The dissent voting behaviour of central bankers: what do we really know?

Abstract

We examine the determinants of the dissent in central bank boards voting records about monetary policy rates in the Czech Republic, Hungary, Sweden, the UK and the US. In contrast to previous studies, we consider 25 different macroeconomic, financial, institutional, psychological or preference-related factors jointly and use Bayesian model averaging (BMA) to formally assess the attendant model uncertainty. We find that the rate of dissent is between 5% and 20% for the examined central banks. Our results suggest that most of the examined regressors, including factors that capture the effects of inflation and output, are not robust determinants of voting dissent. This result suggests that unobserved characteristics of central bankers and different communication strategies drive the dissent, rather than the level of macroeconomic uncertainty.

Keywords:

• monetary policy
• voting record
• dissent

JEL Classification:

• E52
• E58

Funding

We acknowledge support from the Grant Agency of the Czech Republic P402/12/1993.

Notes

1. ¹See Hoeting et al. (1999), for the statistical theory underlying the BMA approach. Montgomery and Nyhan (2010) provide an example in the field of political science that illustrates how BMA can be useful in distinguishing among the importance of various collinear factors for explaining the occurrence of civil wars.

2. ²The data are available on the website of the Central Bank Communication Network, http://www.central-bank-communication.net/links/, and can be freely used by other researchers.

3. ³This dummy is perfectly collinear with the variable team; therefore, it is possible to only include it to address the case of the US Fed. Our sample period includes the Greenspan and Bernanke chairmanships; thus, only one dummy variable is introduced.

4. ⁴Interestingly, the absence of board members from monetary policy meetings is relatively common. Clearly, board members may be absent because of illness; they also may miss policy meetings because they are obliged to represent their bank in meetings with international organizations, such as the European Union or the International Monetary Fund.

5. ⁵For example, the number of regression models for the US Fed is 2⁴⁷; thus, more than 140 trillion regressions would need to be estimated to assess all of these models.

6. ⁶The conditional posterior signs are not presented in this analysis for the sake of brevity, but these results are available from the authors upon request.

7. ⁷Raftery (1995) suggests that the PIP of a regressor must be greater than 0.5 for that regressor to be regarded as an effective determinant of regressand.

8. ⁸Our Hungarian sample starts only in October 2005, so we did not estimate the BMA regressions for Hungary on this restricted ‘no crisis’ sample.