doi: 10.1111/j.1467-6419.2009.00597.x INFLATION AND CENTRAL BANK INDEPENDENCE: A META-REGRESSION ANALYSIS Jeroen Klomp University of Groningen, The Netherlands Jakob de Haan University of Groningen, The Netherlands CESifo, Munich, Germany Abstract. Using 59 studies, we perform a meta-regression analysis of studies examining the relationship between inflation and central bank independence (CBI). The studies considered are very different with respect to the CBI indicator used, the sample of countries and time periods covered, model specification, estimators used and publication outlet. We conclude that there is a significant publication bias. However, we also find a significant genuine effect of CBI on inflation. Differences between studies are not caused by differences in CBI indicators used. Keywords. Central bank independence; Inflation; Meta-analysis No wonder politicians often find the Fed a hindrance. Their better selves may want to focus on America’s long-term prosperity, but they are far more subject to constituents’ immediate demands. That’s inevitably reflected in their economic policy preferences. If the economy is expanding, they want it to expand faster; if they see an interest rate, they want it to be lower. (Greenspan, 2007, pp. 110–111) 1. Introduction During the last two decades, many countries granted their monetary authorities greater independence. It is widely believed that central banks otherwise will give in to pressure from politicians who may be motivated by short-run electoral considerations or may value short-run economic expansions highly while discounting the longer-run inflationary consequences of expansionary policies (Walsh, 2005). 1 If the ability of politicians to distort monetary policy results in excessive inflation, countries with an independent central bank should experience lower rates of inflation. Indeed, beginning with Bade and Parkin (1988), an important line of empirical research focusing on the relationship between central bank independence (CBI) and inflation suggests that average inflation is negatively Journal of Economic Surveys (2010) Vol. 24, No. 4, pp. 593–621 C 2009 Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
doi: 10.1111/j.1467-6419.2009.00597.x
INFLATION AND CENTRAL BANKINDEPENDENCE: A META-REGRESSION
ANALYSISJeroen Klomp
University of Groningen, The Netherlands
Jakob de Haan
University of Groningen, The NetherlandsCESifo, Munich, Germany
Abstract. Using 59 studies, we perform a meta-regression analysis of studiesexamining the relationship between inflation and central bank independence(CBI). The studies considered are very different with respect to the CBI indicatorused, the sample of countries and time periods covered, model specification,estimators used and publication outlet. We conclude that there is a significantpublication bias. However, we also find a significant genuine effect of CBIon inflation. Differences between studies are not caused by differences in CBIindicators used.
Keywords. Central bank independence; Inflation; Meta-analysis
No wonder politicians often find the Fed a hindrance. Their better selvesmay want to focus on America’s long-term prosperity, but they are far moresubject to constituents’ immediate demands. That’s inevitably reflected in theireconomic policy preferences. If the economy is expanding, they want it to expandfaster; if they see an interest rate, they want it to be lower. (Greenspan, 2007,pp. 110–111)
1. Introduction
During the last two decades, many countries granted their monetary authoritiesgreater independence. It is widely believed that central banks otherwise willgive in to pressure from politicians who may be motivated by short-runelectoral considerations or may value short-run economic expansions highly whilediscounting the longer-run inflationary consequences of expansionary policies(Walsh, 2005).1 If the ability of politicians to distort monetary policy results inexcessive inflation, countries with an independent central bank should experiencelower rates of inflation. Indeed, beginning with Bade and Parkin (1988), animportant line of empirical research focusing on the relationship between centralbank independence (CBI) and inflation suggests that average inflation is negatively
related to measures of CBI (see Eijffinger and De Haan, 1996a; Berger et al.,2001; Crowe and Meade, 2007, for summaries). However, this evidence has beencriticized by various authors, claiming that the results are sensitive with respect tothe measure of CBI used (see, for instance, Forder, 1996), the specification of themodel (see, for instance, Posen, 1995; Campillo and Miron, 1997) or the inclusionof high-inflation observations (see, for instance, De Haan and Kooi, 2000).
Using meta-regression analysis (MRA), this paper addresses two issues. (1) Towhat extent has the literature confirmed that there is a negative association betweenCBI and inflation? (2) Can we explain the pattern in the results of empirical researchon the relationship between CBI and inflation? Using 59 studies we find that, onaverage, there exists a significant relation between CBI and inflation. We also findthat the results reported in the studies in our sample suffer from a publication bias.Studies report the strongest relationship between CBI and inflation if they focuson OECD countries (especially when studies control for outliers) and include the1970s. Furthermore, we find that when a bivariate regression is used or if the modelincludes the labour market the significance of the CBI indicator increases. We donot find significant differences between studies based on a cross-country settingand those that use panel models. Differences between studies are also not causedby differences in CBI indicators used.
The remainder of the paper is organized as follows. Section 2 reviews the mainissues in the empirical research on the relationship between CBI and inflation.Section 3 outlines the methodology of the MRA and the studies used in ouranalysis, while Section 4 contains the MRA. Section 5 offers our conclusions.
2. Measuring Central Bank Independence
To examine whether there is any relationship between CBI and inflation, one needsan indicator of the extent to which the monetary authorities are independent frompoliticians. Most empirical studies use either an indicator based on central banklaws in place, or the so-called turnover rate of central bank governors (TOR).
The most widely employed legal index of CBI is from Cukierman (1992) andCukierman et al. (1992),2 although alternative measures have been developed byAlesina (1988) and Grilli et al. (1991) among others (see Arnone et al. (2006) for anextensive comparison of the various CBI indicators). Even though these indicatorsare supposed to measure the same phenomenon and are all based on interpretationsof the central bank laws in place, their correlations are sometimes remarkably low(Eijffinger and De Haan, 1996a).
Legal measures of CBI may not reflect the true relationship between the centralbank and the government. Especially in countries where the rule of law is lessstrongly embedded in the political culture, there can be wide gaps between theformal, legal institutional arrangements and their practical impact (Walsh, 2005).This is particularly likely to be the case in many developing economies. Cukierman(1992) argues that the actual average term in office of the central bank governormay therefore be a better proxy for CBI for these countries than measures based oncentral bank laws. The TOR is based on the presumption that, at least above somethreshold, a higher turnover of central bank governors indicates a lower level ofindependence.3 According to Cukierman’s data, TOR values range from a minimumof 0.03 (which corresponds to an average term in office for the governor of some33 years) to a maximum of 0.93 (which corresponds to an average term in officeof just 13 months). Cukierman’s data suggest that TORs in developing countriescover a much broader range of values than in OECD countries, where values areall below 0.20 turnovers per year.4
The next step is to employ these indicators in a particular model forinflation and estimate it for a specific group of countries and a sample period.Initially, the research focused on industrial countries using legal CBI indicators.Most of the older studies, which generally used simple cross-country bivariateregressions for particular periods, reported that CBI was negatively correlated withaverage inflation (see, for instance, Alesina and Summers, 1993). The estimatedeffect of independence on inflation turned out to be significant – in both astatistical and economic sense – especially during periods with flexible exchangerates.
While researchers found that legal CBI indicators were negatively associatedwith inflation among industrial countries, this was not the case for developingcountries. However, initial findings suggested that in these economies the TOR ofcentral bank governors is positively correlated with inflation, therefore also lendingsupport to the hypothesis that CBI and inflation are negatively related. Countriesthat experienced rapid turnover among their central bank heads (i.e. countries witha low level of CBI) also tended to experience high rates of inflation (see, forinstance, Cukierman, 1992). This is a case, however, in which causality is difficultto evaluate: Is inflation high because of political interference that leads to rapidturnover of central bank officials? Or are central bank officials tossed out becausethey cannot keep inflation down? (Walsh, 2005).
Second, studies on the relationship between CBI and inflation often fail tocontrol adequately for other factors that might account for cross-country differencesin inflation. Countries with independent central banks may differ in ways thatare systematically related to average inflation. A good example of this line ofcritique is the work by Posen (1993, 1995) who argues that both low inflationand CBI reflect the presence of a strong financial sector constituency for lowinflation. Average inflation and the degree of CBI are jointly determined bythe strength of political constituencies opposed to inflation. Posen argues thatonce these constituencies are taken into account, the coefficient of CBI is nolonger significant in models explaining cross-country inflation differentials.5 AlsoCampillo and Miron (1997) claim little role for legal CBI when control variablesrelating to the degree of openness, political instability and a country’s inflation anddebt history are introduced. However, this result has been criticized as Campillo andMiron’s sample includes many developing countries for which legal CBI indicatorsmay not be appropriate. Sturm and De Haan (2001) use TORs in a similar modelas Campillo and Miron and conclude that the coefficient of this CBI indicator issignificant in a multivariate model.6
A recent strand of literature argues that the effects of CBI should not be analysedindependently of labour market institutions. Trade unions may, for instance, beinflation averse. The reason usually given is consistency: unions encompass mostof society, which in its majority is inflation averse, at least according to the standardRogoff (1985) model of monetary policy. Inflation-averse unions will make realvariables in equilibrium a function of the institutional set-up like the degree ofcentral bank conservatism given a certain degree of CBI. The more conservativethe central bank, the lower output will be and the higher the level of unemploymentin equilibrium. In that sense, monetary policy has real effects in these models. Alsothe effects of CBI on inflation will be different in this setting compared to thestandard Rogoff-type of model (see Berger et al. (2001) for a further discussion).A good example of this line of research is the study of Cukierman and Lippi(1999). Using data for 19 OECD economies for the period 1980–1994, they findthat the inflation reducing impact of CBI is stronger at intermediate levels of unioncentralization.
Finally, a few studies have sounded a warning that conclusions on the relationshipbetween CBI and inflation are highly sensitive to influential observations. Forinstance, Temple (1998) finds that if high-inflation countries are added to his sampleof OECD and developing countries, the effect of CBI (proxied by Cukierman’s(1992) legal index) on inflation disappears, while De Haan and Kooi (2000) andSturm and De Haan (2001) report that the TOR indicator only becomes significantif high-inflation countries are included in the sample.
MRA not only recognizes the specification problem but also attempts to estimateits effects by modelling variations in selected econometric specifications. MRAprovides us with the means to analyze, estimate, and discount, when appropriate,the influence of alternative model specification and specification searches. In thisway, we can more accurately estimate the empirical magnitudes of the underlyingeconomic phenomena and enhance our understanding of why they vary acrossthe published literature.
The issue of CBI lends itself perfectly for such an analysis. However, to the bestof our knowledge, such an analysis has not been done so far. We have gathered 59studies that come up with empirical estimates of the effect of CBI on inflation ina cross-country and/or panel setting, using some proxy for CBI. That means thatcountry-specific studies are excluded from the analysis. We started our search forstudies with the surveys of Eijffinger and De Haan (1996a) and the update thereofin Berger et al. (2001). To find more recent (published and unpublished) studieswe used Google and JSTOR. Table A1 in the Appendix contains all the studies weidentified. We stopped searching at 31 December 2006. Table A1 also shows foreach of these studies the percentage of regressions in which there is a significantnegative relationship between inflation and CBI. We have coded all studies includedin our analysis independently; whenever we coded studies differently initially, thesedifferences were discussed until we both agreed about the proper coding.
The average sample size is about 91 observations, while on average about 28countries are included. Most studies examine the effect of CBI on inflation byestimating (variants of) a single-equation model without extensively testing for therobustness of the results. Although most studies report a negative relation betweenCBI and inflation, various papers find a positive or no effect of CBI on inflation.
Drawing on Stanley and Jarrell (1989), we can explain our MRA as follows.Most studies on CBI and inflation involve a standard regression model such as
π = Xβ + ε (1)
where π is the (n × 1) dependent variable vector, i.e. some measure of inflation,X is an (n × m) matrix of explanatory variables, including an indicator of CBI,8
and ε denotes some random error, which is typically assumed to conform to theclassical regression model. As we are primarily interested in the relevance of CBIin explaining inflation, we focus on the estimated t-statistic of the coefficient ofthe CBI indicator. This also forgoes the problem that the coefficients of the variousindicators are not comparable, as their scaling differs. If the TOR is used as a CBIindicator, we multiply the reported t-statistic by −1 so that it becomes comparablewith studies using legal CBI indicators.
Table 1 shows the distribution of the t-statistics across time period, countrysample, and indicator used. The average t-statistic of the CBI indicator of allregressions in our sample is −1.78 if we take all estimates independently. Whenwe account for study differences, the average t-statistic increases to −1.85. Bothaverages indicate that the relation between inflation and CBI is significantlynegative at the 5% significance level.
We can draw some stylized facts from Table 1. First, in OECD countries theaverage t-statistic of the CBI and inflation relation is lower (i.e. more significant)compared to developing and transition countries. In most cases, if we do notdifferentiate between CBI indicators used, these results suggest that the relationbetween inflation and CBI is significant in OECD countries and insignificant indeveloping countries.
Second, in the period 1970–1979 the t-statistic of the CBI indicator becomesmost negative (i.e. significant). This is probably due to the breakup of the BrettonWoods system in 1973. Under the Bretton Woods system of fixed exchange rates,monetary policy in most countries was determined by the fixed exchange rate target.The t-values for most country groups also decline overtime. This is probably dueto the fact that central bank laws have converged over time and have thereforebecome less capable of explaining inflation differentials.9
but also on the indicator used. The significance levels differ much across indicators,when we hold the time period and country group fixed.
However, before we come to any conclusion with regard to the existence of anegative relation between CBI and inflation, we first have to analyse whether thereis a so-called ‘publication bias’ (i.e. journals only publish papers with particularresults).
4. Meta-regression Analysis: Approach
The key research issues are whether there is a publication bias in research on thelink between CBI and inflation, and whether a meaningful CBI effect remains aftera publication bias is filtered out. Drawing heavily on Doucouliagos and Stanley(2009), we can explain a typical meta-regression model as follows:
effecti = β1 + β0SEi +K∑
k=1
αk Z jk + ei (2)
where effecti is the focus of the analysis (in our case, the effect of CBI on inflation),SEi is the standard error of the estimated effect, Zjk is a vector of meta-independentvariables reflecting differences across studies, αk is the meta-regression coefficientwhich reflects the effect of particular study characteristics and ei denotes the meta-regression disturbance term. Without publication bias, the observed effects shouldvary randomly around the ‘true’ value, β1, independently of the standard error. Theterm β0SEi allows for the very common tendency of researchers and reviewers toprefer statistically significant results and for researchers therefore to rerun theiranalysis until they find such significance (Doucouliagos and Stanley, 2009). This isespecially the case for studies with only a small number of observations. To reporta significant relationship, these studies have to find a sufficiently large estimatedeffect, which compensates for the large standard errors associated with the smallnumber of observations. If the number of observations increases indefinitely, thestandard error will approach zero and the reported effects will approach β1, the‘true’ effect (Stanley, 2008; Doucouliagos and Stanley, 2009).
Studies that try to explain the same relationship usually use different sample sizesand model specifications. Hence, the random estimation errors ei in equation (2)are likely to be heteroscedastic. As suggested by Doucouliagos and Stanley (2009),dividing equation (2) by SEi, i.e. a sample estimate of the standard deviation ofthese meta-regression errors, gives the weighted least squares version of equation(2):
ti = β0 + β1
(1
SEi
)+
K∑k=1
αkZ jk
SEi+ ei (3)
where ti represent the reported t-values. The conventional t-test of the intercept ofequation (3), β0, is a test for publication bias.
As follows from Section 2, the variation among the empirical results may beexplained by various study characteristics or model specifications, reflected in Zjk.
The types of design elements that we include in Zjk are as follows:
1. the CBI indicator, sample of countries, and time period used, i.e. aredifferences in results related to the indicators and samples used?
2. the specification of the regression model, i.e. does the inclusion of controlvariables have an effect on the reported significance of the CBI indicator, andif so, which control variables matter?
3. characteristics of the publication, i.e. does the study focus on the relationshipbetween CBI and inflation? Does the publication form (journal, book orworking paper), outlet (does the journal in which the study is published havea social science citation impact (SSCI) score?) or publication year have anyrelationship with the reported results?
4. the estimation method, i.e. is there any systematic difference between cross-country versus panel studies and does it make a difference if a study controlsfor outliers and/or high-inflation observations?
In our analysis, the unit of observation is not a study, but every regression reported.Many studies contain more than one regression, for instance, when they test forthe sensitivity of the choice of a particular CBI indicator. Since the observationsare not independent, ordinary least squares would lead to biased estimates. Thisis corrected by using a hierarchical linear model, which is a particular regressiontechnique that is designed to take into account the hierarchical structure of the data(Raudenbusch and Bryk, 1986). Equation (3) can be rewritten as
ti j = β00 + β1 j
(1
SEi j
)+
K∑k=1
αkZi jk
SEi j+
K∑k=1
γkV0 jk
SEi j+ ωi + u j (4)
The meta-independent variable is split up in two parts. One part explains thedifferences between studies and estimates Zijk, while the other part explains studydifferences V0j. The ωi and uj are the error terms on estimate and study level,respectively.10
5. Meta-regression Analysis: Results
Table 2 gives our first estimation results. About 60% of the total variance iscontributed to the variance on study level. This implies that there is dependencewithin a study and that a multilevel model is the appropriate model to use. Column 1of Table 2 shows the estimation results of the so-called funnel graph asymmetrytest (Doucouliagos and Stanley, 2009). The parameter of the inverse standard errorsis significant, which indicates that the effect of CBI on inflation is significantlynegative. However, the constant term is also significant at a 5% level, meaningthat the effect found in the CBI–inflation literature is subject to a publicationbias.
Table 2. MRA Tests for Publication Bias and Genuine Empirical Effect.
t-statistic CBI coefficient
Coefficient z-value
Fixed parametersConstant −1.651∗∗ −5.67Inverse standard errors −0.073∗∗ −2.02
Random parametersVariance estimate level p-value 0.000Variance study level p-value 0.000Intra-class correlation 0.583
Diagnostic statisticsNumber of observations 356Number of studies 58Maximum likelihood ratio p-value 0.000
∗∗, ∗Indicates significance at 5% and 10% level, respectively.
Next, we include variables to control for different specifications used in thevarious studies examined in the MRA. As in any regression model, the estimatedcoefficients in the MRA model can be biased when important explanatory variablesare omitted. Table 3 presents the definition of the control variables used in theMRA. The first set of variables refers to the CBI indicator used in the regression(ALES, GMT, CUK, BP, TOR, OTHER). The next variables focus on countrysample (OECD, LDCs, TRANS, MIXED) and time periods (1960, 1970, 1980,1990).
In bivariate regressions of inflation and CBI, the impact of omitted variableson inflation is attributed to the CBI indicator. Multivariate studies will thereforeprobably report lower absolute t-statistics of the CBI indicator (BIVARIATE). Inorder to examine which control variables reduce the impact of CBI on inflation,we have constructed dummy variables for a number of commonly used controlvariables.
According to Romer (1993), inflation depends on the openness of an economy.Since the real effects of monetary policy are lower in more open economies,governments in these countries have fewer incentives to inflate. We thereforeconstruct a dummy variable reflecting whether a regression takes this controlvariable into account (OPEN).
ALES A dummy variable equal to 1 if the CBI indicator of Alesina is used,0 otherwise
GMT A dummy variable equal to 1 if the CBI indicator of Grilli et al.(1991) is used, 0 otherwise
CUK A dummy variable equal to 1 if the CBI indicator of Cukierman isused, 0 otherwise
BP A dummy variable equal to 1 if the CBI indicator of Bade–Parkin isused, 0 otherwise
TOR A dummy variable equal to 1 if the TOR indicator is used, 0 otherwiseOTHER A dummy variable equal to 1 if another CBI indicator is used, 0
otherwise
OECD A dummy variable equal to 1 if the analysed countries are all OECDcountries, 0 otherwise
LDCs A dummy variable equal to 1 if the analysed countries are alldeveloping countries, 0 otherwise
TRANS A dummy variable equal to 1 if the analysed countries are alltransition countries, 0 otherwise
MIXED A dummy variable equal to 1 if the analysed countries are mixed, 0otherwise
1960 A dummy variable equal to 1 if data refer to the 1960s, 0 otherwise1970 A dummy variable equal to 1 if data refer to the 1970s, 0 otherwise1980 A dummy variable equal to 1 if data refer to the 1980s, 0 otherwise1990 A dummy variable equal to 1 if data refer to the 1990s, 0 otherwise
BIVARIATE A dummy variable equal to 1 if the inflation and CBI relation isexamined using bivariate regression, 0 otherwise
OPEN A dummy variable equal to 1 if openness of a country is taken intoaccount, 0 otherwise
LABMARKT A dummy variable equal to 1 if some labour market variable is takeninto account, 0 otherwise
ILABMARKT A dummy variable equal to 1 if an interaction of the CBI indicatorwith the labour market is taken into account, 0 otherwise
EXCHANGE A dummy variable equal to 1 if the exchange rate regime is taken intoaccount, 0 otherwise
DEBT A dummy variable equal to 1 if government debt is taken intoaccount, 0 otherwise
POLSTAB A dummy variable equal to 1 if political stability is taken intoaccount, 0 otherwise
GDP A dummy variable equal to 1 if income is taken into account, 0otherwise
INTER A dummy variable equal to 1 if an interaction of the CBI indicatorwith other variables is taken into account, 0 otherwise
LOGINFL A dummy variable equal to 1 if the log of inflation is used as thedependent variable, 0 otherwise
OUTLIER A dummy variable equal to 1 if the author controls for outliers, 0otherwise
NUMOBS Number of observationsPRIMDATA A dummy variable equal to 1 if the author creates his own CBI data,
0 otherwiseSECDATA A dummy variable equal to 1 if the author modified existing CBI data
of others, 0 otherwisePANEL A dummy variable equal to 1 if the author uses panel data, 0
otherwiseFIXEDTIME A dummy variable equal to 1 if the author uses panel data with fixed
time effects, 0 otherwise (if panel data are used)
FIXEDCOUNT A dummy variable equal to 1 if the author uses panel data with fixedcountry effects, 0 otherwise (if panel data are used)
OBJECT A dummy variable equal to 1 if the inflation and CBI regression ofthe study focuses on this issue, 0 otherwise
BOOK A dummy variable equal to 1 if the study is published in a book, 0otherwise
WORKING A dummy variable equal to 1 if the study is a working paper, 0otherwise
PUBYEAR Publication year (1991 = 1, . . . , 2006 = 15)IMPACT SSCI score of a journal
of labour market institutions and CBI affects both the real and nominal effects ofmonetary policymaking and that is why we include a dummy that is one in casethis interaction is included and zero otherwise (ILABMARKT).
Other control variables that various studies have included – generally followingCampillo and Miron (1997) – are the exchange rate regime (EXCHANGE),government debt (DEBT), political instability (POLSTAB) and income (GDP).Stable exchange rate regimes are often argued to reduce inflation; a fixed exchangerate can be considered as an alternative commitment device to counter theinflationary bias of monetary policymaking. A high debt-to-GDP ratio and a highlevel of political instability are determinants of the inflation bias and are thereforeoften argued to lead to higher inflation, while income is often reported to have anegative impact on inflation. We include dummies in our MRA reflecting whetherthese control variables are taken up in regressions. Finally, we take up a dummythat is one if a regression includes an interaction (INTER) of the CBI indicator anda control variable other than the labour market variable, and zero otherwise.
these high-inflation observations to become very influential. Similarly, correctingfor outliers may affect the significance of the CBI indicator. However, the effect ofcorrecting for outliers may differ across country groups. Temple (1998) finds that inhis sample of OECD countries the CBI indicator of Cukierman et al. (1992) is onlysignificant if high-inflation countries are dropped, while De Haan and Kooi (2000)and Sturm and De Haan (2001) report for their sample of developing countriesthat the inclusion of high-inflation countries renders the coefficient of the TORindicator of CBI significant. We therefore include the interaction of our outlierdummy and our dummies for country groupings.
The next variable we include in our MRA is the number of observationsas more observations are expected to lead to higher significance levels of theCBI indicator (NUMOBS). We also test whether the t-statistic of the CBIindicator is different if the author uses his own CBI measure (PRIMDATA) ormodifies an existing index as there may be bias if an author uses his own CBIindicator (SECDATA). We also distinguish between various estimation methods,differentiating between cross-country and panel models (PANEL). In the lattercategory, we have dummies reflecting whether the author controls for time orcountry fixed effects (FIXEDTIME, FIXEDCOUNT).
Finally, we control for publication and study differences. First, we ask whetherthere are any differences between studies that only estimate the relationship betweenCBI and inflation and those that have a broader perspective. We use dummiesreflecting that a study is published in a book (BOOK) or as a working paper(WORKING) (at the time we did this research) instead of in a journal, respectively.We also test for the effect of the SSCI score of journals as the citation impact of ajournal is often considered as a quality indicator (IMPACT). The final variable weinclude in the MRA is the publication year (PUBYEAR), allowing us to analysedifferences over time in reported t-statistics.
We showed in Table 1 that the significance of the t-value of the CBI coefficientvaries across time and place; therefore in the first regression of Table 4 we controlfor time periods and country sample. All control variables are divided by thestandard error of the CBI coefficient as given in equation (4).
When we include multiple variables, the inverse of the standard error no longerrepresents the genuine effect of CBI on inflation. Rather, it is the combinationof all the coefficients on the variables that reflect the corrected effect of CBI oninflation. We find that the CBI coefficient is insignificant in the 1960s, 1980sand 1990s, while it is significant in the 1970s. We confirm the hypothesis thatthe t-statistic of the CBI coefficient is significantly negative in studies includingonly OECD countries. In the next column we control for the CBI indicatorused. We do not find any significant difference between the results of studiesthat are caused by differences in the indicator used. So even though the variousindicators are constructed in a different way, the significance of the relationshipbetween CBI and inflation is not dependent on the selection of a particular CBIindicator.
and an interaction term between the labour market indicator and the CBI indicatorinfluences the t-value of the CBI coefficient and makes the relationship betweeninflation and CBI (more) significant. We do not find that any other variable thatis suggested by Campillo and Miron (1997) influences the significance of theCBI coefficient. This finding therefore does not support Campillo and Miron’sconclusion that the omission of control variables in earlier studies is behind thefact that these older studies found a significant relationship between CBI andinflation.
In Table 5 we add variables to control for the method of estimation and dataissues. Correcting for outliers by deleting countries or time periods from the samplehas only a significant effect in OECD countries, meaning that correcting for outliersin OECD countries makes the relation between CBI and inflation more significant.That the sign of the interaction of outlier correction and country group differsbetween OECD and less developed countries is due to the different impact ofoutliers in these country groups mentioned earlier. The use of the logarithm ofinflation instead of actual inflation as dependent variable has no effect on thesignificance of the CBI coefficient. Not surprisingly, studies that estimate therelation between inflation and CBI using a bivariate regression report a higher levelof significance of the CBI coefficient than studies that take control variables intoaccount. This suggests that bivariate regressions have an omitted variable bias.12
Next we check whether there exists a bias in studies using data that have beenconstructed or collected by the author of the study, or in studies in which the authorhas modified existing data. The estimation results do not support this hypothesis.Also there is no significant difference when panel estimation (with period or countryfixed effects) is used instead of a cross-country estimation.
The final two columns in Table 5 show the results for various publication effects.There is no systematic difference between studies that focus on the relationshipbetween CBI and inflation and those that do not. Likewise, the publication outletdoes not influence differences across studies in a systematic way. There is also nodifference between papers published in journals with different SSCI scores.
Finally, to test the joint significance of the regressors, we performed alikelihood ratio test of a full model, which contains all independent variablesused (except for the fixed effects indicator and the impact factor score becausethese reduce our sample drastically) against a baseline model with only thesignificant variables (i.e. OECD, OECD∗OUTLIER, 1970, LABORMARKT,ILABORMARKT, BIVARIATE). The results indicate that the full model does notperform better than the model that only includes significant variables (p > 0.10).Also we tested the model that only includes significant variables against a modelwith only a constant and the inverse standard errors included. The results show thatthe model with the significant variables included outperforms the model with onlythe constant and the inverse standard error (p < 0.05).
Random parametersVariance estimate level p-value 0.000Variance study level p-value 0.000Intra-class correlation 0.512
Diagnostic statisticsNumber of observations 363Number of studies 55Maximum likelihood ratio p-value 0.000
∗∗, ∗Indicates significance at, respectively, 5% and 10% level. All control variables are divided bythe standard errors.
for the total sample is similar to those for only a part of the sample (results areavailable on request).
Finally, we performed a general-to-specific approach on the variables includedin this study. Stepwise we deleted the variable with the highest p-value, untilall variables were significant at a 10% significance level. The results as shown inTable 6 confirm our previous findings. Together, the variables included have a strongeffect, as evidenced by the p-value of the likelihood ratio. So there is a genuineeffect of CBI on inflation. The results in Table 6 offer a clear interpretation of thepresence of this result. There is a negative significant effect of CBI on inflationin OECD countries. This effect is even stronger if the researcher corrects thesample for outliers and includes a labour market indicator and the interaction ofthe labour market indicator and the CBI indicator. Inclusion of the 1970s in thesample strengthens this negative CBI effect further.
Arnone et al. (2006) and Crow and Meade (2007). However, to the best of ourknowledge, this paper is the first to apply MRA on the vast amount of empiricalstudies examining the impact of CBI on inflation.13 MRA is an effective means toanalyse the influence of, among others, alternative indicators, model specificationand sample selection.
It is widely believed that countries with a more independent central bank will,on average, have lower levels of inflation. Our MRA corroborates the conventionalview by finding a significant ‘true effect’ of CBI on inflation, once we controlfor a significant publication bias. The effect is strongest when a study focuses onOECD countries, the period 1970–1979, considers the labour market, and when therelation is estimated using a bivariate regression. We also find that the literatureon CBI and inflation suffers from a publication bias, i.e. the reported results aresubject to a selection effect. We do not find any significant difference betweenthe results of studies that are caused by differences in the indicator used. Soalthough the CBI indicators are constructed in a different way, the relationshipbetween CBI and inflation is not dependent on the selection of the CBI indicator.Furthermore, we conclude that there is no significant difference between studiesusing regressions in a cross-country setting and those using panel estimation withfixed time and/or country effects. Also there are no significant differences betweenpublications in scientific journals, chapters in books or working papers. Focusingon journal articles, there is no significant difference between high- and low-rankedjournals in terms of their SSCI score.
Acknowledgements
We thank participants at the conference ‘Does Central Bank Independence Still Matter?’(14–15 September 2007) at Bocconi University (Milan, Italy) and the Aarhus Colloquiumfor Meta-Analysis in Economics (27–30 September 2007, Sønderborg, Denmark) and twoanonymous referees for their comments on a previous version of the paper.
Notes
1. One theory underlying this view is the time inconsistency approach to monetarypolicymaking. The basic message of this theory is that government suffers from aninflationary bias and that, as a result, inflation is sub-optimal. Rogoff (1985) hasshown that when monetary policy is delegated to an independent and ‘conservative’central banker, this inflationary bias will be reduced. Conservative means that thecentral banker is more averse to inflation than the government, in the sense that(s)he places a greater weight on price stability than the government does.
2. The only difference between the indicators of Cukierman (1992) and Cukiermanet al. (1992) is the procedure employed to aggregate the various dimensions of CBIinto one measure.
3. Still, this indicator is less than perfect, as it suffers from the limitation that centralbank governors can hold office for quite some time simply by being subservient topolitical leaders (Brumm, 2000).
4. Dreher et al. (2008) have extended the sample of countries and the time period forwhich TORs are available.
5. The empirical evidence that the financial sector is inherently inflation averse isnot compelling. Although Posen (1995) presents supportive evidence, other studiesfind less or no support (De Haan and van’t Hag, 1995; Campillo and Miron, 1997;Temple, 1998).
6. However, they also find that this result is driven by the inclusion of high-inflationcountries in the sample; excluding those countries makes the coefficients of theCBI indicator insignificant. Also De Haan and Kooi (2000) point to the role ofhigh-inflation observations.
7. Examples include Abreu et al. (2005), Doucouliagos (2005), Rose and Stanley(2005) and Nijkamp and Poot (2005).
8. In various studies, especially the older ones, X consists only of some CBI indicator.9. We thank Alex Cukierman for this observation.
10. All regressions have been estimated with STATA using the generalized linear latentand mixed models command with the Newton–Raphson algorithm.
11. Studies focusing on OECD countries often include a dummy for Iceland as thiscountry had a very high rate of inflation in the 1980s and 1990s, but also anindependent central bank.
12. However, as one referee pointed out, this finding could also reflect the fact thatbivariate regressions are older and hence focus on older time periods. In otherwords, there could be a multicollinearity problem between our ‘multivariate’ variableand the sample period dummies. To check for this we calculated the correlationcoefficients between the period dummies and the multivariate indicator. We do notfind any evidence that the period dummies are related to the bivariate regressionindicator. The correlations range between 0.05 and 0.11.
13. Although there is a possibility of reverse causality, most papers have not examinedthis issue. An exception is the study by Dreher et al. (2008) who find that thelikelihood that a central bank governor will be replaced increases with high pastinflation, suggesting that the TOR is indeed endogenous.
References
Abreu, M., de Groot, H.L.F. and Florax, R.J.G.M. (2005) A meta-analysis of betaconvergence: the legendary two-percent. Journal of Economic Surveys 19: 389–420.
Adolph, C. (2004) The uses of autonomy: central bankers’ careers, institutional contextand economic performance. Paper presented at the Annual Conference of the MidwestPolitical Science Association, Chicago, IL, 15–18 April.
Alesina, A. (1988) Macroeconomics and politics. NBER Macroeconomics Annual 13–52.Alesina, A. and Summers, L.H. (1993) Central bank independence and macroeconomic
performance: some comparative evidence. Journal of Money, Credit, and Banking25(2): 151–162.
Al-Marhubi, F. and Willett, T.D. (1995) The anti-inflationary influence of corporatiststructures and central bank independence: the importance of the hump shapedhypothesis. Public Choice 84: 153–162.
Arnone, M., Laurens, B.J. and Segalotto, J.-F. (2006) The measurement of central bankautonomy: survey of models, indicators, and empirical evidence. IMF Working PaperNo. 06/227.
Bade, R. and Parkin, M. (1988) Central bank laws and monetary policies. Unpublished,University of Western Ontario, London, Ontario.
Banaian, K. and Luksetich, W.A. (2001) Central bank independence, economic freedomand inflation rates. Economic Inquiry 39: 149–161.
Banaian, K., Burdekin, R.C.K. and Willett, T.D. (1998) Reconsidering the principalcomponents of central bank independence: the more the merrier? Public Choice97: 1–12.
Berger, H., De Haan, J. and Eijffinger, S.C.W. (2001) Central bank independence: anupdate of theory and evidence. Journal of Economic Surveys 15: 3–40.
Bleany, M. (1996) Central bank independence, wage-bargaining structure, and macroeco-nomic performance in OECD countries. Oxford Economic Papers 48: 20–38.
Bouwman, K., Jong-A-Pin, R.M. and De Haan, J. (2005) On the relationship betweencentral bank independence and inflation: some more bad news. Applied FinancialEconomics Letters 1: 381–385.
Broz, J.L. (2002) Political system transparency and monetary commitment regimes.International Organization 56(4): 861–887.
Brumm, H.J. (2000) Inflation and central bank independence: conventional wisdom redux.Journal of Money, Credit, and Banking 32(4): 807–819.
Campillo, M. and Miron, J.A. (1997) Why does inflation differ across countries? InC.D. Romer and D.H. Romer (eds), Reducing Inflation: Motivation and Strategy (pp.335–357). Chicago: University of Chicago Press.
Cargill, T.F. (1995) The statistical association between central bank independence andinflation. Banca Nazionale del Lavoro Quarterly Review (193): 159–172.
Cecchetti, S.G. and Krause, S. (2002) Central bank structure, efficiency policy, andmacroeconomic performance. Federal Reserve Bank of St Louis Review 84: 47–60.
Chou, Y.K. (2001) The impact of central bank independence and union concentration onmacroeconomic performance in the presence of aggregate supply shocks. Evidencefrom 10 OECD countries (1971–1985). Mimeo, University of Melbourne.
Clark, W.R. (2003) Capitalism, Not Globalism Capital Mobility, Central Bank Indepen-dence, and the Political Control of the Economy. Ann Arbor, MI: University ofMichigan Press.
Crowe, C. (2008) Goal-independent central banks: why politicians decide to delegate.European Journal of Political Economy 24: 748–762.
Crowe, C. and Meade, E. (2007) Central bank governance: what is it and does it matter?Journal of Economic Perspectives 21: 69–90.
Cukierman, A. (1992) Central Bank Strategy, Credibility, and Independence: theory andEvidence. Cambridge, MA: MIT Press.
Cukierman, A. and Lippi, F. (1999) Central bank independence, centralization of wagebargaining, inflation and unemployment – theory and some evidence. EuropeanEconomic Review 43: 1395–1434.
Cukierman, A. and Webb, S.B. (1995) Political influence on the central bank: internationalevidence. The World Bank Economic Review 9: 397–423.
Cukierman, A., Webb, S.B. and Neyapti, B. (1992) Measuring the independence of centralbanks and its effects on policy outcomes. The World Bank Economic Review 6:353–398.
Debelle, G. and Fischer, S. (1994) How independent should a central bank be? In J.Fuhrer (ed.), Goals, Guidelines, and Constraints Facing Monetary Policymakers (pp.195–211). Federal Reserve Bank of Boston Conference Series No. 138.
De Haan, J. (1999) The case for an independent European Central Bank: a comment.European Journal of Political Economy 15: 759–762.
De Haan, J. and van’t Hag, G.J. (1995) Variation in central bank independence acrosscountries: some provisional empirical evidence. Public Choice 85: 335–351.
De Haan, J. and Kooi, W. (1997) What really matters? Conservativeness or independence?Banca Nazionale del Lavoro Quarterly Review (200): 23–38.
De Haan, J. and Kooi, W. (2000) Does central bank independence really matter? Newevidence for developing countries using a new indicator. Journal of Banking andFinance 24: 643–664.
De Haan, J. and Siermann, C.L.J. (1996) Central bank independence, inflation and politicalinstability. Journal of Policy Reform 1: 135–147.
De Haan, J. and Sturm, J.E. (1992) The case for central bank independence. BancaNazionale del Lavoro Quarterly Review (182): 305–327.
De Haan, J., Leertouwer, E. and Wansbeek, T. (2003) Measuring central bankindependence: a latent variables approach. Scottish Journal of Political Economy50(3): 326–340.
Dolmas, J., Huffman, G.W. and Wynne, M.A. (2000) Inequality, inflation and centralbank independence. Canadian Journal of Economics 33(1): 271–287.
Doucouliagos, C. (2005) Publication bias in the economic freedom and economic growthliterature. Journal of Economic Surveys 19: 367–388.
Doucouliagos, C. and Stanley, T.D. (2009) Publication selection bias in minimum-wageresearch? A meta-regression analysis. British Journal of Industrial Relations 47:406–429.
Dreher, A., De Haan, J. and Sturm, J.-E. (2008) Does high inflation cause central bankersto lose their job? Evidence based on a new data set. European Journal of PoliticalEconomy 24: 778–787.
Eijffinger, S.C.W. and De Haan, J. (1996a) The political economy of central bankindependence. Special Papers in International Economics No. 19, Princeton.
Eijffinger, S.C.W. and De Haan, J. (1996b) Comment on central bank independence: onlypart of the inflation story. De Economist 144: 658–666.
Eijffinger, S.C.W. and van Keulen, M. (1995) Central bank independence in anothereleven countries. Banca Nazionale del Lavoro Quarterly Review (192): 39–83.
Eijffinger, S.C.W. and Schaling, E. (1995) Central bank independence: criteria and indices.Kredit und Kapital (Special Issue No. 13): 185–217.
Eijffinger, S.C.W., Schaling, E. and Hoeberichts, M. (1998) Central bank independence:a sensitivity analysis. European Journal of Political Economy 14: 73–88.
Forder, J. (1996) On the assessment and implementation of ‘institutional’ remedies. OxfordEconomic Papers 48: 39–51.
Franzese, R.J. (1999) Partially independent central banks, politically responsive govern-ments, and inflation. American Journal of Political Science 43(3): 681–706.
Franzese, R.J. (2001) Institutional and sectoral interactions in monetary policy andwage-price bargaining. In P.A. Hall and D. Soskice (eds), Varieties of Capitalism:The Institutional Foundations of Comparative Advantage (pp. 104–144). Cambridge:Cambridge University Press.
Franzese, R.J. (2003) Multiple hands on the wheel: empirical modeling partial delegationand shared control of monetary policy in the open. Political Analysis 11(4): 445–474.
Franzese, R.J. and Hall, P.A. (2000) Institutional dimensions of coordinating wage-bargaining and monetary policy. In T. Iversen, J. Portusson and D. Soskice (eds),Unions, Employers and Central Banks: Wage Bargaining and Macroeconomic Policyin an Integrating Europe. Cambridge: Cambridge University Press.
Fuhrer, J.C. (1997) Central bank independence and inflation targeting: monetary policyparadigms for the next millennium? New England Economic Review 1(2): 19–36.
Fujiki, H. (1996) Central bank independence indexes in economic analyses: a reappraisal.Bank of Japan Monetary and Economic Studies 14: 79–99.
Greenspan, A. (2007) The Age of Turbulence. New York: Penguin.Grilli, V., Masciandaro, D. and Tabellini, G. (1991) Political and monetary institutions and
public financial policies in the industrial countries. Economic Policy (13): 341–392.Gutierrez, E. (2003) Inflation performance and constitutional central bank independence:
evidence from Latin America and the Caribbean. IMF Working Paper No. 03/53.Hall, P.A. and Franzese, R.J. (1998) Mixed signals: central bank independence, coordi-
nated wage-bargaining, and European Monetary Union. International Organization52(3): 505–536.
Hayo, B. and Voigt, S. (2005) Inflation, central bank independence and the legal system.ICER Working Paper No. 02–2005, http://www.icer.it/docs/wp2005/ICERwp2-05.pdf(last accessed 20 December 2006).
Heylen, F. and Van Poeck, A. (1996) Central bank independence: only part of the inflationstory. De Economist 144: 45–61.
Ilieva, J. and Gregoriou, A. (2004) Central bank independence and inflation performancein transition economies: new evidence from a primary data approach. ManchesterMetropolitan Business School Working Papers Series, No. 04/06.
Iversen, T. (1999) The political economy of inflation: bargaining structure or central bankindependence? Public Choice 99: 237–258.
Jacome, L.I. and Vazquez, F. (2008) Any link between central bank independence andinflation? Evidence from Latina America and the Caribbean. European Journal ofPolitical Economy 24: 788–801.
Jenkins, M.A. (1996) Central bank independence and inflation performance: panacea orplacebo? Banca Nazionale del Lavoro Quarterly Review (197): 241–270.
Jonsson, G. (1995) Institutions and macroeconomic outcomes – the empirical evidence.Swedish Economic Policy Review 2: 181–212.
Keefer, P. and Stasavage, D. (2000) Bureaucratic delegation and political institutions:when are independent central banks irrelevant? World Bank Policy Research WorkingPaper No. 2356.
Kilponen, J. (1999) Central bank independence and wage bargaining structure – empiricalevidence. Bank of Finland Discussion Paper No. 9/99.
King, D. and Ma, Y. (2001) Fiscal decentralization, central bank independence, andinflation. Economics Letters 72: 95–98.
Loungani, K.P. and Sheets, N. (1997) Central bank independence, inflation, and growthin transition economies. Journal of Money, Credit, and Banking 29: 381–399.
Maliszewski, W.J. (2000) Central bank independence in transition countries. Economicsof Transition 8(3): 749–789.
Mangano, G. (1998) Measuring central bank independence: a tale of subjectivity and ofits consequences. Oxford Economic Papers 50: 468–492.
Nijkamp, P. and Poot, J. (2005) The last word on the wage curve? Journal of EconomicSurveys 19: 421–450.
Oatley, T. (1999) Central bank independence and inflation: corporatism, partisanship, andalternative indices of central bank independence. Public Choice 98: 399–413.
Posen, A.S. (1993) Why central bank independence does not cause low inflation: thereis no institutional fix for politics. In R. O’Brien (ed.), Finance and the InternationalEconomy (No. 7, pp. 41–65). Oxford: Oxford University Press.
Posen, A.S. (1995) Declarations are not enough: financial sector sources of central bankindependence. NBER Macroeconomics Annual 10: 253–274.
Rogoff, K. (1985) The optimal degree of commitment to an intermediate monetary target.Quarterly Journal of Economics 100(4): 1169–1190.
Romer, D. (1993) Openness and inflation: theory and evidence. Quarterly Journal ofEconomics 108: 869–904.
Rose, A.K. and Stanley, T.D. (2005) Meta-analysis of the effect of common currencieson international trade. Journal of Economic Surveys 19: 347–366.
Sakamoto, T. (2003) Social democratic corporatism, central bank independence, andeconomic performance: an empirical analysis of 17 industrialized economies,1961–1998. Paper presented at the 2003 Conference of the American Political ScienceAssociation.
Stanley, T.D. (2008) Meta-regression methods for detecting and estimating empiricaleffects in the presence of publication selection. Oxford Bulletin of Economics andStatistics 70: 103–127.
Stanley, T.D. and Jarrell, S.B. (1989) Meta-regression analysis: a quantitative method ofliterature surveys. Journal of Economic Surveys 3: 54–67.
Stasavage, D. and Keefer, P. (2003) The limits of delegation: veto players, central bankindependence and the credibility of monetary policy. American Political ScienceReview 97(3): 407–423.
Sturm, J.-E. and De Haan, J. (2001) Inflation in developing countries: does central bankindependence matter? New evidence based on a new dataset. Ifo Studien 47(4):389–403.
Temple, J. (1998) Central bank independence and inflation: good news and bad news.Economics Letters 61: 215–219.
Treisman, D. (2000) Decentralization and inflation: commitment, collective action orcontinuity. American Political Science Review 94(4): 837–857.
Walsh, C. (2005) Central bank independence. Prepared for the New Palgrave Dictionary,December 2005 (downloaded from http://econ.ucsc.edu/walshc/; last accessed 13December 2006).