Monetary Policy and Banking Supervision: Is There a Conflict of Interest?

Transcription

1 Monetary Policy and Banking Supervision: Is There a Conflict of Interest? D. Lima I. Lazopoulos V. Gabriel December 17, 2012 Abstract The objective of this paper is to empirically assess whether central banks are less aggressive in their inflation mandate when they are in charge of banking regulation, since tight monetary policy conditions could have an adverse effect on the stability of the banking system. Due to this conflict of interest between the monetary policy makers and bank regulators, it has been argued that banking supervisory powers should be assigned to an independent authority to avoid an inflation bias. We perform an econometric analysis using panel data for 25 industrialised countries from 1975 to 2007 to analyse the impact of a country s institutional mandate of monetary policy and banking supervision on inflation outcomes. Using a fixed effects approach, the estimation results obtained do not provide evidence to suggest that separation of banking supervision and monetary policy has a significant effect on inflation outcomes. Nevertheless, results show that other institutional factors, such as inflation targeting and deposit insurance schemes, are significant determinants of inflation outcomes. Key words: monetary policy, banking regulation, institutional mandates. 1 School of Economics, University of Surrey and Central Bank of Portugal. 2 School of Economics, University of Surrey. Supported by FCT Fundacao para a Ciencia e a Tecnologia, through project SFRH / BD / /

2 1 Introduction The recent financial crisis raised the debate regarding the institutional architecture of monetary policy and banking regulation: 1 In the clear light of morning, the legal or otherwise effective divorce between central bank and supervisory agency, or monetary policy and regulatory policy, was a mistake. (...) The guards in the twin towers of monetary policy and regulatory policy surveyed their compounds as if the other did not exist. (...) Monetary policy was deliberately oblivious to the asset price boom that was somebody else s problem. Regulatory policy was oblivious to macro risks that was the central bankers job. (...) In booms there is a premium to new trends and a discount on old ideas and much good is abandoned for the sake of modernity. Avinash Persaud, Feb 21st 2011, The Economist Historically, the institutional arrangements concerning the monetary policy and banking regulation were mainly influenced by two distinct traditions; the English influence where monetary policy and banking supervision are combined under the central bank, and the German influence where these functions are separated. 2 According to Haubrich (1996), the origin of these different traditions is related to the evolution of the payment system. Countries that adopted the English tradition experienced a rapid expansion of credit through the introduction of alternative forms of money, where the central banks naturally became the guarantors of the smooth functioning of the payment system and the regulators in these marketbased financial systems. In contrast, countries that experienced a slow expansion of credit developed a bankbased financial system of wellcapitalised banks that were regulated by an independent authority following the German tradition. However, in the late nineties there was a tendency for the separation of these functions following the German tradition. 3 According to Masciandaro (2009), in a sample of 91 countries, 94% chose to reform their financial supervisory architecture between 1986 and 2006, which suggests that there has been a tendency to unify the financial system regulators within the same agency that is different from the central bank. In particular, it is shown that the degree of unification of financial regulators is inversely related to the central bank s role in banking regulation and supervision. More recently, the 2008 financial crisis questioned this apparent consensus towards the separation of functions, and many countries, including the European Union and the United Kingdom, are currently implementing reforms regarding the role of the central bank in banking supervision mandates (Dalla Pellegrina et al., 2010). Notwithstanding the institutional mandate trends around the world, there are strong arguments for and against separation of banking supervision from the central bank in the academic literature. The aim of this paper is to empirically examine if monetary policy makers and bank regulators have conflicting interests by assessing whether a combined institutional mandate has led to higher inflation rates, on average, than a separate regime. This important 1 In this paper, following the literature, we have opted to not distinguish between the concepts of banking regulation and banking supervision. 2 For example, countries with an English influence include the United States, United Kingdom, Australia and Hong Kong, whereas countries with German influence include Austria, Germany, Denmark, and Switzerland. 3 Most notably, in 1999 the European Central Bank was assigned the responsibility for the conduct of monetary policy in the euroarea and the national authorities became in charge of the banking regulation and supervision. Likewise, United Kingdom and Australia had opted for the separation of these functions. 2

3 argument in favour of the separate regime states that a central bank responsible for banking regulation will be more flexible in its inflation mandate if it fears that tight monetary conditions may cause bank distress due to adverse effects of high interest rates on the profitability and soundness of the banking sector (Goodhart and Schoenmaker 1993 and 1995). 4 Under these circumstances, it is likely that the flexibility in guiding monetary policy will lead to higher inflation rates. There is evidence in the literature that supports the existence of a conflict of interest between monetary policy makers and bank regulators. Using data from the United States over the period , Ioannidou (2005) examines whether monetary policy responsibilities have implications in the conduct of the bank supervision when the Federal Reserve System (Fed) is responsible for both functions. 5 The results suggest that monetary policy influences Fed s supervisory actions as it turns out to be more flexible in its bank supervisory role when it tightens the monetary policy stance. Moreover, focusing on 25 industrialised countries over the period , Di Noia and Di Giorgio (1999) find evidence that average inflation rate is explained by countries institutional mandate when controlling for central bank independence from the government. The authors conclude that central banks are less effective in controlling inflation when they are responsible for the regulation of the banking sector. In a similar study, Copelovitch and Singer (2008) consider 23 industrial countries from 1975 to 1999 and found empirical evidence that inflation rates have been significantly lower, on average, in countries where the central bank and the banking supervisor are separate agencies. This effect is conditional on the choice of the exchange rate regime and the size of the domestic banking sector. In particular, the separation mandate has a significant negative effect on inflation under floating rates, but this effect is only observed at middle to high levels of banking sector size. Our analysis is based on the works of Di Noia and Di Giorgio (1999) and Copelovitch and Singer (2008), but we introduce a number of innovations. First, the time span of the sample is enlarged from 1999 to The end date falls before the start of the recent recession of 2008, because our aim is to capture normal times. Second, we suggest the use of appropriate methods to estimate panels, such as fixed effects. In addition, we implement two different data specification approaches; one follows Copelovitch and Singer (2008) s approach of fiveyear average inflation rate, while the other uses annual panel data. Fourth, we introduce additional explanatory variables, such as inflation targeting, oil imports as percentage of GDP and bank concentration index. The estimation results show that the separation of banking supervision from the central bank does not have a significant impact on inflation. In this sense, the conflict of interest argument against combination of monetary policy and banking supervision is not supported by our empirical findings. More interestingly, we find that inflation targeting and deposit insurance are the main institutional driving forces of low inflation rates. Bank concentration, as a proxy for the market structure of the banking system, is also significant and it has a negative impact on inflation outcomes. External factors, such as trade and capital account openness, are also 4 The impact of shortterm interest rates on banks profitability and solvency does not only depend on the the length of time for which high interest rates are likely to persist, but also on the structure of the banks balance sheet, i.e. the interest sensitivity difference between assets and liabilities and the weighted average maturity of fixed rate assets relative to variable rate liabilities. 5 Note that the regulatory architecture of the banking system in the United States is such that the Fed along with the Office of the Comptroller of the Currency (OCC) and the Federal Deposit Insurance Corporation (FDIC), all share the supervisory powers, but the Fed is the only regulator responsible for monetary policy. 3

4 important determinants of inflation. In addition, our empirical evidence indicates that central bank independence has only a negative significant impact on inflation rates when we are not controlling for the external factors. The paper proceeds as follows. Section 2 presents the data and describes the methodology used. The estimation results are presented in Section 3, and Section 4 concludes. 2 Empirical Analysis 2.1 Data We consider annual time series data for 25 OECD countries over the period The dependent variable is the annual inflation rate, in logarithmic form. The selection of explanatory variables is based on Copelovitch and Singer (2008), but we also include a number of variables that we believe can be related to inflation. Appendix 1 provides the definition and data sources for each explanatory variable considered in the econometric analysis. The banking supervisory separation variable (thereafter separate) is the main explanatory variable we are interested in, since the purpose of this study is to measure the impact of a separate banking supervision mandate on inflation behaviour. We define separate as a dummy that takes value of 1 if, in a certain country, the functions of monetary policy and banking regulation are conducted by two independent institutions, and value 0 in the opposite case, which we classify as a combined mandate. The classification of countries into these two groups (separate or combined) was based on information disclosed by the World Bank Regulation and Supervision Survey (updated June 2008) 6, which collected information from supervisory authorities located in 142 countries for the year This information was complemented with other sources. We consulted Copelovitch and Singer s classification and in the cases for which there was uncertainty about the type of mandate, the central banks and supervisory agencies websites were also checked. The Courtis (2011) survey about the supervision arrangements around the world was also used to confirm our previous classification. Table 1 shows the evolution of the institutional arrangements and inflation rates in the 25 countries included in our sample along the period In addition, we have also analysed the responses from an older version of the same survey (2000). 4

5 Table 1: Number of countries with separate and combined mandates per year and average inflation rates. Year Separate Mandate Combined Mandate Inflation Rate % % % % % % % % In 1975, the banking supervision responsibilities were assigned to the central bank in 17 countries, whereas only 8 countries had preferred to allocate this policy to a different authority. This distribution remained stable until late 1990 s, when we observe an increase in the number of countries that have opted to separate their banking supervision functions from the monetary policy authority: in 2000, 12 presented a separate institutional mandate and 13 a combined one. In 2007, the last year considered in our sample, the majority of countries (14 out of 25) had preferred to assign their banking supervision powers to a different authority. During the sample period, inflation rates also decreased substantially: in 1975, the global sample inflation was 13,5% on average, and in 2007 it had decreased to 2,2%. The classification of countries in terms of separate and combined institutional mandates is presented in Appendix 2. For the euro area member states, we assume a separate banking supervision regime after their entrance in the European Monetary Union, in 1999, 7 since monetary policy is centralised in the European Central Bank. 8 We expect that a separate banking supervision regime will have a negative impact on inflation rates, as argued by the conflict of interest effect. A pool of additional regressors is considered and they are divided in four categories: institutional, external, economic and banking structure. In addition to the variable separate, we include as institutional factors central bank independence (CBI), explicit deposit insurance, inflation targeting, exchange rate regime, euro membership and a dummy variable to account for the great moderation period. There is a large literature following Cukierman et al. (1992) that suggests a significant negative impact of the degree of independence of the central bank on inflation outcomes. Thus, it is expected that a country with a higher degree of central bank independence will also have, on average, lower inflation rates. However, the empirical evidence is not conclusive, since although the first studies confirm a significant effect of CBI on inflation rates (Grilli et al., 1991 and Cukierman et al., 1992), there are recent papers that do not find a strong effect (Mangano, 1998 and Crowe and Meade, 2007). The CBI variable is based on the work of Arnone et al. (2007). In their study, they update the Cukierman et al. (1992) and Grilli, Masciandaro and Tabellini (1991) measures for central bank political and economic autonomy. Political autonomy is interpreted as the power of central banks to define and implement monetary policy, 7 Except for Greece, that joined the European Monetary Union in It can also be argued that the European Monetary Union is a combined mandate, in the sense that the national central banks are part of the Euro System. We perform a sensitiveness analysis to assess for the opposite case, in which we assume a combine institutional mandate between monetary policy and banking supervision, whenever the national central bank is in charge of supervisory responsibilities. The estimation results do not change substantively. 5

6 whereas economic autonomy evaluates the central banks operational independence. The CBI measures that we find in the literature are usually computed for certain periods of time. This means that is not possible to find a time series for a CBI index. The way researchers circumvent this problem is by assuming that CBI measures do not change significantly across time. We follow this approach, and we assume the CBI measures computed in the late 80 s do not vary till 2003, the year for which Arnone et al. (2007) update the index. Another institutional determinant of low inflation rates is associated to depositors protection. A country with an explicit deposit insurance scheme with the role of protecting depositors from losses in the event of a bank failure, will experience, on average, lower inflation rates, because the central bank can be aggressive on its inflation mandate as it is not concerned about the effect of interest rates on banking stability. The deposit insurance variable takes value of 1 for countries with explicit deposit insurance and of 0 otherwise. The classification uses information from the World Bank Deposit Insurance Around the World Dataset, from 1975 to 2003, and from the International Association of Deposit Insurers (IADI), for the remaining years. To account for the effects of inflation targeting on inflation behaviour we introduce a dummy variable that takes value of 1 at the year that a country adopted inflation targeting and onwards 9 and value of 0 in the remaining cases. The classification is based on Roger (2010). Since this approach pursues an explicit public commitment to control inflation as the principal policy goal we expect that a country that has adopted inflation targeting will experience lower inflation rates. The exchange rate regime variable takes value of 1 for all varieties of hard fixed exchange rates and 0 for floating or managed floating regimes. Data is based on the International Monetary Fund classification, by Ilzetzki, Reinhart and Rogoff (2008). The Great Moderation dummy variable is included to control for the persistent decline of inflation rates in the developed world since the early 1980 s. The structural breaking point is the year 1984 following McConnell and PerezQuiros (2000). Euro membership is controlling for the Euro Area countries specific monetary policy mandate and it takes the value of 1 from 1999 onwards for the Euro Area member countries, except for Greece, that entered the European Monetary Union in In order to capture for the impact of external factors, we consider trade openness, capital account openness and oil imports as percentage of GDP. Trade openness is measured as the sum of imports and exports as a percentage of GDP and the data is taken from the Comparative Political Dataset ( ). Theory predicts an inverse relation between trade openness and inflation: more open economies will benefit from lower inflation, on average. The degree of openness affects inflation through two different channels (Romer, 1993). First, in a theoretical model of monetary policy without precommitment, a more closed economy has a higher incentive to engage in surprise inflation since its impact on the real depreciation is less costly, given that the fraction of imported goods is lower in this economy. 10 Second, openness affects the outputinflation tradeoff: for a given increase in output, the rise in domestic prices will be higher as more open the economy is, given the exchange rates disciplining effect. Thus, monetary policymakers incentives to engage in expansionary policies are lower in more open economies, and therefore inflation is expected to be smaller. Empirical evidence 9 In our sample, the countries that use inflation targeting are New Zealand (since 1990), Canada (since 1991), United Kingdom (since 1992), Sweden (since 1993), Australia (since 1993), Iceland (since 2001) and Norway (since 2001). Finland and Spain adopted inflation targeting, in 1993 and 1995 respectively, but abandoned it when they entered the Euro Area, in In these models it is assumed that domestic and foreign goods are not perfect substitutes. 6

7 supports theory, by showing a strong and robust negative impact of openness on inflation outcomes (Romer, 1993). Capital account openness is measured using the ChinnIto index, developed in Chinn and Ito (2006). 11 This index accounts for restrictions on capital account transactions, current account transactions, requirements of the surrender of exports proceeds and the presence of multiple exchange rates. Similarly to trade openness, empirical evidence shows a negative relationship between financial openness and inflation (Gruben and McLeod, 2002 and Gupta, 2008). For oil imports, we expect that an increase in oil prices will have a positive effect on inflation of oil importer countries. The data for the value of oil imports is from the World Economic Outlook Database, published by the International Monetary Fund. To account for the effect of economic conditions on inflation, we include log of Gross Domestic Product (GDP), log of GDP per capita, currency crisis and banking crisis. Currency and banking crisis are dummy variables that take value of 1 whenever the country is experiencing a currency or a banking crisis. 12 We expect that inflation will be higher on average when the country is going through a banking or currency crisis. The log of GDP and GDP per capita were considered to control for the size and the level of development of countries, respectively. Lastly, to allow for the possibility that the magnitude of the influence of institutional mandates of monetary policy and banking supervision on inflation outcomes may be affected by the structure of the banking system, two variables were included: the size of the banking system, measured by domestic credit over GDP, and the degree of banking concentration, measured by the ratio of total assets held by the three largest banks in each country over the banking system total assets. 13 The size of the banking system may have a negative impact on inflation outcomes, since when the banking system contributes to a larger share of the domestic economy, a supervisory central bank will be more concerned about the monetary policy effects on bank stability (Copelovitch and Singer, 2008). The impact of the degree of bank concentration on inflation is not as straightforward. There are two distinct views in the literature concerning the impact of banking concentration on the monetary policy transmission mechanism. One view, more common, states that higher concentration implies less competition, higher profitability due to greater interest margins between deposits and loans, and therefore a less efficient transmission mechanism of monetary policy to the real economy. In this case, we would expect a positive impact of bank concentration measures on inflation. Alternatively, the efficientstructure theory (Demsetz, 1973) argues that costefficient banks could drive costinefficient banks out of the market and increase their market share, which would lead to higher concentration and greater profitability. In this situation, profitability is generated due to cost efficiency and the transmission of monetary policy is not affected as interest margins are also not affected. Therefore, according to this approach, a higher concentration degree in the banking industry would lead to lower inflation rates. Table 2 summarises the expect impact of each explanatory variable on inflation rates. 11 The ChinnIto index is taken from the Comparative Political Dataset ( ). 12 For banking crisis, data is based on Glick and Hutchison (1999), except for Australia and USA, for which data comes from Caprio and Klingebiel (2003). For the remaining years ( ), data comes from Laeven and Valencia (2010). For currency crisis, data is based on Glick and Hutchison (1999), except for Australia and USA, for which the data comes from Laeven & Valencia (2008). For the years , data comes from Laeven and Valencia (2008). 13 For the period , the data follows Beck, DemirgukKunt and Levine (2006). From 2002 onwards, we update the dataset using Orbis database to compute the index. 7

8 Variable Table 2: Regressors expected impact on inflation. Description Expected Impact on Inflation Institutional Factors Separate CB Yes = 1 CBI Central Bank Independence Index Inflation Targeting Yes = 1 Great Moderation = 0; = 1 Deposit Insurance Yes = 1 Exchange Rate Regime Fixed = 1 + Euro Membership Yes = 1? External Factors Trade Openness Index Capital Accounts Openness ChinnIto Index Oil Imports (% GDP) Controls for oil importing countries + Economic Factors GDP (log) Controls for size of the economies + GDP per capita (log) Controls for wealth of the economies Banking Crisis Yes = 1 + Currency Crisis Yes = 1 + Banking Structure Factors Bank Concentration Index Controls for the banking sector degree of concentration + / Domestic Credit (% GDP) Controls for the banking sector size 2.2 Model Specification In order to estimate the relationship between inflation rates and institutional arrangements of banking regulation and supervision, we adopt the following regression model: π i,t = β 0 + β 1 separate i,t + λx i,t + u i,t (1) where, π i,t is the (logged) inflation rate for country i in year t, separate i,t is a binary variable that takes value of 1 if the country is classified as separate at time t and value of 0 otherwise, X i,t is a vector of control variables included in the model (see Section 3.1 for a description of the variables) and u i,t is the error term. The econometric analysis is developed in two stages. First, we suggest the use of annual panel data, that comprises a timeseries component of 33 years, ranging from 1975 to 2007, and a crosssection pool of 25 advanced countries. Second, we follow Copelovitch and Singer (2008) that compute fiveyear averages for inflation rates, assuming that many of the institutional determinants of inflation rates included in their model change infrequently. It is formed by a timeseries component of 7 time periods and a cross section component of 25 developed countries. For each data specification, annual and 5 years average panel data, we estimate a Full Model (Model 1), that includes all explanatory variables described previously. Then, we test for the collective statistical significance of the exclusion of each block of explanatory variables (external, economic and banking structure regressors), by calculating a F Test. The estimation results are presented for each version of the model (Models 2, 3 and 4, respectively). In 8

9 addition, we estimate a reduced version of Model (1) that includes only institutional factors. 14 The models are estimated by using a Fixed Effects approach. 15 Panel data models have some advantages over crosssection or time series models, because they integrate heterogeneity across units (individuals, firms or countries) that is typical in microeconomic data. This heterogeneity is taken explicitly into account by allowing for individualspecific variables. If it is assumed that heterogeneity (individual effect) is correlated with the regressors, fixedeffects model may be appropriate, whilst if it is not assumed that there is correlation, the randomeffects model may perform better. Since we assume that there are countryspecific characteristics in this model that can be correlated with the explanatory variables, such as culture, geographic location, language, among others, we consider that a Fixed Effects approach is more adequate than a Random Effects approach. Among the main weaknesses of using panel data are estimation and inference problems, such as heteroskedasticity and autocorrelation, since it involves both crosssection and time series data. In order to correct for heteroskedasticity problems we use a fixed effects approach with corrected standard errors. The estimation results are presented in section Estimation Results Annual Panel Data Table 3 shows the regression results based on the estimation of the annual panel data specification for the period The econometric software adopted is STATA, version 12. Five model specifications are estimated: Model 1 is the full version of the model presented in section 3.2 and, as described before, Models 2, 3, and 4 are constrained versions of the same model, in which external, economic and banking structure factors are excluded, respectively. F test statistics for the statistical significance of omiting each block of variables are also presented. Model 5 only considers institutional factors. The F Test shows that the null hypothesis is rejected for every group of variables, meaning that the collective inclusion of each group of variables is statistically relevant to explain inflation. The most striking result is that the variable separate, in which impact we focus our analysis, is not statistically significant, in any of the models considered. In particular, even when only institutional variables are accounted for (Model 5), separate is still not significant and the coefficient sign is contrary to our expectations. Therefore, our results do not provide empirical evidence for the conflict of interest argument. In what regards the inflation targeting variable, the previous empirical literature does not consider it as an explanatory variable. However, our findings show that this variable is relevant in determining inflation outcomes. In fact, inflation targeting is statistically significant in every model specification and it has a negative impact on inflation rates, as predicted by the literature. The estimated impact on inflation rates of using inflation targeting as a monetary policy approach ranges from a minimum of 24% (Model 4) to a maximum of 34% (Model 14 We also account for the interaction of some regressors with separate, such as exchange rate regimes and bank concentration index. The estimation results do not change substantively and the majority of the interactions considered are not statistically significant. 15 In terms of estimation methods, Copelovitch and Singer (2008) adopt pooled Ordinary Least Squares with corrected standard errors. This methodology is not the most appropriate to estimate panel data, because it does not take into consideration unobserved effects. 9

10 5), 16 showing not only a large effect, but also its robustness since it does not vary considerably across different model specifications. Moreover, the estimation results reveal that deposit insurance, bank concentration index, trade and capital account openness are statistically significant variables. As for deposit insurance, its estimated impact on inflation rates is negative as expected and statistically significant across all the regressions. The coefficients magnitudes of deposit insurance are lower than inflation targeting, varying from a minimum of 14% to a maximum of 25%. This means that a country with an explicit deposit insurance scheme will have, all else equal, inflation rates that are at least 14% lower than a country without deposit protection. The variable bank concentration index enters the regressions with a negative sign, suggesting that an increase in the degree of concentration in the banking sector has a negative impact on inflation rates. The variable is statistically significant in Model 1 and Model 3, but not in Model 2, where external factors are omitted, although the coefficient maintains the negative sign. The magnitude of the coefficient in Model 1 is identical to Model 3 and can be interpreted as follows: a one percentage point increase in the bank concentration ratio is associated with a 30% decrease in inflation rates holding all other independent variables constant. As touched on in Section 3.1, the alternative explanation focusing on costefficiency offers a better explanation to our results, since it suggests that a higher concentration in the banking system does not seem to affect the transmission mechanism of monetary policy and, as a result, inflation rates can be lower on average when concentration is higher. Finally, our results suggest that a more open economy in terms of trade and capital flows seems to benefit from lower inflation rates, all else equal. These results are consistent with previous empirical evidence (Romer, 1993; Gupta, 2008) and partly consistent with Copelovitch and Singer (2008), since their results also suggest a negative significant effect of capital account openness on inflation rates. The remaining explanatory variables, such as central bank independence, exchange rate regime, domestic credit as a percentage of GDP, oil imports over GDP and the log of GDP and GDP per capita, are not significant determinants of inflation behaviour in industrialised countries. In particular, it is worth to discuss in more detail the results for central bank independence. There is a vast literature investigating the impact of central bank independence on inflation rates and stating that central bank independence is the main institutional factor affecting inflation outcomes. In our regression models, although central bank independence enters the regression with the expected negative sign, it seems to be insignificant whenever banking structure variables are accounted for. Only when we are not controlling for the structure of the banking system, as in Model 2 and Model 5, central bank independence has a negative and statistically significant effect on inflation rates. The CBI variable is based on the work of Arnone et al. (2007), which update a de jure measure of independence following Cukierman et al. (1992), since it assesses regulations only. As suggested by Cukierman et al. (1992), de facto measures of central bank independence are also important to assess how regulations work in practice. Our results seems to suggest that the degree of legal independence per se is not sufficient to guarantee a significant negative impact on inflation outcomes. In addition, it is interesting to note that its impact is statistically significant only when exter 16 Since the dependent variable in our model is the log of inflation, the interpretation of the magnitude of these regression coefficients cannot be) done directly. Thus, for dummy variables, we have to convert the estimated values, using the formula exp( ˆβ 1. 10

11 Dependent Variable: inflation Constant Table 3: Estimation results for annual panel data Model 1 Model 2 Model 3 Model 4 Model (11.101) (14.639) (0.150) (7.135) (0.053) Separate BS (1 = Yes) (0.099) (0.112) (0.094) (0.074) (0.069) Inflation Targeting (1 = Yes) (0.077) (0.070) (0.064) (0.068) (0.055) CBI (index) (0.150) (0.117) (0.150) (0.140) (0.098) Deposit Insurance (1 = Yes) (0.067) (0.079) (0.063) (0.046) (0.046) Exchange Rate Reg (1 = fixed) (0.065) (0.084) (0.062) (0.044) (0.073) Euro Area Member (1 = Yes) (0.166) (0.138) (0.132) (0.096) (0.074) Great Moderation (omitted) (omitted) (omitted) (0.049) (0.049) Domestic Credit (% of GDP) (0.001) (0.001) (0.001) Bank Concentration Index (0.120) (0.131) (0.113) GDP (log) (0.841) (0.998) (0.532) GDP per capita (log) (1.175) (1.212) (0.716) Banking Crisis (1 = Yes) (0.060) (0.063) (0.030) Currency Crisis (1 = Yes) (0.085) (0.080) (0.036) Trade Openness (0.023) (0.021) (0.027) Capital Account Openness (0.002) (0.002) (0.001) Oil imports (% GDP) (1.600) (1.505) (1.323) Observations No Countries F Test (global significance) (15, 21) (12, 24) (11, 21) (14, 21) (7, 24) F test (exclusion) (3, 21) (4, 21) (2, 21) R squared (within) * p < =.10; ** p < =.05; *** p < =.01 Robust standard errors are in brackets. nal factors are not considered in the model. In summary, the lack of significance of central bank independence may be due to measurement issues (due to the fact that it does not assess how regulations are implemented) or even due to theoretical problems with the association between central bank independence and inflation. Our results also suggest that the relation between CBI and banking structure should be studied further. The Great Moderation variable is omitted in 3 out of 5 regressions due to collinearity prob 11

12 lems. For the remaining models, the variable is statistically significant and the estimated coefficient enters the regression with the expected sign. In synthesis, estimation results in Table 3 suggest that inflation rates in industrialised countries are mainly affected by institutional factors, such as inflation targeting and deposit insurance, but not by the separation of banking supervision powers from the central bank, even when we are not controlling for banking structure, economic size and wealth and external factors. Other factors, such as the degree of openness of the economy and the degree of bank concentration are also important to assess inflation behaviour. Given the estimation results so far, in particular the insignificant effect of a separate banking supervisor in contrast with the significant impact of inflation targeting in every alternative specifications of the regression models, they suggest that the interaction between the separate and inflation targeting should be studied more deeply. The decision of some countries towards the separation banking supervision powers from the central bank was coincident in time with the assignment to monetary policymakers of a more transparent mandate in which price stability was considered a primary goal. Therefore, we assess the interactions between the two variables and test the hypothesis that inflation is lower on average in developed countries due to inflation targeting mandates and not because of different institutions mandates for banking supervision. In order to test this hypothesis, we build three dummies. The first dummy, Sep_Inftarg1, takes value of 1 whenever a particular country has inflation targeting and a combined mandate in a certain period in time. The second dummy, Sep_Inftarg2, takes value of 1 whenever a country does not follow inflation targeting and has implemented a separate mandate. Finally, there is a third dummy, Sep_Inftarg3, that takes value 1 if a country targets inflation and has a separate mandate. The dummies compare to a baseline case, in which a country has a combined regime and no inflation targeting. For all dummy variables, we expect a negative impact on inflation outcomes. If this hypothesis is correct, we expect that Sep_Inftarg1 will be statistically significant, whereas Sep_Inftarg2 will not. Table 4 presents the estimation results for the interaction of separate central bank and inflation targeting. The estimation results show that the variable Sep_Inftarg1 is statistically significant in every model specification and its impact on inflation rates is negative, as expected. This suggests that a country with inflation targeting and a combined regime will have, on average, a lower inflation outcome than a country with no inflation targeting and no separation of functions. In turn, Sep_Inftarg2 is not statistically significant in any regression, which suggests that a central bank that an institutional mandate characterised by a central bank exclusively responsible for monetary policy and without an inflation target mandate has no significant impact on inflation rates. Lastly, the impact on inflation outcomes of Sep_Inftarg3 is significant and negative. These findings provide empirical support to our argument that the main institutional driver of low inflation outcomes in industrialised countries may be the fact that these countries implemented an inflation targeting approach. Therefore, there is additional evidence that the conflict of interest argument is not supported by our estimation results. The estimation results for the remaining explanatory variables are identical to the ones presented in Table 4. 12

13 Table 4: Estimation results for annual panel data Dependent Variable: Model 1 Model 2 Model 3 Model 4 Model 5 (log) inflation Constant (11.241) (14.672) (0.151) (7.040) (0.053) Sep_Inftarg (1: Separate = 0, Inf. Targeting = 1) (0.077) (0.073) (0.063) (0.076) (0.087) Sep_Inftarg (1: Separate = 1, Inf. Targeting = 0) (0.113) (0.127) (0.111) (0.078) (0.074) Sep_Inftarg (1: Separate = 1, Inf. Targeting = 1) (0.143) (0.129) (0.104) (0.097) (0.060) CBI (index) (0.146) (0.111) (0.143) (0.138) (0.095) Deposit Insurance (1 = Yes) (0.069) (0.081) (0.063) (0.046) (0.047) Exchange Rate Reg (1 = fixed) (0.068) (0.087) (0.067) (0.044) (0.073) Euro Area Member (1 = Yes) (0.165) (0.136) (0.131) (0.096) (0.071) Great Moderation (omitted) (omitted) (omitted) (0.048) (0.049) Domestic Credit (% of GDP) (0.771) (0.001) (0.000) Bank Concentration Index (0.122) (0.133) (0.113) GDP (log) (0.846) (1.001) (0.527) GDP per capita (log) (1.182) (1.218) (0.714) Banking Crisis (1 = Yes) (0.063) (0.065) (0.030) Currency Crisis (1 = Yes) (0.085) (0.081) (0.036) Trade Openness (0.024) (0.022) (0.028) Capital Account Openness (0.002) (0.002) (0.001) Oil imports (% GDP) (1.617) (1.525) (1.311) Observations No of Countries F Test (global significance) (16, 21) (13, 24) (12, 21) (15, 21) (8, 24) F test (exclusion) (3, 21) (4, 21) (2, 21) R squared (within) * p < =.10; ** p < =.05; *** p < =.01 Robust standard errors are in brackets Five Years Panel Data In Table 5, the estimation results for the fiveyear panel data specification are presented. Here we follow the data approach adopted by Copelovitch and Singer (2008). The 5 years average is calculated for the all pool of regressors considered in the model, including dummy 13

14 variables for which we adopt a different average criterion. The average of dummy variables is first computed and then it is converted into 0 whenever it is less than 0.5 and into 1 otherwise. Models 2, 3 and 4 consist of Model 1 (Full Model) excluding blocks of variables: external variables, economic variables and banking structure variables, respectively. A F Test was performed to test for the impact of the joint exclusion of each block. The F Test shows that null hypothesis that the coefficients for each group of variables are zero is rejected. Hence, the joint inclusion of each group of variables is statistically relevant to explain inflation. The findings using a 5 years panel do not differ substantively from the ones obtained by using annual data. The estimation results for Model 1 show that the variable separate has a negative impact on inflation rates but this effect is not statistically significant. Even when accounting solely for institutional factors, the effect of separate on inflation is still insignificant. The same result applies to other model specifications. In conclusion, the separation of banking supervision from the central bank does not affect significantly inflation rates according to our results. This outcome differs from Copelovitch and Singer (2008) and Di Noia and Di Giorgio (1999), although they adopt the same data specification. As before, inflation targeting has a statistically significant negative impact on inflation rates. This result is consistent with the different model specifications presented in Table 5. The magnitude of the estimated coefficient ranges from 35% (Model 4) to 52% (Model 2), meaning that an inflation targeter country will have, on average, 35% to 52% lower inflation outcomes each five years than a country that does not target inflation. Furthermore, these results show once again that a country that provides explicit deposit insurance schemes will have on average lower inflation outcomes. The institutional variables of central bank independence and exchange rate regime remain statistically insignificant. The variable that controls for the period of the Great Moderation is not omitted in this data specification and enters the regressions with the expected sign, though it is only statistically significant in Models 3, 4 and 5. In what concerns economic factors, currency crisis has a statistically significant impact on inflation behaviour in Models 1 and 2, whereas banking crisis is statistically significant in Models 2 and 4. Both regressors have a positive impact on inflation rates, although currency crisis effect is stronger than banking crisis. The variables log of GDP and GDP per capita are not relevant predictors of inflation. Moreover, external factors are important in the sense that trade openness has a statistically significant negative effect on inflation. The remaining external factors, such as oil imports as percentage of GDP and capital account openness, are not relevant in explaining inflation behaviour. Finally, bank concentration is an important determinant of inflation outcomes. The degree of concentration in the banking system has a negative and statistically impact on inflation. Conversely, the size of the banking system does not seem to be relevant to explain inflation and its impact is close to zero. As before, the interaction between a separate central bank and a inflation targeting central bank is assessed using a 5 years average panel data. Table 6 shows the estimation results. The estimation results suggest the variable Sep_Inftarg1 is statistically significant in every model specification, except when there is no control for banking structure variables (Model 4). The impact is negative, as predicted. On the contrary, Sep_Inftarg2, the variable that assesses the impact on inflation rates of having a separate regime and no inflation targeting, is insignificant in every model specification, except Model 1, in which we control for external, economic and banking structure factors. Sep_Inftarg3 has a statistically significant negative 14

15 Dependent Variable: (log) inflation Constant Table 5: Estimation results for fiveyear average panel data Model 1 Model 2 Model 3 Model 4 Model (10.277) (13.300) (0.526) (8.095) (0.157) Separate CB (1 = Yes) (0.227) (0.271) (0.258) (0.213) (0.123) Inflation Targeting (1 = Yes) (0.161) (0.115) (0.168) (0.187) (0.125) CBI (index) (0.522) (0.362) (0.576) (0.531) (0.276) Deposit Insurance (1 = Yes) (0.223) (0.294) (0.205) (0.189) (0.112) Exchange Rate Reg (1 = fixed) (0.145) (0.194) (0.229) (0.144) (0.203) Euro Area Member (1 = Yes) (0.202) (0.270) (0.190) (0.173) (0.136) Great Moderation (0.477) (0.665) (0.379) (0.161) (0.109) Domestic Credit (% of GDP) (0.001) (0.001) (0.001) Bank Concentration Index (0.454) (0.603) (0.634) GDP (log) (0.420) (0.497) (0.363) GDP per capita (log) (0.349) (0.182) (0.406) Banking Crisis (1 = Yes) (0.091) (0.112) (0.094) Currency Crisis (1 = Yes) (0.158) (0.146) (0.271) Trade Openness (0.113) (0.108) (0.114) Capital Account Openness (0.004) (0.005) (0.005) Oil imports (% GDP) (1.590) (1.657) (1.797) No of Observations No of Countries F Test , *** (global significance) (16, 21) (13, 24) (12, 21) (14, 21) (7, 24) F test (exclusion) (3.21) (4,21) (2,21) R squared (within) * p < =.10; ** p < =.05; *** p < =.01 Robust standard errors are in brackets. impact on inflation outcomes that is robust in different model specifications. Results are not as clear as in an annual panel data approach, but they still suggest that being an inflation targeter is more important to assure lower levels of inflation in industrialised countries than having an independent banking supervision mandate. 15

16 Table 6: Estimation results for fiveyear average panel data Dependent Variable: Model 1 Model 2 Model 3 Model 4 Model 5 (log) inflation Constant (9.850) (12.109) (0.500) (7.482) (0.122) Sep_Inftarg (1: Separate = 0, Inf. Targeting = 1) (0.277) (0.305) (0.351) (0.200) (0.239) Sep_Inftarg (1: Separate = 1, Inf. Targeting = 0) (0.250) (0.294) (0.235) (0.253) (0.145) Sep_Inftarg (1: Separate = 1, Inf. Targeting = 1) (0.277) (0.241) (0.229) (0.211) (0.164) CBI (index) (0.493) (0.331) (0.516) (0.482) (0.237) Deposit Insurance (1 = Yes) (0.192) (0.310) (0.214) (0.167) (0.104) Exchange Rate Reg (1 = fixed) (0.174) (0.260) (0.145) (0.154) (0.182) Euro Area Member (1 = Yes) (0.261) (0.267) (0.227) (0.208) (0.185) Great Moderation (0.278) (0.490) (0.339) (0.163) (0.107) Domestic Credit (% of GDP) (0.001) (0.002) (0.001) Bank Concentration Index (0.363) (0.541) (0.436) GDP (log) (0.391) (0.494) (0.300) GDP per capita (log) (0.310) (0.195) (0.234) Banking Crisis (1 = Yes) (0.116) (0.174) (0.118) Currency Crisis (1 = Yes) (0.287) (0.412) (0.346) Trade Openness (0.088) (0.063) (0.075) Capital Account Openness (0.004) (0.004) (0.003) Oil imports (% GDP) (1.630) (1.421) (1.890) No of Observations No of Countries F Test 313, (global significance) (17, 21) (14, 24) (13, 21) (15, 21) (8, 24) F test (exclusion) (3, 21) (4, 21) (2, 21) R squared (within) * p < =.10; ** p < =.05; *** p < =.01 Robust standard errors are in brackets. 3 Conclusions The estimation results show that the separation of banking supervision from the central bank does not have a significant impact on inflation. In this sense, the conflict of interest argument against combination of monetary policy and banking supervision is not supported by our em 16