Sacrifice Ratios and Inflation Targeting: The Role of Credibility

• Author: Nicolás De Roux,Marc Hofstetter
• Date: 01 December 2014
• Published date: 01 December 2014
• DOI: http://doi.org/10.1111/infi.12054

Sacrifice Ratios and Inflation

Targeting: The Role of

Credibility

Nicolás De Roux

y

and Marc Hofstetter

z

y

Department of Economics, Columbia University, New York, USA,

and

z

Department of Economics and CEDE, Universidad de los Andes,

Bogota, Colombia

Abstract

Proponents of inflation targeting (IT) claim that it increases the credi-

bility of central banks, which in turn should result in smaller sacrifice

ratios (SRs) –that is, the ratio of output losses to the change in trend

inflation during disinflations. We show that IT does indeed reduce SRs,

but only if the disinflation is long: in a four‐year‐long disinflation, our

estimates suggest that IT reduces SRs by at least 60%. In fast disinfla-

tions, IT does not affect SRs. These results suggest that IT and fast

disinflationsaresubstitutealternativesinenhancingthecredibilityof

disinflationary processes and reducing their costs.

The authors appreciat e the comments and sugges tions from Laurence Ball, t he Editor and

two anonymous referees. Comments from conference participants at the M idwest Macro-

economic Meetings and th e AFSE Conference 2012 are also ackn owledged. We also would

like to thank Nathalie Gonz ález for her excellent research assistance . All remaining errors are

ours.

International Finance 17:3, 2014: pp. 381–401

DOI: 10.1111/infi.12054

© 2014 John Wiley & Sons, Ltd.

I. Introduction

Inflation targeting (IT), first implemented during the early 1990s by New

Zealand and Canada, has become popular among central banks. Nowadays,

close to thirty countries have IT regimes. A large literature has emerged

evaluating different aspects of IT performance. Walsh (2009) surveys the

literature and concludes that, whereas IT appears to have no effect in

advanced economies, it does matter in developing countries. Ball (2010)

arrives at the same conclusion for advanced economies, but claims that

evidence of the effects of IT in developing countries ‘is not yet conclusive’.

A number of papers have addressed an interesting question regarding the

impact of IT on macro‐performance, namely its role as a determinant of

sacrifice ratios (SRs) –that is, output losses during disinflations. On the one

hand, Gonçalves and Carvalho (2009, GC hereafter) claimed that IT econo-

mies suffered smaller output losses during disinflations than non‐targeters in

a sample of OECD countries. They interpreted the result as evidence that IT

makes inflation objectives more credible, and thus allows for less costly

disinflations. On the other hand, Brito (2010) suggests that GC’sfinding is

not robust. In particular, using the same sample of SRs as GC, he shows that

the inclusion of controls for time‐varying effects makes IT an irrelevant

determinant of SRs. Mazumder (2014) arrives at a similar conclusion using a

different sample.

Our paper contributes to this debate in several ways. First, using the same

sample of countries as GC, we construct a new set of disinflations using

Ball’s (1994) identification strategy.

1

Second, having identified a sample of

disinflations, we estimate the SRs àlaBall (1994). Our average SR estimate is

1.7, which is in line with those estimated elsewhere in the literature.

2

Third, Hofstetter (2008), Zhang (2005) and Smith and Senda (2008),

among others, show that Ball’s method of estimating SRs tends to underesti-

mate the disinflation costs. Following their proposals, we construct SRs àla

Zhang and Hofstetter, and show that for this sample, their findings remain

true –that is, Ball’s method tends to underestimate the disinflation costs.

These findings constitute further evidence that output losses precede infla-

tion peaks during disinflations (e.g. Romer and Romer 2004; Hofstetter

2008) and that disinflations have long‐lived effects on eco nomic activit y

(Ball 1997; Zhang 2005; De Roux and Hofstetter 2012; Ball et al. 2013).

1

We explain in the Appendix why our sample ha s some differences from that in GC.

2

In GC (likewise in Br ito), the SRs are not consi stent with other estim ates in the literature

that use this same app roach. For instance , while Ball (1994), Z hang (2005) and Smith and

Senda (2008) obtain average SRs close to 1.5 using Ball’smethod,GC’sfigures are c lose to 6.

See Appendix I and II for details.

382 Nicolás De Roux and Marc Hofstetter

© 2014 John Wiley & Sons, Ltd.