Central Bank Independence and Inflation in Transition Economies: A Comparative Meta-Analysis with Developed and Developing Economies

Abstract

This article empirically examines the effect of central bank independence (CBI) on inflation by means of a comparative meta-analysis of studies of transition economies and of other developed and developing economies. The results of a meta-synthesis indicate that both transition and nontransition studies have successfully identified a negative relationship between CBI and inflation. Moreover, the present meta-regression analysis suggests that a series of study conditions strongly affected the empirical results concerning transition economies. The article also finds that no significant difference exists between the two types of studies in terms of either effect size or statistical significance.

Keywords:

• Central and Eastern Europe
• central bank independence
• former Soviet Union
• inflation
• meta-analysis
• publication selection bias
• transition economies

JEL Classification:

• E31
• E58
• G18
• P24
• P34

ACKNOWLEDGMENTS

We thank Josef Brada (editor-in-chief), Hristos Doucouliagos, Evžen Kocenda, two anonymous referees, as well as participants at the First World Congress of Comparative Economics at Roma Tre University, Rome, June 25–27, 2015 and the MAER-NET 2015 Colloquium at Charles University, Prague, September 10–12, 2015, for their helpful comments and suggestions. We also would like to thank Eriko Yoshida for her research assistance, and Tammy Bicket for her editorial assistance.

FUNDING

This research work was financially supported by grants-in-aid for scientific research from the Ministry of Education, Culture, Sports, Science and Technology of Japan (Nos. 23243032 and 15H01849).

Notes

1. In fact, the Washington Consensus stopped short of touching on the liberalization of interest as one of its policy packages, mentioning almost nothing about banking reform (Iwasaki and Suzuki 2016).

2. While the development of the banking sector in transition economies is shown in Table 1, the developed market economies had already begun their banking and financial sector reforms in 1973. The reforms included the elimination of credit controls, deregulation of interest rates, free entry into the financial services industry, bank autonomy, private ownership of banks, and liberalization of international financial flows (compare with the reform elements listed in the first note to Table 1). However, this liberation movement did not attain satisfactory results in most of the counties until the 1980s and 1990s (Williamson and Mahar 1998).

3. See various issues of EBRD’s Transition Reports (2015, 2016) and Iwasaki and Uegaki (2015, Figure 2).

4. The first comprehensive attempt to build CBI indexes was made by R. Bade and M. Parkin in 1978. See Parkin (2012).

5. Arnone, Laurens, and Segalotto (2006) provide an excellent review of empirical studies in this research field.

6. As for the study of nontransition economies, Cukierman proposed combined measurements of CBI that included (a) legal independence, (b) the turnover rate of the CB governor, and (c) the characterization of CBI by answers to a questionnaire. Cukierman asserted that the turnover rate of the CB governor appears to capture significant variations in independence within less-developed countries (Cukierman 1992, 369–414, 430).

7. The period is divided into two or three subperiods.

8. The final literature search using these databases was conducted in December 2016.

9. This means that the meta fixed-effect model is a special case based on the assumption that ${\delta }_{\theta }^2=0$.

10. For more details on the method of evaluating the quality level of the study, see the Appendix A.

11. Rosenthal’s fail-safe N denotes the number of studies with an average effect size equal to zero, which needs to be added in order to bring the combined probability level of all the studies to the standard significance level to determine the presence or absence of effect. The larger value of fsN in Equation (6) means the more reliable estimation of the combined t value.

12. In addition to MRA using these orthodox estimators, some meta-analysts employ several types of model-averaging approaches, including frequentist model averaging and Bayesian model averaging to tackle the issue of model uncertainty. For instance, see Ahtiainen and Vanhatalo (2012), Babecky and Havranek (2014), and Havranek and Sokolova (2016). We thank the referee for his/her insight regarding this point.

13. The estimation of Equation (8) uses the cluster-robust random-effects estimator and the cluster-robust fixed-effects estimator. With regard to Equation (9), which does not have an intercept term, we report the random-effects model estimated by the maximum likelihood method and the population averaged GEE model.

14. In addition to Babecky and Havranek (2014), meta-studies of the transition literature that employ methodology similar to that of this article include Fidrmuc and Korhonen (2006), Hanousek, Kočenda, and Maurel (2011), Kuusk and Paas (2013), Iwasaki and Tokunaga (2014, 2016), and Tokunaga and Iwasaki (2017).

15. Doucouliagos (2011) argues that Cohen’s guidelines for zero-order correlations are too restrictive when applied to economics and proposes to use the 25th percentile, 50th percentile (median), and 75th percentile of a total of 22,141 PCCs he collected as alternative criteria. According to his new guidelines, for general purposes, 0.070, 0.173, and 0.327 are considered to be the lower threshold for small, medium, and large effects, respectively. In addition, Doucouliagos (2011) also presents field-specific guidelines, in which 0.103, 0.156, and 0.212 are recommended for use as corresponding criteria for the CBI disinflation effect. His new guidelines give a more positive evaluation to the estimates collected here than those in accordance with Cohen’s guidelines.

16. Sample size has a considerable influence on the statistical significance of estimates. Therefore, many meta-studies, from a statistics perspective, use the square root of the degree of freedom as a control variable for the meta-regression model.

17. Appendix B lists the names, definitions, and descriptive statistics of these meta-independent variables.

18. On the other hand, surprisingly, those studies that measured CBI by the governor term tend to more conservatively assess the CBI disinflation effect. One may surmise that this might happen because the governor term is cut off at the upper limit of the legal term for the sake of data, although the governor of a central bank, in fact, could have served a number of terms or resigned in the middle of the term.

19. In Hungary, Slovenia, and Slovakia, banking reforms as a whole have progressed more than in other countries; however, the degree of CBI was relatively low. On the contrary, in Georgia and Armenia, banking reforms have stagnated, whereas the CBI was relatively high. For more details, see Iwasaki and Uegaki (2015, Section 2, pp. 3–7).

20. This is consistent with the estimation result that a dummy variable for estimates of transition economies is not significant in the meta-regression model of Klomp and de Haan (2010), which takes the t value as a dependent variable (Table 4, p. 606).

21. More precisely, the dummy for the transition studies was insignificant in four of seven models that take the PCC as a dependent variable and in six models using the t value.

22. Cukierman, Miller, and Neyapti (2002) argue that it is “too extreme” to conclude that the difference in the degree of legal independence of the central banks in the transition economies does not have much influence on inflation just because the legal index of central bank independence in transition economies does not reflect the true degree of independence (p. 255).

23. The method for assuming the mean of the top 10% most-precise estimates is the approximate value of the true effect along the lines of Stanley (2005).

24. Following Havránek (2015) and Havranek and Sokolova (2016), we also estimated Equation (8) by IV method using the inverse of the square root of the number of observations as an instrument and obtained similar estimation results to those in Table 7, with slightly lower statistical significance of β¹. We thank the referee for his/her suggestion on this point.

Additional information

Funding

This research work was financially supported by grants-in-aid for scientific research from the Ministry of Education, Culture, Sports, Science and Technology of Japan (Nos. 23243032 and 15H01849).

Notes on contributors

Ichiro Iwasaki

Ichiro Iwasaki is a professor at the Research Division of Comparative and World Economics, Institute of Economic Research, Hitotsubashi University, Kunitachi, Tokyo, Japan.

Akira Uegaki

Akira Uegaki is a professor of economics at the Department of Economics, Seinan Gakuin University, Fukuoka, Japan.