Macro-Prudential Policy and Bank Systemic Risk: Cross-Country Evidence Based on Emerging and Advanced Economies

ABSTRACT

This study examines the effects of macro-prudential policy on bank systemic risk by using a cross-country panel data of 65 economies. We find that: (1) macro-prudential policy mitigates bank systemic risk by decreasing the individual risk of banks and systemic linkage between banks; (2) the effect is stronger for banks with larger sizes, wider geographical regions of operation, and more diversified business services; (3) the effect is stronger in countries with less-developed or less-open financial systems, more concentrated banking industries, and emerging economies; (4) the policy-risk relationship is stronger when credit is more crunched, monetary policy is looser, a financial crisis occurs, or after the subprime crisis.

KEYWORDS:

• Macro-prudential policy
• systemic risk
• systemic linkage
• financial development
• financial crisis

JEL:

• G28
• E58

1. Introduction

The 2008 financial crisis made clear that systemic risk poses a significant threat to financial stability. As the interconnectedness of financial institutions increases (Narayan, Narayan, and Prabheesh 2014), maintaining the stability of individual financial institutions does not guarantee the stability of the overall financial system. To mitigate systemic risk, the Basel Committee on Banking Supervision (BCBS) has included macro-prudential supervision in its regulatory framework. Many countries have adopted macro-prudential supervision to prevent risk.

In macro-prudential regulatory practices, policy toolboxes and actions vary across countries. For instance, advanced and developing economies employ different policy instruments based on the characteristics of their financial systems. Apart from the financial system, the financial cycle significantly influences financial stability. Therefore, in implementing macro-prudential policy, policymakers should pay attention to the characteristics of the financial system and financial cycle.

Our aim is to evaluate the effectiveness of macro-prudential policy in mitigating bank systemic risk and analyze the various effectiveness across countries. A cross-country sample and a fixed effects model are employed. Our empirical test consists of two main parts. Firstly, we assess the effectiveness of macro-prudential policy by utilizing the bank systemic risk proxy, ΔCoVaR. We also investigate the impact mechanism through which macro-prudential policy affect bank systemic risk by decomposing the ΔCoVaR into individual risks and systemic linkages. Secondly, we explore the heterogeneous influence of policy from three perspectives: micro, macro, and cyclical. The results demonstrate that macro-prudential policy mitigate bank systemic risk. This effect is influenced by the bank characteristics, financial system characteristics, and financial cycle.

A large number of empirical articles prove that macro-prudential policy enhances financial stability. Lee, Asuncion, and Kim (2016) utilize credit, leverage, and housing price as indicators of financial risk and find that macro-prudential policy enhances financial stability. Scholars also examine the impact of macro-prudential policy on bank asset allocation or risk-taking (Andries, Melnic, and Nistor 2017; Butzbach 2016; Meuleman and Vander Vennet 2020). However, existing studies primarily focus on the different effects of various policy instruments. Few studies examine the mechanisms through which macro-prudential policy impacts the financial risk. Our article empirically tests this question.

Several studies also focus on the factors that influence the macro-prudential policies. Scholars find that the impact of macro-prudential policy on economic output is influenced by various factors, including exchange rate regime, financial openness, and bank corporate governance (Boar et al. 2017; Gaganis et al. 2020; Kim and Mehrotra 2022). However, when discussing the impact of macro-prudential policy on financial risk, few scholars analyze the factors that influence the interaction between the policy and risk. Existing studies have find that macroeconomic and institutional factors influence both the stock market (Narayan, Narayan, and Thuraisamy 2014) and bank systemic risk. Therefore, it is necessary to consider the influences of various macro- and micro-level factors when exploring the macro-prudential policy.

In related literature, Apergis, Aysan, and Bakkar (2021) also find the cross-country variation in the relationship between macro-prudential policy and bank stability. They focus on the setting of regulatory agencies while we examine the financial market and cycle. Our sample includes banks from both advanced and developing countries, which enables us to explore the varying effectiveness of policy on risk across countries. Our study is also related to Ely, Tabak, and Teixeira (2019) which examines the impact of macro-prudential policy on bank individual risk. We extend this research by studying the ultimate goal of macro-prudential policy, systemic risk. Micro-prudential policy focuses on the individual risk while macro-prudential policy targets the systemic linkage between banks. Our paper investigates the effectiveness of macro-prudential policy in decreasing bank systemic risk and analyze the mechanism.

We make three potential contributions to the academic literature. Firstly, we focus on the impact of macro-prudential policy on banks’ systemic risk. Our work connects to studies that show the macro-prudential policy decrease the credit growth, housing price (Lee, Asuncion, and Kim 2016), and bank risk-taking (Meuleman and Vander Vennet 2020). Although credit growth and house prices reflect financial stability, they are not the ultimate goal of the macro-prudential policy. We investigate the impact of macro-prudential policy on bank systemic risk, which is the ultimate policy goal. We find that macro-prudential policy mitigates bank systemic risk, and this impact is dampened when macro-prudential policy is already tight. We aim to contribute to the literature by examining the effects of macro-prudential policy on financial stability from the perspective of bank systemic risk. Our empirical result also shows that macro-prudential policy impact bank systemic risk by reducing individual risk and systemic linkages. This finding is consistent with the theoretical study showing that macro-prudential policy reduces the interlinkages between banks (Freixas, Laeven, and Peydró 2015). We add to these studies by providing an empirical test. The results demonstrate that the macro-prudential policy enhance the safety of individual banks and mitigate the contagion of risk.

Secondly, the paper examines the disparities in macro-prudential policy between developed and developing countries. Previous research validates variation in the impact of macro-prudential on financial stability across developed and developing countries (Akinci and Olmstead-Rumsey 2018; Cerutti, Claessens, and Laeven 2017). Our study not only finds similar results but also offers corresponding explanations. We assess the systemic risk of 843 banks across 65 economies, enabling us to ascertain this explanation. By examining a wide range of influencing factors, the empirical findings demonstrate that the impact of macro-prudential policy on systemic risk is stronger in countries with less-developed or less-open financial systems. Therefore, the macro-prudential policy in advanced countries exhibit diminished effectiveness in mitigating systemic risk because their financial systems are developed and open.

Finally, our empirical analysis benefits from a higher frequency of data. The majority of research about macro-prudential policy use quarterly or annual data for empirical analysis (Apergis, Aysan, and Bakkar 2021; Lee, Asuncion, and Kim 2016). Given the availability of monthly policy information in the IMF database, we are able to utilize monthly panel data. The utilization of high-frequency data enables a better identification of the impact of macro-prudential policy on bank systemic risk. The implementation of macro-prudential policy takes time, making it challenging for policymakers to address systemic risk within one month. Consequently, regressions based on monthly frequency data can alleviate the endogeneity problem arising from reverse causality.

The remainder of this paper is organized as follows. Section 2 provides a brief discussion of the related literature. Section 3 presents the data and the methodology. Section 4 discusses the results. Section 5 is further analysis. The conclusions of this paper are provided in section 6.

2. Macro-Prudential Policy and Bank Systemic Risk

2.1. The Effectiveness of Macro-Prudential Policy

In the theoretical literature, scholars analyze how macro-prudential policy prevent the systemic risk and crises by embedding financial-real linkage into banking or macro models (Galati and Moessner 2018). Silvo (2019) examines the implementation of macro-prudential policy using a New Keynesian dynamic stochastic general equilibrium (DSGE) model and identifies a trade-off between the output gap and inflation stabilization. Adrian and Boyarchenko (2018) compare liquidity and capital constraints, and find that liquidity constraints are more effective, as tightening liquidity requirements lowers the systemic distress without impairing consumption growth. Dubois and Lambertini (2018) emphasize the significant role of countercyclical liquidity regulation in preventing bank loan contraction during crisis.

Empirical studies can be categorized into two groups based on the data characteristics: macro-level and micro-level studies. Macro-level research focuses on the overall financial sector and analyzes the impact of macro-prudential policy on credit growth and housing prices. Previous research findings confirm that macro-prudential policy can play an effective role in stabilizing credit cycles and controlling housing price growth (Cerutti, Claessens, and Laeven 2017; Fendoğlu 2017; Yao and Lu 2020).

Micro-level empirical studies focus on the impact of individual financial institutions on systemic financial risk. Relevant empirical studies find that the implementation of macro-prudential policy reduce the leverage ratio, assets, and non-core liability share (Claessens, Ghosh, and Mihet 2013; Igan and Kang 2011; Jiménez et al. 2017), thereby affecting bank risk (Altunbas, Binici, and Gambacorta 2018). Although literature confirms the impact of macro-prudential policy on bank asset and liability allocation, few studies directly investigate their effect on bank systemic risk. Meuleman and Vander Vennet (2020) as well as Apergis, Aysan, and Bakkar (2021) examine this effect. However, to the best of our knowledge, none of the above studies investigate the impact mechanism.

2.2. Heterogeneous Impact of Macro-Prudential Policy

The implementation of macro-prudential policy differs across countries. Cerutti, Claessens, and Laeven (2017) and Akinci and Olmstead-Rumsey (2018) find that developed economies tend to use macro-prudential instruments linked to housing prices, while emerging economies tend to use macro-prudential tools related to cross-border capital. They find that macro-prudential policy is more effective in emerging countries. Cerutti, Claessens, and Laeven (2017) and Reinhardt and Sowerbutts (2015) also find that financial openness weakens the effects of macro-prudential policy on credit growth. When macro-prudential policy tightens, financial institutions tend to increase international operations to avoid regulations. Similarly, Boar et al. (2017) find that the impact of a macro-prudential policy on credit and output is influenced by the financial openness and financial development. Additionally, the impact of macro-prudential policy is also influenced by bank characteristics. Altunbas, Binici, and Gambacorta (2018) find that macro-prudential policy exerts a greater influence on banks with smaller size, less capital, and a higher share of wholesale financing. Therefore, it is crucial to analyze the impact of the bank characteristics, financial system characteristics, and financial cycle on the macro-prudential policy.

3. Data and Methodology

3.1. Data

Macro-prudential policy data are sourced from the iMaPP database on the International Monetary Fund (IMF) website. The database records macro-prudential policy implementation in 36 advanced and 98 emerging economies. Given the need for market data in the construction of risk proxy, the final regression sample includes 843 listed banks from 65 economies.

3.2. Empirical Specification

We use the panel regression model in Equation 1(1) $\mathit{Ris}{k}_{\mathit{i},\mathit{c},\mathit{t}}={\alpha }_0+{\alpha }_1\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}+\sum _k^K{\beta }_k\mathit{Mi}{c}_{\mathit{k},\mathit{i},\mathit{t}}+\sum _l^L{\gamma }_l\mathit{Ma}{c}_{\mathit{l},\mathit{c},\mathit{t}}+{\mu }_i+{\nu }_t+{\epsilon }_{\mathit{i},\mathit{c},\mathit{t}}$(1) to identify the impact of macro-prudential policy on systemic risk. The model is at the bank level with a monthly frequency.

(1) $\mathit{Ris}{k}_{\mathit{i},\mathit{c},\mathit{t}}={\alpha }_0+{\alpha }_1\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}+\sum _k^K{\beta }_k\mathit{Mi}{c}_{\mathit{k},\mathit{i},\mathit{t}}+\sum _l^L{\gamma }_l\mathit{Ma}{c}_{\mathit{l},\mathit{c},\mathit{t}}+{\mu }_i+{\nu }_t+{\epsilon }_{\mathit{i},\mathit{c},\mathit{t}}$(1)

Indices i, c, and t represent the bank, country, and month, respectively. $\mathit{Ris}{k}_{\mathit{i},\mathit{c},\mathit{t}}$ is the bank systemic risk and $\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}\mathit{\hspace{0.17em}}$denotes the macro-prudential policy index. $\mathit{Mi}{c}_{\mathit{k},\mathit{i},\mathit{t}}$ denotes the bank-level control variables. $\mathit{Ma}{c}_{\mathit{l},\mathit{c},\mathit{t}}$ denotes country-level control variables. Bank and time fixed effects are indicated by ${\mu }_i\mathit{\hspace{0.17em}}$and ${\nu }_t$. Based on the baseline regressions, this study also analyzes the heterogeneous effectiveness of the macro-prudential policy on bank risk using Models (2) and (3).

(2) $\mathit{Ris}{k}_{\mathit{i},\mathit{c},\mathit{t}}={\alpha }_0+{\alpha }_1\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}+{\alpha }_2\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}\cdot \mathit{MicFacto}{r}_{\mathit{i},\mathit{c}}\left(\mathit{MacFacto}{r}_c\right)+\sum _k^K{\beta }_k\mathit{Mi}{c}_{\mathit{k},\mathit{i},\mathit{t}}+\sum _l^L{\gamma }_l\mathit{Ma}{c}_{\mathit{l},\mathit{c},\mathit{t}}+{\mu }_i+{\nu }_t+{\epsilon }_{\mathit{i},\mathit{c},\mathit{t}}$(2)

(3) $\mathit{Ris}{k}_{\mathit{i},\mathit{c},\mathit{t}}={\alpha }_0+{\alpha }_1\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}+{\alpha }_2\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}\cdot \mathit{CyCFacto}{r}_{\mathit{c},\mathit{t}}+{\alpha }_3\mathit{CyCFacto}{r}_{\mathit{c},\mathit{t}}+\sum _k^K{\beta }_k\mathit{Mi}{c}_{\mathit{k},\mathit{i},\mathit{t}}+\sum _l^L{\gamma }_l\mathit{Ma}{c}_{\mathit{l},\mathit{c},\mathit{t}}+{\mu }_i+{\nu }_t+{\epsilon }_{\mathit{i},\mathit{c},\mathit{t}}$(3)

$\mathit{MicFacto}{r}_{\mathit{i},\mathit{c},\mathit{t}}$ represents a bank-level variable comprising size, geographical region of operation, and business diversification. $\mathit{MacFacto}{r}_{\mathit{c},\mathit{t}}$ represents a country-level variable encompassing the degree of financial development, banking concentration, financial openness, and emerging economies.$\mathit{\hspace{0.17em}}\mathit{CyCFacto}{r}_{\mathit{c},\mathit{t}}$ is a cyclical variable that includes credit cycle, monetary cycle, financial crisis, and subprime crisis. $\mathit{MicFacto}{r}_{\mathit{i},\mathit{c}}\left(\mathit{MacFacto}{r}_c\right)$ remains constant over time. To avoid multicollinearity between $\mathit{MicFacto}{r}_{\mathit{i},\mathit{c}}$ $\left(\mathit{MacFacto}{r}_c\right)$ and bank fixed effects, the model incorporates only the interaction term and excludes $\mathit{MicFacto}{r}_{\mathit{i},\mathit{c}}$ $\left(\mathit{MacFacto}{r}_c\right)$. $\mathit{CyCFacto}{r}_{\mathit{c},\mathit{t}}$ varies across banks and time. Therefore, Model (3) incorporates $\mathit{CyCFacto}{r}_{\mathit{c},\mathit{t}}$.

3.3. The Macro-Prudential Policy Index

A macro-prudential policy index is constructed based on the iMaPP database. This database provides data on 17 policy instruments for 134 countries from 1990 to 2019.¹ Most databases on macro-prudential policy record data by dummy-type indicators. Alam et al. (2019) document the tightening (+1) and easing (−1) of each policy instrument in response to specific policy actions. To reflect a country’s policy holistically, we calculate a cumulative indicator of all 17 macro-prudential tools.

3.4. Bank Systemic Risk

This study utilizes a commonly used indicator of systemic risk, the ΔCoVaR (Adrian and Brunnermeier 2016). This indicator is constructed by Equation (4)(4) $\Delta \mathit{CoVa}{R}_{q,t}^{sy\mathrm{stem}|i}=\frac{{\mathit{\sigma }}_{\mathit{system},\mathit{t}}}{{\sigma }_{\mathit{i},\mathit{t}}}\times \underset{\mathrm{systemic}\mathrm{linkage}}{\underbrace{{\rho }_t^{i,s\mathrm{ystem}}}}\mathit{\hspace{0.17em}}\times \underset{\mathrm{individual}\mathrm{risk}}{\underbrace{\left[\mathit{Va}{\mathit{R}}_{\mathit{i},\mathit{t}}\left(5\right)-\mathit{Va}{R}_{\mathit{i},\mathit{t}}\left(50\right)\right]}}$(4) :

(4) $\Delta \mathit{CoVa}{R}_{q,t}^{sy\mathrm{stem}|i}=\frac{{\mathit{\sigma }}_{\mathit{system},\mathit{t}}}{{\sigma }_{\mathit{i},\mathit{t}}}\times \underset{\mathrm{systemic}\mathrm{linkage}}{\underbrace{{\rho }_t^{i,s\mathrm{ystem}}}}\mathit{\hspace{0.17em}}\times \underset{\mathrm{individual}\mathrm{risk}}{\underbrace{\left[\mathit{Va}{\mathit{R}}_{\mathit{i},\mathit{t}}\left(5\right)-\mathit{Va}{R}_{\mathit{i},\mathit{t}}\left(50\right)\right]}}$(4)

${\sigma }_{\mathit{i},\mathit{t}}$ and ${\sigma }_{\mathit{sysystem},\mathit{t}}$ denote the standard deviation of the returns. ${\rho }_t$ denotes the dynamic correlation between the returns. VaR represents the tail risk of a financial institution. $\mathit{Va}{R}_{\mathit{i},\mathit{t}}\left(5\right)\mathit{\hspace{0.17em}}$and$\mathit{\hspace{0.17em}}\mathit{Va}{R}_{\mathit{i},\mathit{t}}\left(50\right)\mathit{\hspace{0.17em}}$equal to the 5% and 50% empirical quartiles of $r_t^i$, respectively. q represents the quartile. Equation (4)(4) $\Delta \mathit{CoVa}{R}_{q,t}^{sy\mathrm{stem}|i}=\frac{{\mathit{\sigma }}_{\mathit{system},\mathit{t}}}{{\sigma }_{\mathit{i},\mathit{t}}}\times \underset{\mathrm{systemic}\mathrm{linkage}}{\underbrace{{\rho }_t^{i,s\mathrm{ystem}}}}\mathit{\hspace{0.17em}}\times \underset{\mathrm{individual}\mathrm{risk}}{\underbrace{\left[\mathit{Va}{\mathit{R}}_{\mathit{i},\mathit{t}}\left(5\right)-\mathit{Va}{R}_{\mathit{i},\mathit{t}}\left(50\right)\right]}}$(4) shows that $\Delta \mathit{CoVaR}$ is influenced by the dynamic correlation coefficient (${\rho }^{\mathit{i},\mathit{system}}$)$,$ the standard deviation (${\sigma }_{\mathit{system},\mathit{t}}/{\sigma }_{\mathit{i},\mathit{t}}$) and the individual tail risk($\mathit{Va}{R}_{\mathit{i},\mathit{t}}\left(5\right)-\mathit{Va}{R}_{\mathit{i},\mathit{t}}\left(50\right)$). ${\rho }^{\mathit{i},\mathit{system}}$ represents the systemic linkage and $\mathit{Va}{R}_{\mathit{i},\mathit{t}}\left(5\right)-\mathit{Va}{R}_{\mathit{i},\mathit{t}}\left(50\right)$ represents the individual risk.

In the subsequent empirical test, the mechanism is analyzed in terms of individual risks and systemic linkages. The formation of systemic risk involves two primary elements: “shock” and “contagion” (Freixas, Laeven, and Peydró 2015). This process can be summarized as follows: negative shocks affect the entire financial system, initially resulting in individual risks for certain financial institutions. These individual risks then propagate and amplify within the financial system.

In addition, the marginal expected shortfall (MES) defined by Acharya et al. (2017) is used as an alternative risk proxy for robustness testing. The definition is as follows.

(5) $\mathit{ME}{S}_q^i=\mathit{E}\left(r_t^i|r_t^{sy\mathrm{stem}}<\mathit{Va}{R}_{q,t}^{sy\mathrm{stem}}\right)$(5)

Table A2 in the Supplementary Appendix presents the definitions and data sources for all variables in the baseline regression.

3.5. Micro, Macro, and Cyclical Factors

The first set of factors that impact macro-prudential policy is the micro characteristics of banks. They are Small, Region, and Diversify. The second set of factors that affect policy is the characteristics of financial system. We conduct four dummy variables: FinDev, Con, Open, and Emerging. We also consider the effects of cyclical factors, including Exp, Loose, Crisis, and Before2008. Table 1 presents the definitions and data sources for variables in the heterogeneity analyses.

Table 1. Source and description of variables.

Variables	Definition	Source
Micro Factors
Small	It takes the value of 1 (0) when the bank’s total assets are less(greater) than the median.	Bloomberg
Region	It takes the value of 1 when the bank’s GICS code is 40,101,015 and 0 when the GICS code is 40,101,010.	Bloomberg
Diversify	It takes the value of 1 (0) when the non-interest income ratio is greater than (less than) the median.	Bloomberg
Macro Factors
FinDev	It takes a value of 1 (0) when the sum of private credit and stock market capitalization as a fraction of GDP is above (below) the median.	World Bank Database
Con	It takes a value of 1 (0) when the Boone index is above (below) the median value.	World Bank Database
Open	It takes the value of 1 (0) when the consolidated foreign claims of banks as a fraction of GDP is above (below) the median	World Bank Database
Emerging	It takes 1 when a bank is in an emerging economy, and 0 when it is not.	iMaPP database
Cycle factors
Exp	It takes the value of 1(0) when the credit growth rate is above (below) the country’s median.	BvD database
Loose	It takes a value of 1(0) when the M2 growth rate is above (below) the median.	BvD database
Crisis	Crisis takes the value of 1(0) when the bank sample is(not) experiencing a banking crisis.	Laeven and Valencia (2018)
Before2008	If the sample is before 2008, it takes the value of 1; otherwise, it takes the value of 0.

Notes: This table lists the name and descriptions of the micro, macro, and cyclical factors used in the heterogeneity analysis. In the construction process of micro and macro factors, we first calculate the mean value for individual bank (country) and compare the mean value to the median of the full sample. The construction process of cyclical factors is mainly based on within-country comparisons. We compare the cyclical variable for country c time t with the within-country median. The credit data used in this paper are domestic credit stocks in a country, including household loans, corporate loans, and other loans.

4. Results

4.1. Descriptive Statistics

Table A3 Panel A in the Supplementary Appendix presents the summary statistics. Panel B compares macro-prudential policy and systemic risk across subgroups. This table shows significant differences in policy as well as risk among subgroups.

4.2. Baseline Results

The baseline results of macro-prudential policy and bank systemic risk are presented in Table 2. We regress the$\Delta \mathit{CoVaR}$ and MES on the iMaPP. The results in Columns 1 and 2 of Table 2 show a significantly negative relationship between macro-prudential policy and bank systemic risk. A tight policy action decreases systemic risk. This result aligns with the existing literature, which also uses the MES to proxy for systemic risk (Meuleman and Vander Vennet 2020). A unit increase in the macro-prudential index corresponds to a tightening of the macro-prudential policy stance. Therefore, regarding the economic implications of the regression coefficients, a restrictive policy action is associated with a 1.35% (0.017 divided by the mean, 1.261) decrease in $\Delta \mathit{CoVaR}$ and a 1.49% (0.032 divided by the mean, 2.1442) decrease in MES.

Table 2. Effects of macro-prudential policy on systemic risk: main regression results.

Variable	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%	Individual Risk%	Systemic Linkage%
	(1)	(2)	(3)	(4)
iMaPP	−0.017***	−0.032***	−0.019***	−0.426***
	(0.002)	(0.004)	(0.005)	(0.049)
Size	0.228***	0.344***	−0.058	10.912***
	(0.032)	(0.048)	(0.061)	(0.866)
Lev	−0.010***	−0.012*	0.017	−0.356***
	(0.004)	(0.006)	(0.011)	(0.075)
ROA	0.006	−0.128***	−0.387***	1.315***
	(0.012)	(0.024)	(0.055)	(0.226)
NII	−0.002**	−0.000	0.004	−0.010
	(0.001)	(0.001)	(0.003)	(0.015)
DtL	0.002	0.001	−0.004	0.059**
	(0.001)	(0.002)	(0.002)	(0.029)
LtA	−0.001	−0.002	−0.001	−0.021
	(0.001)	(0.002)	(0.002)	(0.028)
GDPr	−0.022***	−0.018**	−0.040***	0.267***
	(0.004)	(0.008)	(0.010)	(0.062)
Inf	0.030***	0.051***	0.004	0.840***
	(0.006)	(0.008)	(0.012)	(0.118)
RateRE	−0.618***	−0.745***	−0.862***	−10.569***
	(0.106)	(0.151)	(0.224)	(2.356)
Creditr	−0.001	0.001	−0.004**	0.050**
	(0.001)	(0.002)	(0.002)	(0.022)
CtG	0.333***	0.603***	0.715***	1.015
	(0.070)	(0.104)	(0.116)	(1.391)
Constant	0.443	0.272	5.631***	−36.177***
	(0.305)	(0.457)	(0.615)	(7.118)
Obs	107,995	107,995	107,995	107,995
R-squared	0.497	0.369	0.410	0.307
Firm FE.	YES	YES	YES	YES
Month FE.	YES	YES	YES	YES
Bank Num.	843	843	843	843

Note: This table contains information on the baseline regression results based on Model (1). The total sample consists of 843 banks and spans 2000:M1–2019:M12. The dependent variables are ΔCoVaR, MES, Individual Risk, and Systemic Linkage. The first column shows the results using the preferred systemic risk measure, the ΔCoVaR. The second column confirms our baseline results when using MES as an alternative risk proxy. In the next two columns, we use two subcomponents of the $\Delta \mathit{CoVaR}$, being the Individual Risk and Systemic Linkage. For each systemic risk index, we consider the aggregated macro-prudential policy index. Bank and time fixed effects are included in all models. The standard errors in our estimations are clustered at the bank level. Standard errors are shown in parentheses. Additionally, *, **, *** represent significance at the 10%, 5%, and 1% level, respectively.

In Columns 3 and 4 of Table 2, we estimate the effect of macro-prudential policy on individual risk and systemic linkage to examine how macro-prudential policy influence systemic risk. The regression results suggest that the policy reduces systemic risk by affecting both individual risk and systemic linkages. The macro-prudential policy toolbox contains instruments that target systemic linkages (e.g., operational restrictions for systemically important financial institutions) and instruments that focus on countercyclical regulation (e.g., countercyclical capital buffers). The regression results show that macro-prudential tool boxes can effectively mitigate systemic risk by reducing both individual risk and systemic linkage among financial institutions.

4.3. Robustness Checks

Our study may have endogeneity issues. Sample for empirical studies should be representative. We change the sample by reducing the over-representation of U.S. banks, following the methodology of Hagendorff, Keasey, and Vallascas (2018). To mitigate the impact of measurement errors, a new macro-prudential policy index is constructed using a new database provided by Cerutti et al. (2017). This database provides data for 64 countries and is widely used by scholars (Chen et al. 2022; Gaganis et al. 2021). In addition, the macro-prudential policy index is recalculated by replacing the value of easing actions with 0. In the baseline regression, tightening actions (+1) and easing actions (−1) add to zero and offset each other. The new calculation process for the policy index eliminated this offset problem. Moreover, we replace the independent variable with a lagged term.

Furthermore, return on assets (ROA), credit cycle and systemic risk are closely related. To prevent reverse causality issue, we lag all control variables. We also use a propensity score matching approach to mitigate the interference of omitted variables. Finally, the regression model is adjusted. In order to reduce the number of fixed effect dummy variables, we include only three dummy variables: America, Emerging, and 2008 year. Additionally, a random-effects model, monthly fixed effect, as well as CCE estimation method (Pesaran 2006) are used. In addition, we cluster the regressions at the continental level to account for the correlation between countries. The regression results, shown in Table A4 and Table A5 of the Supplementary Appendix, remain robust.

4.4. Heterogeneous Impact Across Banks

This study categorizes samples based on bank size, geographical regions of operation, and business diversification. The first two columns of Table 3 show that the coefficient of iMaPP is significantly negative, indicating that macro-prudential policy significantly reduce the systemic risk of large banks. The positive coefficients in the second row indicate a greater effect of macro-prudential policy on large banks compared to small banks.

Table 3. Heterogeneity in the policy-risk relationship: bank level.

Variable	Small		Region		Diversify
	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%
	(1)	(2)	(3)	(4)	(5)	(6)
iMaPP	−0.021***	−0.037***	−0.017***	−0.033***	−0.009***	−0.025***
	(0.002)	(0.004)	(0.002)	(0.004)	(0.003)	(0.004)
iMapp*dummy	0.034***	0.040***	0.023***	0.038***	−0.016***	−0.017***
	(0.003)	(0.004)	(0.003)	(0.004)	(0.002)	(0.004)
Constant	0.608**	0.465	0.748**	0.774*	0.679**	0.521
	(0.284)	(0.440)	(0.290)	(0.443)	(0.297)	(0.447)
Control	Yes	Yes	Yes	Yes	Yes	Yes
Obs	107,995	107,995	107,995	107,995	107,995	107,995
R-squared	0.508	0.378	0.501	0.377	0.500	0.371
Firm FE.	YES	YES	YES	YES	YES	YES
Month FE.	YES	YES	YES	YES	YES	YES
Bank Num.	843	843	843	843	843	843

Note: This table reports the effect of macro-prudential policy on bank systemic risk based on Model (2), in which we interact the macro-prudential index with a dummy variable (Small, Region, and Diversify). Bank and time fixed effects are included. Additionally, *, **, *** represent significance at the 10%, 5%, and 1% level, respectively.

Regarding the banks’ geographical regions of operation, Columns 3 and 4 show that the coefficient of iMaPP is significantly negative. Therefore, macro-prudential policy can mitigate the risk for national bank. The positive coefficients in the second row suggest that the policy has a greater impact on national banks’ risk than that on regional banks’ risk. A t-test for ${\alpha }_1+{\alpha }_2$ suggest that although macro-prudential policy has a smaller impact on the risk of smaller and regional banks, it remains effective. Regarding business diversification, the result shows that macro-prudential policy has a stronger impact on the systemic risk of banks with more diversified operations.

Combining the regression results in Table 3, it can be observed that macro-prudential policy has a stronger impact on the risk of banks with larger size, a wider geographical operational scope, or more diversified business services. These institutions possess a significant market share, maintain intricate counterparty relationships, and serve as critical nodes in the financial network. If these banks are in distress, it could pose a huge threat to the financial stability. As a result, large banks are often the primary focus of macro-prudential regulation, particularly if they are on the list of global systemic important banks. Therefore, macro-prudential policy has a greater influence on the individual risks and systemic linkages of these banks than others.

4.5. Heterogeneous Impact Across Countries

In this part, we analyze the variation behind the average policy-risk relationship from four country-level perspectives. The first two columns of Table 4 indicate that the policy effect is weaker in a more developed financial system. This result is in line with the findings of Cerutti, Claessens, and Laeven (2017). The impact of financial development on macro-prudential policy may be twofold: on the one hand, higher levels of development enhance institutional capacity; on the other hand, greater financial development implies a more complex financial system and augments the challenges of policy implementation. The coefficient of the interaction term is significantly positive, indicating that the degree of financial development diminishes the effectiveness of macro-prudential policy.

Table 4. Heterogeneity in the policy-risk relationship: country level.

Variable	FinDev		Con		Open		Emerging
	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
iMaPP	−0.024***	−0.055***	−0.009**	−0.031***	−0.021***	−0.047***	−0.015***	−0.023***
	(0.004)	(0.007)	(0.004)	(0.007)	(0.003)	(0.006)	(0.003)	(0.004)
MaPP*dummy	0.009**	0.022***	−0.013***	−0.014**	0.012***	0.021***	−0.003	−0.013***
	(0.004)	(0.007)	(0.004)	(0.006)	(0.004)	(0.006)	(0.003)	(0.004)
Constant	0.796**	0.738	0.770**	0.631	0.755**	0.672	0.444	0.275
	(0.336)	(0.501)	(0.326)	(0.502)	(0.339)	(0.512)	(0.304)	(0.454)
Control	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Obs	88,295	88,295	90,170	90,170	90,170	90,170	107,995	107,995
R-squared	0.515	0.390	0.519	0.393	0.519	0.394	0.497	0.370
Firm FE.	YES	YES	YES	YES	YES	YES	YES	YES
Month FE.	YES	YES	YES	YES	YES	YES	YES	YES
Bank Num.	773	773	790	790	790	790	843	843

Note: This table reports the effect of macro-prudential policy on bank systemic risk based on Model (2). We interact the macro-prudential policy index with dummy variables (FinDev, Con, Open, and Emerging). Bank and time fixed effects are included. Additionally, *, **, *** represent significance at the 10%, 5%, and 1% level, respectively.

In terms of banking competition, a larger Boone index signifies a higher degree of concentration. Therefore, the coefficient of the interaction terms indicates that macro-prudential policy is more effective in more concentrated financial markets. The result can be explained by the “concentration-fragility hypotheses.” Larger banks tend to have higher systemic risk (Beck, De Jonghe, and Schepens 2013). A highly concentrated banking system often consists of numerous large banks. Macro-prudential policy restricts the specific behavior of large banks. Thus, implementing macro-prudential policy can mitigate the negative impacts of bank concentration.

The results presented in Column 5 and 6 of Table 4 indicate that the relationship between macro-prudential policy and systemic risk is weaker in more open countries. These results align with the findings of Cerutti, Claessens, and Laeven (2017) as well as Reinhardt and Sowerbutts (2015). Macro-prudential regulations have an impact on cross-country capital flows. When the lending policy of the host countries are tightened, foreign banks are not subject to the same policy and increase their cash flows to the host countries. Consequently, this weakens the effectiveness of macro-prudential policy. A t-test for ${\alpha }_1+{\alpha }_2$ is conducted and the results show that the sum of the coefficients is significantly different from zero. In regions with a higher degree of financial openness and financial development, macro-prudential policy can effectively reduce bank systemic risk.

We analyze the distinct impacts of macro-prudential policy on advanced and emerging economies. The results are presented in the final two columns of Table 4. The results in Column 8 indicate that the effectiveness of macro-prudential policy is weaker in advanced economies. The regression results align with previous paragraphs. The effectiveness of macro-prudential policy is weaker in advanced economies because of their greater financial development and openness.

4.6. Heterogeneous Impact at Different Cycle Stages

We also analyze the heterogeneous impact of macro-prudential policy on systemic risk across different stages of financial cycle. Table 5 demonstrates that this policy is more effective during credit-crunch periods. Some macro-prudential policy tools, such as capital buffers, exhibit countercyclical characteristics. In response to the credit crunch, regulators relaxed constraints and alleviated the burden on banks, leading to a reduction in risk within the banking sector. A t-test for ${\alpha }_1+{\alpha }_2$ show that the sum is significantly different from zero. These results suggest that macro-prudential policy also contributes to a reduction in bank risk during periods of credit expansion, although the effectiveness are weakened.

Table 5. Heterogeneity in the policy-risk relationship: cycle stages.

Variable	Exp		Loose		Crisis		Before2008
	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%	$\Delta \mathit{CoVaR}\mathrm{\%}$	MES%
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
iMaPP	−0.018***	−0.035***	−0.015***	−0.031***	−0.020***	−0.046***	−0.016***	−0.031***
	(0.002)	(0.004)	(0.002)	(0.004)	(0.003)	(0.005)	(0.002)	(0.004)
dummy	−0.129***	−0.205***	0.080***	0.063**	0.487***	0.525***
	(0.014)	(0.023)	(0.019)	(0.027)	(0.055)	(0.072)
iMaPP*dummy	0.003***	0.006***	−0.007***	−0.006***	−0.068***	−0.130***	0.040***	0.066***
	(0.001)	(0.001)	(0.001)	(0.002)	(0.025)	(0.039)	(0.008)	(0.017)
Constant	0.579**	0.471	0.382	0.228	0.016	−0.169	0.559*	0.439
	(0.294)	(0.445)	(0.299)	(0.454)	(0.306)	(0.498)	(0.296)	(0.447)
Control	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Obs	107,995	107,995	107,902	107,902	82,392	82,392	107,995	107,995
R-squared	0.499	0.372	0.498	0.370	0.533	0.401	0.499	0.373
Firm FE.	YES	YES	YES	YES	YES	YES	YES	YES
Month FE.	YES	YES	YES	YES	YES	YES	YES	YES
Bank Num.	843	843	843	843	715	715	843	843

Note: This table reports the effect of a macro-prudential policy on bank systemic risk based on Model (3), considering the influence of cyclical factors. We interact the macro-prudential policy index with dummy variables (Exp, Loose, Crisis, and Before2008). The Before2008 term is not included in the model because Before2008 does not vary among individual banks. There is a multicollinearity problem between Before2008 and the time fixed effects. Bank and time fixed effects are included. Additionally, *, **, and *** denote significance at 10%, 5%, and 1% level.

The coefficient of Loose in Table 5 is significantly positive. This reflects the monetary policies’ risk-taking channel (Borio and Zhu 2012). The coefficients of the interaction terms are significantly negative, indicating that macro-prudential policy is more effective in the presence of loose monetary policy. By targeting bank systemic risk, macro-prudential policy can mitigate the negative impact of loose monetary policy on financial stability.

The distinct effects of macro-prudential policy during crisis and non-crisis periods are tested, and the results are shown in Table 5. The significantly negative coefficients of the interaction terms indicate that the effectiveness of macro-prudential policy is stronger during crisis periods compared to non-crisis periods. These results are similar to those reported by Berger et al. (2022).

Lastly, the disparities in policy effectiveness pre- and post- 2008 are analyzed. The empirical results show a negative coefficient of iMaPP index and a positive coefficient of the Before2008 interaction term. This implies that macro-prudential policy can significantly mitigate bank systemic risk, and this effectiveness further improves in the aftermath of the crisis.

5. Further Analysis

The initial level of macro-prudential policy affects the policy effectiveness. Alam et al. (2019) observe, through the analysis of loan to value (LTV) cap data, that when LTV regulations are already tight, the impact of additional tightening on credit are dampened, while those on consumption are strengthened. We follow the method of An et al. (2021) and use a nonlinear model to examine the effects of the initial policy level. The model is conducted as follows:

(6) $\mathit{Ris}{k}_{\mathit{i},\mathit{c},\mathit{t}}={\alpha }_0+{\alpha }_1\mathit{MacroPru}\_\mathrm{lo}{\mathrm{w}}_{\mathit{c},\mathit{t}}+{\alpha }_2\mathit{MacroPru}\_\mathrm{hig}{\mathrm{h}}_{\mathit{c},\mathit{t}}+\sum _k^K{\beta }_k\mathit{Mi}{c}_{\mathit{k},\mathit{i},\mathit{t}}+\sum _l^L{\gamma }_l\mathit{Ma}{c}_{\mathit{l},\mathit{c},\mathit{t}}+{\mu }_i+{\nu }_t+{\epsilon }_{\mathit{i},\mathit{c},\mathit{t}}$(6)

$\mathit{MacroPru}\_\mathrm{lo}{\mathrm{w}}_{\mathit{c},\mathit{t}}\mathit{\hspace{0.17em}}$ represents the case of a looser macro-prudential policy, and $\mathit{MacroPru}\_\mathrm{hig}{\mathrm{h}}_{\mathit{c},\mathit{t}}$ represents the case of a tighter macro-prudential policy. $\mathit{MacroPru}\_\mathrm{lo}{\mathrm{w}}_{\mathit{c},\mathit{t}}$ and $\mathit{MacroPru}\_\mathrm{hig}{\mathrm{h}}_{\mathit{c},\mathit{t}}$ are defined as follows:

(7) $\mathit{MacroPru}\_\mathit{lo}{w}_{\mathit{c},\mathit{t}}\left\{\begin{array}{cc}\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}& \mathrm{if}\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}<\mathit{mea}{n}_c\\ \mathit{mea}{n}_c& \mathrm{if}\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}\ge \mathit{mea}{n}_c\end{array}\right.$(7)

(8) $\mathit{MacroPru}\_\mathit{hig}{h}_{\mathit{c},\mathit{t}}\left\{\begin{array}{cc}0& \mathrm{if}\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}<\mathit{mea}{n}_c\\ \mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}-\mathit{mea}{n}_c& \mathrm{if}\mathit{MacroPr}{u}_{\mathit{c},\mathit{t}}\ge \mathit{mea}{n}_c\end{array}\right.$(8)

The empirical results are presented in Column 1 to 4 of Table A6 in the Supplementary Appendix present. The results show that the coefficient of iMaPP_high is larger than that of iMaPP_low. When the dependent variables are ΔCoVaR, MES, and Individual Risk, a significant difference is observed. Therefore, the results show that when macro-prudential policy is already stringent, the impact on systemic risk of further tightening is reduced. Under stringent macro-prudential policy, banks operate with greater caution, resulting in diminished effects of additional policy constraints on financial stability.

There are multiple tools in the macro-prudential policy toolbox. To test the impact of these diverse policy instruments, we categorize them into three groups: capital, credit, and liquidity.² The empirical results, presented in Columns 5 and 6 of Table A6 in the Supplementary Appendix, shows that all three types of macro-prudential policies are effective in reducing bank systemic risk. Notably, credit policy instruments exhibit a greater impact on risk.

6. Conclusion

This study empirically tests the impact of macro-prudential policy on bank systemic risk using data of 65 countries from 2000–2019. The empirical results reveal that the implementation of macro-prudential policy mitigates individual risk and systemic linkages, and thus suppresses systemic risk. In addition, the effectiveness of these policy varies across financial systems and financial cycle phases. Finally, we find that the initial policy level plays a significant role in shaping the impact of macro-prudential policy on financial stability.

This study has some limitations. First, our policy indicators only capture whether macro-prudential policy is tightened or loosened, calling for additional quantitative studies to test the change of systemic risk when macro-prudential policy regulatory indicators are raised by one unite. Moreover, our risk proxy relies on stock price data, which means our sample is limited to listed banks and may not accurately represent unlisted banks.

Future research could be conducted. This study focuses on the banking sector and overlooks the risk contagion between banks and other financial sectors, such as real estate. With the increased interconnectedness among financial sectors, it is essential to address risk contagion across sectors. It would be interesting to include this industry correlation in future studies.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/1540496X.2023.2266114

Additional information

Funding

National Social Science Foundation of China “A Study on the Formation, Changes and Governance of Commercial Bank Risks Driven by Financial Technology” [NO. 22BJY008].

Notes

1. The list of 17 policy instruments is based on the International Monetary Fund (2014) paper. iMaPP database contains all the instruments mentioned in the IMF paper. These instruments are grouped into 17 categories. Table A1 in the Supplementary Appendix presents the detail.

2. Our classification approach follows the method of Lim et al. (2011).