• No results found

The effect of financial stability transparency on financial stability: does the quality of institutions matter?

N/A
N/A
Protected

Academic year: 2021

Share "The effect of financial stability transparency on financial stability: does the quality of institutions matter?"

Copied!
54
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

1

The effect of financial stability transparency

on financial stability: does the quality of

institutions matter?

by

Tim van Duuren Student number: S2717743 Email: t.o.van.duuren@student.rug.nl MSc International Economics & Business

University of Groningen Faculty of Economics and Business

Supervisor: Prof. dr. J. de Haan Co-assessor: dr. A.C. Steiner

(2)

2

Abstract

This study analyses the effect of financial stability transparency on financial stability and tests whether this relationship depends on the perceived institutional quality by the national population. The empirical analysis is based on panel data from 110 countries for the period 2000-2011. The study uses fixed effects models with different indicators of financial stability. In addition, 3-year lagged independent variables and system GMM estimators are conducted as robustness checks. The results imply that more financial stability transparency increases the degree of financial stability in a country. However, the robustness checks do not confirm all these findings. Furthermore, the study finds evidence that the effect of financial stability transparency on financial stability is conditional on institutional quality. Even though mixed results have been found on which levels of institutional quality financial stability transparency has the strongest effect on financial stability.

Keywords: financial stability, transparency, quality of institutions

JEL Classifications: E58, E61, G28

Acknowledgements

(3)

3

Table of contents

1. Introduction ... 4

2. Literature review ... 7

2.1. Financial stability transparency ... 7

2.2. Relationship between financial stability transparency and financial stability ... 8

2.3. Quality of institutions ... 10

3. Methodology ... 13

3.1. Measuring financial stability and transparency ... 13

3.2. Measuring institutional quality ... 14

3.3. Empirical models ... 15

3.4. Panel data assumptions ... 17

4. Data ... 19 4.1. Data collection ... 19 4.2. Data description ... 21 5. Empirical results ... 24 5.1. Main results ... 24 5.2. Robustness checks ... 30

6. Discussion and limitations ... 34

7. Concluding remarks ... 37

References ... 39

Appendix A – Construction of the FST-index ... 43

Appendix B - Annually FST-index per country ... 44

Appendix C - Averages of the CP-index and the GE-index per country. ... 46

(4)

4

1. Introduction

The importance of financial stability gained renewed interest during the recent Global Financial Crisis (GFC), which exposed financial imbalances and the associated economic costs as a result of these imbalances (Born, Ehrmann, & Fratzscher, 2014). In addition to the existing objectives of long-term price stability and sustainable long-term growth, several central bank mandates were extended with the objective of maintaining financial stability (Horváth & Vaško, 2016). The GFC exposed that monetary policy is not able to simultaneously achieve both price stability and financial stability fully (Fahr & Fell, 2017). Tinbergen (1952) was the first to notice that achieving independent policy objectives can only be realized if the same number of instruments is available for the same number of defined objectives. Based on the work of Tinbergen (1952), Fahr & Fell (2017) argued that financial cycles may be considered as the underlying cause for the inability of monetary policy to achieve price- and financial stability simultaneously. Several studies emphasize the importance of financial cycles, because they do not follow the same pattern as business cycles, they are longer in duration and they have a greater amplitude than business cycles (Borio & Drehmann, 2009; Claessens, Kose, & Terrones, 2011; Schüle, Hiebert, & Peltonen, 2015). Therefore, these differences show that a certain monetary policy may be beneficial for price stability while this policy may have negative effects on financial stability, because the business- and financial cycles may be desynchronized (Fahr & Fell, 2017).

As a consequence, macro-prudential policies gained renewed attention to ensure financial stability because they allow central banks to pursue price stability on the one hand and it provides the ability to influence financial stability on the other hand (Lombardi & Siklos, 2016). Macro-prudential policies are aimed at keeping the systemic risk in the financial system as small as possible by reducing the risks that arise from the interconnectedness of financial actors in the financial system and the presence of financial procyclicality (Claessens, 2015). In addition, the author argues that monetary policies and macro-prudential policies interact with each other via the same transmission channels. This means that achieving price stability may reduce financial stability and vice versa (Claessens, 2015). The authors argue that there is no consensus yet on the optimal combination of monetary policies and macro-prudential policies to achieve price stability and financial stability fully simultaneously. Therefore, it is valuable to examine instruments which could potentially impact one of these objectives (Popoyan, Napoletano, & Roventini, 2017).

(5)

5

central banks, that do not have financial stability as an objective in their mandate, have a keen interest in maintaining financial stability (Horváth & Vaško, 2016).

Several studies have tried to explore the relationship between financial stability transparency and financial stability (Oosterloo, de Haan, & Jong-A-Pin, 2007; Born et al., 2014; Čihák, Muñoz, Teh Sharifuddin, & Tintchev, 2012; Horváth & Vaško, 2016). The first empirical study of Oosterloo et al. (2007) on the relationship between FSRs and financial stability did not discover a significant relationship. Čihák et al. (2012) found that high-quality FSRs are associated with having an impact on the financial soundness of a country, but they did not find conclusive evidence that FSRs solely have an effect on financial stability. In contrast, Born et al. (2014) found empirical evidence that stock markets’ volatility is expected to be lower when the content of the communication tools are perceived to be optimistic. Furthermore, Horváth & Vaško (2016) designed a Financial Stability Transparency index (FST-index) and showed that this index has a positive influence on financial stability. In addition, the authors found evidence that overabundant financial stability transparency has a negative impact on financial stability. Furthermore, van der Cruijsen, Eiffinger, & Hoogduin (2010) also found an optimal level of central bank transparency concerning monetary policy transparency. In general, Kaufmann & Weber (2010) argue that the objective of central bank transparency must focus on the quality of information rather than the quantity, which is in line with the previously mentioned optimal levels of financial stability transparency.

Furthermore, the perceptions of the quality of institutions by the national population are considered to influence the relationship between having an inflation targeting (IT) regime and financial stability (Fazio, Silva, Tabak, & Cajueiro, 2018). The authors found evidence that the effect of IT policies on financial stability is positive at average levels of institutional quality. Furthermore, Klomp & de Haan (2014) found that imposing stricter capital regulations and supervisory control reduces banking risk, and therefore has a higher impact on financial stability. The effect is stronger for countries with higher institutional quality. In addition, the authors found that regulations on bank activities and liquidity decreases banking risk, although this effect is only discovered for higher values of the measure used for institutional quality. As IT monetary policies consist of a commitment to transparency and price stability, and the effect of these policies depend on the level of quality of the institutions (Mishkin, 2004; Fazio et al., 2018), there is reason to assume that financial stability transparency also depends on the quality of institutions. Furthermore, Etzioni (2010) argues that strong transparency is a form of regulation. Therefore, since banking regulations and supervision depend on institutional quality (Klomp & de Haan, 2014), it is highly likely that financial stability transparency also depends on the quality of institutions.

(6)

6

the financial stress index from the International Monetary Fund (IMF) and a dummy variable for the banking crisis as indicators for financial stability. Various alternative measures have been used for financial stability in literature, because there is not one operational measure for financial stability. The indicators for financial stability used in this study are: the ZA-score, ZE-score, and the leverage ratio; this choice is based on previous studies related to financial stability (Berger, Klapper, & Turk-Ariss, 2008; Lee & Hsieh, 2014; Fazio et al., 2018). In addition, this study will use the bank non-performing loans following the work of Horváth & Vaško (2016) to examine the effect of the quality of institutions on the relationship between financial stability transparency and financial stability. The study addresses the following research question:

‘’How does the perception of the quality of institutions affect the relationship between central bank transparency and financial stability in a country?’’

The empirical analysis will be executed by using fixed effects regressions to examine the conditional effect of the quality of institutions on the relationship between financial stability transparency and financial stability for the timeframe of 2000-2011. Data for this paper have been obtained from external sources. The FST-index constructed by Horváth & Vaško (2016) for 110 countries is used. Furthermore, data from the World Bank (2018a; 2019), the International Monetary Fund (2019) and the updated External Wealth of Nations Mark II database of Lane & Milesi-Ferretti (2007) will be used to create the different indicators for financial stability and to obtain data for the control variables. In addition, data on the quality of institutions will also be obtained from the World Bank (2018b) and Transparency International (2019).

The findings of the paper are as follows. Higher financial stability transparency reduces bank non-performing loans in a country. Although, there exists an optimal level of financial stability transparency concerning bank non-performing loans, since too much transparency may increase bank non-performing loans. Furthermore, evidence has been found that the effect of financial stability transparency on financial stability depends on the quality of institutions. Although, mixed results have been found on which levels of institutional quality, the relationship between financial stability transparency and financial stability is influenced. For countries with low institutional quality, financial stability transparency reduces bank non-performing loans, reduces the probability of default for banks, and increases the leverage ratio. At high levels of institutional quality, financial stability transparency increases the operational profitability of banks.

(7)

7

2. Literature review

This section provides an overview of the concepts and theories, which have been mentioned in the introduction. Furthermore, this section explains how these concepts are defined and provides the main literature on the specific research areas to answer the research question proposed in the introduction.

2.1. Financial stability transparency

Firstly, there is no precise definition of financial stability which is commonly used in literature. Houben, van der Molen, & Wierts (2012) define financial stability as “the ability of the financial system to help the economic system allocate resources, manage risks and absorb shocks”. To provide a more specific operational definition of financial stability is challenging, because there might be different relevant authorities involved in the process and safeguarding financial stability often relies on various instruments (Born et al., 2014). In line with the study of Horváth & Vaško (2016), this study narrows its focus down to the banking sector as part of the financial system, because the banking sector heavily influences the soundness of the financial system in a country (Das et al., 2004). In addition, Čihák & Schaek (2010) found empirical evidence that using aggregated banking data for the variables of financial stability which are used in this research and consist of bank non-performing loans, leverage ratio and return on assets and equity, are able to identify systemic banking problems.

In the early stages of upcoming interest in financial stability transparency, its definition was associated with the extent of publishing FSRs, because publishing FSRs was the main communication channel to safeguard financial stability (Oosterloo & de Haan, 2004; Oosterloo et al., 2007). As time went by, the set of central banks financial stability communication tools was extended by interviews, speeches, stress tests and financial stability indicators which are often included in FSRs (Born et al., 2014; Horváth & Vaško, 2016). All these communication tools, which are specifically aimed at providing information on financial stability, became useful instruments to guide and caution the financial markets (Horváth & Vaško, 2016). This section of the paper explains the relevance of the various parts in the FST-index based on existing literature. Exploring the relevance of being transparent will help to gain an understanding of why individual sections are added to the FST-index and how being transparent may affect financial stability (Horváth & Vaško, 2016).

(8)

8

the past might contribute to an increase in the publications of FSRs, because central banks may be more aware of and also want to focus on financial stability after a period of crisis. Besides, they concluded that Gross Domestic Product (GDP) per capita is an indicator for publishing FSRs, because higher GDP per capita and higher financial development are often related. A more sophisticated financial system may create a higher need for a country to care about financial stability compared to countries with an underdeveloped financial system (Oosterloo et al., 2007). Furthermore, their results suggest that being a member of the EU increases the likelihood of publishing FSRs, due to the loss of monetary responsibilities when they joined the euro area. Therefore, EU countries are more likely to focus on financial stability and are less focused on maintaining price stability, since it is the responsibility of the European Central Bank. Horváth & Vaško (2016) extended the work of Oosterloo et al. (2007) and found that being transparent in financial stability related activities could be explained by being transparent in their monetary policies. In addition, the authors also found that the more developed a country is in terms of GDP per capita, the higher the degree of financial stability transparency.

Besides the studies which focused on the determinants of FSRs, Čihák et al. (2012) examined criteria of the composition of FSRs. They argue that forward-looking is essential for the effectiveness of FSRs for achieving financial stability, because it enables the system to anticipate on upcoming risks. In addition, the authors argue that the coverage of FSRs must contain information on key risks and vulnerabilities of the financial system. Furthermore, Čihák et al. (2012) also assessed the content of FSRs on consistency and clarity, however these criteria seem rather subjective. Therefore, Horváth & Vaško (2016) have included the publishing, the frequency, the coverage and the extent of forward looking of FSRs in their FST-index, because these sections ensure the objectivity of the FST-index. Furthermore, they included the additional publication of stress tests and information on the Financial Stability Indicators (FSI) created by central banks. These stress test and indicators provide additional quantitative assessments about the state of the financial system. Furthermore, Born et al. (2014) noticed the increase in interviews and speeches as communication tools for financial stability. Therefore, Horváth & Vaško (2016) also included the presence of a database for speeches and website information on financial stability into their FST-index. Furthermore, Horváth & Vaško (2016) also added a general section to the framework since macro-prudential policies can help to reduce systematic risks with respect to financial stability and transparency in this field may be valuable (Houben et al., 2012). When central banks are not legally obliged to pursue financial stability, this does not mean that they are not concerned about financial stability. However, the extent to which central banks are committed to safeguarding financial stability may vary between legally obliged and non-legally-obliged central banks (Horváth & Vaško, 2016). Furthermore, the authors assigned index points to countries that have an independent financial stability committee with a primary focus on safeguarding financial stability. A more extensive description of the FST-index is provided in Appendix A.

2.2. Relationship between financial stability transparency and financial stability

(9)

9

devoted to the field of monetary policy transparency in the past two decades, since price stability is the main objective of most central banks (Blinder et al., 2008; Geraats, 2009; Dincer & Eichengreen, 2014). The link between monetary policy transparency and financial stability has not been empirically explored, because monetary policy transparency is mostly aimed at creating price stability. There have been some early studies, which tried to find a relationship between publishing FSRs and the financial stability situation of a country, with different results.

Oosterloo et al. (2007) were the first who tried to explore the relationship between an indicator of FSRs, which was seen as an indicator of FSR transparency, and financial stability. The authors did not find a significant relationship between their created FSR transparency indicator and the two measures they used for financial soundness. They used Moody’s weighted average bank financial strength indicator and a financial system indicator created by Das et al. (2004) as measures for financial soundness. Oosterloo et al. (2007) argued that there might not be an existing relationship between FSR transparency and financial stability.

After the period of renewed interest in FSRs, Čihák et al. (2012) conducted research with the main aim to see whether quality of the FSRs played a specific role concerning the impact on financial stability. The authors composed a quality indicator concerning FSRs and rated them on the following criteria: the clarity, the coverage of the key risk in the financial system, and the consistency of the FSRs. They estimated a probit model and a random GLS panel and found no clear evidence for an empirical relationship between publishing FSRs and the financial stability in a country. In contrast, the authors did find that high-quality FSRs are associated with higher financial stability.

Beyond the associated influence of the quality of the FSRs on the relationship of publishing FSRs and financial stability, Born et al. (2014) found empirical evidence for a reduction of uncertainty in the stock market in case of the content of FSRs being optimistic. The authors gave 1.000 FSRs an optimism rating and they answered their hypotheses using event study methodology. The results of Born et al. (2014) may be interpreted in a way that optimistic FSRs have an indirect positive effect on a stable financial environment (Horváth & Vaško, 2016).

(10)

10

Since there is little empirical research done in the field of financial stability transparency, this paper is trying to establish more robust results by using different measures of financial stability than Horváth & Vaško (2016) have used in their research. Based on the results found in previous studies (Čihák et al., 2012; Horváth & Vaško, 2016) the following hypothesis will be tested:

H1) Financial stability transparency is positively related to financial stability in a country

However, too much central bank transparency concerning monetary policy transparency may result in a less stable financial environment (van der Cruijsen, Eiffinger, & Hoogduin, 2010). The authors argue that confusion or information overload and uncertainty could contribute to the creation of an optimal level for monetary policy transparency. Van der Cruijsen et al. (2010) argue that confusion or information overload may make it difficult for economic agents to filter the necessary information needed. Also, the authors argue that uncertainty may reduce the quality of transparency as perceived by economic agents. Horváth & Vaško (2016) found evidence that too much financial stability transparency results in lower financial stability. Following the results of van der Cruijsen et al. (2010) and Horváth & Vaško (2016), the following hypothesis will be tested with various measures for financial stability:

H2) Financial stability transparency has a non-linear relationship with financial stability in a country

2.3. Quality of institutions

(11)

11

study of Das et al. (2004) and Fazio et al. (2018) the following hypothesis will be tested to explore whether the quality of institutions causes financial stability:

H3) The quality of institutions is positively related to financial stability in a country

Klomp & de Haan (2014) empirically tested whether the effect of banking regulation and supervision on banking risk is conditional on the quality of institutions. The authors found that capital regulations and supervisory control have a negative impact on banking risk for countries with low institutional quality and the negative effect becomes stronger when the quality of institutions increases. In addition, they found that the effect of restrictions on banking activities and stricter liquidity regulations on banking risk is conditioned on institutional quality. Although, these findings are only significant for countries with higher levels of institutional quality. Furthermore, the authors tested whether other measures of banking regulation and supervision, which consist of deposit insurance power, private monitoring, and market entry, are conditional on the quality of institutions. However, they did not find significant results that these measures depend on the quality of institutions.

Furthermore, Fazio et al. (2018) found that the quality of institutions has a non-linear effect on the relationship between having an IT regime and financial stability. The optimal level is at average values of the measures for the quality of institutions, because at these values the quality of institutions has a positive impact on the relationship between having an IT regime and financial stability. They argue that there must be a minimum level of trust in the institutions in order to let the policies of institutions be effective. Furthermore, they found that IT countries with high-quality institutions may suffer from the ‘paradox of credibility’. Borio & Lowe (2002) introduced the concept of the paradox of credibility concerning the credibility of monetary policies. The authors found evidence that a high level of credibility can result in a more unstable financial environment. Borio & Lowe (2002) argue that too much credibility may drive a wedge between the actual economic fundamentals, such as inflation, and the perceptions and expectations of economic actors about the state of the economy. This wedge may arise, due to the credibility of institutions which determines the capacity to affect the expectations of the public on the future inflation values (Montes & Peixoto, 2014). Therefore, central banks, which are perceived highly credible, have to put in less effort to achieve the desired inflation level. In other words, fewer movements in interest rates are observed (De Mendonça & De Guimarães e Souza, 2009). However, this stable economic environment may enhance risk-taking of banks, because this stable economic environment may lead to lower interest rates and therefore stimulating the creation of bubbles, which often occur firstly in the housing market (Fazio et al., 2018). Therefore, high credibility of institutions may amplify financial procyclicality and it can harm the financial system as it may create financial imbalances (Borio, 2005). However, it is questionable whether this line of reasoning can be applied to financial stability transparency.

(12)

12

between actual economic fundamentals and expectations of economic actors, which may lead to excessive credit creation and bubbles. FSRs, interviews, and speeches are issued to be transparent about financial stability and macro-prudential policies and this might influence the expectations and decision making of economic actors. These instruments for financial stability transparency are aimed at preserving financial stability, instead of creating financial imbalances. Therefore, the concept of the paradox of credibility does not seem applicable to the effect of the quality of institutions on the relationship between financial stability transparency and financial stability. Nevertheless, the linear-interactive models, which will be presented, may indicate whether high-institutional quality has a negative impact on the relationship between financial stability transparency and financial stability.

Based on the established conditional effect of institutional quality on the relationship between having an IT regime and financial stability and the argument of Fazio et al. (2018) that countries with high-quality institutions are more easily able to cope with shocks that may disturb financial stability, an increase in the FST-index will have a lower effect on financial stability for countries with high-quality institutions. Therefore, the effect of financial stability transparency on financial stability for countries with high-quality institutions may be lower than for countries with low-quality institutions. Henceforth, the following hypothesis will be tested:

H4) The effect of financial stability transparency on financial stability is conditional on the quality of institutions; at low levels of institutional quality financial stability transparency has a stronger positive impact than at high levels of institutional quality

All in all, the conceptual model presented in figure 1 provides a visual overview of the hypotheses which will be tested.

Figure 1: Conceptual model

(13)

13

3. Methodology

The first part of this section elaborates on the measures which are used to capture financial stability, financial stability transparency and the quality of institutions. Also, an overview of the empirical models is presented. Hereafter, the data are tested to justify the specifications of the models.

3.1. Measuring financial stability and transparency

As argued previously, there is no specific operational definition of financial stability. Therefore, several measures are used in this study to capture the extent of financial stability in a country. Furthermore, using different measures of financial stability may increase the validity of the results. In line with several studies (Berger et al., 2008; Lee & Hsieh, 2014; Horváth & Vaško, 2016; Fazio et al., 2018), the ratio of bank non-performing loans to total gross loans and the aggregated bank Z-scores of countries are used as measures for financial stability. In addition, this study explores the leverage ratio of banks as an indicator of financial stability (Berger et al., 2008).

The ratio of bank performing loans to total gross loans, hereafter called non-performing loans (NPL), is the ratio of defaulting loans to the total value of the loan portfolio. Following the study of Horváth & Vaško (2016), the non-performing loans will be transformed as follows:

!"#$% = '(( *+,-.

(/0*+,-.)) (1)

where NPLit is the ratio of aggregated bank non-performing loans for country i in year t. This transformation deals with the skewness of the ratio of the non-performing loans to total gross loans. As a result of this transformation, a low value of non-performing loans will indicate a higher degree of financial stability in a country, because there are fewer bad loans and therefore banks face fewer credit risks (Čihák & Heiko, 2010; Horváth & Vaško, 2016).

(14)

14

average 76% of all bank failures can be predicted by using the ZA-score. Laeven & Levine (2009) use a natural logarithm for the ZA-score, because the ZA-score is skewed. In line with the studies of Laeven & Levine (2009) and Fazio et al. (2018), the natural logarithm of the ZA-score will be used in this study. For purposes of brevity, the study uses the label ZA in referring to the natural logarithm of the ZA-score. The ZA-score is calculated as follows:

23$% = '( 4(567<(567)-.8 ,:;-.)

-. =

(2)

where the ROAit is the yearly average return on assets of banks in country i in time (year) t. In addition, the σ(ROA)it represents the standard deviation of this aggregated return on assets for a country and it is calculated for the period of 2000-2011. Furthermore, LEVit is defined as the aggregated bank’s equity to total assets ratio for a country at time t. In line with Berger et al. (2008) and Laeven & Levine (2009), this study uses the standard deviation of the return on assets of the sample period under study. Lepetit & Strobel (2013) argue that the standard deviation of the return on assets of the whole sample under study is more appropriate to use instead of a moving window for 3 or 5 years, because it does not drop initial observations.

Following Lee & Hsieh (2014), the aggregated return on assets of banks is replaced by the aggregated return on equity (ROE) in Eq. 2, which will result in the ZE-score for return on equity (ZE). The authors argue that return on equity is often used as a measure of profitability and it contains information on operational efficiency and loan loss provision of the banks. In line with the reasoning of the ZA-score, a higher ZE-score will indicate higher profitability and a lower distance to default and therefore it will correspond with a higher level of financial stability in a country. Again, the natural logarithm will be applied to the calculated, because the ZE-score is skewed. The natural logarithm of the ZE-score will be abbreviated to ZE.

In line with Berger et al. (2008) and Lee & Hsieh (2014), this study uses the leverage ratio (LEV) as measure of financial stability. This measure of financial stability implies that a higher leverage ratio is corresponding with lower bank risk, which ultimately contributes to less overall risk in the financial system of a country since banks are more able to cover unexpected losses (Berger et al., 2008).

The financial stability transparency index (FST-index) of Horváth & Vaško (2016) will be used to measure financial stability transparency. The FST-index is based on an 11-point scale and some categories are divided into various subparts. A higher ranking on the FST-index is related to a higher level of financial stability transparency. Furthermore, the squared FST-index (FST-FST-index2) is constructed to measure whether there might be an optimal level of financial stability transparency1.

3.2. Measuring institutional quality

In line with Fazio et al. (2018), this study uses the Corruption Perception Index

(CP-index) of Transparency International (2019) and the Government Effectiveness Index (GE-1 The authors made their constructed FST-index publicly available. The data may be retrieved from:

(15)

15

index) from the World Governance Indicators provided by the World Bank (2018b) as

measures for the quality of the institutions. These measures are used to capture the possibility of a conditional effect of the quality of institutions on the relationship between financial stability transparency and financial stability. Therefore, interaction variables between financial stability transparency and the two indices of the quality of institutions are created

(CP-index*FST-index & GE-(CP-index*FST-index). The CP-index ranks countries based on the

perceived levels of corruption in its public enterprises and public services. The countries are ranked on a 0-10 scale and these rankings are based on expert assessments and opinion surveys. A high value of the CP-index implies that there are lower levels of corruption in the corresponding country. On the other hand, a low value of the CP-index suggests a higher perception of corruption and the lower the perceived quality of institutions will be. Furthermore, in line with Fazio et al. (2018) the GE-index is used as a measure of the quality of institutions to check for the robustness of the results provided by the CP-index. The GE-index consists of the perception of the quality of public- and civil services and the policy formulation and the credibility of the government's commitment to these policies (World Bank, 2019). The countries are ranked on a scale running from -2.5 until 2.5. Countries with a higher value at the GE-index experience a higher level of the quality of institutions.

3.3. Empirical models

This study builds on the work of Horváth & Vaško (2016). However, pooled OLS regressions, which they estimated, are not appropriate in the specified models according to section 3.4. This study estimates the following models presented in Eq. 3, 4, and 5 by using fixed effects panel regressions. Clustered robust standard errors are included in all models to control for heterogeneity and autocorrelation. Firstly, the estimated baseline is as follows:

>?(@ABC?'?AD$% = E$% + G1>IJ?(KLM$%+ G2O"?(KLM$%+ O$% + P$%+ Q$% (3)

where Finstabilityit represents one of the four indicators of financial stability for country i in year t. The indicators consist of the natural logarithm of aggregated non-performing loans

(NPL), the natural logarithm of the aggregated ZA-score (ZA), the natural logarithm of the

aggregated ZE-score (ZE), and the aggregated leverage ratio in percentage of banks (LEV). β1 captures the relationship between financial stability transparency and financial stability and β2 captures the relationship between the quality of institutions, which is reflected by the CP-index, and financial stability. Cit is a set of country-specific control variables. Following the work of Horváth & Vaško (2016), this study controls for variables which are considered to be early warning indicators of a financial crisis in regression estimations as proposed by Frankel & Saravelos (2012). The country-specific controls included in the model are:

a) gross domestic product per capita (GDPPC)

b) the annual gross domestic product growth in percentage (GDPG) c) inflation measured in percentage change of consumer prices (INFL) d) real interest rate change in percentage (REALINT)

(16)

16

f) the nominal exchange rate change in percentage, which is constructed by taking the change in the yearly average exchange rate of the national currency against the U.S. Dollar (EXCH)

g) stock market capitalization to GDP in percentage (MARKCAP)

h) and a measure of financial openness as proposed by Lane & Milesi-Ferretti (2007) which is the sum of foreign assets and liabilities divided by GDP (FINOPEN).

Furthermore, Bit are a set of bank-specific controls as proposed by Fazio et al. (2018). These bank-specific controls consist of the ratio of non-interest income to total income in percentage

(NONINT) as a proxy for non-traditional activities of banks. Also, banks overhead costs to

assets ratio in percentage (COST) is included to control for the cost performance of the banks. Last, a banking concentration measure which measures the total assets of the three largest banks in percentage (CONCEN). Finally, εit represents the error term.

In addition to the baseline regression presented in Eq. 3, the squared values of the FST-index are added to capture whether there is a non-linear effect of financial stability transparency. Therefore, the second model estimated is:

>?(@ABC?'?AD$% = E$% + G1>IJ?(KLM$%+ G2O"?(KLM$%+ G3>IJ?(KLMS

$%+ O$%+ P$% + Q$% (4)

where Finstabilityit is one of the variables used for measuring financial stability, which is the same as in Eq. 3. β1 captures the relation between FST-index and financial stability. β2 captures the relation between the quality of institutions, which is the CP-index, and its relation to financial stability. In addition, β3 is the coefficient of the squared FST-index and it determines whether there is a limitation to the level of financial stability transparency. Cit and Bit are the same set of controls as have been used in Eq. 3. Finally, εi,t represents the error term.

Eq. 5 presents the model which will be estimated to capture the conditional effect of the quality of institutions on the relationship between financial stability transparency and financial stability. Therefore, the model estimated is as follows:

>?(@ABC?'?AD$% = E$% + G1>IJ?(KLM$%+ G2O"?(KLM$% +

G3>IJ?(KLM ∗ O"?(KLM$% + O$%+ P$%+ Q$% (5)

(17)

17

a relationship investigated is supposed to be conditional, can be detected based on a linear-interactive model. When the direction of the sign of the coefficient changes, this may indicate that high or low levels of institutional quality have a negative impact on the effect that financial stability transparency has on financial stability.

To control for the robustness of the results of the models in Eq. 3 and 4, three-year lagged independent variables and system GMM estimations are performed in line with Horváth & Vaško (2016). Lastly in line with Fazio et al. (2018), the GE-index will be used instead of the CP-index in Eq. 5 to perform the robustness check for the conditional relationship.

3.4. Panel data assumptions

Several assumptions have to be tested to justify the specifications of the empirical models which are introduced in section 3.3. These tests are displayed in tables D1 to D11 in Appendix D. Firstly, table D1 shows the pair-wise correlation matrix of the variables included in the models. Dormann et al. (2013) argue that there might be a threshold of 0.7 at which collinearity may distort model estimation and subsequent prediction. It should be noticed, the variables GDPPC, CRE, CP-index, and GE-index seem to have relatively large correlations with each other, with values close or above the proposed threshold. According to Hill et al. (2012), detecting collinearity from a pairwise correlation matrix may be difficult when there are many variables used in the regression. The authors argue to perform an additional test to diagnose collinearity and they propose the Variance Inflation Factor (VIF) test. The VIF tests are presented in table D2. According to Dormann et al. (2013), a model experiences severe collinearity when the VIF of the independent variables exceeds a value of 10. There is no reason to assume that the models suffer from severe collinearity, because all the individual VIF values are below the rule of thumb value of 10 (Dormann et al., 2013). Furthermore, the mean VIF values of the independent variables are between 1.9 and 2.1, which indicates that the predictor for correlation is low (Curto & Pinto, 2011). It should be noticed that the interaction variables and squared independent variables are excluded in the analyses. When these variables would be included, the VIF analyses would not give a representative VIF score because of the high level of collinearity with the individual variables. Allison (2012) argues that the collinearity between the interaction terms and the individual variables is not affecting the p-values of these interactions. Lastly, a condition number test, which is also a commonly used test for detecting collinearity, is performed and presented in Appendix D table D3 (Dormann et al., 2013). The results imply the same for both sets of independent variables, because they are all below the threshold value of 30 which may be considered as the rule of thumb value for severe collinearity (Dormann et al., 2013). Therefore, these additional tests shows that the models do not suffer from severe collinearity.2

Furthermore, Breusch-Pagan Lagrange multiplier tests for random effects have been estimated in order to investigate whether it is appropriate to use a pooled OLS regression. Table D4 presents the Breusch-Pagan Lagrange multiplier tests for the models estimated in Eq. 3.

2 The ratio of total banks assets to GDP, which controls for the size of the banking system, has been removed from

(18)

18

The tests reject the null hypothesis of zero variance across entities. As a consequence, a pooled OLS regression is not appropriate. The results of the Breusch-Pagan Lagrange multiplier tests for the models in Eq. 4 and 5 are identical. All the different specifications show statistical significance at a 1% significance level.3 Therefore, using a pooled Ordinary Least Squares (OLS) regression is not appropriate in all the models and specifications under study. This finding is in contrast to the work of Horváth & Vaško (2016), who used pooled OLS regressions to estimate their models. In addition, Hausman tests are performed to determine whether fixed- or random effects estimators are preferred for the various models. Hill et al. (2012) argue that when the null hypothesis is rejected, which implies that the difference in the coefficients is not systematic, the random effects model cannot be used to capture the true parameters because random effects are correlated with the regressors. Tables D5-D8 show results of the Hausman test and the significance of the Hausman test implies that fixed effects estimators should be applied when the NPL is the dependent variable. In addition, fixed effects estimators should be applied when the ZA-score is used as dependent variable in Eq. 5. The significant Hausman test implies that for these models, a random effects approach is not appropriate. The fixed effects estimators are still consistent in the case the random effects estimators are inconsistent (Hill et al., 2012). Therefore, this study uses fixed effects estimators for all the models, because for some specifications the random effects estimators are perceived to be inconsistent.

Furthermore, the Modified Wald tests for groupwise heteroscedasticity for the fixed effects models have been executed to detect the presence of heteroscedasticity in Eq. 3. Table D9 presents the result of the models estimated in Eq. 3. The results of the test are uniform and are statistically significant at a 1% significance level for all models in the proposed equations.4 The significance implies that clustered robust standard errors must be included to control for heteroscedasticity in all the specifications of the models. Clustered robust standard errors also control for autocorrelation. In addition, a Wooldridge test for Eq. 3 is presented in table D10 to examine whether the models contain autocorrelation (Drukker, 2003; Wooldridge, 2010). In all the different models, autocorrelation is detected by using the Wooldrigde test, because all specifications of the models are statistically significant at a 1% significance level, which implies that autocorrelation is present.5 Due to the relatively large number of the cross-section dimensions relative to the time-dimensions, autocorrelation may not be a massive problem for the models are specified in section 3.3. According to previous work of several authors (Berger et al., 2008; Lee & Hsieh, 2014; Horváth & Vaško, 2016; Fazio et al., 2018), there is no explicit economic reasoning to include a lagged variable of one of the four indicators of financial stability in the models.

Furthermore, table D11 shows the Fisher-type unit-root test based on the augmented Dickey-Fuller test. This test, which investigates whether the variables contain unit roots, is appropriate for an unbalanced panel. If variables contain unit roots, estimates may be unreliable

3 For brevity purposes, the results of the Breusch-Pagan Lagrange multiplier test for the models estimated in Eq.

4, and 5 with the CP-index and the GE-index included are available upon request since they are all significant.

4 The results of the Modified Wald tests for groupwise heteroscedasticity for Eq. 4, and 5 with the CP-index and

the GE-index included are available upon request since they are all significant.

5 The results of the Wooldridge tests for Eq. 4, and 5 with the CP-index and GE-index included are available upon

(19)

19

due to spurious regressions. The results in table D11 presents that the null hypothesis is rejected at a 1% significance level for all variables. Therefore, the panels do not contain unit roots and the stationarity condition is preserved.

Furthermore, the GMM estimator technique requires several assumptions, which have to be tested. This study uses the system GMM estimator of Blundell & Bond (1998). Roodman (2009a) argues that that too many instruments may lead to a loss of power. Therefore, the number of instruments for the GMM-style instruments are restricted to 2. This leads to a lower number of instruments in comparison to the number of countries which are included in the regressions. Thereafter, a Sargan-Hansen test (J-test) has been executed to determine whether the over identifying restriction are valid. The results of this test can be found in table 5. The Sargan-Hansen test cannot be rejected for the models estimated in Eq. 3 and 4. Therefore, the number of instruments is valid. In addition, the AR(2) test is conducted to examine whether the model contains second order autocorrelation. The AR(2) is insignificant for all models in table 5. Therefore, it can be concluded that the models do not contain second order autocorrelation (Arelano & Bond, 1991). This study used the two-step procedure, since Roodman (2009b) argues that the two-step procedure is asymptotically more efficient. Lastly, Windmeijer (2005) corrected standard errors have been applied to control for heteroscedasticity and autocorrelation. In addition, these standard errors deal with the downward biased problem (Windmeijer, 2005).

4. Data

A large database has been constructed containing annual data for 110 countries for the period 2000-2011. The created dataset consists of 1320 observations for the countries included in the sample. This section elaborates on the process of data collection and it provides insights into the trends of the data.

4.1. Data collection

(20)

20

All data used to construct the four measures of financial stability have been obtained via the extensive Global Financial Development database constructed by the World Bank (2018a). The aggregated leverage ratio, which is defined as bank’s equity divided by total assets, has been derived from the World Bank’s (2018a) calculation of the bank Z-score. When the aggregated leverage ratio would be calculated using bank data derived from Bankfocus of Bureau van Dijk, inconsistencies between the construction of the Z-scores from the World bank and the leverage ratio may arise. The process of collection of bank data from Bankfocus of Bureau van Dijk is time consuming and highly prone to human error. Therefore, deriving the leverage ratio used by the World Bank (2018a), which is an internationally trusted institution, may secure the quality and consistency of the variables for financial stability used in this study. The World Bank (2018a) used unconsolidated data provided by Bankfocus and Bankscope, which is the previous version of Bankfocus, of Bureau van Dijk. Fazio et al. (2018) argue that some banks have control over others which may lead to double counting of the financial statements. The use of unconsolidated data avoids this problem and therefore preserves the quality of the data. The leverage ratio based on the calculation of the bank Z-score is derived as follows. The bank Z-score calculated by the World Bank (2018a) is constructed as the (aggregated ROA/leverage ratio)/standard deviation of ROA, where the standard deviation is related to the aggregated ROA in the period of 1960-2018 for which at least 5 years of the aggregated ROA are reported. Since the aggregated ROA and the bank Z-score are reported and the standard deviation of the ROA can be calculated, the aggregated equity to total assets ratio for banks in all countries can be calculated easily.

The obtained leverage ratio has also been used to construct the ZA-score and ZE-score as in Eq. 2. In addition, the banks ROA and ROE, which have been aggregated at country-level, are also obtained via the Global Financial Development Database provided by the World Bank (2018a). The Z-score calculated by the World Bank (2018a), which is equal to the ZA-score in this study, is not used. The Z-ZA-score calculated by the World bank is based on the standard deviation of the ROA of 59 years. The standard deviation of the aggregated ROA based on 59 years may not be a representative standard deviation in this study.

Furthermore, data on non-performing loans are also included in the Global Financial Development database and based on data reported by the IMF on Financial Soundness Indicators. The data are considered to be high-quality data, because they are directly reported to the IMF by the designated authorities of each country.

Data for the quality of institutions have been obtained from the CP-index of Transparency International (2019) and the GE-index of the World Governance Indicators (2018b). These two indices are updated annually by internationally trusted organizations and are publicly available. Therefore, the indices may be considered as high-quality data since they avoid personal data mining issues (Fazio et al., 2018).

(21)

21

the nominal exchange rate (EXCH), and the data to construct a measure for financial openness

(FINOPEN) required data from other sources. Data for the stock market capitalization have

been obtained from the Global Financial Development database constructed by the World Bank (2018). In this dataset, data on stock market capitalization have been constructed using data from World Federation of Exchanges and Standard and Poor's Emerging Market Database and Emerging Stock Markets Factbook. The level of the stock market’s capitalization is divided by GDP data obtained from World Development Indicators of the World Bank (2019). The nominal exchange rate change is collected via data provided by the International Financial Statistics database of the IMF. The nominal exchange rate change is constructed by taking the annual average exchange rate of a country’s national currency against the U.S. Dollar. Data on foreign assets and liabilities, which are needed to construct the measure for financial openness, have been retrieved from the updated External Wealth of Nations Mark II developed by Lane & Milesi-Ferretti (2007). Furthermore, the bank-level controls non-interest to income ratio

(NONINT), bank overhead costs to total cost ratio (COST), and the banking concentration

measure (CONCEN) have been obtained from the Global Financial Development database provided by the World Bank (2018a).

4.2. Data description

(22)

22

Table 1. Descriptive statistics of the variables

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

Obs. Mean St. Dev. Minimum Maximum

NPL 947 -3.177 1.143 -6.907 -0.708 ZA 1,265 2.473 0.686 0.172 4.280 ZE 1,220 0.979 0.734 -1.318 2.317 LEV 1,265 8.613 4.504 0.373 23.96 FST-index 1,320 2.328 2.572 0 9 CP-index 1,110 4.849 2.341 0.400 10 GE-index 1,207 0.372 0.965 -2.271 2.437 GDPPC 1,309 17,831 21,311 193.9 93,463 MARKCAP 1,044 54.39 51.25 0.453 247.3 CREDIT 1,231 59.99 49.10 0.186 197.4 REALINT 1,011 5.642 8.586 -17.15 43.79 INFL 1,252 5.716 5.924 -2.198 34.48 GDPG 1,319 4.110 3.961 -7.979 14.23 FINOPEN 1,267 3.828 8.795 0.337 75.76 EXCH 1,308 0.953 9.122 -16.61 37.30 NONINT 1,285 38.28 13.38 3.406 92.86 COST 1,282 3.412 2.533 0.0492 82.10 CONCEN 1,206 70.40 19.60 20.85 100

Figure 2: Increase in the FST-index from 2000-2011

(23)

23

with high values of the FST-index experience a lower level of the ZA-score. Furthermore, figure 3C shows an identical trend for countries with high and low values of the FST-index. Again, the level of the ZE-score differs between those two subgroups. Countries with high values of the FST-index seem to have lower values of the ZE-score. In addition, figure 3D presents the increase of the average leverage ratio of banks. The increase in the leverage ratio after 2008 can be explained by the introduced Basel III Accord, which increased capital standards and forced banks to hold more high-quality capital. A similar trend is observed for countries with low and high values of the FST-index, however, the level of the leverage ratio differs for these subgroups. The leverage ratio is on average higher for banks with low values of the FST-index than for banks in countries with high levels of the FST-index. Figures 3B, 3C, and 3D show that countries with lower levels of the FST-index are on average considered as more stable than countries with high levels of the FST-index. These observations are in contrast with Figure 3A, where countries with a high FST-index are perceived to be more stable. Therefore, measures of financial stability may be considered as high quality, because they may measure different aspects of financial stability. However, the non-performing loans have a relatively large number of data-points missing in comparison to the other measures of financial stability as is presented in table 1.

Figure 3. Measures of financial stability over the period of 2000-2011

(24)

24

D shows that the two indices are highly correlated with 93%. For example, the Scandinavian countries are in both indices at the top of the sample group, which means that there are low levels of corruption perceived and there are high perceptions of government effectiveness. Looking at the figures in Appendix C which provides the averages of the two indices over the period of 2000-2011, heterogeneity among the countries is observed because perceived corruption is not centered in specific regions in the world since high levels of corruption are present in Asia (Bangladesh), Africa (Sudan), Europe (Ukraine) and South-America (Guatemala). The same dispersion of high- and low quality of institutions can also be observed when looking deeper into the GE-index. Therefore, the measures of the quality of institutions are considered to be highly useful. However, some data-points in the CP-index are missing because in the early 2000s, some countries of the sample were not included in the CP-index. The GE-index also lacks some data-point, because it has not been constructed for the year of 2001.

5. Empirical results

This section of the paper presents the results for the various models, which were introduced in section 3.3. The first part of this section is devoted to the analysis of the empirical models which are used to answer the research question and the proposed hypotheses. The second part of this section presents multiple robustness checks for the results presented.

5.1. Main results

As argued previously, fixed effects panel regression with clustered robust standard errors are estimated in this section to answer the proposed hypotheses. The results of the models in Eq. 3 are presented columns 1-4 in table 2. Column 1 and 2 show that the coefficients of the FST-index concerning non-performing loans and the ZA-score as dependent variables are significant at a 5% significance level. The negative coefficient of the FST-index in column 1 indicates that higher financial stability transparency is associated with higher financial stability, since a lower value of non-performing loans is corresponding to higher financial stability. In addition, the positive coefficient for the FST-index in column 2 indicates that an increase in financial stability transparency is associated with higher financial stability. An increase in financial stability transparency will decrease the probability of default for banks. Furthermore, the coefficients of the FST-index in column 3 and 4 are also positive, but they are insignificant. Taking all results into account, mixed evidence is found concerning hypothesis 1, which is stating that the financial stability transparency is positively related to financial stability. Based on the results of column 3 and 4, hypothesis 1 is rejected since there is no statistical support that the coefficients have a positive effect on the ZE-score and the leverage ratio. However, hypothesis 1 is accepted based on the results presented in column 1 and 2. All in all, mixed evidence has been found in support of hypothesis 1. The acceptance or rejection of the hypothesis depends on the measure used for financial stability.

(25)

25

significance level of 10%, when non-performing loans are used as dependent variable. This negative coefficient indicates that higher quality of institutions is associated with higher financial stability. For the other indicators of financial stability, the coefficients of the CP-index are insignificant. In addition, the signs of these coefficients do not provide a uniform direction, since column 3 presents a negative coefficient, which indicates that higher quality of institutions would cause a decrease in financial stability. Although, there is weak evidence provided by column 1 that the quality of institutions enhances financial stability, hypothesis 3 is rejected based on the other coefficients of the CP-index. These outcomes are in line with the results of Fazio et al. (2018), who do not find conclusive evidence for a direct relationship between the same measures of the quality of institutions and financial stability.

Columns 5-8 in table 2 present the results of the models which has been estimated in Eq. 4. Column 5 shows there is a non-linear relationship between financial stability transparency and financial stability, because the squared FST-index has a positive coefficient at a significance level of 5%. Therefore, the coefficient is indicating that there is an optimal level of financial stability, since higher non-performing loans is equal to a lower level of financial stability. Regarding the other indicators of financial stability, column 6, 7, and 8 present insignificant coefficients for the squared FST-index. Mixed results are presented and therefore the acceptance of hypothesis 2 depends on the measure used for financial stability. Hypothesis 2, which is stating that there exists a non-linear relationship between financial stability transparency and financial stability, is rejected based on the results of column 6, 7, and 8. Based on the results of the coefficient of the squared FST-index of column 1, hypothesis 2 will be accepted. Furthermore, the magnitude of the coefficients for the squared FST-index, which are presented in column 6, 7, and 8, is considerably low. Due to the inclusion of the squared FST-index in the models, the coefficient of the FST-index has become significant at a 1% significance level, when non-performing loans are used as the dependent variable. In addition, the coefficient of the FST-index on the leverage ratio is significant at a 10% significance level. Lastly, the coefficient of the CP-index is significant at a 5% significance level. These slight changes in the significance levels may arise due to the inclusion of the squared FST-index which may be, to a certain extent, collinear with its original value. Therefore, these changes have to be interpreted with caution (Brambor, Clark, & Golder, 2006).

(26)

26

Table 2. The effect of financial stability transparency and the quality of institutions on financial stability.

(1) (2) (3) (4) (5) (6) (7) (8)

Variables NPL ZA ZE LEV NPL ZA ZE LEV

FST-index -0.0513** 0.0244** 0.0286 0.119 -0.214*** 0.0495** 0.0410 0.304* (0.0252) (0.0101) (0.0176) (0.0991) (0.0710) (0.0226) (0.0448) (0.170) CP-index -0.269* 0.0351 -0.0584 0.201 -0.298** 0.0398 -0.0560 0.236 (0.149) (0.0336) (0.0788) (0.282) (0.141) (0.0331) (0.0835) (0.284) FST-index2 0.0237** -0.00361 -0.00179 -0.0266 (0.00944) (0.00329) (0.00666) (0.0263) Country-specific controls Bank-specific controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Country fixed effects

Number of observations Yes 515 Yes 607 Yes 587 Yes 607 Yes 515 Yes 607 Yes 587 Yes 607 Number of countries 66 72 72 72 66 72 72 72 R-squared 0.326 0.147 0.191 0.092 0.348 0.152 0.191 0.096

Notes: This table presents fixed-effects regressions of Eq. 3 and 4. Columns 1-4 present the results of Eq. 3 and columns 5-8 present the results of Eq. 4.

(27)

27

Table 3. The conditioned effect of the quality of institutions

(1) (2) (3) (4) (5) (6) (7) (8)

Variables NPL ZA ZE LEV NPL ZA ZE LEV

FST-index -0.177** 0.0421** -0.0168 0.377** -0.0788** 0.0291*** 0.0142 0.196* (0.0783) (0.0204) (0.0373) (0.189) (0.0350) (0.0102) (0.0199) (0.102) CP-index -0.306** 0.0405 -0.0729 0.280 (0.153) (0.0358) (0.0803) (0.307) FST-index*CP-index 0.0252* -0.00368 0.00952 -0.0537 (0.0138) (0.00451) (0.00627) (0.0433) GE-index -0.693 0.173 -0.0207 1.088 (0.430) (0.122) (0.200) (1.026) FST-index*GE-index 0.0618* -0.00461 0.0299 -0.136 (0.0344) (0.283) (0.0191) (0.125) Country-specific controls Bank-specific controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Country fixed effects

Number of observations Yes 515 Yes 607 Yes 587 Yes 607 Yes 497 Yes 606 Yes 588 Yes 606 Number of countries 66 72 72 72 67 72 72 72 R-squared 0.343 0.150 0.198 0.108 0.345 0.156 0.178 0.101

Notes: This table presents fixed-effects regressions of Eq. 5. The dependent variables used for these regressions are the natural logarithm of the transformed

(28)

28

Figure 4 presents the marginal effects of financial stability transparency on the different indicators of financial stability against the CP-index, which are estimated using Eq. 5. There is a statistically significant effect of the FST-index on financial stability when the upper and lower bounds of the confidence intervals are both below or above zero. Figure 4A shows evidence of a negative coefficient of the marginal effect of the FST-index, which is statistically significant at a significance level of 5% between the range of 0.0 to 5.0 on the CP-index. In contrast, for high values of the CP-index, the coefficient of the marginal effect may turn into positive territory. However, the positive coefficient of the marginal effect for high levels of the CP-index is statistically insignificant at the significance level of 5%. Furthermore, figure 4A shows that the coefficient increases as the value of the CP-index increases. It should be recalled that a lower value of non-performing loans is indicating higher levels of financial stability. Therefore, based on figure 4A, there is sufficient support for hypothesis 4, which is stating that financial stability transparency on financial stability is conditional on the quality of institutions and it has a stronger positive impact at lower levels of the CP-index than at higher levels. Berry, Golder, & Milton (2012) argue that acceptance of the hypothesis depends on the number of observations for which the range of values of the CP-index are statistically significant. Therefore, the density of the percentage of observations is also displayed in the plot. To provide an exact number, 59.8% of total observations falls within the significant range for the coefficient of the FST-index. However, to provide a minimum percentage of the observations, which are required to support the hypothesis which fall within the significant range, is somewhat subjective since there are no strict guidelines (Berry, Golder, & Milton, 2012). Therefore, the validity of the significant range depends on own interpretations on the density of observations within the significant range. This study assumes that 59.8% of the total observations within the significant range may be considered as valid.

Furthermore, figure 4B shows a positive coefficient of the marginal effect of the FST-index on the ZA-score against the CP-FST-index, which is statistically significant at a significance level of 5% within the range of 0.0-5.5 for the index. In addition, for high values of the CP-index the coefficient may turn into negative territory. However, the negative coefficient for high levels of the CP-index is statistically insignificant at the same significance level of 5%. Furthermore, figure 4B shows that the coefficient declines in value as the CP-index increases. Based on that the marginal effect of the FST-index on financial stability against the CP-index is statistically significant at a 5% significance in a range of 0 to 5.5, the relationship of the CP-index and the FST-CP-index is considered as conditional. In addition, since the negative coefficient for the high values of the CP-index is statistically insignificant at a 5% significance level, hypothesis 4 is supported. Based on reasoning of Berry, Golder, & Milton (2012), 68.2% of the observations fall in the significant range and this range is therefore considered as supportive evidence to accept hypothesis 4 based on figure 4B.

(29)

29

marginal effect is higher of the FST-index leads to higher profitability concerning the return on equity. Therefore, hypothesis 4 is rejected based on figure 4C, when the ZE-score is used as dependent variable. The discussion section provides an alternative explanation for an increase in the profitability when the quality of institutions increases at high levels of institutional quality.

In addition, figure 4D shows a positive coefficient of the FST-index for low values of the index. The coefficient is statistically significant from a range of 0 to 3.8 on the CP-index. The coefficient may become negative for high values of the CP-index, however there is no statistical evidence to support the change of direction of the sign. The observations which fall within the significant range make up 45.6% of all observations for the regression, which makes the interpretation of the significant coefficient reliable. Based on the statistically insignificant coefficients at the highest values of the CP-index and the relatively large percentage of observations within the significant range, there is support for hypothesis 4. Therefore, there exists a conditional relationship between the CP-index and FST-index, which is the strongest at low values of the CP-index.

Figure 4. Marginal effects plots of the FST-index on the measures of financial stability against

the levels of the CP-index

Note: The red line indicates the coefficients of the marginal effect of the FST-index on financial stability

(30)

30

The linear-interactive models do not present indications for a non-linear effect of the quality of institutions on the effect of financial stability transparency on financial stability. Berry, Golder, & Milton (2012) argue that linear interactive models could indicate non-linearity of the moderator variable when the sign of the coefficient changes direction. Figure 4A, 4C, and 4D present this change of signs in the coefficients of the FST-index. However, this change in figure 4A, 4C, and 4D is not supported by statistical significance. Therefore, the figures suggest that the non-linear interactive plots do not present a new insight into a possible existing non-linear effect of the quality of institutions. In addition, figure 4B does not present a change in direction of its sign, and therefore a non-linear effect of the quality of institutions on the relationship between financial stability transparency and financial stability is excluded.6

5.2. Robustness checks

In line with the study of Horváth & Vaško (2016), three-year lags of the independent variables are used to check for the robustness of the results estimated by Eq. 3 and 4. In addition, system GMM estimations will be used to investigate whether the results of the fixed effects regression are robust. Table 4 presents the coefficients of the 3-year lagged independent variables. The coefficients of the FST-index and the squared FST-index for the model are, when the dependent variable non-performing loans is used, significant at a significance level of 1%. The positive coefficient for the FTS-index and the negative coefficient for the squared FST-index are in line with the results for this measure of financial stability, which have been found in section 5.1. For the other measures of financial stability, the coefficients of the FST-index and the squared FST-FST-index are all insignificant. In addition, all the coefficients of the CP-index are also insignificant for the four measures of financial stability. All in all, only the results of the coefficients of the FST-index and the squared FST-index, when the dependent variable is non-performing loans, are robust to this alternative estimation of Eq. 3 and 4.

Furthermore, system GMM estimations are presented in table 5. Column 1 shows that the coefficient of the FST-index is negative and significant at a 10% significance level, when non-performing loans is the dependent variable. Furthermore, column 1 shows a negative coefficient for the CP-index, which is statistically significant at a 1% significance level. In addition, column 5 shows that the coefficient of the FST-index is negative and the squared FST-index is positive and significant at a significance level of 5% and 1%, respectively. Besides, column 5 presents a negative coefficient of the CP-index, which is significant at a 1% significant level. Although the significance level differs, the results of the model with non-performing loans as the dependent variable are similar to the results of the fixed effects regression estimated in the previous section. Furthermore, column 6 presents a positive coefficient for the CP-index at a significance level of 10%. However, the level of significance for this coefficient has not been found by other estimation techniques used in this study. Furthermore, the coefficients of the FST-index in column 2-4 and 6-8 have turned negative and the coefficients of the squared FST-index in column 6-8 have turned positive, however they are all insignificant.

Referenties

GERELATEERDE DOCUMENTEN

In the marketing literature many studies had already showed that research shopping and show rooming behaviour exists in multi-channel environment with non-mobile online versus offline

The expectation is still that firms that deliver high quality audits reduce earnings management more than firms that deliver less quality audits (refer to hypothesis one), only

Factor 3, ‘Using the professional organisation as major referent’ correlated negatively with both factor 4, ‘Belief in self-regulation’, and factor 5, ‘Belief in

Besides this, I also found that independence of the board of the parent has no significant effect on the relationship between foreign subsidiaries financial reporting quality and

A higher audit quality of the parent company would mean that the auditor constraints the level of earnings management within the parent company and the subsidiary, and thus would

This study aims to bridge the gap between the impact of both financial leverage and liquidity on disclosure levels on a quantitative basis and the actual impact on the quality

According to De Groot (2010), risk reporting consists of three components, namely the risk profile, the description of the risk management system and the

Foreign banks only influence the degree of financial stability if they are operating as a bank and increase the host country’s share of foreign banks to the total number of