Evaluating inflation persistence considering model uncertainty and structural break

ABSTRACT

There is little consensus about the historical change in inflation persistence. One of the reasons is that researchers cannot choose a uniform model structure due to model uncertainty. In this study, a model to investigate inflation persistence is proposed. Unlike previous studies, I try to overcome model uncertainty by being agnostic about the structure of the model. Estimation results indicate that inflation has become less persistent over time, implying that inflation persistence is related to the way how monetary policy is conducted.

KEYWORDS:

• Inflation persistence
• model uncertainty
• Gibbs sampling

JEL CLASSIFICATION:

• E31
• C22
• C11

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ The examples of lag structure are as follows: Brainard and Perry (2000) use eight quarters, Levin and Piger (2004) do six quarters, Pivetta and Reis (2007) do three quarters, Cogley and Sargent (2005) do two quarters, and Beechey and Österholm (2012) do one quarter of lag.

² Since this study is focusing on the relation between inflation persistence and monetary policy regime, I do not allow for stochastic volatility.

³ For example, if ${\delta }_1=1$ and ${\delta }_j=0$ for $j=2,\cdots ,k$ (in other words, $\delta =[1,0,\cdots ,0{]}^{\prime }$), then the model is reduced to $AR(1)$ process.

⁴ According to the model, policymaker’s relative preference to stabilize inflation is the key determinant of the time for which inflation disequilibria persist, which is inflation persistence.

⁵ Empirical evidence on the regime shifts in monetary policy is found at Clarida, Galí, and Gertler (2000) and Boivin and Giannoni (2006).

⁶ Considering that Brainard and Perry (2000), Levin and Piger (2004), and Pivetta and Reis (2007) select eight, six, and three quarters of lag, respectively, twelve quarters of maximum lag length is regarded sufficient.

⁷ To check the robustness of the result, ${\lambda }_j=0.4$ and ${\lambda }_j=0.6$ are tried as well. The results are not very sensitive to the choice of this prior distribution.

⁸ I omit the figures of the other economies to save space. The implications are not different from the case of the US. I will provide the other figures upon a reader’s request.