Asymmetric effect of macroprudential policy

(1)

University of Amsterdam

_{MSc Economics | International Economics and Globalization}

Asymmetric Effect of

Macroprudential Policy

Author: Aydan Gasimova

Student ID: 10603220

Supervisor: Konstantinos Mavromatis

Second Reader: Naomi Leefmans

(2)

1

2 Abstract

The 2008 financial crisis has demonstrated the exposure of the global financial system to systemic risk. Macroprudential policy, a new type of financial regulatory policy has been introduced in a large number of countries

to address vulnerabilities to systemic risk and the procyclicality of the financial system.

This paper analyzes the impact of adoption of 12 Macroprudential policy instruments in 100 countries over a period of 14 years. By differentiating between periods of economic upturns and downturns, the paper seeks to find asymmetry in the impact of Macroprudential policies. Asymmetric effects are captured by including interaction variables of GDP growth and policy variables, such that the effect of the policies in periods of economic upturns is

differentiated from the periods of recessions. Using panel data based on a recent IMF survey, the effects of Macroprudential instruments on the national growth of Credit to GDP ratios are estimated.

The results suggest that while some Macroprudential instruments have a notable impact on the growth of Credit to GDP, others do not. Moreover, evidence is found in favor of the presence of asymmetric effects in Macroprudential policy instruments, with no particular instrument capable of both limiting unsustainable Credit activity in times of booms and promoting it in recessionary periods. However, a number of individual instruments are found to be only

effective in either one of the economic conditions. Hence they are identified as either effectively preventative or effectively stimulating for the financial system. This study brings forward implications for the policy makers

considering selective usage of Macroprudential regulation depending on whether the economy is currently experiencing an upturn or a downturn

.

(4)

3 1. Introduction

Since the outbreak of the 2008 financial crisis economists and regulators alike have been trying to understand the underlying causes of the financial bust. It is a common belief that the build-up of financial imbalances coupled with loose monetary policy has fueled the crisis (Galati & Moessner, 2013). Needless to say, the consequences for the real economy have been severe and to this day the world economy is not yet fully recovered.

During the crisis we observed how failures of single important organizations in the financial industry not only caused turmoil in the global financial system but also affected the real economy. The concept of system-wide risk provides an explanation for this. One can think of systemic risk as the exposure of the financial system to the feedback effects among the key players in the financial industry (FSB, 2011). As a result of financial liberalization, which has been taking place for the past few decades, the interdependence of financial institutions has reached unprecedented levels (Borio, 2007). Hence, failures of single large and systemically important financial institutions trigger malfunctions in the whole industry which lead to further defaults of other organizations, thereby threatening the financial system as a whole. According to some economists the importance of systemic risk has been perpetually underestimated (Galati & Moessner, 2013). However, in view of the recent financial crisis the general opinion has changed. As the significance of system-wide risk is no longer a debated issue, questions of appropriate policy responses arise. The most prominent suggestion in this respect is to devise a Macroprudential policy (from this point on referred to as MPP), (Galati & Moessner, 2013).

MPP is a macro-based approach to financial regulation and supervision. It is distinctive from conventional policy in its intrinsic goals. Unlike other types of policy, it acknowledges the procyclicality of the financial system and its

vulnerability to systemic risk (Galati & Moessner, 2013). Thus, it aims to mitigate the systemic risk by building up “buffers” in times of booms in order to benefit from them in times of financial turmoil. Macroprudential objectives may be achieved through a number of channels which include instruments related to credit, liquidity, and capital. Exact choice of Macroprudential tools depends on the given institutional and policy setting. For instance, due to structural differences in financial systems of emerging and advanced economies they have had a different experience with Macroprudential regulation (Blanchard, 2010).

There is a widespread consensus today on the importance of adoption of MPP. For instance the countercyclical capital buffer is a significant Macroprudential element in the new regulatory framework known as Basel III (CGFS, 2011). The unanimous acceptance of MPP has reached the level of the IMF, which recognizes it as a legitimate necessity in terms of current policy action. Work on the design of Macroprudential instruments is at the top of the policy agenda in the

(5)

4

EU both on a national and union level. The year 2011 saw the creation of the European Systemic Risk Board which mainly focuses on the imposition of capital regulations (Lim, 2011). Nevertheless, it is a rather new approach to policy making and it is still at the early stages of its development.

Why is there a need for a new policy approach and why can the traditional approaches not be used?

Firstly, the financial industry today is more susceptible to systemic risk than ever before. The trend towards

globalization, as manifested in high degrees of cross border integration of financial intermediaries, enlarges the scale of systemic risk. Therefore, it has become of utmost importance for financial regulation to account for the presence of system-wide risk (Borio 2007).

Moreover, our understanding of drivers behind financial crises has changed. It was not so long ago when financial crises were merely associated with high and unstable inflation (see for instance Blanchard et al, 2010). The prime goal of policymakers was then to secure stable inflation at low levels. Monetary policy was expected to foster

macroeconomic stability by aiming exclusively at price stability, which in turn would ensure financial stability. While price stability was achieved in the early 1990s, it did not stop financial imbalances from building up. The Japanese recession, the Asian crisis and the most recent subprime-mortgage crisis have clearly shown that tight monetary policy does not provide security against financial busts (Borio, 2007). Macroprudential approach is a step beyond the traditional monetary policy.

Lastly, recent research has linked movements in the financial and real business cycles. Financial booms and busts have significant macroeconomic consequences (Borio, 2011a; Borio 2012; Claessens et al 2011). MPP tools which enable policy-makers to directly influence the supply of credit are needed to smoothen out the movements in the asset price cycles which are now directly linked to financial stability and macroeconomic volatility.

While the use of Macroprudential regulation has been on the rise across many advanced countries around the world following the aftermath of the recent financial crisis, these policies have been used extensively in developing and emerging countries for many years. Hence a recent wave of empirical literature has been looking at the effectiveness of Macroprudential tools across a variety of countries. This thesis aims to add to the literature on the topic of MPP effectiveness by evaluating the asymmetry in the effects of MPP instruments on credit developments by answering the following research question:

Is there an asymmetry in the effect of Macroprudential policy on the growth of the Credit to GDP ratio such that the magnitude of its impact varies depending on the economic conditions at the time of policy use?

(6)

5

While, other studies evaluate the effect of Macroprudential policies on indicators of financial risk, which among other measures, include proxies for credit activity, they do not differentiate between the effects of the Macroprudential policy used in a recessionary period from that used in a boom period. Unlike previous studies, this thesis assesses the asymmetry in the impact of Macroprudential regulation, arguing that MPP is not aimed to reduce credit activity at all times, but only during periods of booms, in order to prevent it from growing unsustainably. Similarly, it is

simultaneously aimed to dampen the potentially harmful fall in credit activity when the economies are struggling. The thesis arrives at the answer to this research question by conducting Allerando-Bond GMM estimation on a panel data describing the usage of 12 Macroprudential instruments over a period of 14 years and spanning 100 countries. The asymmetric effects of Macroprudential policy are accounted for using an interaction variable with an indicator for positive versus negative economic growth conditions. The results of this thesis show that individual instruments vary greatly in whether or not they affect the growth of Credit to GDP, how big their effects are and whether they are in line with the Macroprudential goal of promoting credit activity when the economy is doing poorly and simultaneously curbing excessive credit activity when it is doing well.

Based on the results of the econometric analysis, none of the Macroprudential tools were found to be strictly

countercyclical in their ability to simultaneously drive credit activity down in times of booms and promote it in times of economic downturns. The results also highlight some policy instruments which seem to be only effective in one of the two economic conditions but not the other, suggesting that in order to achieve its objective of limiting the procyclicality of the financial system, regulators should compose a Macroprudential toolkit which would employ a combination successful preventative measures in periods of positive economic growth and a different combination of stimulating measures in recessionary periods.

The remainder of this thesis is structured as follows. Chapter 2 provides an overview of the existing theoretical and empirical literature on the topic of Macroprudential policy and its effects. Chapter 3 illustrates the gaps in the literature and formulates the research question and Hypotheses for the study. Chapter 4 discusses the methodology used to answer the research question and Chapter 5 presents and discusses the results of the analysis. Finally, Chapter 6 provides a summary of the findings, discusses the limitations of the research and provides suggestions for future research on this topic.

(7)

6 2. Literature review

Before moving on to the discussion of the impacts of MPP, it is important to provide some theoretical background and economic reasoning to explain the dynamics behind this policy paradigm. The idea of prudential policies is divided into macro and micro perspectives. Microprudential policy is a traditional approach to financial regulation and supervision. The main objective of a microprudential approach is limiting the distress of individual institutions, bearing the interests of investors and depositors. Whereas, macroprudential policy aims at securing stability in the financial system as a whole (including the interactions with the real sector), rather than targeting individual components (FSB, 2011). Major difference between the two policies is also in the way they view risk. While risk is integrated as an endogenous factor in the Macroprudential framework, in the context of micro-based regulation it is merely an exogenous variable which is not determined based on the behavior of individual agents. Thus,

microprudential policy disregards individually rational actions which might have an adverse aggregate effect, deeming systemic risk unimportant (Caruana, 2009). Many economists have argued that a purely micro-based

approach to financial regulation and supervision is outdated and has been part of the problems leading to the financial crisis of 2008 (Galati & Moessner, 2013).

The term “Macroprudential” has become a buzzword in the wake of the recent financial crisis (Clement, 2010). The literature does not provide a clear-cut definition of MPP. However, there is a consensus on the objectives, scope and the type of instruments it uses (FSB, 2011). Ultimately, the MPP agenda is to limit the risks and costs of systemic crises. According to Hanson et al. systemic crises develop in the following way. As multiple financial institutions are hit with a common shock, excessive balance sheet shrinkage occurs on the part of these institutions, leading to further social costs. As individual firms try to sell their assets simultaneously in the time of a financial bust it results in a credit-crunch or a fire sale of securities and eventually contracts the economy. According to models based on fire sales and credit crunches, financial institutions are more eager to shrink assets instead of recapitalizing in times of financial turmoil which is exacerbated by their tendency to operate with insufficient capital buffers in normal times. Therefore, the Macroprudential approach aims to counterbalance these tendencies (Hanson et al, 2011)

The literature distinguishes two dimensions of Macroprudential policies: the cross-Sectional dimension and the time dimension. The cross-Sectional dimension describes how risk is distributed within the financial system at a given point in time, while the time-dimension focuses on development of aggregate risk over time. The main challenge of MPP in the cross-Sectional dimension is addressing common exposures of financial intermediaries to the possibility of a joint failure, thus mitigating the dangers of system-wide risk. In the context of this dimension, Macroprudential policies calibrate prudential tools based on individual institutions’ contribution to systemic risk. On the other hand,

(8)

7

the main concern of MPP in the time dimension is to build up buffers in times of financial booms when it is cheaper to do so, in order to benefit from less dramatic falls in the financial cycle when a crisis is underway (Caruana, 2009).

How can this rather abstract policy paradigm be operationalized?

Formulating a set of comprehensive tools as part of a functional MPP is a complex task which entails a number of challenges. High degrees of financial innovation cause continuous changes in the financial industry, which give rise to new types of risks. Therefore, a comprehensive policy cannot be devised in the form of a rigid toolkit, as it needs to be able to adapt to the changing environment. Moreover, the effectiveness with which MPP is achieving its objectives largely depends on how well it is coordinated with other types of policies in place, institutional and more generally macroeconomic environment. Financial regulatory policies in general are essential in addressing systemic risk, but they alone are simply not enough to solve complicated problems associated with it (Caruana, 2009). The interactions of monetary and macroprudential policies are essential determinants of financial stability (Angelini et al, 2011). For instance, Maddaloni and Peydró (2013) have shown that tight monetary policy achieves its goals most optimally, when stringent bank capital or loan-to-value ratio restrictions are in place. MPP is closely interlinked with other types of public policy. Hence, effective Macroprudential frameworks need to be designed to suit given institutional and governance environment, to ensure optimal choice of regulatory tools to address systemic risk (FSB, 2011).

Opinions on what specific regulatory instruments should be included in the Macroprudential toolkit diverge. Different regulatory instruments aim at reducing different types of risks. Common differentiation is between borrower-based policies and policies aimed at financial institutions. The former aims at reducing risks undertaken by individual borrowers in the economy and the latter aims to limit the risks which are taken by banks and other financial institutions. For instance, Macroprudential instruments that require the revision of financial accounting standards aim to reduce problems of procyclicality by demanding higher disclosures and setting prudential filters on financial institutions. Precautionary tools such as insurance mechanisms and failure and resolution management are suggested to lessen the adverse effects of crises. Some tools impose limits on credit extensions, based on increases in asset value. A prime example of such tools especially used in the housing sector are the loan-to-value (LTV) ratio caps, which determine the maximum allowed levels of loans to be extended by financial intermediaries relative to their value. Cyclically dependent funding liquidity requirements are used to set standards in liquidity provision. MPP can set the limits to concentration of risk to prevent extreme systemic crises. Tougher capital regulations imposed on financial institutions act as lean against the wind type of policy, obliging financial institutions to retain a certain portion of their capital as a forward-looking safeguard (Galati & Moessner, 2013). Many of the aforementioned instruments of MPP are enforced by the central banks, which emphasizes the link between monetary and MPP.

(9)

8

As a number of countries that adopt Macroprudential tools has been on the rise in the past years, investigation of the impact of such policy reform is of essence. Recent years have seen valuable additions to the empirical literature on the topic of MPP. Researchers have focused on the effectiveness of particular MPP tools in achieving their objectives. Empirically this has been done through analyzing the effect of implemented Macroprudential regulation on indicators of systemic risk, financial stability and procyclicality of the financial system. Commonly, used measures of financial stability and systemic risk are credit growth, house price inflation and leverage in the economy. In spite of the fact that research in this area suffers from lack of empirical data, significant results have been found in support of MPP (Cerutti, 2015). Provided below is the summary of the state of the art empirical knowledge in the field.

Empirical studies of the effect of Macroprudential instruments are conducted both on single country and international levels. The variation in Macroprudential policy across time and countries is typically captured using indicator

variables corresponding to the stringency of a given policy instrument or as a dummy variable corresponding to weather a certain policy instrument is in place or not. While some studies look at the effect of a single instrument in more detail, others investigate joint and individual impacts of a vector of instruments comprising Macroprudential policy.

Several researchers found evidence for the effectiveness of borrower-based Macroprudential policy instrument s. Prominently, a Hong Kong based study by Craig and Hua (2011) found that LTV ratios have a significant effect on slowing property price inflation, whereas, a study by Wong et al. (Wong, Fong, Li and Choi, 2011) found that LTV policies have a dampening effect on real estate sector leverage. Another study which found support for the effectiveness of borrower-based Macroprudential instruments is by Igan and Kang (2012) who looked at the implementation of DTI and LTV ratios as part of the Macroprudential framework in Korea. They find that these Macroprudential instruments have a significant effect on the growth of transaction volumes, additionally, the imposition of the ratio requirements was found to have slowed down house price inflation.

A study by Lim et al. (2011, 2013) is prominent in that it looks at the effect of a comprehensive set of Macroprudential instruments in a relatively large international sample of 42 countries. The study finds that LTV and DTI limits, ceilings on credit growth, reserve requirements and dynamic provisioning rules are effective in reducing the procyclicality of credits and leverage. In their 2015 paper Lim et al. look not only at the aggregate effect of the changes in policies, but also at the difference between the effects of tightening and loosening of the policies.

(10)

9

investigation of the MPP effects in a sample of 16 Central, Eastern, and Southeastern Europe countries. Their study suggests that out of a comprehensive set of Macroprudential instruments some but not all had a significant effect on slowing down house price inflation. In particular, raises of the minimum capital adequacy requirements, marginal reserve requirements targeted at specific excesses were found to slow house price inflation. Similar to Lim et al. the study finds asymmetries in the intensity of the effects of loosening versus tightening of Macroprudential instruments. Specifically, loosening and tightening of countercyclical instruments are found to differ in the intensity of their effect. Jimenez et al. (2012) evaluated the effect of dynamic provisioning requirements on the smoothness of the credit cycle in Spain, specifically by examining the impact of the policy on the supply of credits in “bad times”. Using micro-level data they find that Macroprudential policy can contain the downswing of the credit supply. Another study observing the effect of Macroprudential tools on the credit cycle was done by Tovar Mora et al. (Tovar Mora, Garcia-Escribano and Vera Martin, 2012). The study concludes that average reserve requirements, as a composite of a Macroprudential framework have a moderate and transitory effect on credit growth and are complementary to monetary policy in a sample of 5 countries in Latin America.

A recent study by Cerutti et al. (Cerutti, Claessens and Laeven, 2015) documents and analyses the use of a versatile set of 12 Macroprudential instruments in a global study of 119 countries. The study observes the effect of the presence of the different Macroprudential instruments on house price inflation and credit growth. The study finds that the use of policies in general leads to lower credit growth rates, whereby the effect is stronger in less developed and less open economies. Some evidence of increased cross-border borrowing suggests that introduction of MPP tools triggers regulatory arbitrage. The study also finds asymmetry in the effect of the instruments depending on the phases of the financial cycle, as such, MPP is found to be more effective in preventing excessive credit growth and less effective in inhibiting credit busts.

To that extent the authors argue that the effects of Macroprudential policies vary by the intensity and phase of the financial cycle. Macroprudential policies may be more effective when the financial cycle is more intense (when increases in either credits or house prices for instance are greater). Moreover, it is argued that Macroprudential policies are meant to be ex-ante tools, with their main purpose being helping to reduce the boom part of the financial cycle rather than the crunch part of it. The authors find support for the precedent by looking at the interaction effect between the presence of certain policy instruments and credit growth variables.

(11)

10 3. Research Question

This Section builds on the insights from the literature review and elaborates on the existing gaps in the literature on the topic of Macroprudential policy. In line with these gaps and the scope of this paper, the research question of this thesis is identified.

Although, the literature on Macroprudential regulation is rapidly growing, overall evidence of its effect is still limited (Cerrutti et al., 2015; Galati, 2014). This is in large, due to the shortage of data on the use of the Macroprudential regulation. Most data used by researchers either suffers from a lack of comprehensive overview of the use of different instruments or from the low frequency and granularity of the available data on different instruments. Empirical research in this area is hindered by the fact that time series documenting the use of the policies cover a rather short period of 10 to 15 years.

As described in the literature many studies in the field have been interested in analyzing the effectiveness of Macroprudential regulation. That being said, most empirical studies do so by defining a risk variable, and analyzing how Macroprudential regulation affects said variable. An important limitation of many studies is that only the bulk effect of Macroprudential policy is researched. The effectiveness of Macroprudential regulation is then often

measured by the extent to which the presence, introduction or stringency of Macroprudential instruments manages to contain growth in the risk variable. The argument is that unsustainable or excessive growth in the risk variable is meant to be contained by Macroprudential regulation.

A popular example of the aforementioned risk variable used in the empirical literature is credit growth. Although, traditionally, credits are seen as intrinsically beneficial for economies, it is argued that excessively high levels of credits can be a cause of financial crises. Too much credit can be bad, as it leads to excessive risk taking and bubbles. The burst of such bubbles can have devastating effects for economies, as the 2008 crisis has shown. However, such one-sided interpretation of the effectiveness of Macroprudential regulation has major limitations. Risk is not originated in Credit growth itself; rather it is a result of the procyclicality associated with it. The procyclicality of credits implies that, there is less credit growth in economic downturns and more of it in economic booms. Hence, procyclicality of credits growth may lead to the unavailability of credits in bad times and an excessive amount of it in good times. As a result of procyclicality, economic crises are met with shortages in credit which is indispensable in stimulating the economy and assisting its recovery. Thus, credit growth can be both a positive and a negative sign depending on the context.

Consider the following example in support of the argument above. There are two hypothetical countries, of which one has introduced a Macroprudential instrument and the other has not. Suppose that in an upturn of the business cycle,

(12)

11

exactly the same movement in credit growth happens in both countries. In the country with the Macroprudential policy in place a credit crunch occurs whereas in the country where the policy has not been introduced, this does not happen. If one analyzes this example by considering credit growth unilaterally as a risk variable, it may appear that the country with the policy in place managed to reduce credits, thus identifying the policy in this country as effective. The problem with this is that credit growth falling on average over the whole observed period is not a good measure of the policy’s effectiveness. Instead, a more accurate measure of the policy effect needs to differentiate between the stages of the business cycle at which the policy is implemented. The purpose of the Macroprudential policy is to dampen the excessive growth of credits in times of economic growth and stimulate credits when the economy struggles. Therefore, increases in so-called risk variables with regard to the Macroprudential instruments should not be viewed from a one-sided perspective. In order to be able to make conclusions regarding the effectiveness of the policy it is then imperative to analyze how symmetrical the effect of the policy is in good versus bad times. Existence of asymmetric effects could have important implications for the use of individual instruments. Hence, close

examination of the nature of asymmetric effects in Macroprudential regulation is necessary. Asymmetry in this context implies difference in the direction (sign) or the magnitude of the effect of the same policy instruments depending on an outside factor, such as the economic conditions at the time of policy adoption.

Several studies (Jiménez et al., 2014; Lim et al., 2013; Cerutti et al., 2015) have previously found evidence in support of Macroprudential instruments having asymmetric effects depending on the timing in the financial and business cycles. As such Lim et al. (2013) find that the coefficients of dummy variables representing the presence of Macroprudential instruments are different between the overall sample and the sub-sample where the policies are implemented during periods of economic expansions. This confirms the need to take account of different phases of the business cycle at the time of policy implementation when drawing conclusions on Macroprudential policy effects. On the other hand, Cerutti et al. (2015) only lightly touch upon the subject of asymmetry in the effect of the policies and look at it from the perspective of the desired outcome of containing the financial cycle. More specifically, the authors examine if the introduction of the same policy instruments manage to curb the boom in credit growth as good as they manage to sustain the fall in credit growth. Whereas, Jiménez et al. (2014), look at the asymmetric effects of the policies on the credit cycle in the context of the economic conditions at the time of policy implementation, thereby differentiating between policy implementation in good and bad times. Overall, empirical evidence shows that

Macroprudential policy does not have a symmetric impact in economic upturns and downturns.

This thesis aims to contribute to the debate on the Macroprudential policy effectiveness, by examining more closely the asymmetric effects of Macroprudential instruments. In order to do this, the basic idea of differentiation between bad and good times of policy usage as proposed by Jiménez et al. is taken further by applying it to the dataset collected

(13)

12

and used by Cerutti et al. (2015) in their research. Hence, this paper aims to add to the current state of knowledge in the field by focusing on and examining the asymmetric effects of a comprehensive set of Macroprudential policy tools on the smoothening of the financial cycle (as reflected in national credit activity) in relation to the timing of the business cycle (reflected in economic growth).

Thus, the following research question is posed.

- Is there an asymmetry in the effect of Macroprudential policy on the growth of the Credit to GDP ratio such that the magnitude of its impact varies depending on the economic conditions at the time of policy use?

In order to examine the research question, the effect of Macroprudential Policy on the growth of Credit to GDP ratio is observed in the context of positive and negative GDP growth as a proxy for good and bad economic times. In this thesis, the research question is answered through the use of four Hypotheses listed below. Based on the results of empirical analysis those Hypotheses were accepted or rejected.

Hypothesis 1: Macroprudential policy has a significant impact on the growth of Credit to GDP ratio

The purpose of this Hypothesis is to test whether or not Macroprudential policy is an impactful tool. It is expected that overall Macroprudential policy would have a significant effect on growth of Credit to GDP, since it is a measure of the financial cycle developments which Macroprudential policy is meant to influence. A more elaborate explanation of growth of Credit to GDP as a proxy for risk in the financial system and its relationship with Macroprudential policy is described in Section 4.11. Accepting this Hypothesis based on the empirical analysis would suggest evidence of Macroprudential policy as an effective tool when it comes to influencing credit activity.

Hypothesis 2: The effect of Macroprudential policy on growth of Credit to GDP differs depending on whether the period in which the policy was used is characterized by positive or negative growth

The purpose of the second Hypothesis is to test for the existence of asymmetry in the effect of Macroprudential policy by detecting effects differing based on the economic conditions at the time of policy utilization. If the results of the analysis suggest that Macroprudential policy implemented in times of positive GDP growth had a different impact than the policy which was implemented in times of negative GDP growth, it can be concluded that the policy has an asymmetric effect.

(14)

13

Hypothesis 3: Macroprudential policy utilized during periods of positive growth has a negative effect on the growth of Credit to GDP

This Hypothesis tests the effectiveness of Macroprudential regulation as a preventative measure. Acceptance of this Hypothesis based on the empirical findings would suggest that Macroprudential regulation is capable of successfully preventing the unsustainably high levels of growth in credit activity as it is used in times of booms.

Hypothesis 4: Macroprudential policy utilized during periods of negative growth has a positive effect on the growth of Credit to GDP

This Hypothesis tests the notion that Macroprudential policy is effective not merely as a preventative measure but also as a stimulating countercyclical tool which would promote the growth of Credit to GDP during recessionary periods.

(15)

14 4. Methodology

This Chapter describes the methodology used to answer the research question. The Chapter is structured as follows. Section 4.1 sets up the economic model and defines its variables. Section 4.2 describes the dataset used to analyze the model. Section 4.3 describes the empirical strategy and the econometric procedure used to estimate the model.

4.1 Model and Variables

This Section defines, describes and motivates the choice of economic variables included in the model used to answer the posed research question. The description of each variable involves a brief discussion of the sources used to collect the data for these variables. The Section first presents the model and identifies the variables. Each variable of the model is then discussed in the following Subsections, where the motivation for including each variable and their particular measures are discussed. Subsection 4.1.1 elaborates on the choice and use of the response variable, followed by the discussion of Macroprudential policy as the explanatory variable in Section 4.1.2, the use of an Interaction variable as a measure of asymmetric effects in Macroprudential policy in Section 4.1.3 and the discussion of control variables presented in Section 4.1.4.

This thesis analyzes how the usage of the various Macroprudential instruments relates to developments in the credit market by looking at growth in the Credit to GDP ratio. The effect of Macroprudential policy instruments on the Credit to GDP growth is estimated using the base regression model below.

∆ (𝐶𝑟𝑒𝑑𝑖𝑡 𝐺𝐷𝑃 )𝑖,𝑡 = 𝛽1∆ ( 𝐶𝑟𝑒𝑑𝑖𝑡 𝐺𝐷𝑃 )𝑖,𝑡−1 + 𝛽2𝑀𝑃𝑃𝑖,𝑡−1+ 𝛽3𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡−1+ 𝛽4𝐺𝑜𝑜𝑑𝑡𝑖,𝑡−1× 𝑀𝑃𝑃𝑖,𝑡−1+ 𝛽5𝑅𝑖,𝑡−1+ 𝛽6𝐺𝑜𝑜𝑑𝑡𝑖,𝑡−1 + 𝛽7𝐺𝑜𝑜𝑑𝑡𝑖,𝑡−1× 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡−1+ 𝜇𝑖+ 𝑡 + 𝜀𝑖𝑡

The left-hand-side variable is ∆ (𝐶𝑟𝑒𝑑𝑖𝑡_𝐺𝐷𝑃 )

𝑖,𝑡which captures the growth in the Credit to GDP ratio. In turn, ∆ ( 𝐶𝑟𝑒𝑑𝑖𝑡

𝐺𝐷𝑃 )_{𝑖,𝑡−1}

on the right hand-sight is the first lag of the dependent variable included to account for the expected presence of the effect of autocorrelation in the growth of Credit to GDP ratio time series. 𝑀𝑃𝑃𝑖,𝑡−1 is the explanatory variable of the

model, a vector of the one year lagged values of the Macroprudential Policy instruments indicating the presence of each instrument. 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡−1 is a control variable capturing last period’s real GDP growth, while 𝐺𝑜𝑜𝑑𝑡𝑖,𝑡−1 is an

indicator variable based on GDP growth which is a proxy for differentiating between the states of economic growth and rcessions. Hence 𝐺𝑜𝑜𝑑𝑡𝑖,𝑡−1× 𝑀𝑃𝑃𝑖,𝑡−1 is an interaction variable used to measure the asymmetry in the effect of

(16)

15

Macroprudential instruments in economic upturns as opposed to economic downturns. 𝐺𝑜𝑜𝑑𝑡𝑖,𝑡−1× 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡−1 is

another interaction variable which is included in order to capture the full effect of GDP growth on the response variable. Besides economic growth the model also includes a measure of Real Interest Rates, 𝑅𝑖,𝑡−1 as a proxy of

Monetary Policy of the previous period. Additionally, the model also includes µi country fixed effects, capturing

variation in Credit to GDP growth innate for each country in the model. Finally, t is the time effect accounting for the aggregate time specific movement in the response variable and 𝜀𝑖𝑡 is the error term of the model.

4.1.1 Response Variable

The response variable of the model embodies the measure which is the basis for assessing the effectiveness of the policy. Since the effects of separate Macroprudential instruments may be reflected in a myriad of different financial and real indicators a careful choice of a response variable is of utmost importance to the study of Macroprudential effectiveness. For this, it is important to consider the objectives behind Macroprudential regulation. Independently the instruments in the Macroprudential toolkit tackle different problems, however ultimately they pursue a common general goal of strengthening financial stability. In this regard, many studies choose to observe the movement in the financial cycle as a display of financial soundness and risk. Arguing that one of the main objectives of Macroprudential policy is to contain the busts and booms of the financial cycle, a variable representing the movement in the financial cycle is conventionally chosen as the response variable in studies of Macroprudential policy effectiveness. Although, theories of financial cycles are still at their outset, it is generally believed that the movement in the financial cycle is captured in the variation of three financial variables. The financial cycle is said to be comprised of the cycles of credit, housing and equity prices (Borio, 2014). Credit Growth and House Price Inflation are two of the most popular

response variables, which are also, used by the Cerutti et al. (2015). Response variables in models of Macroprudential policy effectiveness are commonly referred to as risk variables.

As argued above high levels of credit during economic booms and their shortage during economic crunches may be unsustainable and lead to financial imbalances. Therefore, similar to previous studies, this thesis assumes a credit variable as a proxy of the risk in the financial system. Specifically, growth in private sector Credit to GDP ratio is used as the dependent variable of the model. The source of empirical data for this variable is World Bank’s indicator of “Domestic credit provided by financial sector (% of GDP)”. The official description of this indicator and its composition by the World Bank is included in the Appendix A.2. Credit is defined by the World Bank as Loans and Financial non-equity security.

This choice of a response variable is due to the following reasons. Most other studies including the study by Cerutti et al. (2015) on which the analysis in this thesis is loosely based, choose credit growth as the variable of interest, instead of Credit to GDP. Hence, one of the ways in which this paper differentiates itself from the existing literature is by

(17)

16

considering growth in Credit to GDP as the response variable, as opposed to growth in the volumes of credit.

Considering that the data on Macroprudential regulation which lies at the basis of this thesis’ analysis is also that used by Cerutti et al. (2015) in their study, it is especially interesting to assess the effectiveness of the policy from the perspective of a different variable.

Although, growth in Credit to GDP is less frequently chosen as the risk-variable in studies of Macroprudential policies, it is nonetheless a valid indicator of risk in the financial system. According to Borio (2014), positive deviations of the ratio of private sector Credit to GDP and asset prices from their historical norms are recognized as being indicative of systemic risk. A case can be made in favor of the use of growth credit levels in relation to the size of the economy in question as opposed to focusing on the nominal credit growth.

While by country credit growth captures changes in aggregate levels of credits in each country, credit to GDP variation illustrates how volumes of credit change in relation to the size of the economy. Credit to GDP can be seen as a

measure of how leveraged the economy is. Said more loosely, how heated up the economy is. A positive change (higher level) in this variable indicates that credit has grown more than proportionally in relation to GDP. Hence, there is more money chasing relatively less real output. Using Credit to GDP instead of plain credit growth is useful, because we look at the credit relative to the level of real output. An increase in credit is, as such, is not very telling, because if GDP grows at the same rate, then this growth in credit may solely be a result of the growing real output. An increase in Credits in relation to GDP, however, indicates that there is more money chasing fewer goods, which is a better indicator of risk. To illustrate consider an example of two countries, one advanced and one developing. The developing country, by definition is at a relatively lower level of GDP and likely to experience more economic growth. Suppose that the economic growth in said country gives rapid rise to the private sector credits, resulting in high values of credit growth, while in the advanced country the same level of credit growth occurs with only a very modest rise in its (already relatively high) GDP level. If one was to compare the level of financial risk in the two countries merely on the basis of Credit Growth, both countries could be seen as being at rather equal levels of risk. However, such judgment is incomplete, considering that the credit growth in the advanced country is potentially more likely to be risky/unsustainable than that of the developing country. This is because high credit growth in an environment of low economic growth is less likely to stem from the investment demand of local firms. On the other hand, comparison of Credit to GDP changes in the two countries would result in a different conclusion; the developing country’s change in the overall Credit to GDP ratio would be much smaller than that of the advanced country.

4.1.2 Macroprudential Policy as an Explanatory Variable

The research question of this thesis necessitates a measure of Macroprudential Policy to act as the main explanatory variable in the model. Since the starting point for the empirical analysis in this thesis is the dataset collected by Cerutti

(18)

17

et al. (2015), usage of MPP is measured in line with the aforementioned dataset. Namely, the notion of MPP is captured into a measurable variable by being broken down into a composite of 12 policy instruments listed below:

1. Limits on Domestic Currency Loans 2. Concentration Limits

3. Countercyclical Capital Requirement 4. Dynamic Loan/Loss Provisioning 5. Debt-to-Income Ratio

6. Limits on Foreign Currency Loans 7. Limits on Interbank Exposures 8. Leverage Ratio For Banks 9. Loan-to-Value Ratio

10. Reserve Requirements Ratios (including Special Reserve Requirement ratios adjusted countercyclically 11. Capital Surcharges on SIFI (Systemically Important Financial Institutions)

12. Tax on Financial Institutions

Technical definitions of each of these instruments and the way in which their corresponding variables were

constructed are provided in the Appendix A.1. In short, the researchers have collected annual-level cross-country data across a sample of 119 countries and 14 years with the means of a survey (Cerutti et al., 2015). The survey used to collect the dataset is included in Appendix A.1. The researcher’s resulting dataset reveals information regarding the presence of each of the above MPP instruments across different years and places but not about the intensity or stringency with which these instruments were used. Therefore, each of these instruments is defined as a dummy variable which takes the value of 1 or 0 to signify respectively the presence or absence of a given policy instrument in a given country for a given year.

Hence, the variable of MPP is defined as a vector of dummy variables, meaning that its effect is assessed by looking at the effect that the introduction or presence of various instruments of the Macroprudential toolkit have on the response variable.

The previous year’s values of the Macroprudential instruments are included instead of the present year’s values because, as with any policy, introduction or presence of the MPP in a given country is not expected to have an

immediate effect on the Credit to GDP ratio. The assumption is that the credit market needs time to react and adjust to a new regulation or a sudden abolishment of one. Therefore, by including the lags of the predictor variable instead,

(19)

18

this adjustment period is accounted for.

Due to the inclusion of an interaction term containing the MPP variable, the interpretation of coefficient β2 is discussed

together with that of β4 in the following Subsection. However, in line with Hypothesis 1 we expect either β2 or β4 to

have a significant impact on the response variable.

4.1.3 Interaction Variable, Measure of Asymmetry

Following from the discussion above it is clear that the analysis will assess how the documented usage of the various Macroprudential instruments (𝑀𝑃𝑃𝑖,𝑡−1) relates to developments in credit markets (∆ (

𝐶𝑟𝑒𝑑𝑖𝑡

𝐺𝐷𝑃 )_𝑖,𝑡). This helps in

assessing the overall impact of Macroprudential regulation on the credit market but is not enough to answer the research question. To answer the research question our model needs to allow for the assessment of the difference (asymmetry) in the effect of MPP in regards to positive versus negative change in economic growth.

The asymmetry in the effect of MPP is captured in the model by using an interaction variable. This interaction variable is included in the model to differentiate between the effects of the Macroprudential instruments in different states of the economic climate (the so-called bad economic times as opposed to good times). The interaction variable consists of the Macroprudential vector variable of 12 instruments defined above and the Good Times variable defined as follows. Good Times is a dummy variable aimed at differentiating between different states of economic growth. The dummy variable is designed to take the value of 1 when a certain year has been characterized by positive GDP growth and a value of 0 when in that year, the country in question experienced negative GDP growth.

The presence of a significant effect of the interaction between two predictor variables on the response indicate that the effect of one predictor variable (a vector of MPP instruments) on the response variable is different at different values of the other predictor variable (Good Times versus Bad Times). This idea is tested by adding a term to the model in which the two predictor variables are multiplied. Adding an interaction term to a model changes the

interpretation of the coefficients of variables involved. Without the interaction variable Macroprudential instruments have unique effects on growth of Credit to GDP. However, the inclusion of the interaction variable implies that the effect of the Macroprudential instruments on growth of Credit to GDP differs based on the state of the economic climate.

Since the two variables in the interaction term are both dummy variables, if the previous period was characterized by negative economic growth, the Good Times variable takes the value of 0, meaning that, β2 captures the full effect of the

(20)

19

other hand, the sum of β2 and the coefficient of the interaction term β4 is a measure of the effect of the policies if they

were introduced in periods of economic growth. Hence, with the usage of this interaction term the effect of MPP is dissected into two scenarios, one in which it was used in the context of an economic upturn and the other where it was used while the economy was regressing.

Based on Hypothesis 1, either β2 or β4 should have a significant impact on the growth of the Credit to GDP ratio. More

specifically, from Hypothesis 2 it follows that β2 should be positive, meaning that MPP introduced or used in times of

economic turmoil promote growth in the Credit to GDP ratio, thereby achieving their goal of reducing procyclicality in the Credit markets. On the other hand, following from Hypothesis 3, the sum of β2 and β4 should have a significant

negative effect, showcasing the effectiveness of the policy in containing credits during booms.

4.1.4 Control Variables

 GDP growth: Lagged GDP Growth, 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡−1, is included as a control variable in order to account for the

relationship between credit growth and economic development. It is important to account for the GDP growth in addition to the Good Times variable, as the latter only indicates whether last year’s period was

characterized by positive or negative growth, while GDP growth captures the variation in economic development thereby incorporating the extent to which real economic growth is a determining factor of growth in Credit to GDP. The variable is defined as % of Real GDP change from one year to another and the data for this variable is collected from the World Bank. The extended definition is included in Appendix A.2. Lagged GDP Growth is expected to have a positive effect on the volume of credits both according to intuitive reasoning and results from previous studies such as Cerutti et al. (2015). As we include both the lagged GDP Growth and a dummy variable constructed on its basis (Good times), it is important to also include an interaction term for these two variables. This is necessary to capture the full effect that GDP growth has on the response variable when the values of growth are above zero. A simple illustration in Figure 4

demonstrates how these effects can be broken down.

However, since the response variable contains GDP in its denominator, the direction of the effect of growth on the response is more complicated than in the case of Credit growth as the response. Since it is likely that Growth will affect both the nominator and the denominator of the ratio in the response variable, it is hard to speculate what the expected effect of this variable will be. This does not have major implications for

answering the research question since the effect of GDP growth on the growth of the Credit to GDP ratio is not the point of interest for the present research. However, it limits the interpretive value of including this control variable. Nevertheless the variable is included, in order to avoid (or minimize) omitted variable bias.

(21)

20

 Interest Rate: In addition, Interest Rate variable 𝑅𝑖,𝑡−1 is included as a proxy of each country’s monetary policy

of the previous period. The Interest Rate variable is meant to capture the variation in the Monetary Policy which is likely to have a strong impact on the growth Credit to GDP. Interest Rates included are also one period lagged, due to a similar logic as in the case of the MPP variable. Since interest rates represent changes in Monetary Policy, it is expected that credit markets require an adjustment period, meaning that the effect of Monetary Policy on Credit to GDP will not be contemporaneous.

4.2 Data Description

This Section aims to describe, summarize and draw insights from the dataset which was used in the empirical analysis of this paper. The Section is structured as follows: first general information regarding the overall dataset is presented, and then the summary statistics of the response and control variables are discussed, followed by cross-country comparison of the variables. Finally, patterns observed in MPP usage are discussed.

The final dataset consists of a panel of time series ranging over 14 years between 2000 and 2013 with a yearly frequency, consisting of data on 109 countries. However the total number of observations is 1526. However it should be noted that this panel is not balanced, meaning that the availability of data across time periods is not equal for all countries in the sample. Due to the large cross-Section of countries, for the purposes of data description, they are classified into three groups, to allow for a more accessible summary of the variables. 59 Countries in the sample are classified as emerging economies, 22 as developing and 26 as advanced. The classification of countries into the three categories of the level of economic development was done in line with Cerutti et al. (2015). The full list of countries included in the analysis and their categorization can be found in Appendix. A.3

The summary statistics for the main variables described in Section 4.1 are presented in Table 1. The response variable of the model is characterized by a large variation, ranging from -70.4% up to 380.3% and averaging at 71.7% with a standard deviation of 63.7%. The unusually high percentage points for this variable can be explained by the fact that the country in question issues more credit each year than its overall output as measured by GDP. Within this dataset, the nation of Cyprus had attained the highest record in 2007 by reaching the 380.3% mark. The explanation behind the negative values of this variable is trickier as it has to do with the way in which the data was originally composed by the World Bank.

According to the official definition by the World Bank their measure of Credit to GDP accounts for domestic credit provided by the financial sector including all credit to various sectors on a gross basis, with the exception of credit to

(22)

21

the central government, which is net. The fact that the domestic credit to the central government is counted on a net basis allows for negative values in this variable. Hence, a negative value for credit to GDP implies that the central government has itself extended more credit to the financial sector than the other way around, making it possible for nations like Timor-Leste and Botswana to maintain a negative percentage of Credit to GDP for a considerable portion of the observed time period.

Table 1: Summary Statistics (The Whole Sample, 109 Countries)

Variable Mean Std. Dev. Min Max Observations

Credit to GDP 71.651 63.861 -70.378 380.337 1483

Limits on Domestic Currency Loans 0.102 0.303 0.000 1.000 1526

Concentration Limits 0.594 0.491 0.000 1.000 1526

Countercyclical Capital Requirement 0.016 0.127 0.000 1.000 1526

Dynamic Loan/Loss Provisioning 0.074 0.262 0.000 1.000 1526

Debt-to-Income Ratio 0.104 0.306 0.000 1.000 1526

Limits on Foreign Currency Loans 0.110 0.313 0.000 1.000 1526

Limits on Interbank Exposures 0.229 0.421 0.000 1.000 1526

Leverage Ratio For Banks 0.123 0.328 0.000 1.000 1526

Loan-to-Value Ratio 0.182 0.386 0.000 1.000 1525

Reserve Requirements Ratios 0.314 0.464 0.000 1.000 1526

Special Reserve Requirements 0.168 0.374 0.000 1.000 1526

Capital Surcharges on SIFI 0.009 0.092 0.000 1.000 1526

Tax on Financial Institutions 0.110 0.313 0.000 1.000 1526

GDP Growth 4.163 4.164 -14.814 34.500 1514

Good Times 0.895 0.306 0.000 1.000 1526

Interest Rate 6.299 8.693 -60.798 54.681 1220

Notably, the distribution of the Good Times variable shows that 89.5 % of the time in the dataset was characterized by economic upturns (i.e. positive growth) and 10.5% were characterized by recessions (i.e. negative growth). When divided by country groups this indicator shows that good economic times were experienced by advanced countries 83.16 % of the time in the observed period. For Developing countries the proportion of good times was 92.86% and 91.28% for the emerging countries. This shows that, on average countries overall experience negative GDP growth for

(23)

22

10.49% of the observed time period, with advanced countries spending highest portion of time (16.84%) in periods of negative GDP growth and Developing countries spending the least portion in bad times (7.14%).

In terms of the control variables, the variation is also large. GDP growth averages at 4.16% with an equal % in standard deviation. The minimum GDP growth recorded during the observed period was -14.814% (Lithuania in 2009) and the highest was 34.5% (Azerbaijan in 2006). The interest rate representing Monetary Policy varies between -60.8% (Angola in 2000) and 54.68% (Trinidad and Tobago in 2009) and averages at 6.29%.

In order to illustrate how the trajectories of the aforementioned variables differ for the three country categories, Figures 1, 2 and 3 illustrate the average pattern of development over time in the growth of Credit to GDP, GDP growth and Interest Rates across the different country categories. On average the higher levels of Interest rates are in emerging economies, developing economies and advanced economies accordingly (see Figure 2). A similar pattern is seen with the average GDP growth and growth of Credit to GDP on Figures 1 and 3. The financial crisis is marked on these Figures by the spike of interest rates and a sharp drop in GDP growth around 2008 and 2009. Rises and falls in the values of these variables are the most dramatic in emerging countries and the least dramatic in advanced economies. Rises and falls in the values of these variables are the most dramatic in emerging countries and the least dramatic in advanced economies, showing that countries with lower levels of economic development are hit hardest by external shocks.

Figure 1: Average GDP Growth across Country Categories

-10 -5 0 5 10 15 20 25 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Average GDP Growth

Emerging Countries Developing Countries Advanced Countries

(24)

23

Figure 2: Average Interest Rates across Country Categories

Figure 3: Average Growth of Credit to GDP (%) across Country Categories

0 5 10 15 20 25 30 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Average Interest Rates

Emerging Countries Developing Countries Advanced Countries -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Gtowth of Credit to GDP

Emerging Countries Developing Countries Advanced Countries

(25)

24

According Cerutti et al.’s (2015) analysis of their Macroprudential dataset countries generally increased their usage of Macroprudential measures over time. The most frequent Macroprudential instrument used is concentration limits, with 75% of country and year combinations of overall uses of the tool recorded. The second most commonly used instrument is Limits on Interbank exposures, with 29% of the total uses. Following in the descending order of uses are (countercyclical) Reserve Requirements (21%), Loan to Value ratios (21%). Debt to Income Ratios (15%), Bank Leverage ratios (15%), Taxes on Financial institutions (14%), Limits on Foreign Currency Loans (14%),Limits on Domestic currency Loans (12%),Dynamic Loan-Loss Provisioning(9%), Countercyclical Capital Requirements, Limits on Foreign Currency Loans (1%).

Cerutti et al. (2015) also find that in general usage of Macroprudential policies is most common in emerging countries, which is in line with the previously discussed Figures illustrating the higher levels of vulnerability of emerging markets to shocks. Accordingly, the second category of countries most frequently using Macroprudential policies is that of developing countries. As to the distribution of most commonly used policy instruments by country groups, Loan to Value ratios are most commonly seen in the Macroprudential toolkit of advanced economies, while Reserve Requirements and Limits on Foreign Currency Loans are most frequent players in the policies of emerging countries. Developing countries on the other hand mostly use Dynamic Loan Loss Provisioning measures and Limits on

Domestic Currency loans. These preferences of policy instruments by country groups are peculiar as they highlight the differences in the type of risks which are prevalent in the different country groups.

4.3 Econometric Methods

This Section describes the econometric procedures conducted in order to estimate the effects of the variables discussed in the two preceding Sections. Restated below is the regression model of this paper.

(𝐶𝑟𝑒𝑑𝑖𝑡 𝐺𝐷𝑃 )𝑖,𝑡 = 𝛽0+ 𝛽1( 𝐶𝑟𝑒𝑑𝑖𝑡 𝐺𝐷𝑃 )𝑖,𝑡−1 + 𝛽2𝑀𝑃𝑃𝑖,𝑡−1+ 𝛽3𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡−1+ 𝛽4𝐺𝑜𝑜𝑑𝑡𝑖,𝑡−1× 𝑀𝑃𝑃𝑖,𝑡−1+ 𝛽5𝑅𝑖,𝑡−1 + 𝛽6𝐺𝑜𝑜𝑑𝑡𝑖,𝑡−1+ 𝛽7𝐺𝑜𝑜𝑑𝑡𝑖,𝑡−1× 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡−1+𝛽8𝑢𝑖+ 𝛽9𝑡 + 𝜀𝑖𝑡

The model specifies the dataset as a panel, i.e. the variables in the model vary both by time and cross-Sections (countries). Use of panel data as opposed to only cross-time or only cross-entity data has obvious advantages. For

(26)

25

instance, it allows for the use of a larger pool of observations, thereby enhancing the accuracy with which the parameters of the model are estimated. However, estimation of a panel using a (multiple) OLS regression is likely to result in pitfalls. This is because, such a regression would not account for the presence of entity or country-specific effects in the model (Stock & Watson, 2003).

Individual countries included in the sample may have structural differences which are likely to have a large impact on the outcome variable. These are unobserved variables that vary between the different countries but do not vary over time and thus have a varying impact on the response from one country to another. If such structural differences exist, this results in omitted variable bias which leads to an inconsistent and biased OLS estimator (Stock & Watson, 2003). Hence, Fixed Effects Panel Data regression is used which accounts for structural differences between the different countries in the sample. The results of this estimation are reported in Appendix A.4. The Table in Appendix A.4 also reports the results of the F-test which tests for the existence of country-specific effects. The null-Hypothesis of the F test is that all the country-specific effects are zero. The null-Hypotheses can be rejected at 1% significance, with the P-value very close to 0 which leads to the conclusion that significant country-specific effects exist and should be accounted for.

Although the Fixed Effects estimation controls for country specific effects, as with any OLS estimators the validity of the estimated parameter coefficients hinges upon three core assumptions:

1. All variables have finite fourth moments – This assumption implies that the sample is not biased by the presence of outliers

2. {𝑋𝑖, 𝑌𝑖} are i.i.d draws – meaning that the sample is random and is not subjected to selection bias, representing

some segments of the population more heavily

3. Exogeneity (conditional mean zero assumption) – Exogeneity in an empirical model implies that the error term is not correlated with the independent variable, hence the mean of the error conditional on the independent variable is zero, or 𝐸(𝜀𝑖|𝑋𝑖) = 0.

However, given the model at hand the assumption of strict exogeneity is likely to be violated, the reason being probable reverse causality. When looking at the effect of Macroprudential regulation on the growth of Credit to GDP ratio a case could be made for how the cause and effect relationship is not just directed from the policy variables towards the growth of credit relative to GDP, but also the other way around. Although Macroprudential regulation is designed to contain excessive movement in the credit cycle, it is possible that the regulation is introduced as a response to excessive credit growth in the first place. This would mean that countries with higher growth of Credit to GDP ratios are more likely to have Macroprudential regulation in place and vice versa. If reverse causality is indeed an

(27)

26

issue, this means that the independent variable is correlated with the error term which violates the conditional mean zero assumption of the pooled OLS estimator, rendering the parameter estimates inconsistent and biased (Stock & Watson, 2003).

Endogeneity is commonly dealt with either by using so-called ad-hoc solutions or solutions involving instrumental variables. Ad-hoc solutions either involve using a different variable as a proxy for the one which is suspected to be endogenous or using the lags of the independent variable as they are assumed to be less subjected to the problem of reverse causality. In the context of the latter solution, instead of relying on the OLS estimator, the General Method of Moments estimation is used (Wooldridge, 2010).

The Allerando-Bond estimator is essentially an instrumental variable method designed to consistently estimate dynamic panel data models. When using this estimator the panel structure generates instruments for the difference of the lagged dependent variable while assuming that the idiosyncratic error term is serially uncorrelated. It assumes that the dependent variable and its first two lags are valid instruments for the first difference of the same variable. This thesis employs the Allerando-Bond estimator which is performed in two steps making use of the General Method of Moments (GMM) using David Rodman’s xtabond2 Stata command. In the first step the weighing matrix is based on the MA (1) structure of the first differenced error terms. In the second step the weighting matrix is replaced by the differenced residuals obtained from the preliminary consistent first-step estimator (Wooldridge, 2010). The results of this estimation are presented in Table 2 and are discussed in detail in the following Chapter.

(28)

27 5. Results

This Chapter describes the results obtained from the estimations used to answer the research question. The results of the Fixed Effects OLS estimation are summarized in Appendix A.4. Although there are some similarities in the

observed effects when compared to the GMM estimation results, less policy instruments are statistically significant in the OLS results than in the GMM results and the magnitude of the effects of MPP, interaction and control variables vary between the two estimations. As discussed in Section 4.3 the OLS results are likely to be biased, therefore only the results of the GMM estimation are interpreted as meaningful in this Chapter. Table 2 provides the main regression results from the Allerando-Bond GMM estimation conducted using the Stata xtabond2 command. Due to a number of missing variables in the different values of the regression variables, the command automatically dropped 5 countries from the estimation and has therefore conducted the analysis for 95 countries in the sample, with the average number of years observed per country being 10.51. The overall number of observations used for the estimation is 998. All variables presented in Table 2 except for Year are one period lagged and all variables starting with “i_” indicate and interaction of the Good Times variable with one of the Macroprudential instruments.

The discussion of the results is structured as follows. First the effects of the MPP instruments and the associated interaction variables are summarized and interpreted. Then the effects of control variables are discussed. Finally, all results are summarized and used to reject or accept the Hypotheses posed in Chapter 3.

(29)

28

Table 2: Allerando-Bond Estimation Results (Main Regression)

∆Credit to GDP (%) Coef. Std, Err T P-value 95% Conf. Interval

Limits on Domestic Currency Loans -0.228* 0.124218 -1.83 0.067 -0.471 0.016 Concentration Limits 0.296** 0.110855 2.67 0.008 0.079 0.514

Countercyclical Capital Requirement 0.400 0.259471 1.54 0.123 -0.108 0.909

Dynamic Loan Loss Provisioning 0.038 0.212504 0.18 0.857 -0.378 0.455

Debt to Income Ratio 0.049 0.163937 0.3 0.766 -0.272 0.370

Limits on Foreign Currency Loans -0.343** 0.115866 -2.96 0.003 -0.570 -0.116 Limits on Interbank Exposures -1.040*** 0.179002 -5.81 0.000 -1.391 -0.689

Leverage Ratio For Banks -0.131 0.131614 -1 0.318 -0.389 0.127

Loan to Value Ratio 0.175 0.133371 1.31 0.19 -0.087 0.436

Reserve Requirements 0.118 0.133516 0.88 0.378 -0.144 0.379

Capital Surcharges on SIFI 0.380* 0.195868 1.94 0.052 -0.004 0.764

Tax on Financial Institutions 0.041 0.188897 0.22 0.83 -0.330 0.411

i_Limits on Domestic Currency Loans 0.087 0.061527 1.41 0.158 -0.034 0.208

i_Concentration Limits 0.303*** 0.072648 4.16 0.000 0.160 0.445 i_Countercyclical Capital Requirement -0.490* 0.247125 -1.98 0.047 -0.975 -0.006

i_Dynamic Loan Loss Provisioning -0.129 0.139261 -0.93 0.353 -0.402 0.144

i_Debt to Income Ratio -0.059 0.142424 -0.41 0.68 -0.338 0.220

i_Limits on Foreign Currency Loans -0.280*** 0.059792 -4.69 0.000 -0.397 -0.163

i_Limits on Interbank Exposures -0.172 0.169143 -1.02 0.309 -0.503 0.160

i_Leverage Ratio For Banks -0.186 0.11288 -1.65 0.1 -0.407 0.035

i_Loan to Value Ratio -0.074 0.107811 -0.69 0.492 -0.285 0.137

i_Reserve Requirements -0.319** 0.102512 -3.11 0.002 -0.520 -0.118

i_Tax on Financial Institutions -0.101 0.12928 -0.78 0.433 -0.355 0.152

Interest Rate -0.007*** 0.000605 -10.93 0.000 -0.008 -0.005 Growth 0.016* 0.008081 1.93 0.054 0.000 0.031 Goodt 0.175** 0.054624 3.21 0.001 0.068 0.282 Growth*Goodt -0.007 0.008668 -0.81 0.418 -0.024 0.010 ∆Credit/GDPt-1 -0.117*** 0.002935 -39.72 0.000 -0.122 -0.111 Year 0.006* 0.002433 2.52 0.012 0.001 0.011

Asymmetric effect of macroprudential policy

University of Amsterdam

MSc Economics | International Economics and Globalization