Abstract
Central bank independence is important for monetary stability. This paper investigates the relationship between central bank independence (CBI) and inflation for 30 OECD countries over 1980s and 2000s. I explore different measures of CBI and develop an empirical CBI index based on the fixed effects within the monetary reaction function. The empirical results suggest that there is two-way causality between CBI and inflation. The results, by extension, imply an asymmetry in de facto and de jure measures of CBI, and may shed light on the role of institutions in monetary system across countries.
Word Count: 6883
Introduction
Nowadays, it is widely believed that a more independent central bank leads to lower inflation. Economists hold the consensus that there is an inflation bias in the construction of monetary policy and this can be reduced by setting up politically, economically and personally independent central banks (Barro and Gordon 1983, Bade and Parkin 1984 and Rogoff 1985). The interaction of central bank independence and inflation would be crucial for us to understand the role of institutions in monetary system.
Nevertheless, the empirical evidence for this consensus is rather limited. On the one hand, Grilli et al. (1991), Cukierman et al. (1992), Jonsson (1995) and Eijffinger et al. (1998) found that many central banks steadily increased their institutional autonomy and enjoyed higher price stability. On the other hand, Bade and Parkin (1984) and Alesina and Summers (1993) suggested there is no correlation between central bank independence and the variability of inflation. The later studies (Swinburne and Castello-Branco 1991, Eijiffinger 1996), questions the reliability of both empirical conclusions on various grounds. In this paper, I aim to address these empirical problems and provide an updated study on the hypothesis that central bank independence (CBI) reduces inflation.
The first problem comes from the measures of CBI. As Eijiffinger (1996) points out, the negative correlations between CBI and inflation are particularly sensitive to the numerical values of CBI indices. The existing de jure measures of the central bank independence are suspected to be biased as they have unavoidable subjective elements by construction. The turnover rate of chief governors tries to capture the de facto independence but also proves problematic, since low turnover rate does not necessarily imply high independence. In this paper, I will introduce a new empirical CBI index (as the individual country-specific characteristics) through the fixed effects model, and the result suggests it is a strong proxy for actual CB independence. For comparison purpose, I will test the CBI hypothesis using three measures of CBI: the de jure measures of CBI, the de facto measure of CBI and my empirical CBI index.
The second deficiency against this inverse relationship conclusion argues the endogeneity and omitted variable bias exists in this relationship. I will include real GDP per capita and trade openness as my control variables. In order to eliminate the further omitted variable bias, I will adopt the difference in difference specification in addition to multivariate OLS regression. I will also use the Instrumental Variables (IV) method to eliminate the endogeneity effects and improve the robustness of results.
The third major flaw in the cross-sectional approach is the sample homogeneity. While the earliest studies of independence focusing on a fairly narrow subset of industrial countries delivered this inverse relationship result, later studies covering a wider set of developing and industrial countries found more equivocal results. To mitigate the sample selection bias, I will include both developed and developing countries in my sample, which includes 30 OECD member countries. They are classified as: developed economy (Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Israel, Japan, Korea, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom and United States) ; developing economy (Chile, Hungary, Mexico, Poland, Turkey). For the time sample, I will use years from 1987 to 2006 and primarily consider two sub periods: period 0 (1987 to 1991) and period 1 (2002 to 2006).
Using the empirical CBI index I have constructed, I found that there is two-way causality between CBI and inflation. While the empirical CBI index has a positive correlation with inflation, the turnover rate of governors (after instrumented for the empirical CBI index) demonstrates a negative relationship with inflation. The intuition is that, countries previous high inflation leads to high central bank independence (captured by the empirical CBI index), which in turn reduces the turnover rate of governors and results in lower inflation in later periods.
The remainder of this paper is organized as follows: Section B offers theoretical background. Section C discusses data and estimation strategy. Section D discusses the results, and Section E concludes.
B. Theoretical Background
The modern theory for central bank independence stems from the time inconsistency problem in the ongoing "Rules versus Discretion" debate. Barro and Gordon (1983) argue that governments are tempted to expand output over its natural level during the electoral periods to get elected and cause an inflationary bias to the economy later, which indicates that what is optimal for now might become suboptimal later. Individual workers recognize that politicians have incentives to use monetary policy to achieve political objectives and form inflationary expectations, which will result in union wage bargaining. At the same time producers expect to incur higher costs so they will set higher prices which finally lead to inflation.
According to Thomas Oatley (1999), by transferring monetary policy control from the government to the independent central banks, the inflationary expectations can be reduced. Following this logic, nations with independent central banks are likely to have lower inflation on average than nations with central banks actually controlled by governments. Figure 1 illustrates the effects of central bank independence.
Figure 1.　Inflation Biases of a Government and a Central Bank
C. Data Description and Estimation Strategy
Inflation
Data for annual inflation from 1987 to 2006 are taken from the OECD Online Statistics. I choose consumer price for all items over consumer price for energy or food as the first index is a relatively complete measure of a country's inflation. Data are generally available up to 2006; except for Poland the data for inflation from 1987 to 1989 are unavailable. I take the five year averages of inflation for each country in the sample and mark them as follows:
INF0: average inflation rate over 1987 to 1991 (period 0)
INF2: average inflation rate over 1992 to 1996 (period 2)
INF3: average inflation rate over 1997 to 2001 (period 3)
INF1: average inflation rate over 2002 to 2006 (period 1)
Inflation rates over period 0 and 1 are of my primary interests; nevertheless I also derive the average inflation rates between the two periods for later comparative analysis.
Measuring Central Bank Independence
The measurement of central bank independence has been developed over the past three decades but remains as a key challenge facing researchers. After Bade and Parkin introduced the first legal index for CBI in 1980, Alesina in 1988, Grill (1991), Cukierman (1992) and Eiffinger-Schaling (1993) developed the so called de jure CBI proxy by assessing a set of legal characteristics obtained from an institution's statutes. Among them, the GMT (Grill-Masciandoro-Tabellini) and CWN (Cukierman-Webb-Neyapt) indices are the most widely used since they specify the criteria explicitly and provide the relatively large country samples. However, the de jure CBI measure often fails to capture the operational aspect of CB independence and thus deviates itself from the actual independence. In the following subsections, I will first explore the de jure and de facto CBI measures accordingly, and then provide a new CBI proxy.
1. The De Jure Central Bank Independence
The de jure CBI measure followed the well-recognized methodology of Cukierman, Webb, and Neyapti (1992), which considers four different aspects of a central bank's legal independence : 1. The appointments, dismissal and term in office of the central bank chief governor 2. the role of the central bank in policy formulation 3. policy objective 4. provisions of lending to government
Based on the above criteria, Cukierman, Webb, and Neyapti (1992, henceforth CWN) measured independence in 72 countries using 16 different legal characteristics from central bank statutes in the 1980s. Crowe and Meade (2008) suggest that the CWN measure "is easy to assess not only the current state of play but also the evolution of governance practices over time" and producing new measures of CBI for 96 countries followed the same methodology. They updated the CWN indices by using a database of central bank laws held by the IMF which is current through 2003. Collecting data from the study of Crowe and Meade (2008), I now have five distinct measures for the de jure CBI in two periods (1987-1991 and 2002 -2006):
Q: the overall measure for the degree of independence after weighting all four aspects specified above
q1: CBI aspect 1
q2: CBI aspect 2
q3: CBI aspect 3
q4: CBI aspect 4
However, CWN measure is not a perfect proxy for actual central bank independence. Cukierman (1992) notes legal independence is not the same as actual independence and argues that there are at least two reasons for this divergence. Firstly, central bank laws are far from complete, so it is impossible to specify the limits of the authority of the central bank and the political authorities in all situations. These limits might be instead determined by a third factor such as traditions and personalities of central bankers. Secondly, even if the scope of authority, procedures, objectives, ect. are described explicitly in law, actual practice may be different. The empirical evidence for central banks in less developed countries shows the divergence in the actual CBI and legal CBI could be so large that makes the legal index a particularly misleading measure for actual CBI.
2. The De facto Central Bank Independence
Realizing the limitation of legal index, Cukierman et al. (1992) provides a proxy for de facto independence in the study of central banks in emerging economies: the turnover rate (TOR) of central bank governors. TOR measures the rate of change of chief executives of a nation's central bank and the measure is considered to capture the realistic degree of CBI in relatively new democracies since in those economies, actual independence often differs from the 'legal' independence implied by the central bank charter. Theoretically, a high turnover rate is associated with low CB independence as governments will exert their influence on the structure of central banks while changing the governors frequently. A shorter tenure due to the frequent replacements of chief governors also means that the Central Bank governors have less incentive to implement policies from a long term perspective, thus are less likely to introduce independent policies if there is pressure from the government. I collect the turnover rates of governors based on CWN and Crowe-Meade's samples from 1980 to 1989, and 1995 to 2004 respectively.
Although the turnover rate of chief governors reflects the degree of management independence of central banks, there are several drawbacks of this measure. One drawback of this proxy is that while the turnover rate certainly matters for the overall CB structure, it ignores other aspects of a Central Bank's operations. Its right to set up monetary goals and ability to determine its own approach to meet those targets both count towards the degree of a central bank's actual independence and the turnover rate of governors is unable to capture those factors. Moreover, a low level of turnover rates does not necessarily means a more independent central bank. Rather, if the central bank is largely dependent upon the government, it would be of the government's interest to keep a servile central banker for a long time. Further, a high turnover rate of governors is a common phenomenon for central banks in developing countries (Cukierman 1992, Sturm 1995), but the opposite proves true for developed countries. Generally, developed economies have much lower level and narrower spread of the turnover rates compared to that in less developed countries, so TOR would be less useful to explain the divergent degrees of CB independence among the 30 OECD countries in my sample, which contains a majority number of well developed economies. Therefore I also construct a new CBI index for sample analysis.
3. The Empirical Central Bank Independence
As I argued above, actual independence is determined by many factors from which a lot are often subjectively evaluated and hardly quantifiable, thus I assume it is not viable to measure actual independence directly. Berger and Haan (2000) argued that it is the diverging interpretations of national central laws rather than the divergent definitions of legal independence that lead to the differences in Grill-Masciandoro-Tabellini (GMT) index and the Curkerman Index (LVAU), both of which are constructed to indicate the legal CBI. Besides, the traditional score-based indices are supposed to be somewhat biased towards each economist's own country, because the greater acquaintance with the case brings recognition of the greater freedom of behavior acquired in current practice by the national central bank compared to the formal rule. As pointed by Grill-Masciandoro-Tabellini (1991), there is no non-arbitrary way of aggregating there criteria or attributes to a composite index. Therefore the subjective nature of indices of this kind will more or less induce statistical inaccuracy if we examine its relationship with macroeconomic variables such as inflation.
Instead manipulating score-based CBI indices, I adopt an innovative approach to capture the empirical CB independence from the monetary reaction channel. A central bank conducts its own monetary policy and primarily controls interest rates to stabilize inflation in a relatively low level. Following Taylor (1993), a forward-looking monetary reaction function is well suited to explain the factors affecting Central Banker's behavior and the market interest rates.
The Taylor-rule formulation above can also be expressed as the general form:
it =(1-Ï)α+(1-Ï)βE[pt+n]+(1-Ï) rE[yt+m]+Ïit-1+εt
where it is the short-run interest rates, E[Pt+n] is the expected inflation rate at time t+n, and E[Yt+m] is the output gap at time t+m.βand r represent the respective weights to the stabilization objectives for inflation and output. it-1 is the lagged interest rate to account for the smooth effect (Jondeau and Bihan 2000) andεt is the random shock,εt ~i.i.d (0,σε). Theoretically a more restrictive (lower interest rate i) monetary policy will respond to raising inflation and economic growth. Talyor (1993) proposed setting β=r=0.5. That is, inflation and output gap are of the same importance for central banks to consider and thus symmetric. However, I expect a higher value ofβas my sample consists of inflation-targeting OECD countries.
Model
I will make two main assumptions for this reaction model. First, the central bank is assumed to only respond to expected deviations of next-period inflation from its target and to the output gap. Second, the reaction of monetary authorities to inflation and output gap not only depends on these two variables, but also on an individual country-specific effect. I interpret this effect as actual CBI which captures the differences in central bank independence come forward in different structural pressures to lower or raise money market rate [1] . This interpretation has both theoretical and empirical grounds. Theoretically, a higher CB independence means the central bank has more freedom to conduct monetary policy in the quest for price stability. Empirically, Eijffinger and Rooij (1996) found that CB independence would lower the pressure for central banks to change interest rates over 1977 to 1990 in 10 industrialized countries.
My investigation expands the country sample to include both developed and less developed economies and gives an update of the effect of CB independence on the response of the money market rate. I improve the model by including last period's interest rate as one independent variable, as the last-period policy stance may also affect the current policy stance. [2] Therefore, my monetary policy reaction model takes the following specification for a particular country I at time t:
Δii,t = β0+β1Pi,t+β2Pi,t-1+β3Yi,t+β4Yi,t-1+β5ii,t-1+CBIi+εi, t with i= 1…N, t=1…T
Variable Description:
Δii,t = change of market interest rate of country I in time t;
Pi,t = inflation rate of country I in time t;
Yi,t = real GDP growth rate of country I in time t;
ii,t-1 =lagged interest rate in time t-1;
CBIi = actual central bank independence of country I;
εi, t = the stochastic term of country I in time t. [3]
My investigation focuses on Period 0 (1987-1991) and Period 1 (2002-2006).
Data
Data for short term nominal interest rates (1986-1991, 2001-2006) are collected primarily from OECD Online Statistics. For period 0, some data are missing and instead I use the 3 months interest rates and Treasury Bills from Euro Area Statistics and IMF International Finance Statistics (IFS) datasets. The consistency of data from three different datasets are largely ensured as the OECD constructs their short term interest rates data based on the 90-days interbank offer rate, which coincides well with the other two datasets. Then I calculateΔii for each country.
As specified before, data for inflation rates are taken from the Consumer Price Index in the OECD Online Statistics from two periods: Period 0 (1986 to 1991) and Period 1 (2001 to 2006).
Data for the real GDP growth rate for each country are obtained from the Penn World Tables Version 7.0.
Data for the three variables are generally available through both periods; however, in the case of short term nominal interest rates, there is no data for Poland and Chile have in period 0. I still keep them in the panel as they increase the total degrees of freedom for my sample and are able to provide empirical CBI values in period 1.
The Panel Data Approach
Facing data spanning 2 dimensions, I resort to the use of panel data to infer the country-specific effect CBIi which can not be observed directly. Assuming central bank independence does not vary a lot over the five year range for each country in sample, I apply the fixed effects model to derive the empirical CBI indices for 30 OCED countries over period 0 and 1.
For convenience we introduce the following notation:
β=(β1,β2, β3, β4, β5)'
Xi,t= (Pi,t, Pi,t-1,Yi,t, Yi,t-1)'
The monetary response function then takes the following form:
Δii,t = β0+β'Xi,t +CBIi+εi, t with i= 1…N, t=1…T
where CBIi is the empirical independence of central bank in country I which distinguishes itself from the individual country fixed effect onΔii,t.
To estimate the actual independence, I create 30 dummies for each country in sample and each coefficient of those dummies represents the respective fixed country effect. Estimations for CBIi in period 0 and period 1 are conducted sequentially.
Results
The estimation results of reaction function using panel data are given in the table 1 and the reaction function equations for two periods:
Period 0: Δii,t = 13.1 +0.56Pi,t+0.12Pi,t-1-0.15Yi,t+0.34Yi,t-1-1.1ii,t-1+CBIi+εi, t
(5.26) (9.97) (9.97) (-0.97) (2.22) (-6.46)
Period 1: Δii,t = 6.0 +0.52Pi,t+0.09Pi,t-1+0.09Yi,t+0.11Yi,t-1- 0.76ii,t-1+CBIi+εi, t
(5.42) (10.98) (1.37) (2.07) (2.67) (-13.24)
Absolute t-values for estimated coefficients are given in the brackets. These results confirm that a higher market rate will be in response to high economic growth and inflation. In addition, magnitude of the coefficients suggests that the response to inflation is stronger than the response to inflation in both periods, which confirms my expectation for inflation targeting countries. Regressors inflation Pi,t and lagged interest rate it-1 demonstrate statistical significant, both at 1%. R-squared are 71% and 93% for equations in period 0 and 1, so my model gives a good fit for the data in both periods.
The empirical individual country fixed effects for Period 0 and Period 1 are summarized in table 1. STATA automatically dropped dummies for Chile and Poland in Period 0 due to insufficient data. From Table 1, we see divergent CBI evolution over the two periods. Some OECD countries improved their Independence and some had even lower CBI in Period 1. It is noticeable that only a few countries have statistically significant fixed effects in Period 0 given all exhibit strong significance (1% level) in Period 1. This implies countries in period 0 experience random shocks more frequently than in period 1 and the monetary system is relatively more stable in the latter period.
Table 1: Empirical Central Bank Independence Index
Period 0 Period1
Period 0 Period1
Australia -3.068 -4.380***
(-1.13) (-4.76)
Japan -8.735** -5.999***
(-3.08) (-5.48)
Austria -7.694** -5.438***
(-2.82) (-5.17)
Korea -5.504 -5.494***
(-1.59) (-5.81)
Belgium -6.194* -5.586***
(-2.29) (-5.33)
Luxembourg -6.801* -6.367***
(-2.19) (-6.58)
Canada -5.454* -5.546***
(-2.06) (-5.41)
Mexico 3.623 -3.577***
(0.78) (-4.35)
Chile NA -5.954***
(.) (-6.31)
Netherlands -6.828* -5.525***
(-2.51) (-5.20)
Denmark -5.244* -5.442***
(-2.04) (-5.19)
New Zealand -2.128 -3.422***
(-0.81) (-3.90)
Finland -4.338 -5.170***
(-1.65) (-5.20)
Norway -3.672 -4.768***
(-1.39) (-4.91)
France -5.922* -5.528***
(-2.20) (-5.29)
Poland NA -3.831***
(.) (-4.63)
Germany -7.807** -5.060***
(-2.85) (-4.74)
Portugal -5.456 -6.098***
(-1.85) (-5.60)
Greece -4.468 -6.826***
(-1.75) (-6.66)
Spain -2.542 -6.623***
(-0.88) (-6.40)
Hungary 0 -3.347***
(.) (-4.00)
Sweden -6.105* -5.402***
(-2.34) (-5.48)
Iceland -1.177 -2.879***
(-0.40) (-3.46)
Switzerland -9.182** -6.192***
(-3.38) (-5.69)
Ireland -4.755 -7.184***
(-1.71) (-7.31)
Turkey 0.599 0
(0.18) (.)
Israel -9.684** -3.589***
(-3.38) (-3.99)
United Kingdom -4.312 -4.239***
(-1.61) (-4.49)
Italy -4.543 -5.577***
(-1.69) (-5.12)
United States -8.449** -6.030***
(-3.16) (-5.74)
t statistics in parentheses, * p<0.05, ** p<0.01, *** p<0.001
Control Variables
I select real GDP per capita (GDP) and trade openness (OPEN) as my control variables. They are both taken from the Penn World Tables Version 7.0 and are expressed in US dollars at current prices. Trade openness (OPEN) is measured as the sum of exports and imports divided by GDP while real GDP per capita (GDP) is the GDP per capita adjusted for inflation. Again, I calculate the five-year averages of both variables during 1987-91 (period 0) and 2002-06 (period 1).
Political Indicators
To resolve the endogeneity problem in the relationship between CBI and inflation, I want to control for institutional reform by instrumenting for the change in the independence index using these two measures. Based on the study of Crowe and Meade (2008), I take 2 political indicators "rule of law" (rl) and "voice and accountability" (va) from the World Bank's Governance Matters as proxies for the overall governance for each country. The Worldwide Governance Indicators (WGI) in Governance Matters produced by Kaufmann, Kraay and Mastruzzi for 213 countries are based on 31 different sources that provide information on various aspects of governance, so WGI should be quite informative about the changes of individual countries' relative positions in global governance. As WGI data are available from 1996 onwards, I take the five-year average of rule of law (rl1) and voice and accountability (va1) in period 1 only.
Estimation Strategy
The basic hypothesis is that central bank independence will reduce inflation, ceteris paribus. I propose to test this with the following basic specification, where I, CBI, Q and TOR respectively denote period I, empirical central bank independence (or country fixed effects), measures of de jure central bank independence CWN and the turnover rates of governors:
INF (i) = C + C1*CBI (i) +C2*Q (i) +C3*TOR (i) +C4*GDP (i) +C5*OPEN (i) + Ut
i= 0, 1, 2, 3 and Ut ~ iid (0, σu)
Coefficients C1, C2, C3 capture the effect of central bank independence on inflation based on 3 different measures described previously. Using the updated measures of central bank independence, I will first apply the basic OLS regressions to estimate the relationship between independence and inflation for Period 0 and 1 across 30 OECD countries. As I concern about the potential endogeneity and omitted variable bias affecting the value and consistency of the OLS estimator, I will refine the model by taking the first difference for each dependent and independent variable over period 1 and 0, through which the impact of time-invariant omitted variables is eliminated. As I also want to investigate whether the individual aspects of central bank governance affect inflation, I include four relatively parsimonious measures of legal independence (q1, q2, q3 and q4) into specification for additional examinations. Finally, I take advantage of instrumental variables (IV) to address the potential endogeneity bias further and test whether my IVs are strong enough to improve the significance level of regression results.
D. Results and Discussion
Basic OLS Regression
The results for the cross-country OLS estimations are given in the first 2 columns of Table 2. While the de jure independence Q (i.e. the CWN index) surprisingly shows a positive sign in period 0, the turnover rates of governors TOR exhibits the weak or non-existent relationship between CB independence and inflation for both periods. However, this outcome may arise due to the endogeneity and omitted variable bias in this linear specification, and I will employ further specifications to reduce the bias in order to offer statistically reliable results.
The most puzzling part of the results comes from the empirical CBI, which indicates a strong positive relationship with inflation at 1% significance level. The coefficient of CBI suggests that a 1% increase in the empirical CBI will be associated with a 3.4% and a 1.9% rise in inflation rates in Period 0 and Period 1 respectively. Again, this may be due to endogeneity and omitted variable bias in specification. I will continue to collect relevant results from the next first difference specification to identify the reasons behind the unexpected positive sign of CBI.
First Difference Specification
The existing two time dimensions Period 0 and 1, allow me to employ a first difference specification, eliminating unobservable and omitted country fixed effects that are likely to be correlated with the independent variables. The specification now becomes:
ΔINF = C+ C1*ΔCBI +C2*ΔQ +C3*ΔTOR +C4*ΔGDP +C5*ΔOPEN
Estimation results are reported through Column 3 to 6 in Table 2, and reveal mixed results. On the one hand, after I revisited this relationship by removing time-constant potential missing variables, the negative relationship between central bank independence and inflation predicted by theory can be now verified by the de facto measure TOR (Column 5). None of the five de jure measures are statistically strong to deliver low inflation (Column 3 and 6).
On the other hand, as results in column 4 suggest, the empirical CBI measure still exhibits a significant positive correlation with inflation. Referring to the complete result table 2 in Appendix, I find that under the first difference specification, the empirical CBI always shows strong positive correlation with inflation, usually within 5% significance level. This positive correlation can no longer be simply justified on the ground of missing variable bias. Rather, the underlying reason would be more likely to lie in the interaction between these two variables and I will address it by setting up a causality hypothesis in the next section.
Table 2: Inflation and Central Bank Independence*
Inflation
(1)
(2)
(3)
(4)
(5)
(6)
Dependent variable:
Inflation, period 0
Inflation, period 1
Change in Inflation, ΔINF
CBI
3.367***
1.870***
2.301
1.966*
(3.95)
(4.73)
(2.08)
(2.53)
Q
30.29*
4.766
-72.62
(2.11)
(1.73)
(-1.58)
TOR
26.83
4.776
41.20
163.4*
(1.24)
(0.75)
(1.61)
(2.31)
GDP**
-0.406
0.0238
0.633
1.10**
1.23
2.76
(-0.66)
(0.40)
(1.35)
(3.35)
(0.62)
(1.60)
OPEN
-0.01
-0.02
-0.22
-0.10
-1.10
-0.26
(-0.16)
(-0.84)
(-1.54)
(-0.91)
(-1.79)
(-0.41)
Q1
9.031
(0.16)
Q2
0.611
(0.01)
Q3
-7.362
(-0.16)
Q4
-39.42
(-0.31)
_cons
18.32
8.433**
-15.71
-21.30***
-9.462
-30.29
(1.50)
(3.69)
(-1.72)
(-4.47)
(-0.30)
(-1.06)
N
28
26
24
28
26
30
*t statistics in parentheses, * p<0.05, ** p<0.01, *** p<0.001
**real GDP per capita (GDP) is divided by 1000 for larger regression coefficients.
Causality Hypothesis
We know econometrically, correlation does not mean causality, so independence captured by my empirical CBI index does not necessarily cause the inflation to rise. As Daniel West (2007) suggest, there is possible two-way causality between inflation and actual CBI. Curkeirman's assertion that "high inflation and low central bank independence do, at least, partially reinforce each other" gained empirical support since some countries with low CB independence do experience high inflation rates, which will in turn weaken central banks' credibility and result in even higher inflation levels.
However, this statement does not form a universal rule since there are many countries successfully escaping this vicious cycle. One famous counterexample is Germany, where high inflation leads to the radical reform of its central bank statues and finally strengthens its commitment to stabilize price. Thus I would like to point out: the way inflation and CB independence interacts with each other will vary from country to country and this might also depend on the time dimensions we focus on.
Looking closely at my OECD country sample, most central banks gained their independent status over the last two decades, with the Reserve Bank of New Zealand being the first one to do so in 1989. That is, while Period 0 (1987-1991) sees most central banks remained dependent (on governments), Period 2 (1992-1996) and Period 3 [4] (2002-2006) fall on the time when the crucial changes in central banks' status took place. The possible explanation for the positive sign is that countries, like Germany, reform their central banks and obtain independence after suffering from high inflation in the period 0. I propose the following hypothesis: High inflation Granger causes high CB independence.
Panel Granger Causality
The Granger Causality test is usually designed for time series analysis and requires stationary annual or quarterly data. Due to the relatively time-invariant nature of central bank independence and the five-year averages I take for every variable, I can only obtain 4 data from 1987 to 2006, which means the time series approach is not viable. In this paper, I apply the panel granger causality test methodology, which is rarely seen from literature and remains a new research area for econometrics of panel. [5]
I will test the bivariate relation between the change in empirical CBI and the change in average inflation with different period lags, using the first difference specification. I specify those variables as follows:
ΔCBI: Changes in CBI from period 0 (1987-1991) to period 1 (2002-2006)
ΔINF: Changes in average inflation from period 0 (1987-1991) to period 1 (2002-2006)
ΔINF2: Changes in average inflation from period 0 (1987-1991) to period 2 (1992-1996)
ΔINF3: Changes in average inflation from period 0 (1987-1991) to period 3 (1997-2001)
ΔINF1: Changes in average inflation from period 2 (1992-1996) to period 3 (1997-2001)
If the causality hypothesis holds true, I would expect a positive sign ofΔCBI after it is regressed against the first difference of average inflationΔINF2 andΔINF3. In addition, a more recent change in inflation will theoretically have a larger impact on current degree of CBI, so I also expect to see a larger coefficients forΔINF3. Further, it would be interesting to know the impact ofΔINF1 onΔCBI, that is, whether the inflation change during the intermediate periods is big enough to impact on the degree of central bank independence.
The results reported in Table 3 confirm my causality hypothesis while both ΔINF2 and ΔINF3 post positive signs which are significant at 5% and 1% level respectively. ΔINF3 has a coefficient of 0.127 which is larger than ΔINF2 (0.112), and this implies that a 1% increase in last period's inflation will lead to a 0.127% rise in the degree of central bank independence. Column 4 shows that the change of inflation from period 2 to 3 have no significant impact onΔCBI. One possible explanation would be that some OECD countries already began to reform their central banks during 1992 to 2001, and the inflation impact is thus mitigated by the independence gain.
The mechanism is structured as follows:
reform
high inflation--- high CBI (captured by empirical CBI)---low turnover rates of governors-low inflation
Table 3 Panel Granger Causality
ΔCBI
(1)
(2)
(3)
(4)
ΔINF
0.0928*
(2.74)
ΔINF2
0.112*
(2.40)
ΔINF3
0.127**
(2.86)
ΔINF1
0.136
(1.00)
_cons
0.524
0.154
0.478
0.103
(1.04)
(0.33)
(0.98)
(0.18)
N
28
28
28
28
t statistics in parentheses ="* p<0.05 ** p<0.01 *** p<0.001"
IV Approach
Finally I control for the endogenous reform by instrumenting for the change in two independence measures: turnover rates of governors and CWN legal independence. I select measures of overall governance "the rule of law" and "voice and accountability" as my first two instrumental variables [6] . Further, I try the empirical CBI index as the third IV for the turnover rates of governors as it is constructed to reflect actual central bank independence.
The estimated effect of turnover rates TOR becomes stronger after instrumentation for the 3 IVs compared to that estimated via ordinary least squares, and remains the same positive sign. Referring to Table 3 in Appendix, I found that only the combination of all three IVs improved the significance level of TOR, which implies that derived empirical CBI is a strong proxy for TOR, the measure of the de facto independence.
The de jure independence demonstrated a statistically stronger result after controlling for the first two IV, increasing its significance level from 12.7% to 7.5%. However, once empirical CBI is added to control for Q, the significance level decreases dramatically, which casts further doubts on the reliability of the de jure measures of independence.
Table 4: IV Regressions
(1)
(2)
INF
INF
TOR
131.1*
(2.44)
GDP
0.162
0.243
(0.26)
(0.09)
OPEN
-0.547*
1.111
(-2.57)
(0.88)
Q
-295.7
(-1.78)
_cons
-0.0307
43.76
(-0.00)
(0.69)
N
24
30
t statistics in parentheses ="* p<0.05 ** p<0.01 *** p<0.001"
E. Concluding Remarks
In this paper, I present an updated study on central bank independence and its relationship with inflation for 30 OECD countries. With respect to actual CBI, I offer an alternative measure based on the fixed effects in monetary reaction function. In the first difference specification, the empirical CBI shows a puzzling positive correlation with inflation levels. Applying the innovative Panel Granger Causality methodology, I find that high inflation induces high central bank independence captured by the empirical CBI index. In later IV regression analysis, the empirical CBI index proves to be a reliable instrumental variable for TOR (turnover rates of governors), and the strong negative relationship between CBI (captured by TOR) and inflation are finally revealed.
While this paper generally confirms the central bank independence hypothesis, the work can be meaningfully extended. One can refine the analysis by expanding country sample with more emerging economies to further reduce sample homogeneity. In addition, it would be interesting to try more objective measures of central bank independence via other monetary transmission channels, such as the credit channel where the financial sector acts as a pressure group for central bank (see Posen 1995). Furthermore, CBI is neither a sufficient nor necessary requirement for low inflation. Countries with independent central banks may still experience high levels of inflation, as there are many other factors we should consider, for instance, the role of central bank transparency.
To conclude, I would like to offer my own reading of the result. One implication is an asymmetry across measures of central bank independence. This asymmetry certainly reflects in their construction methodology, for instance, the de jure independence measure CWN ignores the operational aspect of CBI. Further, it also leads to asymmetric results for CBI hypothesis. The empirical CBI captures the impact of last period's inflation while the turnover rates of CB governors influence current inflation level. One needs to select CBI measures properly for a particular country sample and distinguishes the interaction effects of CBI and inflation for each measure. For example, turnover rates are a better proxy for CBI in developing countries and different measures may yield divergent results. Further research is needed to reduce the asymmetry.
The results may further shed light on the role of institutions play in shaping the macroeconomic environment, an important question in economic development. It is conceivable, for instance, that countries with higher turnover rates of governors will be very likely to experience higher inflation. However, one should not ignore the country-specific characteristics which affect central bank's behavior. For instance, the degree of a society's inflation-aversion (Forder, 1996) may also affect its price levels but it differs by country. This is another area for further research.
