Central bank disclosure as a macroprudential tool for financial stability

Central bank disclosure as a macroprudential tool for financial stability

Accepted Manuscript Title: Central bank disclosure as a macroprudential tool for financial stability Authors: Helder Ferreira de Mendonc¸a, Claudio Ol...

529KB Sizes 0 Downloads 75 Views

Accepted Manuscript Title: Central bank disclosure as a macroprudential tool for financial stability Authors: Helder Ferreira de Mendonc¸a, Claudio Oliveira de Moraes PII: DOI: Reference:

S0939-3625(18)30141-9 https://doi.org/10.1016/j.ecosys.2018.07.001 ECOSYS 674

To appear in:

Economic Systems

Received date: Revised date: Accepted date:

13 March 2018 27 June 2018 13 July 2018

Please cite this article as: de Mendonc¸a HF, de Moraes CO, Central bank disclosure as a macroprudential tool for financial stability, Economic Systems (2018), https://doi.org/10.1016/j.ecosys.2018.07.001 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Central bank disclosure as a macroprudential tool for financial stability Helder Ferreira de Mendonçaa,b,* and Claudio Oliveira de Moraesa,c,**

Fluminense Federal University, Department of Economics

b

National Council for Scientific and Technological Development (CNPq)

c

Central Bank of Brazil

SC RI PT

a

* Corresponding author. Address: Rua Dr. Sodré, 59, Vila Suíça, Miguel Pereira, Rio de Janeiro, Brazil. CEP: 26900-000; Tel.: +55(24)2484-4143; E-mail address: [email protected]

M

A

N

900; E-mail address: [email protected]

U

** Address: Avenida Presidente Vargas, 730 Centro-RJ, Rio de Janeiro, Brazil, CEP: 20071-

Highlights

Disclosure of information (transparency) is essential to financial stability.



We build two indices based on central bank communication on credit development.



Central bank communication is shown as a macroprudential tool for financial stability.

CC

EP

TE

D



Abstract

A

The Basel Capital Accord (pillar 3) states that disclosure of information (transparency)

is essential to financial stability. This study analyzes, through inflation reports, the disclosure of information from the Central Bank of Brazil concerning the credit market. We consider credit risk and capital buffers as measures of financial stability in this analysis. Furthermore, in order to measure the perception of the monetary authority on the credit market, we built two indices based on the central bank’s communication on credit development. We performed a 1

panel data analysis based on a sample of 125 banks for the period from June 1999 to September 2014 (7,000 observations). The findings suggest that central bank communication regarding expectations concerning the credit market contributes to financial stability. Therefore, this kind of communication of central banks (about credit development) may

SC RI PT

constitute an important macroprudential tool to improve financial stability.

Keywords: Financial stability, Macroprudential regulation, Central bank communication, Credit risk, Capital buffer JEL classification: E58, G38

U

1. Introduction

N

The subprime crisis has shown that neither price stability nor traditional

A

microprudential regulation are strong enough to support financial stability. In this context, it is

M

an important issue for policymakers to find macroprudential instruments that are able to mitigate the risks to financial stability. Nowadays, central banks make use of transparency in

D

their actions and forecasts of economic variables as part of their management tools. In

TE

particular, the disclosure of predictions regarding the credit market is an important factor to reduce information asymmetry in the financial market.

EP

Central banks have increased disclosure of their actions. According to the Basel

CC

Capital Accord (pillar 3), disclosure of information (transparency) is essential to financial stability. Nevertheless, transparency regarding financial stability has only recently gained

A

relevance. As a result, efforts to separate monetary policy communication from macroprudential issues are still ongoing in central banks (Born et al., 2012). In this context, the primary source of information on financial stability provided by central banks is the Financial Stability Report (FSR). However, in the Brazilian case, the FSR does not present a forward-looking view of the financial market (see Horváth and Vaško, 2016), but the 2

Brazilian Inflation Report has a specific section (“inflation outlook”) that takes into account an assessment of the monetary authority regarding the expected behavior of the credit market. Moreover, since the subprime crisis, the Central Bank of Brazil (CBB) has used this section in

SC RI PT

its Inflation Report to warn about macroprudential risks caused by credit growth. This study analyzes the effect of central bank communication concerning the developments in credit markets on credit risk and capital buffers. Hence, based on the content in the section “Outlook for Inflation” in the Inflation Report of the Central Bank of Brazil (CBB), we built two indices regarding the credit market. We also consider control variables

U

that can influence credit risk and the capital buffer through a panel data analysis for 125 banks

N

from 1999 to 2014. Quarterly data is gathered from economic-financial analyses provided by

A

the CBB.

M

The findings suggest that the forecasts on credit market behavior released by the CBB can affect banks’ expectations regarding credit risk and solvency risk (capital buffer). In other

D

words, this kind of communication from the CBB can lead to a reduction of asymmetry

TE

among banks, which contributes to the stability of the financial system. Furthermore, our

EP

results confirm the pro-cyclical nature of the financial system. Based on content in the Inflation Reports (section “Outlook for Inflation”), two baseline indicators permit us to extract

CC

the signal issued by communication regarding the expected scenarios for the credit market. The first index is a classification that takes into account three views of the monetary authority

A

in relation to credit market conditions: improvement, worsening, or maintenance of credit conditions. The second index considers changes in the monetary authority’s perspective on the credit market within a semester. Moreover, we use two measures for credit risk (loan provision and nonperforming loan) and one for solvency risk (capital buffer) as proxies for financial stability. 3

After the subprime crisis, there was an urgency to develop prudential policies that not only consider individual failures, but that are also able to preserve the entire financial system (Hanson et al., 2011; Galati and Moessner, 2013). Policymakers needed to find time-varying

SC RI PT

tools to avoid excesses in the credit market and the accumulation of financial imbalances (BoE, 2011; Borio, 2011; IMF, 2011). In order to improve systemic resilience and thus prevent the buildup of financial vulnerabilities in banks, Basel III suggests the use of a countercyclical capital buffer (BCBS, 2010; Drehmann et al., 2011). However, there is no clarity regarding macroprudential tools that are able to correct market failures endangering

U

financial system stability (Claessens, 2015). This is an important issue, because Basel III

N

indicates that central bank communication and the use of capital buffers are able to affect

A

market expectations (BCBS, 2010).

M

The problem of asymmetric information is one of the main issues for an analysis of the effects of central bank communication on market expectations. The communication of the

D

monetary authority on future market behavior affects agents’ expectations and therefore

TE

represents an economic policy tool (Blinder et al., 2008; Dincer and Eichengreen, 2014).

EP

Several studies have analyzed the importance of central bank communication on the financial market and find that there exists an effect on financial assets (Reeves and Sawicki, 2007;

CC

Ehrmann and Fratzscher, 2007; Born et al., 2014). This study represents an innovation to the aforementioned literature because, instead of considering a general measure of central bank

A

communication, we build specific indices focusing on credit development. Therefore, it is possible to observe the real impact this kind of communication may have on credit risk and the capital buffers of banks, and consequently on the financial system. In brief, from the findings of this study, it is possible to outline, configure and build a macroprudential tool that is able to mitigate the vulnerabilities of the financial system without creating new distortions. 4

This paper is highly topical. While macroprudential policy has featured prominently in the policy debate in the wake of the global financial crisis, little is yet known regarding the effectiveness of macroprudential instruments. Only a few years ago, the then Deputy

SC RI PT

Governor of the Bank of England, Charles Bean, argued that “we are still in the Stone Age in respect of deploying macroprudential policies. There is lots of scope for academia to help us out here, on both the theoretical and empirical fronts.” Moreover, only very few papers have investigated the impact of central bank communication on financial stability and particularly little is known about the experience of emerging market countries.

U

This paper is organized as follows. Section 2 describes the data and variables as well

N

as the empirical model and the methodology. Section 3 presents the estimation results for

A

credit risk and capital buffers, focusing the analysis on the effect caused by the central bank’s

M

communication concerning credit development on financial stability. Section 4 concludes the

TE

2. Data and methodology

D

paper.

EP

With the objective of analyzing how central bank communication on credit development (CBCCD) affects financial stability, we build two indices of CBCCD concerning

CC

the perception of the monetary authority on the credit market. Furthermore, in order to consider financial stability in the models, this study makes use of two categories of risk

A

traditionally found in the literature, i.e. credit risk and the solvency risk of banks or capital buffers. The next subsection presents the procedure used to obtain indices of CBCCD and financial stability, as well as the control variables used in the models. The second subsection provides the models and methods that are used in order to analyze whether the CBCCD can contribute to financial stability. 5

2.1 Central bank communication on credit development, credit risk and capital buffers This study takes into account sixty-two Inflation Reports from June 1999 (release of

SC RI PT

the first Inflation Report) to September 2014. This report is part of the Brazilian inflation targeting system; its content provides the most complete overview of the economy and includes a subsection analyzing the credit market. It is important to note that the release of the reports occurs before the last day of the quarter (see Table A.1 in the Appendix), and thus this information is available to banks before they close their balance sheets. As pointed out by

U

Doblas-Madrid and Minetti (2013), in order to mitigate information asymmetry, a well-known

N

problem in the credit market, banks take into account all new information immediately. In

A

other words, the possible effect from the content of the report on the banks can be observed in

M

the same quarter.

To extract the relevant information for the construction of our communication indices,

D

we follow the methodology proposed by Rosa and Verga (2007). One advantage of this

TE

methodology is that it is simple and captures the intention of the policymaker well through its

EP

statements. Hence, a glossary was drawn up in order to identify the language used by the monetary authority in the Inflation Report to signal the expected behavior of the credit market

CC

(see Table 1). In general, it is possible to observe that the CBB is more talkative (use of different keywords) when there is an expectation of increase in the credit market. This

A

observation allows us to conjecture that the CBB has adopted a strategy of communication (even unconsciously) in order to warn the public regarding macroprudential risks associated with credit growth. The first CBCCD index measures the CBB’s expectation regarding future credit behavior. This index corresponds to the central bank’s communication on the credit 6

development time variant (COM1), which is a result of the classification of the CBB’s expectation in relation to an increase or decrease in the credit market. In this context, based on the content published in the CBB’s Inflation Report, the index can take three values: “+1” for

SC RI PT

the case where there is an expectation of an increase in the credit market; “-1” for the expectation of a decrease in the credit market, and “0” for the case where there is no expectation of fluctuation in the credit market.

U



N

Based on the COM1 index, Figure 1 allows us to observe optimism of the monetary

A

authority in relation to the credit market (expectations of growth in the credit market) most of

M

the time. Of the 62 quarters, an expansion of the credit market was signaled in 29, and a contraction was indicated only in 6 quarters. Furthermore, it is possible to see three different

D

scenarios:

TE

– Second half of 1999 to the first half of 2003: a deceleration of credit was aggravated by the

EP

crisis of confidence generated by Luiz Inácio Lula da Silva being favored to win the presidential election.

CC

– Second half of 2003 until the end of 2011: the expectation of expansion of the credit market over this period is due to the fact that before the subprime crisis there was significant

A

international liquidity and afterwards there was a strong countercyclical economic policy. – From 2012: the monetary authority signals a slowdown in the credit market. This behavior is a result of the worsening political and economic crisis in the country during the government of President Dilma Rousseff.

7



Because the financial market has the information which is disclosed in the middle of

SC RI PT

the year as the main benchmark for its annual forecast, the second communication macroprudential index used in this study (COM2) takes into account the perspective of the monetary authority in one semester. Since COM1 refers to the monetary authority’s perspective in a quarter, COM2 corresponds to the accumulation of COM1 regarding the previous

period

to

the

current

period

(accumulated

CBCCD

index),

that

is,

COM 2t  t 1 COM 1 . Therefore, this index takes values between -2 and +2. The

U

t

N

interpretation of this index is straightforward; a value of +1 means that the monetary authority

A

signals the prospect of expansion of the credit market, while the value +2 means that it signals

M

the continued expansion of the credit market. On the other hand, the values -1 and -2 represent

D

situations in which the monetary authority indicates a contraction in the credit market. Finally,

TE

the value 0 refers to a situation in which the signaling of the monetary authority does not allow identifying a perspective of the credit market expanding or contracting (situation in

EP

which the monetary authority indicates neutrality in two subsequent periods) or when there exists a period of expansion (contraction) followed by a period of retraction (expansion).

CC

Based on the COM2 index, Figure 2 allows us to observe that in 21 quarters there is an

indication of continuing credit expansion, and that in the other 15 quarters there is an

A

expectation of credit growth. This suggests that most of the period is marked by the perspective of an expanding credit market. The expectation of a continued contraction in the credit market is observed in only two quarters. Moreover, in another seven quarters the contraction is not continued. As observed in Figure 1, the credit market downturn expectations

8

coincide with those for deepening confidence crises in the governments of Luís Inácio Lula da Silva and Dilma Rousseff.

SC RI PT



Credit risk and bank solvency risk have the greatest impact on the stability of the financial system.1 Credit risk and the capital buffer (the proxy for bank solvency risk) are measures that the literature considers representative of the instability of the financial system.2

U

This study makes use of data from 125 Brazilian financial institutions (97% of total financial

N

system assets) for the period from June 1999 to September 2014. Data were obtained from 56

A

quarterly reports on “information for economic-financial analysis” provided by the CBB. It is

M

important to emphasize that the financial market in Brazil is one of the largest and most complex structures among the developing economies (Park, 2012). In addition, the period

D

under consideration corresponds to the inflation targeting period in the country and therefore

TE

takes into account the practice of regular communications from the monetary authority (de

EP

Mendonça and Faria, 2012).

Regarding credit risk, this study makes use of two measures that are enshrined in the

CC

literature: the loan loss provisions/gross loans ratio (CR1), which represents the expected loss of banks with respect to loans; and the nonperforming loan/total loans ratio (CR2), which

A

represents the expected loss of loans that were delayed. The capital buffer (BUF) is measured by the difference between the capital of financial institutions and the minimum capital 1

Regarding the impact of banking solvency and credit risk on financial stability, see Bernanke (1983), Kaminsky and Reinhart (1999), Jiménez and Saurina (2006), Brunnermeier (2009), Altunbas et al. (2010), Jordà et al. (2013), Reinhart and Rogoff (2013) and Jokivuolle et al. (2015). 2 On the relation between capital buffers and credit risk with financial stability, see Hume and Sentance (2009), Foos et al. (2010), Drehmann et al. (2011), Pesola (2011) and Dell’Ariccia et al. (2011).

9

required by regulators (Basel index). Figure 3 shows the path of the three measures (CR1, CR2 and BUF), taking into account the average of information from financial institutions in the sample. In general, it is observed that both credit risk measures (CR1 and CR2) have a

SC RI PT

downward trend that is interrupted only during periods when there were crises (the crisis of confidence due to the presidential election of 2002 and the subprime crisis). Regarding the capital buffer, it is observed that the beginning of the period is marked by a significant increase in this indicator in possible anticipation of the changes introduced by the Basel Accord of 2004. After this period there is a gradual decrease (except between 2008 and 2009

N

U

as a result of the subprime crisis) due to the expansion of credit in the economy.

M

A



2.2 The relation between CBCCD and financial stability

D

An important aspect about the relationship between central bank communication and

TE

financial stability is that banks operate in a market with asymmetric information and thus

EP

make use of all available information in their decision-making (Büyükkarabacak and Valev, 2012). It is important to highlight that central bank’s predictions regarding the credit market

CC

work as a guide to banks’ expectations. Moreover, the release of the Inflation Report (see Table A.1 in the Appendix) always comes before the bank’s quarterly balance sheets are

A

issued, in which exposures to credit risk and solvency risk (capital buffer) are revealed. Therefore, the possibility of an effect caused by CBCCD on credit risk and capital buffers is not negligible. In other words, it is expected that this kind of communication can reduce the asymmetry of information in the financial system, which in turn increases financial stability due to less credit risk and an increased capital buffer. 10

In general, credit risk, as well as solvency risk, can be mitigated but not fully eliminated from the financial system, which in turn suggests that the risk observed in the previous period is relevant to the explanation of the current period. Because the CBCCD is a

SC RI PT

source of information that banks take into account in the decision-making process related to credit risk and solvency risk, the above communication measures (COM1 and COM2) are introduced as explanatory variables along with those traditionally used in the literature (GAP, IR, LIQ, CRED, ROA and ROE).3 Therefore, the baseline model is given by: (1)

FSi ,t   0  1FSi ,t 1   2COM t  3 Zi ,t   i ,t ,

U

where i=1, …, 125 is the cross-section unit (banks), t=1, …, 56 is the time index (quarterly),

N

FSi,t is a vector of variables regarding financial stability (CR1, CR2, and BUF), COMt is a

A

vector of variables regarding CBCCD (COM1 and COM2), Zi,t is a vector of explanatory

M

variables (GAP, IR, LIQ, CRED, ROA, and ROE), and εi,t is the stochastic error term.

D

As pointed out by Pool et al. (2015), there is a relationship between the business cycle,

TE

the credit market and bank risk-taking. Therefore, in order to evaluate the effect of central bank communication regarding credit development (CBCCD) on credit risk and capital

EP

buffers, we take into account the possible effect from the business cycle and bank risk-taking. Hence, we include in the set of control variables (Zi,t) the output gap (GAP, i.e., the difference

CC

between GDP and potential output) and the monetary policy interest rate (IR) to the previous specifications. Furthermore, at least in theory, there is a positive relationship between liquidity

A

and credit risk (see Imbierowicz and Rauch, 2014). Hence, a liquidity indicator (LIQ) is also included as a control variable in the model. In addition, loan growth can be an important driver of the riskiness of banks (see Foos et al., 2010), and thus we introduce the credit growth

3

See Stolz and Wedow (2011), Gambacorta and Mistrulli (2004), Dahl (2012) and Tabak et al. (2013).

11

rate (CRED) into the set of control variables.4 Finally, because the performance of a financial institution can affect financial stability, we take into account variables traditionally used as proxies for the performance of a financial institution (ROA and ROE; see Berger and

SC RI PT

Bouwman, 2013). All data is obtained from the CBB (Table A.2 in the Appendix presents a description of the variables and sources of data).

Because the largest banks (in terms of assets) have more capacity to destabilize the financial system, this study considers, in the total sample, a subsample of large and medium banks that are not controlled by the government. Therefore, considering the same period as in

U

the total sample, a subsample of the 19 largest Brazilian private banks according to total assets

N

(information from the financial stability report of the CBB) is used. 5 Public banks (for

A

example, BNDES and Caixa Econômica Federal) are not included in the sample because the

M

government uses them for economic policy actions and thus they do not follow market behavior.

D

In addition to considering the credit risk measures (CR1 and CR2) and solvency risk

TE

(BUF) to analyze the effect of CBCCD on financial stability, a new measure is used to

EP

observe the phenomenon. In order to take into account both the characteristics related to credit risk and to solvency risk measures, the default coverage rate (COV) is introduced into

CC

equation 1 (FS variable). COV is a variable computed through the information gathered from the balance sheet of each bank and is the loan loss provisions/nonperforming loan ratio.6 In

A

this paper, the main idea is that by using COV it is possible to certify the results that are

“Credit developments” is a key element of the central bank communication indices (qualitative variable) associated with the central bank’s perspective on the credit market. In a different way, the credit growth rate variable (CRED) is an objective measure of how banks have increased or decreased lending. 5 The largest banks are those whose relative participation exceeds 15% of total financial system assets (CBB, 2011). 6 The measure “bank provision to nonperforming loan” is available from the IMF’s Global Financial Stability Report. 4

12

observed when utilizing the variables CR1, CR2 and BUF in equation 1. This study makes use of dynamic panel data analysis. It is important to note that the use of a lagged dependent variable in the models may lead to a correlation problem with the

SC RI PT

error term, which therefore causes bias and inconsistency in OLS estimators (Baltagi, 2005). Furthermore, due to the fact that the control variables used in the model can be determined at the same time as the risk measures, the possibility of endogeneity cannot be neglected. The solution proposed by Arellano and Bond (1991) is the estimation of first-difference GMM panel data (D-GMM). However, Blundell and Bond (1998) show that the use of this method

U

implies weak instruments, and has a bias (for large and small samples) and low accuracy. In

N

order to deal with these problems, Arellano and Bover (1995) and Blundell and Bond (1998)

A

proposed the system GMM panel data (S-GMM). In general, the S-GMM method combines

M

regression equations in differences and in levels into one system and uses lagged differences and lagged levels of the variables in the model as instruments (Bond et al., 2001).

D

Due to the above-mentioned features, this study makes use of the S-GMM method.7 It

TE

is important to note that when there are too many instruments, they tend to over-fit the

EP

instrumented variables, creating bias in the results (Roodman, 2009). Therefore, in order to avoid the use of an excessive number of instruments in the regressions, the number of

CC

instruments/number of cross-sections ratio is limited to be less than 1 in the regressions (de Mendonça and Barcelos, 2015). 8 In addition, in order to check for the validity of the

A

instruments in the models, the test of over-identifying restrictions (J-test) is performed, as suggested by Arellano (2003). Moreover, tests of first-order (AR1) and second-order (AR2) 7

Regarding the use of the S-GMM in macroprudential analysis, see Guidara et al. (2013) and Stolz and Wedow (2011). 8 Besides the lagged values of the variables in equation (1), we also use other variables present in the banking literature as instruments: the regulatory capital assets ratio (CAR) and the total amount of credit (TCRED); see Table A.3 in the Appendix.

13

serial correlation are performed. It is important to highlight that although we perform two-step GMM estimation, our sample is not characterized by a small number of periods t (t=56) and is not small relative to the number of banks (i=125). Hence, there is no risk of over-fitting the

SC RI PT

instrumented variables and biasing the results, thereby making the two-step system GMM estimator consistent (see Hayakawa, 2012).

Besides the above-mentioned procedures, in order to eliminate the possibility of endogeneity in the estimations regarding the effect of CBCCD indices on risk measures, the Granger causality test is performed (see Table A.4 in the Appendix). Furthermore, Table A.5

U

in the Appendix shows the correlations among the variables used in the models. In particular,

N

a negative correlation is observed between the variables related to communication (COM1 and

A

COM2) and those on credit risk (CR1 and CR2). Moreover, the correlation between CBCCD

M

measures and the capital buffer (BUF) is positive. Tables A.2 and A.6 in the Appendix present

TE

of banks, respectively.

D

the descriptive statistics of the samples (total and largest Brazilian private banks) and the list

EP

3. Empirical evidence

This section presents an analysis concerning the effect of CBCCD on credit risk and

CC

bank solvency risk (capital buffer). The first subsection provides empirical evidence on the effect of communication measures (COM1 and COM2) on credit risk measures (CR1 and

A

CR2). The second subsection presents the effects of CBCCD measures on the capital buffer, while the third subsection introduces the nonlinear effects of communication measures (COM12 and COM22) on both credit risk measures and capital buffer. Moreover, this subsection provides evidence regarding the effect of CBCCD on the default coverage rate (COV). 14

3.1 Credit risk Taking into account the total sample (125 banks for the period from June 1999 to

SC RI PT

September 2014 – 56 quarters), the results regarding the effect of CBCCD on the measures of credit risk (CR1 and CR2) are presented in Table 2. In general, all regressions accept the null hypothesis in the Sargan tests (J-statistic) and thus the over-identifying restrictions are valid. Furthermore, both serial autocorrelation tests (AR(1) and AR(2)) do not denote the presence of serial autocorrelation.

U

The statistical significance observed for all coefficients related to CBCCD allows us to

N

infer that credit risk (CR1 and CR2) undergoes an effect after the central bank communicates

A

its predictions on the future behavior of the credit market. Moreover, the fact that all

M

coefficients on communication are negative suggests that there is a reduction in credit risk. In other words, a communication of an improvement in the credit market conditions by the

D

central bank (an increase in COM1 and COM2) can decrease banks’ expectation of loss, which

TE

in turn leads to a decrease in credit risk.

EP

The coefficients on output gap (GAP) are negative and significant in all models (see Table 2), which in turn suggests pro-cyclical behavior due to a decrease in the credit risk

CC

perception by banks in periods of economic growth. In relation to the monetary policy interest rate (IR), the fact of all coefficients being positive and significant denotes the presence of a

A

risk-taking channel. In other words, information asymmetries (adverse selection and moral hazard) arise in the credit market when, for example, the central bank raises the short-term interest rate (Altunbas et al., 2014). The coefficients on liquidity (LIQ) are positive and significant, thus indicating that liquidity and credit risks are positively correlated (see Ghenimi et al., 2017; Imbierowicz and 15

Rauch, 2014). It is important to highlight that liquidity assets have low yield. Hence, a possible consequence of an increase in liquidity is an increase in the bank’s risk appetite. Furthermore, the findings support the idea that the creation of liquidity can precede banking

SC RI PT

crises (see Berger and Bouwman, 2009).



The coefficients on the credit growth rate (CRED) are negative and significant. This

U

result indicates that new credit is associated with a lower default probability than the average

N

credit portfolio. Furthermore, the positive sign and the significance observed for the return on

A

assets (ROA) denotes the recognized relationship between risk and return. Finally, the

M

statistical significance and positive sign of the lagged credit risk variable indicate that there is a persistence of credit risk behavior.

D

The results of the credit risk estimation regarding the sample of the largest banks are

TE

presented in Table 3. Likewise, as observed for the case of the total sample, the tests (J, AR(1)

CC

models.

EP

and AR(2)) indicate that there is no problem of overidentification or autocorrelation in the

A



The effect of CBCCD on the largest banks is similar, in terms of statistical significance

and sign, to that observed for the total sample. Furthermore, the result shows that the magnitude of the coefficients for communication in the case of the largest banks is smaller. This can be explained by the fact that there is a greater technical capacity of the staff of the 16

largest banks, which in turn implies more accurate forecasts for credit market behavior. In general, the coefficients for the other variables in the models do not present significant changes in relation to the results found for the total sample. Regarding the results related to

SC RI PT

liquidity, one explanation could be the liability structure of large banks. Specifically, large banks are less reliant on short-term funding.

In brief, the empirical evidence indicates that CBCCD has an effect on credit risk. Moreover, CBCCD may reduce credit risk when there are signals towards a possible growth of the credit market. This result is observed for the financial system (total sample) as well as

N

U

for the case of the largest banks.

A

3.2 Capital buffer

M

The estimations regarding the effect of CBCCD on capital buffers follow the same structure as in the previous subsection. The results regarding the total sample and the sample

D

of the largest banks are presented in Table 4. Once again, the J, AR(1) and AR(2) tests do not

TE

indicate problems of overidentification or autocorrelation in the models.

EP

The results for both samples (total and largest banks) indicate that the CBCCD is associated with an increase in the capital buffer. The coefficients on the communication

CC

indices are positive and significant for all models. In other words, the information released by the central bank that there is an expectation of credit expansion leads banks to increase the

A

capital buffer. This increase can be justified by the forward-looking behavior of banks to accommodate a possible expansion of credit. Regarding the effect of the business cycle on the capital buffer, the coefficients on GAP are negative and significant in all models. This observation suggests that when there is an economic boom, this increases the risk appetite of banks, which in turn reveals a pro17

cyclical attitude of banks. This is an important observation because, according to Claessens (2015), the correct use of macroprudential policy must avoid or minimize the pro-cyclicality of banks’ behavior. In relation to the impact of monetary policy on the capital buffer, the

SC RI PT

coefficients on IR are positive and significant. In brief, banks react to the monetary policy interest rate by increasing the capital buffer (see de Moraes et al., 2016).



U

In the case of the estimations with the largest banks, the magnitude of the coefficients

N

on communication is lower than that observed for the financial system as a whole (total

A

sample). This result is similar to that seen in the estimations for credit risk, and therefore it is

M

possible that the greater technical capacity of the staff of the largest banks allows a greater capacity to anticipate changes in the market. Regarding the coefficient on LIQ, the signal is

D

positive and significant for all models. Therefore, the greater the liquidity of the bank’s assets,

TE

the lower is the risk of assets and, thus, the greater the capital buffer. Regarding the CRED

EP

variable, the coefficient is significant and negative in all models. This result suggests that, for example, when there is positive credit variation, the capital buffer is reduced because of the

CC

greater leverage with respect to bank credit. Regarding ROE, the coefficient is also negative but not significant in both samples. Indeed, there is a substitution effect in this case. An

A

increase in ROE implies a decrease in the capital buffer because it represents a cost for the shareholders (Jokipii and Milne, 2008). This observation suggests that banks with higher and more stable profits require smaller capital buffers. Finally, the positive and significant coefficient on the lagged capital buffer indicates that the persistent effect from this variable is not negligible. 18

3.3 Robustness analysis The results in the previous subsections indicate that CBCCD leads to a decrease in

SC RI PT

credit risk as well as an increase in the capital buffer. In other words, they suggest that this kind of communication can work as a macroprudential tool for the stability of the financial system. In order to investigate whether there is a limit to the effect of CBCCD on credit risk and capital buffer, we add a quadratic term on the COM1 and COM2 variables (COM12 and COM22, respectively) to equation (1). In other words, if the coefficients’ signs on COM12 and

U

COM22 are different from those on COM1 and COM2, there is a limit to the strategy of

N

increasing CBCCD. In this case, when CBCCD exceeds the threshold considered to be

A

optimal, increases in CBCCD will increase credit risk and decrease the capital buffer.

M

Focusing the analysis on the main variables of interest, that is, the coefficients on COM1, COM2, COM12 and COM22, we observe that the findings denote that there is no

D

ambiguity in the results (see Table 5). In other words, the results indicate that an increase in

TE

CBCCD decreases credit risk and increases the capital buffer. In particular, although the

EP

coefficients on the quadratic terms are smaller than on the non-quadratic terms, the benefits of

CC

increasing CBCCD are not eliminated.

A



We perform an additional analysis to confirm the robustness of our results, substituting

CR1, CR2 and BUF with the default coverage rate (COV) in the models. We use this dependent variable because it can be understood as a combination of the elements present in both credit risk and capital buffer. The results are presented in Table 6. 19



SC RI PT

It is important to highlight that the coefficients on CBCCD are statistically significant and negative in all models. One possible explanation for this result is that the default coverage rate represents a credit risk measure. Therefore, under an environment where there is a perspective of growth of the credit market, banks reduce the loan loss provision/loan default ratio. Finally, with the exception of GAP, the coefficients on the other explanatory variables in

U

the models (IR, LIQ, CRED, ROE and COV(-1)) do not present significant changes from those

A

N

observed in the estimations presented in the previous subsections.

M

5. Conclusion

This paper investigates whether central bank communication on developments in

D

credit markets (macroprudential instrument) affects financial stability. Making use of data

TE

from the Brazilian economy, we built two measures of communication on credit markets and

EP

analyzed whether these measures influence two measures of risk in the financial system. We performed several regressions based on dynamic panel models with quarterly data from 125

CC

Brazilian banks for the period 1999-2014. We found evidence of a significant effect of communication on credit developments on our measures of risk. Thus, central bank

A

communication can be an important macroprudential tool. The findings in this study permit us to observe how central bank communication on

credit development affects credit risk and the capital buffer. The results indicate that the disclosure of information by the central bank regarding the expectation of an increase in the credit market leads banks to decrease credit risk and increase the capital buffer. On the other 20

hand, a signal of reduction in the credit volume leads banks to assume higher credit risk and a lower capital buffer. In other words, due to the anticipation of the credit cycle, the central bank’s communication on credit development contributes to banks forming their expectations

SC RI PT

in order to maintain financial stability. Furthermore, in line with Guttentag and Herring (1984), the findings support a pro-cyclical nature of the financial system.

It is important to note that the main motivation of the monetary authority’s communication is not to anchor bank expectations with respect to financial stability. However, the results presented in this study suggest that reducing the asymmetry of information through

U

prospective communication on the credit market contributes to financial stability. In short, the

N

disclosure of information relating to credit market expectations may constitute an important

A

CC

EP

TE

D

M

A

macroprudential tool.

21

References ALTUNBAS, Y., GAMBACORTA, L., MARQUES-IBANEZ, D. (2010). “Bank risk and monetary policy.” Journal of Financial Stability, 6(3), 121–129.

SC RI PT

ARELLANO, M., BOND, S. (1991). “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations.” Review of Economic Studies, 58(2), 277-297.

ARELLANO, M., BOVER, O. (1995). “Another Look at the Instrumental Variables Estimation of Error-components Models.” Journal of Econometrics, 68(1), 29-51.

U

BoE - Bank of England. (2011). “Instruments of macroprudential policy: a discussion paper”,

N

December.

A

BALTAGI, B.H., (2005). “Econometrics analysis of panel data” (second edition). Wiley, New

M

York.

BCBS (2010) - Basel Committee on Banking Supervision - “Countercyclical capital buffer

D

proposal - consultative document”. July.

TE

BERGER, A.N., BOUWMAN, C.H.S. (2009). “Bank liquidity creation.” Review of Financial

EP

Studies, 22(9), 3779-3837.

BERGER, A.N., BOUWMAN, C.H.S. (2013), “How does capital affect bank performance

CC

during financial crises?”, Journal of Financial Economics, 109(1), 146-176.

A

BERNANKE, B. (1983). “Non-monetary effects of the financial crisis in the propagation of the great depression.” American Economic Review. 73(3), 257–276.

BLINDER, A., EHRMANN, M., FRATZSCHER, M., DE HAAN, J., JANSEN, D. (2008). “Central bank communication and monetary policy: a survey of theory and evidence.” Journal of Economic Literature, 46(4), 910-45. BLUNDELL, R., BOND, S. (1998). “Initial conditions and moments restrictions in dynamic 22

panel data models.” Journal of Econometrics, 86(1), 115-143. BOND, S., HOEFFLER, A., TEMPLE, J. (2001). “GMM Estimation of empirical growth models.” Economics Papers W21, Economics Group, Nuffield College, University of

SC RI PT

Oxford. BORN, B., EHRMANN M., FRATZSCHER, M. (2014). “Central Bank Communication on Financial Stability.” Economic Journal, 577, 701-734.

BORN, B., EHRMANN M., FRATZSCHER, M. (2012). “Communicating About Macroprudential Supervision – A New Challenge for Central Banks.” International Finance,

U

15(2), 179-203.

N

BRUNNERMEIER, M. (2009) “Deciphering the Liquidity and Credit Crunch 2007-2008."

A

Journal of Economic Perspectives, 23(1), 77-100.

M

BÜYÜKKARABACAK, B., VALEV, N. (2012). “Credit information sharing and banking crises: An empirical investigation.” Journal of Macroeconomics, 34(3), 788-800.

TE

18-41.

D

CBB – Central Bank of Brazil (2011). “Banking system.” Financial Stability Report, 10(2),

EP

CLAESSENS, S. (2015) “An Overview of Macroprudential Policy Tools”. Annual Review of Financial Economics, 7(1), 397-422.

CC

DAHL, D. (2012). “Coincident correlations of growth and cash flow in banking.” Journal of Banking & Finance, 36(4), 1139-1143.

A

DELL'ARICCIA, G., RABANAL, P., CROWE, C., IGAN, D. (2011). “Policies for Macrofinancial Stability; Options to Deal with Real Estate Booms.” IMF Staff Discussion Notes 11/02, International Monetary Fund. de MENDONÇA, H., BARCELOS, V. (2015). “Securitization and credit risk: Empirical evidence from an emerging economy.” North American Journal of Economics and 23

Finance, 32(C), 12-28. de MENDONÇA, H., FARIA, I. (2013). “Financial market reactions to announcements of monetary policy decisions: Evidence from the Brazilian case." Journal of Economic

SC RI PT

Studies, 40(1), 54-70. de MORAES, C.O., MONTES, G., ANTUNES, J. (2016). “How does capital regulation react to monetary policy? New evidence on the risk-taking channel.” Economic Modelling, 56(C), 177-186.

DINCER, N., EICHENGREEN, B. (2014). “Central Bank Transparency and Independence:

U

Updates and New Measures.” International Journal of Central Banking, 10(1), 189-259.

N

DOBLAS-MADRID, A., MINETTI, R. (2013), “Sharing information in the credit market:

A

Contract-level evidence from U.S. firms.” Journal of Financial Economics, 109(1), 198-

M

223.

DREHMANN, M., BORIO, C., TSATSARONIS, K. (2011). “Anchoring Countercyclical

TE

7(4), 189-240.

D

Capital Buffers: The role of Credit Aggregates.” International Journal of Central Banking,

EP

EHRMANN M., FRATZSCHER, M. (2007), “Communication by central bank committee members: Different strategies, same effectiveness.” Journal of Money, Credit, and

CC

Banking, 39 (2–3), 509–541.

A

EHRMANN, M., SONDERMANN, D. (2012). “The News Content of Macroeconomic Announcements: What if Central Bank Communication Becomes Stale?” International Journal of Central Banking, 8(3), 1-53.

FOOS, D., NORDEN, L., WEBER, M. (2010). “Loan growth and riskiness of banks.” Journal of Banking & Finance, 34(12), 2929-2940. FUJIWARA, I. (2005). “Is the central bank's publication of economic forecasts influential?” 24

Economics Letters, 89 (3), 255-261. GALATI, G., MOESSNER, R. (2013). “Macroprudential Policy – A Literature Review.” Journal of Economic Surveys, 27(5), 846-878.

Journal of Financial Intermediation, 13(4), 436-457.

SC RI PT

GAMBACORTA, L., MISTRULLI E. (2004). “Does bank capital affect lending behavior?"

GHENIMI, A., CHAIBI, H., OMRI, M.A.B. (2017). “The effects of liquidity risk and credit risk on bank stability: Evidence from the MENA region.” Borsa Istanbul Review, Research and Business Development Department, Borsa Istanbul, 17(4), 238-248.

U

GUIDARA, A., LAI, V.S., SOUMARÉ, I., TCHANA, F.T. (2013). “Banks’ capital buffer,

N

risk and performance in the Canadian banking system: Impact of business cycles and

A

regulatory changes.” Journal of Banking & Finance, 37(9), 3373-3387.

Economics, 46(2), 489-516.

M

GUTHRIE, G., WRIGHT, J. (2000). “Open Mouth Operations.” Journal of Monetary

D

GUTTENTAG J., HERRING R. (1984). “Commercial Bank Lending to Less Developed

TE

Countries: From Overlending to Underlending to Structural Reform.” Brookings

EP

Discussion Papers in International Economics 16, Washington D.C.: Brookings Institution. HANSON, S., KASHYAP, A., STEIN, J. (2011). “A macroprudential approach to financial

CC

regulation.” Journal of Economic Perspectives, 25(1), 3-28. HAYAKAWA, K. (2012). “The asymptotic properties of the system GMM estimator in

A

dynamic panel data models when both B and T are large.” Institute of Economic Research, Hitotsubashi University, Mimeo.

HORVÁTH, R., VAŠKO, D. (2016). “Central bank transparency and financial stability.” Journal of Financial Stability, 22(C), 45-56. HUME, M., SENTANCE, A. (2009). "The global credit boom: Challenges for 25

macroeconomics and policy," Journal of International Money and Finance, 28(8), 14261461. IMBIEROWICZ, B., RAUCH, C. (2014). “The relationship between liquidity risk and credit

SC RI PT

risk in banks.” Journal of Banking & Finance, 40(C), 242-256. IMF - International Monetary Fund - Financial Stability Board, Bank for International Settlements. (2011), “Macroprudential policy tools and frameworks: progress report to G20”, 27 October.

JIMÉNEZ, G., SAURINA J. (2006). “Credit Cycles, Credit Risk, and Prudential Regulation.”

U

International Journal of Central Banking, 2(2), 65-98.

N

JOKIPII, T., MILNE, A. (2008). “The cyclical behaviour of european bank capital buffers.”

A

Journal of Banking and Finance, 32(8), 1440–1451.

M

JORDÀ, Ò, SCHULARICK M., TAYLOR, A. (2013). “When Credit Bites Back.” Journal of Money, Credit and Banking, 45(2), 3-28.

D

JOKIVUOLLE, E., PESOLA, J., VIREN, M. (2015). “Why is credit-to-GDP a good measure

TE

for setting countercyclical capital buffer?” Journal of Financial Stability, 18(C), 117-126.

EP

KAMINSKY, G., REINHART, C. (1999).”The twin crises: the causes of banking and balance-of-payments problems.” American Economic Review, 89(3), 473–500.

CC

KING, M. (1997). “Changes in UK monetary policy: Rules and discretion in practice.” Journal of Monetary Economics, 39(1), 81-97.

A

PARK, J. (2012). “Brazil’s capital market: Current status and issues for further development.” International Monetary Fund, W.P. 12/224.

PESOLA, J. (2011). “Joint effect of financial fragility and macroeconomic shocks on bank loan losses: Evidence from Europe.” Journal of Banking & Finance, 35(11), 3134-3144. POOL, S., de HAAN, L., JACOBS, J.P.A.M., (2015). “Loan loss provisioning, bank credit 26

and the real economy.” Journal of Macroeconomics, 45(C), 124-136. REEVES, R., SAWICKI, M. (2007). “Do Financial Markets React to Bank of England Communication?” European Journal of Political Economy, 23(1), 207-27.

SC RI PT

REINHART, C., ROGOFF, K. (2013). “Banking crises: An equal opportunity menace.” Journal of Banking & Finance, 37(11), 4557-4573.

ROODMAN, D. (2009). “How to do xtabond2: An introduction to difference and system GMM in Stata.” Stata Journal, 9(1), 86-136.

ROSA, C., VERGA, G., (2007). “On the consistency and effectiveness of central bank

U

communication: Evidence from the ECB.” European Journal of Political Economy, 23(1),

N

146-175.

A

ROZKRUT, M., RYBINSKI, K., SZTABA, L., SZWAJA, R. (2007). “Quest for central bank

M

communication: Does it pay to be ‘talkative?’” European Journal of Political Economy, 23 (1), 176-206.

D

STOLZ, S., WEDOW, M. (2011). “Banks' regulatory capital buffer and the business cycle:

TE

Evidence for Germany.” Journal of Financial Stability, 7(2), 98-110.

EP

SCHMIDT, S., NAUTZ, D. (2012). “Central Bank Communication and the Perception of Monetary Policy by Financial Market Experts.” Journal of Money, Credit and Banking,

CC

44(2-3), 323-340.

A

TABAK, B.M., LAIZ, M.T., CAJUEIRO, D.O., (2013). “Financial stability and monetary policy – the case of Brazil.” Revista Brasileira de Economia, 67(4), 431–441.

ULLRICH, K. (2008). “Inflation Expectations of Experts and ECB Communication.” North American Journal of Economics and Finance, 19(1), 93-108.

27

Figure 1 Central bank communication on credit development – COM1 1.50 1.25 1.00 0.75

SC RI PT

0.50 0.25 0.00 -0.25 -0.50 -0.75

-1.25 -1.50 2000

2001

2002

2003

2004

2005

COM1

2006

2007

2008

N

1999

U

-1.00

2009

2010

2011

2012

2013

2014

Trend - Hodrick-Prescott filter

D

M

A

Notes: COM1 – CBB’s communication – expectation in relation to an increase (positive value “+1”), decrease (negative value “-1”), or no fluctuation (value “0”) in the credit market.

A

CC

EP

TE

Figure 2 Central bank communication on credit development – COM2

28

3

2

1

SC RI PT

0

-1

-2

-3 2000

2001

2002

2003

2004

2005

COM2

2006

2007

2008

2009

2010

2011

2012

2013

2014

U

1999

Trend - Hodrick-Prescott filter

M

A

N

Notes: COM2 – CBB’s communication – accumulated index – expectation in relation to an increase (positive value “+1”), continued increase (positive value “+2”), decrease (negative value “-1”), continued decrease (negative value “-2”), or no fluctuation (values “0”) in the credit market.

D

Figure 3 Credit risk (CR1 and CR2) and capital buffer (BUF)

TE

9

8

6

CC

5

EP

7

4

A

3

2

2000 2001

2002

2003

2004

2005

2006

2007

CR1

2008 CR2

2009

2010

2011

2012

2013

2014

BUF

Notes: CR1 – expected loss of banks with respect to loans (loan loss provisions/gross loans ratio); CR2 – expected loss of loans that were delayed (nonperforming loan/total loans ratio); BUF – difference between the capital of financial institutions and the Basel index.

29

Table 1 CBB communication to the credit market and its classification (glossary) Table 1 CBB communication to the credit market and its classification (glossary) Index

+1

0

M

A

N

U

SC RI PT

Main statements of BCB: most important keywords Driven by the increase in credit volume - The economy should benefit from the increase in the volume of credit - The impulse given by credit and its multiplier effects - With respect to credit growth this is an important structural advance – Favorable outlook for the retail trade in the coming months is based on the performance of the labor market, wages, the expansion of credit, and the recovery in consumer confidence - Looking ahead, the expansion of employment, income, and credit growth are factors that boost the activity over the next six months – Over the next quarters, the expansion of employment and income and credit growth will continue to boost aggregate demand - Even if the difficulties from the international financial turbulence add uncertainty to the scenario these are still no observable or significant changes in the context of strong credit expansion over the next quarters - the main scenario assumes continued recovery of the domestic credit market - in the credit market, volumes rebounded intensely Consumer credit should be relatively constant in the next months due to the growth of default A stronger recovery in the credit supply, however, will only be possible with lower interest rates and a decrease in the unemployment rate - One moderation in the rhythm of credit expansion - Moderation in the credit growth - Moderation of the credit market growth, particularly of directed operations - the central scenario assumes moderate credit expansion - the central scenario assumes moderate credit expansion, particularly in the case of individuals segment Slow credit growth - Low credit growth - The example of the credit market, the effects of the deteriorating economic situation are already being felt in the labor market - The conditions of access to bank credit remain restrictive - The growth credit, which was an important element to support aggregate demand, now has played a central role in deceleration.

-1

A

CC

EP

TE

D

Note: “+1” corresponds to the CBB’s expectation of credit growth. “-1” corresponds to the CBB’s expectation of credit decrease. “0” is the CBB’s expectation of credit stability. Statements are gathered from CBB’s Inflation Report (“inflation outlook” section).

30

Table 2 Central bank communication on credit development effect on credit risk

-0.0685*** (0.0102) 0.0864*** (0.0198) 0.4827*** (0.0172) -1.4862*** (0.0837) 0.0443** (0.0210) 0.6258*** (0.0072)

IR LIQ CRED ROA CR1(-1)

4830

0.5197*** (0.0154) 4799

0.5154*** (0.0143) 4775

N.Instr/N.Cross Sec. 0.557

0.530

0.539

0.530

J-statistic P-valor AR(1) P-valor AR(2) P-valor

66.359 0.140 -0.445 0.000 -0.001 0.920

61.955 0.242 -0.486 0.000 0.023 0.131

61.062 0.237 -0.487 0.000 0.019 0.213

4769

D

Obs.

M

CR2(-1)

SC RI PT

GAP

-0.0823* (0.0435) -0.0832*** (0.0122) 0.0886*** (0.0210) 0.5209*** (0.0218) -1.8113*** (0.0862) 0.0598* (0.0296) 0.6283*** (0.0062)

CR2 Model 1 Model 2 -0.6261*** (0.1353) -0.3903*** (0.0764) -0.0464*** -0.0682*** (0.0160) (0.0161) 0.0883** 0.0607* (0.0346) (0.0325) 0.6712*** 0.6936*** (0.0413) (0.0425) -1.9197*** -2.1163*** (0.1748) (0.2507) 0.0629* 0.0607** (0.0357) (0.0325)

U

COM2

Model 2

N

CR1 Model 1 -0.3193*** (0.0883)

A

Regressors COM1

EP

TE

69.060 0.131 -0.457 0.000 -0.003 0.852

A

CC

Notes: Marginal significance levels: (***) denotes 0.01, (**) denotes 0.05, and (*) denotes 0.1. Total sample. Dependent variables: CR1 and CR2; main independent variables of interest: COM1 and COM2. White’s heteroskedasticity consistent covariance matrix was applied in the regressions. Robust standard errors are in parentheses. S-GMM – uses two-step of Arellano and Bover (1995) without time period effects. Tests for AR(1) and AR(2) check for the presence of first-order and secondorder serial correlation in the first-difference residuals.

31

Table 3 Central bank communication on credit development effect on credit risk

IR LIQ CRED ROA CR1(-1) CR2(-1)

810 0.895 14.708 0.143 -0.450 0.000 -0.035 0.275

EP

TE

D

Obs. N.Instr/N.Cross Sections J-statistic P-valor AR(1) P-valor AR(2) P-valor

813 0.947 15.851 0.147 -0.459 0.000 -0.027 0.417

Model 2

SC RI PT

-0.0274*** (0.0103) 0.0084 (0.0131) -0.2024*** (0.0073) -0.8164*** (0.1750) 0.0651* (0.0388) 0.8379*** (0.0156)

-0.1556*** (0.0210) -0.0642*** (0.0185) 0.0059 (0.0078) -0.0816*** (0.0051) -0.9158*** (0.1032) 0.0987*** (0.0132) 0.7767*** (0.0039)

M

GAP

Model 2

-0.0501*** (0.0166) 0.0149* (0.0090) -0.0738*** (0.0058) -0.8176*** (0.1391) 0.1465* (0.0844)

-0.0809** (0.0318) -0.0180 (0.0250) 0.0176* (0.0095) -0.0794*** (0.0060) -0.8504*** (0.1425) 0.1545* (0.0835)

0.7259*** (0.0188) 829 0.947 14.384 0.212 -0.497 0.000 -0.009 0.812

0.7207*** (0.0184) 829 0.947 14.918 0.186 -0.497 0.000 -0.008 0.838

U

COM2

CR2 Model 1 -0.1574*** (0.0572)

N

Regressors COM1

A

CR1 Model 1 -0.1187*** (0.0495)

A

CC

Notes: Marginal significance levels: (***) denotes 0.01, (**) denotes 0.05, and (*) denotes 0.1. Largest banks sample. Dependent variables: CR1 and CR2; main independent variables of interest: COM1 and COM2. White’s heteroskedasticity consistent covariance matrix was applied in the regressions. Robust standard errors are in parentheses. S-GMM – uses two-step of Arellano and Bover (1995) without time period effects. Tests for AR(1) and AR(2) check for the presence of first-order and second-order serial correlation in the first-difference residuals.

32

Table 4 Central bank communication on credit development effect on capital buffer

IR LIQ CRED ROE

-0.0109*** (0.0036) 0.0446*** (0.0059) 0.0206*** (0.0021) -1.5385*** (0.0792) -0.0600***

-0.0633*** (0.0104) 0.6706*** (0.0021) 4844 0.560 60.098 0.400 -0.528 0.000 -0.021 0.168

EP

TE

D

M

(0.0112) BUF (-1) 0.6226*** (0.0027) Obs. 4845 N.Instr/N.Cross Sections 0.526 J-statistic 55.474 P-valor 0.419 AR(1) -0.523 P-valor 0.000 AR(2) -0.020 P-valor 0.192

U

GAP

0.1794*** (0.0243) -0.0171*** (0.0033) 0.0351*** (0.0045) 0.0176*** (0.0018) -1.4670*** (0.0483)

N

COM2

Model 2

A

Regressors COM1

largest banks sample Model 1 Model 2 0.1120*** (0.0095) 0.0332*** (0.0041) -0.0373*** -0.0330*** (0.0064) (0.0051) 0.0012 0.0014 (0.0027) (0.0013) 0.0077*** 0.0067*** (0.0022) (0.0017) -0.1080*** -0.1177*** (0.0165) (0.0202)

SC RI PT

total sample Model 1 0.5021*** (0.0596)

-0.0030***

-0.0034***

(0.0009) 0.7832*** (0.0166) 700 0.944 12.082 0.280 -0.490 0.000 -0.020 0.618

(0.0006) 0.8089*** (0.0250) 700 0.944 12.113 0.278 -0.482 0.000 -0.023 0.569

A

CC

Notes: Marginal significance levels: (***) denotes 0.01, (**) denotes 0.05, and (*) denotes 0.1. Dependent variable is BUF; main independent variables of interest: COM1 and COM2. White’s heteroskedasticity consistent covariance matrix was applied in the regressions. Robust standard errors are in parentheses. S-GMM – uses two-step of Arellano and Bover (1995) without time period effects. Tests for AR(1) and AR(2) check for the presence of first-order and second-order serial correlation in the first-difference residuals.

33

Table 5 Central bank communication on credit development effect

-0.0503*** (0.0048) 0.0383*** (0.0079) 0.2064*** (0.0048) -0.6354*** (0.0285) 0.0176* (0.0105) 0.7232*** (0.0027)

IR LIQ CRED ROA CR1(-1)

-0.0787*** (0.0192) -0.0445*** (0.0050) 0.0415*** (0.0084) 0.2229*** (0.0058) -0.5036*** (0.0214) 0.0160* (0.0085) 0.7210*** (0.0033)

CR2(-1)

TE

BUF(-1)

CC

Obs. N.Instr/N.Cross Sections J-statistic P-valor AR(1) P-valor AR(2) P-valor

A

-0.0949** (0.0458)

EP

COM1² COM2²

0.4327*** (0.0460)

-0.3609*** (0.0833) -0.1047*** -0.0663*** (0.0166) (0.0172) 0.0660* 0.0685* (0.0394) (0.0373) 0.7144*** 0.6738*** (0.0405) (0.0417) -2.0381*** -1.9641*** (0.1933) (0.2409) 0.1148*** 0.0865** (0.0383) (0.0391)

0.5213*** (0.0106)

4502 0.661 78.792 0.174 -0.437 0.000 -0.019 0.213

0.0234* (0.0135) -0.0192*** (0.0035) 0.0277*** (0.0049) 0.0153*** (0.0017) -1.6994*** (0.0509)

-0.0071* (0.0038) 0.0392*** (0.0060) 0.0157*** (0.0020) -0.9220*** (0.0417)

0.5227*** (0.0150) -0.0767*** (0.0084) 0.6600*** (0.0022) 0.1005*** (0.0238)

D

ROE

Model 2

SC RI PT

GAP

-0.5662*** (0.1287)

BUF Model 1

A

COM2

Model 2

U

-0.2152*** (0.0472)

M

COM1

CR2 Model 1

Model 2

N

CR1 Model 1

Regressors

-0.0315** (0.0125) 4502 0.675 80.225 0.167 -0.426 0.000 -0.023 0.126

-0.2949* (0.1587)

4734 0.574 61.831 0.341 -0.485 0.000 0.019 0.203

-0.1176*** (0.0450) 4775 0.539 61.034 0.238 -0.487 0.000 0.021 0.160

4736 0.557 66.705 0.155 -0.571 0.000 -0.023 0.173

-0.0597*** (0.0116) 0.6726*** (0.0021)

0.1890*** (0.0237) 4844 0.578 69.427 0.166 -0.520 0.000 -0.019 0.213

Notes: Marginal significance levels: (***) denotes 0.01, (**) denotes 0.05, and (*) denotes 0.1. Total sample. Dependent variables: CR1, CR2 and BUF; main independent variables of interest: COM1 and COM2. White’s heteroskedasticity consistent covariance matrix was applied in the regressions. Robust standard errors are in parentheses. S-GMM – uses two-

34

A

CC

EP

TE

D

M

A

N

U

SC RI PT

step of Arellano and Bover (1995) without time period effects. Tests for AR(1) and AR(2) check for the presence of firstorder and second-order serial correlation in the first-difference residuals.

35

Table 6 Central bank communication on credit development effect on default coverage rate Regressors COM1

Model 1 -0.2845** (0.1109)

Model 2

0.4615*** (0.0499) 0.4411*** (0.0252) -0.3568*** (0.0189) -2.3621*** (0.1974) 0.2079*** (0.0108) 0.2627*** (0.0054) 4017

IR LIQ

N

CRED

A

ROE

M

COV(-1) Obs.

0.602

J-statistic P-valor AR(1) P-valor AR(2) P-valor

66.062 0.218 -0.439 0.000 0.017 0.281

TE

D

N.Instr/N.Cross Sections 0.602

EP CC A

U

GAP

-0.1826** (0.0803) 0.4922*** (0.0547) 0.4510*** (0.0243) -0.3485*** (0.0204) -2.3472*** (0.2139) 0.2585*** (0.011) 0.2610*** (0.0054) 4017

SC RI PT

COM2

65.581 0.231 -0.447 0.000 0.023 0.127

Notes: Marginal significance levels: (***) denotes 0.01, (**) denotes 0.05, and (*) denotes 0.1. The dependent variable is COV; main independent variables of interest: COM1 and COM2. White’s heteroskedasticity consistent covariance matrix was applied in the regressions. Robust standard errors are in parentheses. S-GMM – uses two-step of Arellano and Bover (1995) without time period effects. Tests for AR(1) and AR(2) check for the presence of first-order and second-order serial correlation in the first-difference residuals.

36

Appendix Table A.1 Inflation Reports

A

CC

EP

TE

D

M

Note: Dates are extracted from the CBB’s Inflation Reports (full publication).

37

Dec 24 29 28 30 30 27 28 20 27 22 22 22 22 20 19

SC RI PT

31 30 28 31 31 29 30 28 27 30 31 30 28 28 27

Sep 30 29 28 30 30 30 29 28 27 29 25 30 29 27 30 29

U

Jun 30 30 29 28 30 30 29 28 28 25 26 30 29 28 27 26

N

Mar

A

Release dates Year 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

N U SC RI PT

Table A.2 Description of the variables, sources of data and descriptive statistics

IR

COV LIQ CRED ROE ROA

COM1

COM2

Mean

Standard deviation

Minimum

Maximum

Observations

T-sample L-sample

T-sample L-sample

T-sample L-sample

T-sample L-sample

T-sample L-sample

5.83

4.72

8.93

5.09

0.00

0.00

100.00

75.10

5612

1006

4.92

8.74

4.68

0.00

0.007

100.00

38.29

5625

1006

1.55

5.73

0.54

-2.38

0.30

90.82

5.34

5814

1006

0.08

0.08

3.05

3.05

-9.88

-9.88

7.76

7.76

7000

1064

13.89

13.89

4.64

4.64

7.15

7.15

26.32

26.32

7000

1064

2.80

3.00

12.57

17.84

0.00

0.01

369.55

369.55

4877

860

29.92

27.09

21.32

14.52

0.01

0.01

100.00

75.92

5841

1006

0.17

0.24

4.15

4.33

-1.00

-1.00

255.91

131.54

5437

947

4.95

7.24

15.98

13.25

-483.43

-184.38a

138.43

105.79a

5723

1002

0.94

0.85

2.94

1.54

-49.66

-15.06

30.16

12.28

5737

1002

0.34

0.34

0.66

0.66

-1.00

-1.00

1.00

1.00

7000

1064

0.68

0.68

1.15

1.15

-2.00

-2.00

2.00

2.00

7000

1064

6.20

3.32

M

ED

GAP

PT

BUF

CC E

CR2

provisions/gross IEFA/CBB – devised by authors IEFA/CBB and document nonperforming loan /total “7019 Balance Sheet” – loans ratio devised by authors difference between the capital of financial institutions and CBB – 50 largest banks – the minimum capital required devised by authors by regulators (Basel index) TSMS/CBB – “code 1252 GAP – difference between Quarterly GDP-1990=100 GDP and potential output – seasonally adjusted (Hodrick-Prescott filter) data” devised by authors. TSMS/CBB – “code 4189 Monetary policy interest rate – accumulated in the month in annual terms. loan loss IEFA/CBB and document provisions/nonperforming “7019 Balance Sheet” – loan ratio devised by authors IEFA/CBB – devised by liquid assets/total assets ratio authors IEFA/CBB – devised by credit growth rate authors Net income/shareholder’s IEFA/CBB – devised by equity ratio authors IEFA/CBB – devised by Net income/total assets ratio authors Macroprudential communication time variant – Inflation Report/CBB – result of the classification of devised by authors the CBB’s expectation in relation to the credit market Accumulated macroprudential communication index – Inflation Report/CBB – t devised by authors loan loss loans ratio

A

CR1

Data source

A

Variable Variable description name

COM 2t  t 1 COM 1

Notes: IEFA/CBB – Information for Economic-Financial Analysis/Central Bank of Brazil. TSMS – Time Series Management System. T-sample is the total sample. L-sample comprises the largest Brazilian private banks. a exclude outliers (banks) BRJ (2001 – quarter 1), KDB (2013 – quarter 3), and KDB (2014 – quarters 1 and 2).

38

Table A.3 List of instruments (S-GMM) Table 2. Central bank communication on credit development effect on credit risk CR1(-2), CRED(-1 to -2), COM1(-1 to -2), ROA(-1 to -3), GAP(-1 to -2), IR(-1 to -3).

Model 2

CR1(-2), ROA(-1), CRED(-1 to -2), COM2(-1 to -2), GAP(-1 to -2), IR(-1 to -3).

CR2 Model 1

CR2(-2), CR1, CRED(-1 to -2), ROA(-1 to -2), GAP(-1 to -3), IR(-1 to -2).

Model 2

CR2(-2), CRED(-2 to -3), ROA(-1 to -2), CR1, IR(-1 to -2), GAP(-1 to -2).

Table 3. Central bank communication on credit development effect on credit risk

SC RI PT

CR1 Model 1

CR1 Model 1

CR1(-2 to -6), COM1(-1), CRED(-1 to -2), ROA(-1 to -2), GAP(-1 to -3), COM2, BUF, IR(-1), CR2.

Model 2

CR1(-2 to -6) COM1(-1 to -3) CRED(-1) ROA(-1 to -2) CR2(-1 to -2) CR2 IR(-3 to -4) BUF(-2) CAR.

CR2

CR2(-2 to -5), LIQ(-3), IR(-1 to -4), ROA(-2), GAP(-3), COM2(-1), CRED(-2), logTCRED(-3 to -4), ROE(-3), COM1(-2), BUF(-2). CR2(-2 to -5), LIQ(-3), IR(-1 to -4), ROA(-2), GAP(-3), COM2(-1), CRED(-2), logTCRED(-3 to -4), ROE(-3), Model 2 COM1(-2), BUF(-2). Table 4. Central bank communication on credit development effect on capital buffer

N

U

Model 1

Total sample

BUF(-2), CR2(0, -2 to -3), LIQ(-1), CR1, COM1(-1), GAP(-1 to -2), IR(-2).

Model 2

BUF(-2), CR2(0, -1 to -3), LIQ(-1), COM1(0 to -1,) COM1, GAP(-1 to -3), IR(-2), LLP, CAR(-1).

A

Model 1

M

Largest Banks

CRED(-1 to -3), ROE(-1 to -2), COM1(-1 to -2), BUF(-4 to -5), CAR(-1 to -2), CR1(-1), IR(-2), ROA(-2), COV(-3) logTCRED(0,-3).

Model 2

CRED(-1 to -3), ROE(-1 to -2), COM1(-1 to -2), logTCRED(0, -2 to -3).

BUF(-4 to -5), CAR(-1 to -2), CR1, IR(-1), COV(-3),

TE

D

Model 1

Table 5. Central bank communication on credit development effect CR1

CR1(-2), CRED(-1 to -5), COM1(-1 to -5), ROA(-1 to -3), GAP(-1 to -2), IR(-1 to -3), TCRED(-1 to -2) BUF(-1 to -3), BUF, COM2(-1 to -3), ROE(-1 to -2).

EP

Model 1

CR2 Model 1 Model 2

CR2(-2), CR1, CRED(-1 to -3), ROA(-1 to -3), GAP(-1 to -3), IR(-1 to -3), ROE(-1). CR2(-2), CRED(-2 to -3), ROA(-1 to -2), CR1, IR(-1 to -3), GAP(-1 to -2).

A

BUF Model 1

CR1(-2), CRED(-1 to -5), COM1(-1 to -5), COM2(-3), ROA(-1 to -3), GAP(-2 to -3), IR(-1 to -3), TCRED(-1 to -3), BUF(-1 to -2), BUF, ROE(-1 to -3).

CC

Model 2

Model 2

BUF(-2), CR2(0, -4), CR1, LIQ(-2), COM1(-1 to -3), GAP(-1 to -2), IR(-1 to -3), CAR(-1), COM2(-1). BUF(-2), CR2(0, -1 to -3), LIQ(-1 to -2), GAP(-1 to -3), IR(-2 to -3), CR1, CAR(-1), COM1(-1), COM2.

Table 6. Central bank communication on credit development effect on default coverage rate COV Model 1 Model 2

COV(-2), CAR(0, -1 to -4), CRED(-1 to -4), COM1(-1 to -4), logTCRED(-2). COV(-2), CAR(0, -1 to -4), CRED(-1 to -4), COM1(-1 to -4), logTCRED(-1). Note: Numbers within parentheses denote the lags associated with each variable.

39

Table A.4 Granger Causality Tests Obs

lags

F-statistic

Prob.

COM1 does not Granger cause CR1

5200

3

5.289

0.001

0.563

0.639

5200

3

4.657

0.003

1.748

0.155

6.25333

0.000

1.71818

0.143

5.03886

0.001

1.12636

0.342

2.162

0.071

0.597

0.665

3.003

0.017

CR1 does not Granger cause COM1 COM2 does not Granger cause CR1 CR1 does not Granger cause COM2 COM1 does not Granger cause CR2

5101

4

CR2 does not Granger cause COM1 COM2 does not Granger cause CR2

5101

4

CR2 does not Granger cause COM2 COM1 does not Granger cause BUF

5298

4

BUF does not Granger cause COM1 5298

4

U

COM2 does not Granger cause BUF

SC RI PT

Null hypothesis:

Table A.5 Correlation matrix

1.000 -0.349 -0.138 0.474 -0.286 -0.380 -0.772 0.287 -0.216 -0.217 -0.152

1.000 -0.148 -0.063 0.242 0.233 0.613 0.329 0.141 0.420 0.367

GAP

EP

IR

COM1

COM2

COV

LIQ

CRED

ROA

ROE

1.000 0.869 0.248 -0.076 0.178 0.375 0.327

1.000 0.288 -0.140 0.231 0.401 0.342

1.000 0.286 0.096 0.341 0.291

1.000 0.058 0.277 0.275

1.000 0.570 0.528

1.000 0.989

1.000

D

BUF

1.000 -0.159 0.168 0.235 0.080 -0.119 -0.093 0.049 0.034

TE

CR2

1.000 -0.346 -0.402 0.001 0.697 -0.169 0.024 0.080

A

CC

CR1 CR2 BUF GAP IR COM1 COM2 COV LIQ CRED ROA ROE

CR1 1.000 0.690 0.162 -0.145 0.703 -0.180 -0.297 -0.119 0.745 -0.336 -0.042 -0.005

M

A

N

0.574 0.681 BUF does not Granger cause COM2 Note: The number of lags is based on the Akaike, Schwartz and Hannan-Quinn criteria.

40

Table A.6 List of banks BCO FORD S.A.

CAIXA GERAL

ALFA

BCO GERADOR S.A.

CITIBANK*

BANCO AZTECA DO BRASIL S.A.

BCO GMAC S.A.

CREDIT AGRICOLE

BANCO BRACCE S.A.

BCO GUANABARA S.A.

CREDIT SUISSE*

BANCO CNH INDUSTRIAL CAPITAL S.A

BCO INDUSCRED DE INVESTIM. S/A

DEUTSCHE*

BANCO FIDIS

BCO KDB BRASIL S.A.

FATOR

BANCO IBM S.A.

BCO KEB DO BRASIL SA

BANCO MONEO S.A.

BCO LA NACION ARGENTINA

BANCO PORTO REAL DE INVEST.S.A

BCO LA PROVINCIA B AIRES BCE

BANCO RANDON S.A.

BCO LUSO BRASILEIRO S.A.

BANCO SEMEAR

BCO MAXINVEST S.A.

BANCO TOPÁZIO S.A.

BCO MODAL S.A.

BANCO VIPAL

BCO POTTENCIAL S.A.

BANCOOB*

BCO RABOBANK INTL BRASIL S.A.

BANESTES

BCO REP ORIENTAL URUGUAY BCE

BANIF

BCO RIBEIRAO PRETO S.A.

BANRISUL

BCO STANDARD INV S.A.

BARCLAYS

BCO SUMITOMO MITSUI BRASIL S.A.

JP MORGAN CHASE*

BANCO DO BRASIL*

BCO TOKYO-MITSUBISHI BM S.A.

MÁXIMA

BBM

BCO TOYOTA DO BRASIL S.A.

MERCANTIL DO BRASIL

BCO A.J. RENNER S.A.

BCO TRIANGULO S.A.

MERCEDES-BENZ

BCO ABN AMRO S.A.

BCO TRICURY S.A.

MIZUHO

BCO ARBI S.A.

BCO VOLKSWAGEN S.A*

MORGAN STANLEY

BCO BRJ S.A.

BCO VOLVO BRASIL S.A.

NATIXIS BRASIL S.A. BM

BCO CAPITAL S.A.

BCO WOORI BANK DO BRASIL S.A.

NOVO BCO CONTINENTAL S.A. - BM

BCO CARGILL S.A.

BCO YAMAHA MOTOR S.A.

ORIGINAL

GOLDMAN SACHS HONDA HSBC*

ICBC DO BRASIL BM S.A. INDUSTRIAL DO BRASIL INDUSVAL ING

INTERMEDIUM ITAU*

A

N

U

J.MALUCELLI

M

D

BCO CATERPILLAR S.A.

SC RI PT

ABC-BRASIL*

JOHN DEERE

BD REGIONAL DO EXTREMO SUL

OURINVEST

BES

PANAMERICANO*

BIC

PINE

BMG

PSA FINANCE

BNDES

RENDIMENTO

BNP PARIBAS*

RODOBENS

BCO DA CHINA BRASIL S.A.

BOFA MERRILL LYNCH

SAFRA*

BCO DAYCOVAL S.A

BONSUCESSO

SANTANDER*

BCO DE LAGE LANDEN BRASIL S.A.

BPN BRASIL BM S.A.

SCANIA BCO S.A.

BCO DES. DE MG S.A.

BR PARTNERS

SCOTIABANK BRASIL

BCO DES. DO ES S.A.

BRADESCO*

SOCIETE GENERALE*

BCO DO EST. DE SE S.A.

BRASIL PLURAL

SOCOPA

BCO DO EST. DO PA S.A.

BRB

SOFISA

BCO DO NORDESTE DO BRASIL S.A.

BROOKFIELD

VOTORANTIM*

BCO FIBRA S.A.

BTG PACTUAL*

VR

BCO FICSA S.A.

CAIXA ECONOMICA FEDERAL

BCO CEDULA S.A. BCO COOPERATIVO SICREDI S.A.* BCO CSF S.A.

A

CC

EP

BCO DA AMAZONIA S.A.

TE

BCO COMMERCIAL INV. TRUST S.A.

(*) Denote Large Sample (L-Sample)

41