scores are then further normalized through the standard z-score procedure (Kreyszig, 1979), and this measure is used in my analysis as this improves the normality. The data was obtained from the Orbis Bank Focus database.
creditor runs and liquidity problems which force close down despite the equity being positive. Despite these limitations, I argue that the Z-score is the most appropriate measure for bank risk-taking.
Measuring Social Capital
There are as many methods to measuring social capital as there are definitions (Stone, 2001; Burt, 1997; Burt, 2009; Grootaert and Van Bastelaer, 2002; Onyx and Bullen, 2000; Lin, 1999; Stiglitz, Sen and Fitoussi, 2009). However, none of these methods can serve as an actual means to measure the social capital, as all have their flaws and make them unusable. The general method in the field of economics is to use indicators of trust, norms, and civic cooperation (Guiso et al., 2004; Martinez, 2012; Knack and Keefer, 1997; Xie et al., 2016). Fine (1990) notes in his book that the general economic view on the social capital topic is disconcerting, and Farr (2004) adds on this criticism that measure of trust is insufficient. These critiques raise the question, whether these indicators so often used, are the correct ones to measure social capital. According to Lattin and Lindstrom (2009) there are many more factors, that come into play, when measuring social capital within and between countries. They consider social capital as an aggregate of three subsystems: economic variables, social/cultural variables, and legal and political institutions. This is also more in line with the definition of social capital used in this thesis; the networks of relationships among people who live and work in a particular society, enabling that society to function effectively. Simply a measure of trust would not encompass the entire definition of social capital as used here. Furthermore I think this is a far superior method of measuring social capital; however, the data used by Lattin and Lindstrom (2009) is not recent enough to be usable.
Therefore, I will be using 2 rather similar rankings, for robustness reasons, the first being the social capital ranking of the Legatum Institute, an international think tank based in London and a registered UK charity. The Legatum Institute sees social capital as the strength of personal relationships, social network support, social norms, and civic participation in a society through which it has an impact on economic performance and life satisfaction. This view is in accordance with the definition of social capital as used in this paper as this indicator, as they both share that the social network in a society enables that society to function.
that is required to make sure the economy can run free of interruptions. This view of social capital is consistent with the social capital definition used here for the same reason mentioned above.
The Legatum Institute indicator of social capital is derived from three major themes; social cohesion and engagement (bridging social capital), community and family networks (bonding social capital), and political participation and institutional trust (linking social capital). These three major themes are further divided up into several subgroups of variables. The indicator factors in helping strangers, volunteering and donating to charitable organisations and how these have an impact on overall life satisfaction and economic performance. The scores are derived from 7 data points: perceptions of social support, volunteering rates, helping strangers, charitable donations, social trust, marriage and religious attendance. The data for these 7 sub-indicators are collected from real data and only where necessary is survey data used.
The SolAbility indicator of social capital is derived from 19 data points, which are made scalable and comparable that make up a single score for each country. A higher score indicates a higher level of social capital or social cohesion. The indicators selected to measure social capital have been selected from five themes (health, equality, crime, freedom and age structure). The data of some of these indicators are qualitative. The indicator was named ‘’social cohesion’’ in 2013 and thereafter ‘’social capital’’. The data from 2013 and before will however not be used as the methodology and results are significantly different therefore making them unusable. From 2014 onwards the indicator for social capital was combined from 19 data points, these are: doctors per 1000 people, hospital bed availability, nurses per 1000 people, child mortality rate, birth per woman, teen moms, overweight, life satisfaction index, Press Freedom Index, peace index, people reported to police, theft, homicide rate, prison population rate, aging society, suicide rate, public health spending, women in parliament & Human Rights Index. The scores of both indicators were further normalized to help with the normality issue through the standard procedure (Kreyszig, 1979).
helping others and charitable donations as measured by the Legatum Institute indicator, holds the most suggestive power for the word capital (Robison et al., 2010). Measurements of the amount of doctors and nurses per 1000 people and prison population can be seen as more the output of social capital. The so called the social safety net in the society which allows people to act as they want whilst making sure that others in the society have that same prosperity. This can be identified as the social part in social capital (Robison et al., 2010). Thus whilst both measuring social capital from a similar view, the indicators seem slightly so to measure different aspects of the social capital term. Slightly so because the other variables used by both the indicators are more difficult to put in any social or capital group and can be said to measure the same phenomena.
There are some limitations to this manner of measuring social capital, despite the definition in these two cases are wider than just trust they still might not be inclusive enough to capture the entire value of social capital. Moreover, it is not feasible to assume that each sub-indicators and variable that makes up a score used for the social capital are consistent in all countries, especially after something major happened in a country and only in that country that might cause a change in the social capital indicator.
Measuring Strength of Legal System
Other Variables, Macroeconomic and Bank-Level controls
This study controls for the macroeconomic environment and furthermore so at bank-level. To control for the macroeconomic environment at country level the GDP per capita and inflation rates were used. These values for the variables were obtained from the Orbis Bank Focus database for the period 2012-2016.
Controls at the bank-level include: an index that indicates whether a bank can be deemed too big to fail, the ratio of loan loss reserves, and the size of the bank. There is no unanimity among scholars for a particular size banks should have, or amount of assets or deposits on its books to be deemed too big to fail, therefore I have set the value arbitrarily at 10% of the total deposits in the entire country. Obviously, this definition has limitations as the number of 10% might be too low or too high or a bank might be considered more important due to it having clients that are of large importance to the government or the economy in general. Furthermore, with the newer regulations for bailing out banks (Powell, 2013) the number might also be inaccurate. The data for the total deposits in the countries, and the total amount of deposits that a bank has in that country is obtained from the Orbis Bank Focus database. According to Boyd and Runkle (1993) and De Nicoló (2001) bank size is an important factor influencing bank risk of failure. Predicted by theory is, that smaller firms getting less size related benefits, such as economies of scale, and are therefore more likely to fail than larger firms. However empirical results find, that large banks are deemed to be more vulnerable to failures, which can be attributed to macroeconomic externalities, and that size related benefits are offset by banks’ higher risk-taking (Greenwood, Landier and Thesmar, 2015). Therefore, the variables of bank size and the loan loss provision are often used to control, when estimating bank risk (Bushman and Williams, 2012; Laeven and Levine, 2002; Houston et al., 2010). All variables mentioned above were normalized to help with normality issues through the standard procedure (Kreyszig, 1979).
middle income into less developed countries and upper middle income and high income into the group of more developed countries following Makino et al. (2004).
Descriptive Statistics
Table II shows the descriptive statistics of all the variables used in the study. The total sample of this study consist of 11066 banks from 148 countries. It shows that the SolAbility indicator for social capital has 43312 observations, whereas the Legatum Institute for social capital has 54165 observations. The bank risk-taking indicator, the Z-score has 44259 observations. Not shown in the table is that standard Z-score is highly skewed to the right, meaning that the right tail of the distribution is longer than the left. The Z-score has a central peak and appears to be leptokurtic. To deal with this skewness the Z-score the natural logarithm is used as the risk indicator. The normalized natural logarithm of the Z-score and the standard natural logarithm of the Z-score correlate perfectly, thus for the ease of interpretation the normalized natural logarithm is used.
Table II
Descriptive statistics for bank- and country-level for 148 countries.
Variable Mean St Dev Min Max Obs
Bank-level Z-score 1.634 .557 -2.836 4.338 44,259 Too-big-to-fail .070 .256 0 1 44,630 Bank size 5.565 .881 .772 9.534 45,150 Country-level Social Capital 1 59.648 9.238 29.928 68.952 54,165 Log GDP per capita 4.432 .488 2.398 5.184 45,150 Inflation (%) .031 .063 -.037 1.217 43,312 Social Capital 2 40 6.692 21.5 63.3 43,312 LLR (%) .028 .056 -1.246 1.085 41,965 AverageR .826 .810 -2.069 2.156 54,165 AverageRL .959 1.004 -1.898 2.043 54,165 AverageCC .782 .962 -1.609 2.308 54,165
of 3 and a standard deviation around 1. The reason for the slight difference is, first that those papers look at different time period in which there were different regulations for banks that influence the Z-score, and in the case of the Houston et al. (2010) paper the period considered is far longer (average for 2000-2007). The fairly high standard deviation, and the wide range of the Z-score shown by these statistics imply, that there is a considerable amount of cross-sectional variation in the level of bank risk.
Table III shows the pairwise correlation coefficients between all variables. The correlations between the law indicators are high, which is something to be expected. As a country has a high strength of law on one front, say control of corruption, it seems very reasonable and likely that the country would score high on a different front, say rule of law. This same reasoning works for the overall indicator of strength of law and the dummy variable indicating whether the countries strength of law is above average, as it would be very likely that if it scores high on at least one of sub-scores that it would score high on the overall indicator of strength of law and thus be above the average. Rather interesting to note is the negative correlation between the two social capital indicators. This indicates that despite having a rather similar methodology, they score countries differently when it comes to social capital. Furthermore, the indicator for GDP per capita is related to the dummy variable indicating whether a country is developed; this is also very reasonable, as I decided, whether a country was developed by looking at the income class it was in. These signs of multicollinearity can therefore be ignored. There are, however, two other signs of multicollinearity: the correlation between bank size indicator (log of total bank assets) and strength of law indicators and the correlation between the Legatum Institute indicator of social capital and the strength of law indicators. Correlations coefficients above 0.8 may reflect the issue of multicollinearity in the regression according to Kennedy (2008). However, Hsiao (2003) states that panel data models reduce the problem of multicollinearity as it contains more observations.
V. RESULTS
This study tests three main hypotheses with respect to two indicators for social capital. In this part, the results of the multi-level modelling will be presented. The tables below present the results for the relationship between the social capital indicators and bank risk-taking.
Social Capital & Bank Risk-taking
are added. The control variables are added one by one, with first the second level, bank level, variables added and then the third level, country level, variables. In all estimations, the standard errors are corrected for heteroskedasticity by using heteroskedastic robust standard errors. Regardless of the model, the main results stay the same for the Legatum Institute indicator. Models 1-3 in Table 4 show a negative impact of social capital on the Z-score. It is however, only significant in the first and third model. The strength of the effect increases with the amount of control variables added to the regression. Meaning, that countries with higher levels of social capital have more risk-taking banks. The results of the third model are only significant at the ten percent significance level. Indicating that the relation is not particular strong. This contradicts my hypothesis, and the results of Xie et al. (2016), who find the opposite: namely that trust and civil norms, their measure of social capital, have a significantly positive impact on the Z-score, implying a negative impact on the bank risk-taking. These results do, however, support work in other disciplines and research fields. Glaeser et al. (1996) and Ichino and Maggi (1999) for example do find, that higher levels of social cohesion and interaction lead to riskier behaviour in accordance with the results. The coefficients in the second and third model are less strong than the coefficient in the first model this could be attributed to the fact that Legatum indicator and the indicator of the strength of law are highly correlated. However, after removing the strength of law indicators from the regression the coefficient is still insignificant (results reported in Table 1 in the appendix).
Table III
Pair wise Correlation Matrix
1 2 3 4 5 6 7 8 9 10 11 12
[1] Z-score -
[2] Too-big-to-fail -.02 -
[3] Bank size -.01 .46 -
[4] Social Capital 1 .02 -.34 -.30 -
[5] Log GDP per capita .03 -.32 -.14 .75 -
The fact that the impact of the two indicators, which measure the same thing, the social capital within a country show different effects is interesting. It may although not come at a complete surprise as the descriptive statistics show the indicators to be negatively correlated. Furthermore, as stated earlier the Legatum Institute indicator measures more the capital in the social capital, whereas the SolAbility indicator highlights the word social more. Thus both measure social capital, and have the same view on what social capital entails, but both measure it different differently as they highlight different parts of the social capital concept. This difference in methodology can explain the difference in the results. Moreover, the methodology was significantly changed three years ago for the SolAbility indicator of social capital allowing only the last three years to be used whereas the last five years for the other indicator. This last point however, cannot explain the significant effect shown by the Legatum Institute indicator and the insignificant effect of the SolAbility indicator. The outcome does show that a precise and overarching definition and measurement instrument are needed for the term social capital. The difference between outcomes indicates, that the methodology used, and the definition used for social capital are of extreme importance when doing research with regards to the topic.
Subsample Analysis: Excluding banks in the United States of America
Subsample robustness test are conducted to prevent some large banks to exert an overly strong influence on the results. Given that the USA has the largest number of banks in the sample (≈81%), I follow Houston et al. (2010) and re-estimate the models after dropping the data from US banks.
The exclusion of the USA has a slight dampening effect on the impact social capital on the Z-score for the SolAbility indicator of social capital. The results indicate that the impact of social capital on the Z-score is still negative for the Legatum Institute indicator, however it is now stronger than in the previous sample.
Table IV
Results Regression social capital on bank risk-taking
This table shows the results of the regression with and without the inclusion of the firm and country fixed variables, using the entire sample covering the period 2012-2016 for the first 3 models and 2014-2016 for the last 3 models. Data regarding bank characteristics are retrieved from the Orbis Bank Focus database and the data regarding the social indicators are retrieved from the Legatum Institute and the SolAbility Sustainable Intelligence Company. Data regarding country strength of law characteristics are retrieved from Worldwide Governance Indicators. Columns 1,2 and 3 show results from the Z-score regressed on the first indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL and AverageCC. Columns 4,5 and 6 show results from the Z-score regressed on the second indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL and AverageCC. Z-score is Equals log of (ROA + CAR)/σ(ROA), where ROA = π/A is return on assets and CAR = E/A is capital-asset ratio. Bank size is the natural logarithm of total assets. LLR(%) is loan loss reserves divided by gross loans. Too-big-to-fail is a dummy variable that takes a value of one if the bank’s share in the country’s total deposits exceeds 10%. Log GDP per capita is the log real GDP per capita, in US dollars. Inflation(%) is in percentage the inflation rate. AverageR , AverageRL and AverageCC are strength of law indicators for respectively regulatory quality, rule of law and corruption control. The first indicator of social capital is the Legatum Institute measure of social capital. The second indicator of social capital is the SolAbility measure of social capital. ***, **, * Represent significance at the 1%, 5% and 10% levels
respectively. Robust standard errors are shown in parentheses under the coefficients.
Z-score (1) (2) (3) (4) (5) (6) Social Capital 1 -.034** (.017) -.029 (.025) -.059* (.034) Social Capital 2 .022** (.009) .013 (.011) .014 (.011) Too-big-to-fail .040** (.018) .046** (.022) .039 (.027) .044 (.026) Bank size -.219*** (.019) -.176*** (.038) -.125*** (.044) -.134*** (.051) LLR (%) -.059* (.033) -.046* (.025) -.032* (.017) -.031* (.017) Inflation (%) -.010* (.007) -.010** (.005) Log GDP per capita -.030 (.050) -.033 (.047) AverageR .102 (.114) .079 (.112) AverageRL -.068 (.244) -.103 (.237) AverageCC .152 (.187) .139 (.180) Constant -.332*** (.053) -.136** (.068) -.027 (.079) -.309*** (.047) -.175*** (.054) -.054 (.091)
The results for the SolAbility indicator of social capital are similar to those without the exclusion of the US banks. They once again indicate a positive impact of social capital on the Z-score. The results are again not significant in the last two models. Indicating again, that there is no significant impact of social capital on the Z-score, for the SolAbility indicator, even with the exclusion of the USA.
on bank risk-taking when regressing with the Legatum Institute indicator, this relation being more significant with the exclusion of the USA from the sample. The relation for the SolAbilty indicator and bank risk-taking is not significant at any level. Meaning that for the Legatum Institute indicator it is shown that higher levels of social capital are associated with lower Z-scores, thus higher levels of social capital lead to more bank risk-taking and for the SolAbility indicator no significant relation is found.
Table V
Results Regression social capital on bank risk-taking subsample
This table shows the results of the regression with and without the inclusion of the firm and country fixed variables, using the subsample that excludes the USA covering the period 2012-2016 for the first 3 models and 2014-2016 for the last 3 models. Data regarding bank characteristics are retrieved from the Orbis Bank Focus database and the data regarding the social indicators are retrieved from the Legatum Institute and the SolAbility Sustainable Intelligence Company. Data regarding country strength of law characteristics are retrieved from Worldwide Governance Indicators. Columns 1,2 and 3 show results from the Z-score regressed on the first indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL and AverageCC. Columns 4,5 and 6 show results from the Z-score regressed on the second indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL and AverageCC. Z-score is Equals log of (ROA + CAR)/σ(ROA), where ROA = π/A is return on assets and CAR = E/A is capital-asset ratio. Bank size is the natural logarithm of total capital-assets. LLR(%) is loan loss reserves divided by gross loans. Too-big-to-fail is a dummy variable that takes a value of one if the bank’s share in the country’s total deposits exceeds 10%. Log GDP per capita is the log real GDP per capita, in US dollars. Inflation(%) is in percentage the inflation rate. AverageR , AverageRL and AverageCC are strength of law indicators for respectively regulatory quality, rule of law and corruption control. The first indicator of social capital is the Legatum Institute measure of social capital. The second indicator of social capital is the SolAbility measure of social capital. ***, **, * Represent significance at the 1%, 5% and 10% levels respectively. Robust standard errors are shown in parentheses under the coefficients.
The Effects of a Strong Legal System on Social Capital
In this section, I consider the effect that the strength of the legal system has on the impact of social capital on bank risk-taking. I will test the previous regression with the inclusion of the interaction variable, to see, if a strong legal system strengthens the effects of social capital on bank risk-taking. Therefore, the dummy variable Strength of Law and the interaction variable Inter Strength of Law are added to the model for the Legatum Institute indicator and for the SolAbility indicator. The dummy variable indicates whether the legal system of that country is above average, it then receives a value of 1, and otherwise it takes a value of 0. The overall measure of the averages of control of corruption, rule of law and regulatory quality are used to determine if a country is above or below the average. The interaction variable is a term measured by dummy times the indicator of social capital. Once again, a subsample without the USA is used, as its use was justified in the results above.
The first model presented in Table 6 is the regression between the social capital indicators, bank risk-taking, and the interaction variable on the rule of law. The first model finds a positive relation, and indicates, that a strong legal system does strengthen the impact of social capital on bank risk-taking. The inclusion of the macroeconomic control variables, however switches the sign of the coefficient and furthermore decrease the significance to levels where it is not significant. Meaning that the results indicate that there is no added effect of the strength of the regulatory system to the impact social capital has on bank risk-taking. This is not in line with my hypothesis and the findings by Xie et al. (2016). Therefore, my second hypothesis cannot be supported.
Table VI
Results regression impact strength of legal system on social capital
This table shows the results of the regression of the impact the strength of the legal system has on social capital with and without the inclusion of the firm and country fixed variables, using the entire sample covering the period 2012-2016 for the first 3 models and 2014-2016 for the last 3 models. Data regarding bank characteristics are retrieved from the Orbis Bank Focus database and the data regarding the social indicators are retrieved from the Legatum Institute and the SolAbility Sustainable Intelligence Company. Data regarding country strength of law characteristics are retrieved from Worldwide Governance Indicators. Columns 1,2 and 3 show results from the Z-score regressed on the first indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL, AverageCC Strength of Law and Inter Strength of Law. Columns 4,5 and 6 show results from the Z-score regressed on the second indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL and AverageCC, Strength of Law and Inter Strength of Law. Z-score is Equals log of (ROA + CAR)/σ(ROA), where ROA = π/A is return on assets and CAR = E/A is capital-asset ratio. Bank size is the natural logarithm of total assets. LLR(%) is loan loss reserves divided by gross loans. Too-big-to-fail is a dummy variable that takes a value of one if the bank’s share in the country’s total deposits exceeds 10%. Log GDP per capita is the log real GDP per capita, in US dollars. Inflation(%) is in percentage the inflation rate. AverageR , AverageRL and AverageCC are strength of law indicators for respectively regulatory quality, rule of law and corruption control. Strength of Law is a dummy variable indicating whether a countries’ strength of law is above average. Inter Strength of Law is a interaction term of Strength of Law and Social Indicator 1 & 2. The first indicator of social capital is the Legatum Institute measure of social capital. The second indicator of social capital is the SolAbility measure of social capital. ***, **, * Represent significance at the 1%, 5% and 10% levels respectively. Robust standard errors are shown in parentheses under the coefficients.
Z-score (1) (2) (3) (4) (5) (6) Social Capital 1 -.065*** (.022) -.055 (.034) -.067 (.041) Social Capital 2 .017 (.013) .010 (.018) .015 (.017) Inter Strength of Law 1 .060 (.039) .043 (.040) -.012 (.048) Inter Strength of Law 2 .006 (.020) .004 (.025) -.001 (.026) Too-big-to-fail .039** (.018) .047** (.023) .039 (.028) .045* (.027) Bank size -.220*** (.019) -.177*** (.038) -.127*** (.045) -.133*** (.051) LLR (%) -.059* (.033) -.046* (.026) -.033* (.018) -.031* (.018) Inflation (%) -.011* (.006) -.011* (.006) Log GDP per capita .034 (.050) .038 (.049) AverageR .136 (.114) .114 (.113) AverageRL -.008 (.238) -.042 (.227) AverageCC .170 (.189) .162 (.182) Constant -.441*** (.067) -.310*** (.093) -.182 (.183) -.353*** (.059) -.258*** (.064) -.180 (.171)
Subsample Analysis: Excluding banks in the United States of America
variables. The results of the first two models of the subsample indicate that there is a positive effect. Meaning that there is a positive effect of the regulatory system on the impact of social capital on bank risk-taking, this is in line with the second hypothesis. The effect in the last model with the inclusion of the macroeconomic controls show a negative effect, which is not in line with the second hypothesis. These results of the first two models are however of little significance and therefore offer little to no proof to the second hypothesis. The result of the third model is also of no significance. The results of the SolAbility indicator in the models 4, 5 and 6 show a coefficient for the interaction variable that is not significant. Indicating that that the interaction variable strength of law has no effect even with the exclusion of the USA.
The Effect of Being Developed on Social Capital
In this section I consider the effect of a country being developed has on the impact social capital has on bank risk-taking. To test the hypothesis, that social capital has a larger effect on bank risk-taking in less developed countries I have created the interaction variable Inter Developed between the dummy that indicates whether a country is developed and both the indicators of social capital. The dummy variable gives a value of 1 to countries that are developed and a value of 0 to those that are not. The developed countries are made up of upper middle and high-income countries whereas the lesser developed countries are made up of countries that fall into the lower and lower middle-income classes.
The first model presented in Table 8 is the regression between the Legatum Institute indicator, bank risk-taking and the interaction variable. The first model finds a positive relation and indicates the fact, that a developed country does strengthen the impact of social capital on bank risk-taking. The inclusion of the bank-level control variables however switches the sign of the coefficient and furthermore decrease the significance to levels where it is not significant. The sign is switched again with the inclusion of the macroeconomic controls added to the model, the significance remains below the threshold. This means that there is no added impact of social capital on bank risk-taking through the country being developed. This is not in line with my third hypothesis.
not significant. Showing that there is no effect of a country being developed or not found on the impact of social capital on bank risk-taking.
Table VII
Results regression impact strength of legal system on social capital subsample This table shows the results of the regression of the impact the strength of the legal system has on social capital with and without the inclusion of the firm and country fixed variables, using the subsample that excludes the USA covering the period 2012-2016 for the first 3 models and 2014-2016 for the last 3 models. Data regarding bank characteristics are retrieved from the Orbis Bank Focus database and the data regarding the social indicators are retrieved from the Legatum Institute and the SolAbility Sustainable Intelligence Company. Data regarding country strength of law characteristics are retrieved from Worldwide Governance Indicators. Columns 1,2 and 3 show results from the Z-score regressed on the first indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL, AverageCC, Strength of Law and Inter Strength of Law. Columns 4,5 and 6 show results from the Z-score regressed on the second indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL, AverageCC, Strength of Law and Inter Strength of Law . Z-score is Equals log of (ROA + CAR)/σ(ROA), where ROA = π/A is return on assets and CAR = E/A is capital-asset ratio. Bank size is the natural logarithm of total assets. LLR(%) is loan loss reserves divided by gross loans. Too-big-to-fail is a dummy variable that takes a value of one if the bank’s share in the country’s total deposits exceeds 10%. Log GDP per capita is the log real GDP per capita, in US dollars. Inflation(%) is in percentage the inflation rate. AverageR , AverageRL and AverageCC are strength of law indicators for respectively regulatory quality, rule of law and corruption control. Strength of Law is a dummy variable indicating whether a countries’ strength of law is above average. Inter Strength of Law is a interaction term of Strength of Law and Social Indicator 1 & 2. The first indicator of social capital is the Legatum Institute measure of social capital. The second indicator of social capital is the SolAbility measure of social capital. ***, **, * Represent significance at the 1%, 5% and 10% levels respectively. Robust standard errors are shown in parentheses under the coefficients.
Table VIII
Results regression impact state of development on social capital
This table shows the results of the regression of the impact the state of development of a country has on social capital with and without the inclusion of the firm and country fixed variables, using the entire sample covering the period 2012-2016 for the first 3 models and 2014-2016 for the last 3 models. Data regarding bank characteristics are retrieved from the Orbis Bank Focus database and the data regarding the social indicators are retrieved from the Legatum Institute and the SolAbility Sustainable Intelligence Company. Data regarding country strength of law characteristics are retrieved from Worldwide Governance Indicators. Columns 1,2 and 3 show results from the Z-score regressed on the first indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL, AverageCC, Developed and InterDeveloped. Columns 4,5 and 6 show results from the Z-score regressed on the second indicator of social capital, Too-big-to-fail, Bank size, LLR(%), Inflation(%), Log GDP per capita, AverageR, AverageRL and AverageCC, Developed and InterDeveloped. Z-score is Equals log of (ROA + CAR)/σ(ROA), where ROA = π/A is return on assets and CAR = E/A is capital-asset ratio. Bank size is the natural logarithm of total assets. LLR(%) is loan loss reserves divided by gross loans. Too-big-to-fail is a dummy variable that takes a value of one if the bank’s share in the country’s total deposits exceeds 10%. Log GDP per capita is the log real GDP per capita, in US dollars. Inflation(%) is in percentage the inflation rate. AverageR , AverageRL and AverageCC are strength of law indicators for respectively regulatory quality, rule of law and corruption control. Developed is a dummy variable indicating whether a country is developed. The first indicator of social capital is the Legatum Institute measure of social capital. The second indicator of social capital is the SolAbility measure of social capital. InterDeveloped is the interaction term of Developed and Social Indicator 1 & 2. ***, **, * Represent significance at the 1%, 5% and 10% levels respectively. Robust standard errors are shown in parentheses under the coefficients.
Z-score (1) (2) (3) (4) (5) (6) Social Capital 1 -.079*** (.025) -.051** (.029) -.074** (.031) Social Capital 2 .012 (.012) .004 (.016) .016 (.011) Inter Developed 1 .066** (.041) -.027 (.051) .019 (.061) Inter Developed 2 .013 (.017) .010 (.018) -.001 (.018) Too-big-to-fail .040** (.019) .046** (.022) .040 (.028) .044* (.026) Bank size -.219*** (.019) -.176*** (.040) -.127*** (.046) -.134*** (.051) LLR (%) -.059* (.033) -.046* (.026) -.032* (.017) -.031* (.018) Inflation (%) -.010* (.006) -.011* (.006) Log GDP per capita .028 (.057) .038 (.056) AverageR .099 (.125) .088 (.123) AverageRL -.068 (.250) -.109 (.238) AverageCC .149 (.189) .141 (.179) Constant -.523*** (.076) -.384** (.097) -.071 (.192) -.411*** (.065) -.315*** (.066) -.026 (.164) VI. CONCLUSION
