Inflation Targeting: Is IT to blame for Banking System Instability?

Transcription

1 Inflation Targeting: Is IT to blame for Banking System Instability? Dimas M. Fazio a,, Benjamin M. Tabak b,c, Daniel O. Cajueiro d,e a London Business School, University of London, London, United Kingdom b Federal Senate of Brazil, CEP , Brasília, DF, Brazil c Department of Economics at Universidade Católica de Brasília, CEP , Brasília-DF, Brazil d Department of Economics at Universidade de Brasília, FACE, CEP , Brasília-DF, Brazil e National Institute of Science and Technology for Complex Systems at Universidade de Brasília, FACE, CEP , Brasília-DF, Brazil This version: January 13, 2015 Abstract In light of the financial crisis, the practice of inflation targeting (IT) has been blamed for authorities failure to respond to increasing systemic risk and developing asset bubbles. However, utilizing a rich database containing nearly 5500 commercial banks from 70 countries (among which, 22 are IT) for the period , this paper demonstrates that on average, inflation targeting national banking systems (i) are more stable; (ii) possess sounder systemically important banks; and (iii) are as affected as (or even less affected than) other banks during periods of global liquidity shortages. Our results do not change significantly when we compare countries with the same legal origins or control for the delegation of bank supervision responsibility to bodies other than the central bank. These conclusions show that IT cannot be blamed for contributing to financial fragility. Keywords: inflation targeting, financial stability, monetary policy, financial crisis, bank regulation JEL classification: D40, E52, E58, G21, G28 The authors would like to thank, without implicating: Iftekhar Hasan, Bill Francis, Tuomas Peltonen, Eduardo Lima, Aquiles Faria, Marcio Garcia, Katherine Hennings, Marta Baltar, Solange Guerra, Naércio Menezes, Luiz Alberto Esteves, Marcelo Zeuli, Carlos Hamilton Araújo, Francisco Figueiredo, Rafael Terra, and the participants at the XVI Annual Inflation Targeting Seminar of the Banco Central do Brasil. The authors are grateful to financial support from CNPQ foundation. The opinions expressed in this paper are those of the authors and do not necessarily reflect those of the Federal Senate or its members. All remaining errors are our own. Corresponding author. Telephone number: addresses: dimasfazio@gmail.com (Dimas M. Fazio), benjamin.mtabak@gmail.com (Benjamin M. Tabak), danielcajueiro@gmail.com (Daniel O. Cajueiro)

2 1. Introduction There is strong evidence that inflation targeting (IT) has helped countries reduce inflation during the first years of implementation and anchor long-run expectations (Mishkin, 1999; Johnson, 2002; Bean, 2009) A traditional view of inflation targeting holds that price stability leads to more predictable returns and, therefore, to improved financial performance. In fact, periods of high inflation may lead to excessive borrowing and large loan defaults when prices start to decline. Nevertheless, some specialists have recently argued that central bankers, by focusing on reducing inflation, overlooked the banking system and the development of asset bubbles (Blanchard et al., 2010). According to them, this behaviour might have increased financial fragility and it might have even created the necessary conditions for the financial crisis of 2007 and 2008 to occur. Despite this ongoing discussion on whether the inflation targeting is to blame for the recent financial turmoil, empirical studies that test these viewpoints are rare. For instance, Frappa and Mésonnier (2010) examine 17 developed economies (9 of which are IT) and demonstrate that IT countries are more likely to experience housing bubbles and, therefore, greater financial instability. Fouejieu (2013), however, focus on 13 emerging economies (7 of which are IT) and find that although IT reduces financial stability, central banks (CBs) react more rapidly to financial imbalances, which undermines the argument of IT critics. Nevertheless, we believe that these papers suffer from the following weaknesses: (i) they examine only a few specific countries and (ii) consider only aggregate measures of financial fragility. Our study fulfills this literature gap by providing empirical evidence on whether the risk-taking level of banks from IT countries are higher than banks from non-it countries. In other words, we test whether the combination of price stability, and enhanced CB s communication and accountability can reduce the likelihood of a crisis in the financial sector. We employ a database containing 5468 commercial banks from 70 countries (22 of which are IT, according to the International Monetary Fund (IMF)) during the period The richness of these data allows us to produce meaningful interpretations of results whose validity is not limited to a specific case or country. In addition, to our knowledge, this is the first study to analyze the effects of IT on banking stability utilizing disaggregated data. Our empirical approach answers the following questions. First (Q1), we ask whether banks from IT countries are, on average, more or less fragile. We answer this question by regressing 2

3 a measure of financial stability on a dummy variable that equals one if the bank operates in an IT country according to the IMF. We observe a large, positive and significant coefficient for this variable, indicating that banks are more stable under IT. Second (Q2), we test how this advantage varies for different levels of inflation by interacting the IT dummy with a price index variable. We observe a negative coefficient for this interaction, meaning that banks are more sensitive to higher inflation, and that the IT advantages are no longer statistically significant for high levels of inflation (index around 10%). Our third question (Q3) is whether banking systems from IT countries are more vulnerable during periods of global uncertainty (i.e., periods with a higher T- bill and Eurodollar (TED) spread). Our results dismiss this idea, since there is some evidence that banks from IT countries are more resilient to global liquidity shocks, or as vulnerable as others at the very least. Finally, our last question (Q4) compares the risk-taking levels of systemically important banks from IT and non-it countries. In other words, are regulators from IT countries less preoccupied with systemically important financial institutions? We do not find evidence to support this claim; these large and interconnected banks are more stable in countries that adopt IT. These results do not necessarily say that the traditional channel from price stability to financial stability is in place. We only conclude with these results that the criticism of IT may not be entirely justified. Other factors may explain why some banking systems suffered more than others did; inflation targeting, however, does not seem to be one of them. In fact, low and predictable inflation together with a more transparent monetary policy by the CBs seems to affect positively financial stability. 1 In addition, one must bear in mind that the both the U.S. and countries from the Eurozone were the most affected by the recent financial crisis, even though they are not explicit inflation targeters according to the IMF. We also address others potential alternative explanations for our results. One potential criticism regards the presence of a potential identification problem: countries may have thought necessary to stabilise the financial system as a pre-requisite for the IT policy. According to this view, the IT policy is not the main responsible for the increased banking soundness in our results, but hypothetical reforms in the financial sector prior to the IT implementation. We do not believe that this criticism holds for two reasons. First, the pre-crisis dichotomy between monetary policy and 1 See Geraats (2002) for a survey regarding the impact of CB transparency and accountability on the economy as a whole. The empirical evidence is favorable to our findings. 3

4 financial stability made authorities conduct these two policies separately. CBs were not used to including financial frictions/fragility in their general equilibrium models (Mishkin, 2011), which mean that they did not believe that financial stability was a condition to promote price stability. Only recently has financial stability become as important as price stability for CBs around the world (Cukierman, 2013). The second reason is that our estimations control for both economic and financial development country-specific factors. When studying countries throughout the world, there is always the possibility of omitted variable bias. There are factors, such as economic and legal environments, that may generate differences in the relationships we uncover. Therefore, we also present robustness results disaggregated by legal origin groups, as defined by (La-Porta et al., 1997, 1998). These authors divide countries into groups depending on the origins of their legal systems, i.e., English (Common Law) or French, German, and Nordic (these last three are considered Civil Law). The results are consistent overall. A few coefficients are no longer significant, but they maintain the same signs from the main model. In addition, we are able to observe some heterogeneity of our conclusions depending on the legal system. Therefore, banking systems that practice IT outperform - or perform similarly to - other banking systems with similar legal and economic environment. Finally, some countries possess supervisory bodies that are separate from the CB. Criticism of inflation targeting may be valid only in countries where the CB is also responsible for bank supervision. Otherwise, the argument that IT CBs are not concerned with financial stability might not apply. Thus, we utilise information about whether the supervisory body is outside the CB from the well-known (Barth et al., 2001, 2004, 2008, 2013) database. However, this database possesses some shortcomings: (a) it does not include all countries in our sample; and (b) the survey question asking whether the supervisory body is separate from the CB changed from the 2000 and 2003 to the 2008 and 2011 surveys. The majority of our results hold when we re-estimate our models after introducing this new dimension. The remainder of this paper is organised as follows. Section 2 presents the opposing views in the literature regarding the relationship between inflation targeting and financial stability. Section 3 discusses the data, sources and variables employed in our empirical models to answer the four questions posed by this paper. Section 4 presents our main empirical results together with some discussion on them. In addition, there are robustness tests to our results in Section 5. Finally, we 4

5 conclude in Section Inflation Targeting and Financial Stability According to Mishkin (2004) and Heenan et al. (2006), inflation targeting consists of four elements: (i) an explicit CB mandate to pursue price stability as the primary objective of monetary policy and accountability for performance in achieving the objective; (ii) explicit quantitative targets for inflation; (iii) policy actions based on a forward-looking assessment of inflation pressures that considers a wide range of information; and (iv) increased transparency of monetary policy strategy and implementation. Therefore, in addition to setting an explicit target, countries that adopt this policy also experience enhanced CB communication and accountability that may increase the transparency of monetary policy and better anchor agent expectations. 2 Recent evidence suggests that IT is a monetary policy that reduces inflationary pressures and anchors price expectations in countries that implement it. In fact, these countries have passed through an initial period of disinflation by setting annually decreasing targets until a target of approximately 3% could be set (Roger, 2010). According to Gonçalves and Carvalho (2009), these disinflation periods have not been as costly in terms of output losses as in non-it countries. The authors argue that this result from the enhanced communication and accountability of monetary authorities under IT regimes. However, these results are contested by Brito (2010) who demonstrates that Gonçalves and Carvalho (2009) select only ad hoc four IT disinflation periods and compare them to earlier costly disinflation periods under different economic conditions. When either the sample is expanded or economic conditions are controlled, the results are inverted: IT countries disinflation is costly in terms of output. Notwithstanding the ambiguity of the real costs of disinflation, lower levels of inflation have, ceteris paribus, positive effects on financial stability. The traditional view argues that price stability is also beneficial to financial stability. According to the European Central Bank, (p)rice 2 According to a survey by Geraats (2002), even though the benefits of monetary policy transparency may depend on the assumptions made by the theoretical model (see Lepetyuk and Stoltenberg, 2013), there is large empirical evidence that supports the advantages of reducing asymmetric information between the CB and the private sector. For instance, Crowe (2010) shows that IT reduces the informational asymmetry between the CB and the private sector. The author finds that inflation forecasts improve significantly in IT countries compared to non-it countries. The results are even more meaningful for private sector agents who had higher forecast errors previous to the IT implementation. 5

6 stability contributes to achieving high levels of economic activity and employment by (...) contributing to financial stability. 3 The mechanism through which enhanced financial stability can be reached is simple. According to the Schwartz hypothesis, periods of unstable price levels can lead to incorrect inferences about the future real returns of investments. This may result in flawed lending/borrowing decisions, increasing loan defaults, compromising the banking system s loan portfolio, and increasing bankruptcies. Inflation targeting improves the credibility and predictability of monetary policy and reduces the degree of uncertainty about the price level over the long run. Bordo and Wheelock (1998) argues that for most of the financial crises that occurred during the XIX and XX centuries, the most severe financially distressing events occurred after unexpected and substantial disinflation. Despite the positive view of IT overall, that is, that the policy succeeds in stabilizing inflation, the recent financial crisis has increased criticism of this policy. The idea that price stability is not a sufficient condition for financial stability has gained strength. The main argument is that by focusing on assuring price stability, CBs might have overlooked changes in the banking system and asset bubbles (Blanchard et al., 2010). According to Borio et al. (2003), the CB s commitment to anti-inflationary policy may produce a paradox of credibility in which unsustainable booms take longer to be discovered and properly addressed. Several specialists suggest improvements to the IT regime, such as including a financial stability goal (White, 2006; Issing, 2009). 4 Others, such as Bean (2009), raise doubts in the achievement of both financial and price stability with only one instrument (Tinbergen s Rule). In addition, Bean (2009) argues that it may be difficult to (i) identify whether an increase in asset prices is fundamentally justified or not; and (ii) determine whether the necessary (i.e., often large) increase in interest rates to counter these financial imbalances will be socially desirable in terms of output loss. Another dimension of the IT critics argument relates to the monetary policy instrument employed to reach the inflation target: the short-term interest rate. For instance, a growing body of literature blames the loose monetary policy in place during the years before the crisis for increased bank risk-taking. In a speech, King (2012), the Bank of England s Governor between 2003 and That the decision to adopt and maintain an IT regime did not depend on financial stability considerations in the years prior to the crisis will be crucial to the identification hypothesis of our empirical model. 6

7 2013, questions whether higher overnight rates in the UK during the pre-crisis period, i.e., whether the monetary authorities should not have been so devoted to IT policy, would have avoided the banking distress that the country has faced. Borio and Zhu (2008) argue that the main problem with expansionary monetary policy is the risk-channel of monetary-policy, i.e., low short-term interest rates create incentives for investors to take more risks to obtain the returns promised to debtors. There is substantial empirical evidence to support this view in the banking literature. Dell Ariccia et al. (2013) employ bank-level data for the US banking system to demonstrate that ex-ante risk taking is negatively correlated with the federal funds rate. This pattern is consistent with the results that Jimenez et al. (2014) and Ioannidou et al. (2009) observe in analyses employing detailed, borrower-specific data on Spain and Bolivia, respectively. Nevertheless, others disagree that IT should be blamed for inciting risk-taking through loose monetary policy. First, in the same speech, King (2012) recognises that the UK s banks would have suffered deeply from contagion of the US financial crisis regardless of whether the Bank of England had adopted tighter monetary policy. Second, not all developed countries that maintained lower interest rates during the pre-crisis were employing IT. For instance, monetary policy in the US, where the crisis originated, is not an explicit IT. Third, agents would increase risk-taking only if they believe that the policy rate would remain low over a prolonged period. This idea is explicated in Jimenez et al. (2014) and Ioannidou et al. (2009). According to these authors, when rates decrease, banks tend to take more risks on new loans. However, the risk of outstanding loans actually decreases because of a wealth effect. We believe that an IT country would not commit to persistent lower interest rates because inflation may deviate from its target, unless the country experiences a liquidity trap. 3. Data and Methodology We draw bank-specific balance sheet data from Bankscope, a financial database distributed by BVD-IBCA, and convert these data into US dollars to guarantee accounting uniformity among countries. Initially, the data included the population of unconsolidated financial statements of commercial and specialised credit government banks (that act as a commercial bank) in the database (both listed and not listed). The use of unconsolidated financial statements avoids double-counting financial statements, since some banks may control others that are also present in Bankscope. We, 7

8 therefore, have access to data regarding state-owned, private, and even foreign subsidiaries banks that operate within each country covered in the database. 5 Finally, mergers and acquisitions are not a problem since from the moment where such operation is realized, the acquired bank stop to report and the acquiring bank incorporates it. To avoid problems with missing data for a relevant variable in a specific year, trienniums are utilised. In other words, we average relevant balance sheet data for each triennium, and if there is missing data for a specific bank and year, the observation for that triennium is the average value of the remaining two years. This procedure allows us to keep the observation in question in the sample. Observations with missing data for all years of the triennium or those with only one observation per triennium are not considered in our analysis. We filter our data as follows: We exclude bank-triennium observations periods with missing, negative or zero values for the relevant balance sheet data; We drop country-triennium observations whose banking systems aggregate market-share is less than 60% of the original data; We drop countries with fewer than 10 different banks (too small to consider). We drop banks with missing country-specific variables. The final sample is an unbalanced panel including 5468 banks in 70 countries during the period (5 trienniums) totaling 20,652 observations. This dataset is among the most representative in the banking literature in terms of the number of years and banks. Essentially, we are interested in determining whether banks from IT countries are riskier. To answer this question, we estimate a fixed effects model in which the dependent variable is a proxy for individual bank stability, and the main independent variable is a dummy variable that equals one if the country is IT. According to the literature, we also employ robust standard errors clustered by country to avoid understated standard errors when the group, i.e., banks, is more detailed than the main regressor s level of variation, i.e., countries. 6 5 For instance, Citibank will have one financial statement in the US, related to its operations in that country only; and one financial statement in each country in which it has foreign subsidiaries 6 See Donald and Lang (2007) and Bertrand et al. (2004) for detailed information. 8

9 Next, we explain our dependent variables, independent variables, and sources for other variables in detail. For more detailed information on these variables, see Table 1. Note that the subscript i refers to banks, t to trienniums, and k to countries. Place Table 1 About Here 3.1. Dependent Variables Our benchmark measure of financial stability is the natural logarithm of the Z-score. Many studies that evaluate bank risk-taking behaviour employ the Z-score as a measure of financial soundness (Mercieca et al., 2007; Laeven and Levine, 2009; Houston et al., 2010; Demirguc-Kunt and Huizinga, 2010). According to Roy (1952), the Z-score measures how far a specific bank is from insolvency. When equity is insufficient to cover losses (E < π), a bank is insolvent. Specifically, the Z-score is equal to the number of return over assets (ROA) standard deviations a bank s ROA must decrease to exceed equity. As in Laeven and Levine (2009), since the Z-score is skewed, we use its natural logarithm, which is normally distributed, as our proxy for financial stability of a bank: Financial Stability it = ln(z-score it ) = ln ROA it + Capital Ratio it (1) σ(roa) it This measure is often employed in cross-sectional OLS models in which the mean and standard deviation of ROA for the whole period can be calculated. However, we propose to calculate this measure for each triennium to maintain the Z-score as a panel variable. Therefore, rather than eliminating the time dimension of the analysis, this approach reduces the number of periods from 15 to 5. To deepen our analysis, we also divide the Z-score in its components: one part relates to the returns: 7 7 There is a problem in applying the natural logarithm of the Z-score or the Risk-Adjusted Returns, because these variable can take negative values as well. These negative values are possible when the bank s profitability offsets its capital ratio (in the case of the Z-score), or when the average ROA is negative (in the Risk-Adjusted Returns case), indicating that the bank is near insolvency. To solve this problem, instead of just deleting these observations, we 9

10 Risk-Adjusted Profits it = ln(rar it ) = ln ROA it ; (2) σ(roa) it another part measures the return s variability: and the remaining one, to the banks capital ROA Volatility it = ln(σ(roa) it ); (3) Equity Ratio it = Equity Capital it Total Assets it (4) Since this last variable is truncated between 0 and 1, we apply a logistic transformation before employing it as the dependent variable Independent variables Our main impendent variable is a dummy variable equal to one if the bank operates in an inflation targeting country k during triennium t (IT kt ). Currently, 26 countries have adopted inflation targets. In addition, 3 other countries have previously adopted this policy but abandoned it, just before joining the Eurozone and having to adopt the block s monetary policy. Table 2 lists these countries as well as the years in which IT took place. These data were drawn from the IMF website, Roger (2010) and the authors own research. Note that the majority of IT countries are emerging economies. Place Table 2 About Here In addition, we measure global illiquidity by employing the TED spread, which equals the difference between the 3-month LIBOR rate and the 3-month Treasury bill rate. Periods during which follow Bos and Koetter (2009) who employs an additional independent variable, the negative Z-score/RAR indicator (NZI/NRI). This variable takes the value 1 when the Z-score/RAR 0 and is equal to the absolute value of the Z- score/rar when Z-score/RAR < 0. We also alter the Z-score/RAR to take the value 1 when they are negative. Since we delete observations whose dependent variable are in the lower and upper percentile in our benchmark estimations, this problem will only occur in the Risk-Adjusted Profits specifications. The Z-score does not assume negative values from its 1st to 99th percentile. 10

11 this variable is large are marked by turmoil in world financial markets. In robustness tests, we employ two alternative measures: the S&P 500 Volatility Index (VIX) and the St. Louis Financial Stress Index. 8 We also include bank balance-sheet variables. First, as a proxy for bank size, we consider the within-country standardised natural logarithm of the assets of bank i at triennium t (SIZE S td. it ). In other words, we take bank total assets, subtract the average total assets of the country where it is operating, and divide by its standard deviation. Note that we construct this variable prior to cleaning our database. Another balance-sheet variable is the liquidity ratio (in %), which is liquid assets divided by total assets. Finally, we consider the costs to assets ratio (in %) to control for bank s cost performance. We derive a systemically important financial institution (SIFI) dummy that equals 1 (one) if the bank claims both total assets and wholesale funding higher than the 90th percentile of the country s banking system. Wholesale funding, in this case, is defined as the difference between deposits and other short-term borrowing and total customer deposits. These two variables measure important SIFIs characteristics, i.e., their size and interconnectedness (see Basel Committee on Banking Supervision, 2010, 2011). The choice of the 90th percentile threshold to define SIFI institutions is arbitrary. Nevertheless, in robustness tests, we modify the SIFI variable to be equal to one if: (i) both total assets and wholesale funding are higher than the 75th percentile; (ii) total assets alone are higher than the 90th/75th percentile; or (iii) wholesale funding alone is higher than the 90th/75th percentile. 9 We include countries economic activity and financial depth variables as additional controls. The first set of variables includes: (i) two indices created by the Heritage Foundation, i.e., property rights and financial freedom; (ii) two variables reported in the World Bank s World Development Indicators (WDI): the consumer price index (CPI), and the degree of economic openness, defined as the ratio between total trade and the GDP (in %); and (iii) the Economic Cycle derived from an recursive HP filter applied to the natural log of each country s real GDP (Constant 2005 US$). 10 In other words, the corresponding GDP trends are calculated using information available only up to 8 The overall conclusion do not change significantly if we employ these other two variables instead of the TED Spread. These robustness results are available upon request. 9 Tables are available in the appendix and upon request. 10 λ =

12 the year in question. 11 The percentual deviation between actual GDP and its HP-calculated trend is defined as the cyclical component of the GDP. We then take the average of this cycle by triennium and employ it as an explanatory variable in our models. The second set of variables includes banking market aggregates calculated from the balancesheet data contained in Bankscope, i.e., (a) the density of deposits (ratio of aggregate deposits to land area) and (b) the ratios of aggregate equity to assets (in %), domestic credit to private sector (as % of GDP) in the WDI, and a banking market competition proxy known as the Lerner Index. This variable is the price mark-up over marginal costs as a percentage of prices. Values close to zero indicate little overpricing and, therefore, a competitive banking market. Value close to 1, however, indicate extreme collusion. The literature has employed this competition measure widely due to the simplicity of estimating it. See Appendix A for more information on estimation. Table 3 displays the Lerner index by country and triennium. Place Table 3 About Here 4. Empirical Results In this Section, we present the results from our estimations. In all tables, column [1] provides the results without financial depth or economic activity controls. In column [2], we include only the former, while the latter are included in column [3]. Finally, column [4] displays the results including both set of control variables. In all specifications, we drop observations in which the dependent variable is higher than the 99th percentile and lower than the 1st percentile to avoid outliers. 12 The hypothesis that a country s choice to adopt and maintain an IT policy is exogenous to bank soundness considerations is crucial to our analysis. Given the elements of IT policy as explained in Section 2 and recent criticisms that it does not consider financial stability, we believe that this 11 Because the WDI database has data on real GDP since 1960 for most of the countries, we will not have any problem to calculate the filter for our period of interest ( ). 12 Overall results do not change if we keep these observations extreme observations. These results are available upon request. 12

13 hypothesis is valid. In other words, prior to the financial crisis, there was a consensus about the dichotomy between monetary policy and financial stability (Mishkin, 2011). Therefore, IT was not employed due to financial stability concerns, nor has the last preceded the former s implementation Q1 and Q2 - Are banks from IT countries more stable? Does it depend on the inflation index? Table 4 displays the regressions of our bank-specific financial stability variable against a series of variables and a dummy variable equal to one for inflation targeting countries. Since the fixed effects estimator of IT kt is driven by its implementation, we drop IT countries that have adopted this policy in 1999 or in the years before in the analysis of this table (total of 9 IT countries). 13 In addition, Table 5 shows the specifications in which we add an interaction between the IT dummy and the consumer price index to access for the credibility of the IT policy. In this case, since our interest is the interaction coefficient, we need not drop any country from the analysis. Place Tables 4 and 5 About Here The IT dummy is significant and positive in all specifications, i.e., countries that adopt inflation targets have, on average, less fragile banks. This result holds even when we control for financial depth and economic activity proxies. In Table 5, the interaction between IT and CPI is negative, as expected, but not significant in the last three columns. Banks from IT countries are, therefore, weakly more sensitive to inflation than banks from other countries. Note that the IT marginal effect on financial stability, i.e., FinS tab it IT kt is decreasing in the CPI and its hypothesis test rejects the null hypothesis at 10% significance level until a CPI around 14% (in column [1]) and 12% (in column [4]). On the one hand, it appears that there is a channel from price stability (in addition to increased CB communication and accountability) to banking system soundness. On the other hand, this effect disappears for high levels of inflation, meaning that lower credibility by the CB may hamper the stability channel. 13 Note that if we have not removed them from the analysis, they would be also considered as a control group. Nevertheless, when we do consider them in the estimation, the coefficients do not change significantly (they are almost the same as the ones presented in Table 4. This finding shows that the differences between IT and non-it banks in financial stability is what drives the results. 13

14 Place Tables 6 and 7 About Here When we disaggregate the dependent variable (Table 6), we observe that the IT coefficient remains significant only when the dependent variables are Risk-Adjusted Profits and ROA Volatility. The effect of an IT policy does not appear to significantly increase bank s capitalization. Table 7 shows that the the dummy s interaction with CPI kt is significant is often significant for all the dependent variables. The interaction coefficients are negative for Risk-Adjusted Profits and Capital Ratio and positive for ROA Volatility. Therefore, one can clearly see the price stability channel: CB commitment to maintain an inflation target allows banks to obtain higher and less variable returns, i.e., profits are more predictable, and therefore risk-taking is lower. A low credibility of this commitment by the CB reduces the overall positive effect of IT Q3 - Are banks from IT countries more stable during times of global illiquidity? Table 8 compares the results of including an interaction between the TED spread and the inflation target dummy to the benchmark specification. Place Table 8 About Here This interaction is positive, but only significant in column [1]. In other words, banks from IT countries are as (or even less) vulnerable as (than) those from other countries. This is another piece of evidence for the price stability channel because banks from IT countries are resilient to global liquidity shocks. Because the financial crisis of is when financial distress was at the highest levels, this result suggests that IT cannot be blamed for accumulating financial imbalances, contrary to the arguments of IT critics. Place Table 9 About Here When we separate the financial stability proxy into risk-adjusted profits and equity ratio, we observe that the interaction between TED Spread and IT continues to be significant at 10% in the first 14

15 column of both the risk-adjusted profits and the ROA Volatility s equation, adding more evidence to our results. In all specifications of the equity ratio equation, this interaction is insignificant. In an addition test, we also re-estimate our models employing the VIX S&P500 and the St. Louis Financial Stress Index, instead of the TED Spread. Both of these variables are also employed as global financial distress proxies. The interactions between the IT dummy and these global illiquidity proxies remain positive but not significant in the Z-score specifications. One can conclude that our results are robust to the measure of distress chosen Q4 - Are systemically important banks from IT countries more stable? Another relevant question is whether IT countries let systemically important banks take more risks than those SIFIs from non-it countries. Table 10 displays the results of including a systemically important dummy that equals one if the bank has both total assets and wholesale funding higher than the 90th percentile of the country s banking system. We also include an interaction between this dummy and the IT dummy. Place Table 10 About Here Systemically important banks from IT countries are more stable, and this result holds for all specifications. This is a startling result that counters the argument that when the authorities pursue price stability, they let these banks increase their leverage and risk-exposure. What can be inferred from these results is that, on average, large banks do not experience moral hazard. Place Table 11 About Here Indeed, this result appears to come from both returns and capitalization. SIFIs have on average higher risk-adjusted returns (significant in column [2] only), higher capitalization (significant in both columns [3] and [4]), and lower ROA Volatility (significant in columns [2] and [4]). When we modify the threshold of both total assets and wholesale funding in order to define the SIFI dummy variable, results do not change significantly. If SIFI is equal to one when both total 14 Results are available upon request. 15

16 assets and wholesale funding are higher than the 75th percentile, the interaction coefficient is still positive, but not statistically and economically significant (the coefficient s magnitude is around 5%). When we consider as SIFI banks with total assets higher than the 90th or 75th percentile, the coefficients are positive, but only significant when we apply the 90th percentile size threshold. Finally, when we only consider the wholesale funding to define a SIFI, the coefficients are negative, but insignificant (75th percentile), and positive and insignificant (90th percentile). Thus, we have not found any specification in which SIFIs from IT countries are closer to insolvency than those of non-it countries. Results seem to point the other way around Robustness Tests 5.1. Legal Origins In this Section, we compare the banking systems of countries with the same legal origins. We also present the results from the last Section disaggregated by legal origin. It is largely accepted that countries with the same legal origin are subject to similar (if not necessarily equal) constraints, which may affect the development and functioning of banking markets among these countries (Levine, 1998; Beck et al., 2003). In addition, differences in CB communication and credibility may exist among these countries. For instance, Horvath and Vasko (2013) demonstrate that German and Nordic CBs are often more transparent than countries with other legal origins. We aggregate countries into three groups: English, French and German/Nordic. To divide countries by their legal origins, we follow the classifications of La-Porta et al. (1997, 1998, 2008) and Djankov et al. (2003) and complement this information by checking the CIA World Factbook and additional research. Ex-socialist countries are transition economies that are returning to their original legal system, as La-Porta et al. (2008) argument. Therefore, we list these transitional economics according to their original legal origin instead of assigning them to a socialist legal system. Table 3 lists the countries in our sample and their legal origin. In this Section, we determine whether our previous results hold when we compare countries with similar legal systems. Therefore, we present the same regressions from the previous Section divided by legal origin (English, French, and German/Nordic) Results are available upon request. 16 We choose to merge German and Nordic legal origins because of the low number of countries of the latter. 16

17 Place Tables 12, 13, 14, and 15 About Here Table 12 indicates that the coefficients for inflation targeting are positive for all specifications and legal origins, but not always statistically significant. Thus, IT countries have stronger banking systems even when compared to other countries with similar characteristics. Note, however, that the coefficients of German/Nordic countries are greater in magnitude, indicating that the difference between IT and non-it countries is larger for this group of countries. The coefficients for English legal origin countries are the next largest in magnitude followed by the French. The results from Table 14, however, are slightly different. The coefficients of the interaction between TED Spread and IT are not significant in all specifications of both English and German legal origins. Note that these results do not validate the alternative hypothesis because insignificant interactions imply that IT countries are neither more nor less financial fragile than their legal origin peers. In Table 15, the interaction between SIFI it and IT kt is positive and significant for all four models for English and French legal origins. For German/Nordic countries, however, all coefficients are insignificant, but still positive. Again, systemically important banks from IT countries are more stable than (English, French) or as stable as (German) non-it peers. Overall, the results are consistent with the benchmark model. Although there are some differences, especially in the statistical significance of coefficients, IT countries appear to have more sound financial systems even when compared to countries with similar legal environments Supervision as a responsibility of the CB As an additional robustness test, we consider whether the country has an institutional body, other than the CB, that supervises commercial banks. IT critics argue that CBs from IT countries may be too focused on reducing inflation to direct sufficient attention to the functioning of the financial system. This argument would be weaker when the supervisory body is outside the CB. Therefore, we test whether banks from IT countries whose CB also holds supervisory power over the financial system are more fragile. We employ a dataset from Barth et al. (2001, 2004, 2008, 2013) that characterises regulation and supervision for several countries. The variable that we utilise is collected by question from the 2008 and 2011 surveys, which asks whether the 17

18 CB is responsible for supervising banks. This dataset, however, has some shortcomings: (a) it does not cover all the countries in our database; 17 and (b) the survey questions regarding whether the supervisory body is separated from the CB changes from the 2000 and 2003 to the 2008 and 2011 surveys. 18 Due to this last problem, we choose to work as follows. We consider the answer to the 2011 survey as the information for the last triennium ( ), and the answer to the 2008 is considered to be the information for all the remaining trienniums. We do so, because of the change in methodology after the 2003 survey that we need to address with further investigation. Anyway, this procedure is an approximation and can be considered as a first look into this question. Since there are some missing data and countries that are not covered by the database, the number of countries in our analysis reduces to 68 and the number of banks reduces to We then create a new dummy variable (CB Sup. Respons. kt ) that equals one if the CB is responsible for bank supervision in that country. We interact this variable with the IT dummy and rerun our estimations to see whether a CB serving both functions (bank supervision and inflation targeting) would be careless toward the financial system. Results are shown in Tables 16, 18, and Place Tables 16, 17, 18, and 19 About Here We do not find evidence to support this recent criticism of IT. Banks in countries where CBs hold both functions are: (i) still overall more stable, as seen by the positive and mostly insignificant coefficients of the IT kt - CB Sup. Respons. kt interaction in Table 16; (ii) less affected by high levels of inflation, as shown by the positive coefficient of CPI kt IT kt CB Sup. Respons. kt, even though they are not significant; (iii) as resilient to international financial shocks as others, despite the insignificantly negative coefficients of the triple interaction between TED Spread t, IT kt, and CB Sup. Respons. kt (Table 18). On the other hand, (iv) the advantage of systemically important 17 See the papers Barth et al. (2001, 2004, 2008, 2013) to access the country coverage of each survey. 18 In 2000 and 2003, the question was whether there is more than one supervisory body, which does not necessarily mean that the CB is not responsible for supervising banks. 19 In Table 16, we drop 3 countries whose interaction between IT kt and CB Sup. Respons. kt equals one for all the five trienniums. Because of the use of the fixed effects, we avoid considering these three countries as control groups, instead of treatment groups, if we delete them from the analysis. 18

19 banks in IT countries is reduced in countries where CBs are also responsible for bank supervision, as seen by the large positive coefficient of S IFI it IT kt CB Sup. Respons. kt Dealing with the financial crisis One may ask whether our results are mainly explained by the comparison between IT countries and developed economies that were deeply affected by the financial crisis. We address this argument by re-estimating our three specifications, but excluding from the analysis banks operating in the G7 countries. This leaves us with 2,812 banks operating in 63 countries. Even though the coefficients are slightly smaller in magnitude, results from Q1 are still statistically significant in the first three columns. Results from Q3 and Q4, however, loose significance at a 10% level. Thus, the banking system, and more specifically SIFIs, still performs better under an IT regime, while during periods of illiquidity all banks are affected regardless of whether their country adopt an IT policy or not. Finally, we have rerun the estimations, but including a dummy equal to one for the fourth triennium ( ). We interact this variable with the IT dummy to access the effects of the financial crisis on our estimations. The results are as follows: the crisis dummy alone is negative and significant; its interaction with the IT dummy is positive and significant in all specifications, except in column [2]; and the IT coefficient is still positive and significant. We conclude that the crisis explains, but it does not drive, our conclusion that IT banks are more stable than their non-it counterparts Discussion on the Euro-Zone We are also aware that the European Central Bank has adopted a regime that has several attributes of IT, even though, according to Svensson (2008) the ECB s policy cannot be considered truly IT yet. In fact, Issing (2009) states that the main differences between ECB s framework and 20 Even though the coefficient of this triple interaction is not significant, the overall effect of IT on SIFIs soundness must consider the analysis of both the coefficients of S IFI it IT kt, and S IFI it IT t CB Sup. Respons. kt. For example, the partial effect of IT on fragility in column [3] and for SIFI it = 1 is: Fin. Fragility it IT kt = CB Sup. Respons. kt When CB Sup. Respons. kt = 1, then Fin. Fragility it IT 21 kt Results are available upon request. =

20 that of an IT are: (i) the lack of combination between a forward-looking assessment of inflation and the actions taken by monetary policies due to structural changes and uncertainties over an extended period of time; (ii) the relevance given to money in the ECB policy. We test a specification in which we consider countries from the Eurozone as inflation targeters - IT dummy equal to one - but in which we also include a Eurozone dummy to disentangle the effect from these countries. By construction, the IT dummy results remain the same. The Eurozone dummy, however: (Q1) is negative and larger in magnitude than the IT dummy; 22 (Q3) its interaction with TED spread is negative but insignificant; (Q4) its interaction with the systemic dummy is negative and significant, and about two times higher than the interaction between IT and SIFI. 23 Perhaps, these results show the importance to fully commit to inflation target, especially by taking into account inflation expectations and forecasts. In addition, one has to bear in mind that, even though the ECB may adopt some aspects of an IT regime, supervision and bail out duties are attributed to sovereigns and not the union. Therefore, part of the responsibility for the increased instability may not be due to the monetary policy per se Inflation Targeting and Countrywide Financial Stability The main analysis of this paper asks whether banks are closer to insolvency if they are from IT countries. Even though this does not necessarily reflect the risk of the entire banking system, we have learned from the crisis how the insolvency of even one bank can contagion to other banks and trigger severe financial turmoil. Even so, we propose another robustness test, in which we are interested in determining whether our results vary if we regress a weighted average of our financial stability variable, the Z-score, by country and triennium on the IT dummy and other controls. Note that in this case the country s Z-score can be interpreted as the inverse probability of default of the entire banking system. Our results do not change considerably. We still find a positive IT coefficient, but only significant in the first three specifications. When we interact the IT dummy and the inflation index, 22 Note that this result is driven by only one country, Slovenia, who has adopted the Euro during our analysis period. The other Euro members have been adopting this currency for the whole period of analysis and they were dropped from our Q1 analysis of this subsection. 23 Results are available upon request. 20

21 we find that the IT coefficient alone is always positive and significant (ranging from 0.85 to 0.60) and the interaction coefficient is always negative and significant (ranging from to -0.06). The positive effect of IT in this case becomes insignificant at a consumer price index ranging from 7% to 5%. These thresholds are relatively lower than the ones from the bank-specific model, but the overall interpretation remains the same: more inflation, less credibility and less financial stability benefits, as well. Finally, when we interact the IT dummy with the TED Spread, its coefficient is positive but not significant in all cases. Again, our previous conclusion holds: banks from IT countries are not more vulnerable than those from non-it countries in times of global turmoil. 6. Final Summary In this paper, we provide new evidence on the relationship between inflation targeting and financial stability. We employ data for nearly 5500 banks from 70 countries during the period to assess whether banks from IT countries are more fragile than banks in non-it countries. Our motivation is recent criticism of IT policy. Specialists argue that by not considering the impacts of monetary policy on financial stability, IT disregards the development of financial imbalances, such as bubbles and excessive leverage, which increases the probability of financial distress. We note that this view may not be entirely correct. Indeed, IT countries have sounder banks, on average, which means that they can stabilise their banking sectors by reducing price volatility and providing stronger communication and accountability. We also find that banks from IT countries are as affected as other banks by global financial turmoil. These findings support the price to financial stability channel. Finally, systemically important banks in IT countries appear less susceptible to be risk taking, i.e., their returns are higher and less variable compared with SIFIs of other countries. Our results are somewhat robust to two alternative specifications. First, when we compare banks only with those that are subject to similar legal constraints, the advantage of price stability is still evident. Second, results are almost the same even if the country has a CB whose responsibility is both supervising banks and maintaining the inflation on the target. We conclude that there is no trade-off between financial stability and the directives of the IT policy even in this case. Nevertheless, we argue neither that CBs must only attain and maintain price stability nor that no trade-offs between monetary policy and financial stability exist. We only affirm that over the 21