Inflation targeting on unemployment rates: a quantile treatment effect approach

Abstract

This article explores the treatment effects of inflation targeting (IT) on unemployment rates across a large panel of 74 countries over the 1980–2010 period. By addressing the ‘self-selection’ problem of policy adoption via a variety of propensity score matching algorithms, we first find that, on average, IT exerts no discernible effect on unemployment rates in the full sample. However, when the full sample is split into subgroups, the results strongly indicate that the adoption of IT is associated with higher (lower) unemployment rate in the industrial (developing) countries. Further outcome from a novel quantile treatment effects approach points out that the higher the unemployment rate is in the first place, the more harmful (beneficial) the implementation of IT becomes in the industrial (developing) subsample.

Keywords:

• inflation targeting
• unemployment rate
• propensity score matching
• entropy balancing
• quantile treatment effect

JEL Classification:

• E4
• E5
• G1
• G2

Funding

Ho-Chuan Huang gratefully acknowledges financial support from the National Science Council (Taiwan, R.O.C.) [grant number NSC 100-2410-H-032-038-MY2].

Notes

¹ Note that, in Flood and Rose (2010) and Miller et al. (2012), the number of IT countries is increased to 27.

² Please see Miller et al. (2012) for a recent survey.

³ We are unaware of any studies providing QTEs of IT on other key macroeconomic variables either.

⁴ Specifically, the medium radius equals the SD of the estimated propensity scores, and the wide radius and the narrow radius are specified to be equal to double and a half the SD(s), respectively.

⁵ The description below follows closely that of Hainmueller (2012) and Hainmueller and Xu (2013), which provide more technical details about the entropy balancing method.

⁶ For brevity, the whole list of countries is omitted. Interested readers are referred to Table 1 of Lin (2010, p. 196) for more details.

⁷ According to Louviere et al. (2000), the probit regression with pseudo- approximately 0.2 is comparable to an OLS adjusted of 0.7.

⁸ The coefficient of main interest is that of IT dummy, and for brevity, the estimates of intercept are omitted.