The intermittent Phillips curve: finding a stable (but persistence-dependent) Phillips curve model specification

Abstract

We establish that the Phillips curve is persistence-dependent: inflation responds differently to persistent versus moderately persistent (or versus transient) fluctuations in the unemployment rate gap. This persistence-dependent relationship appears to align with business-cycle stages and is thus consistent with existing theory. Previous work fails to model this dependence, thereby finding numerous "inflation puzzles" – e.g., missing inflation/disinflation – noted in the literature. Our specification eliminates these puzzles; for example, the Phillips curve has not weakened, nor was inflation "stubbornly low" in 2019. The model's coefficients are stable, and it provides accurate conditional recursive forecasts through the Great Recession. There are important monetary policy implications.

Post‐COVID inflation dynamics: Higher for longer

“…21 For unemployment, there is no standard alternative benchmark, so we also report the forecast accuracy of an AR(1) model. 22 Our findings-forecast improvements are mainly episodic-are in line with the literature (e.g., Ashley & Verbrugge, 2023;Stock & Watson, 2010), and not surprising, because our unemployment terms only influence forecasts during two portions of the business cycle. 23 For this type of comparison, the UCSV model and the RW model perform poorly.…”

“…3 See also Forbes et al (2022). There are numerous antecedents to this finding in the nonlinear Phillips curve literature, which typically finds that the Phillips curve is "convex"; these are reviewed in Ashley and Verbrugge (2023) (see also the recent findings of Ball et al, 2022). 4 An alternative way to state this is that anchored inflation expectations evidently exert only a modest/"slow" force on inflation.…”

“…Given our aims in the present paper, and the fact that we use quarterly rather than monthly data, we use a modification of the baseline Phillips curve equation of Ashley and Verbrugge (2023). We are ultimately interested in reliable forecasts, so model parsimony was a chief consideration.…”

“…1 Importantly, we carefully model the relationship between unemployment and these inflation components. Recent research (Ashley & Verbrugge, 2023;Verbrugge & Zaman, 2023a) has demonstrated benefits to specifying the relationship between aggregate inflation and the unemployment rate as frequency-dependentthat is, to using a specification that allows inflation to respond differently to business-cycle movements in the unemployment rate than it does to low-frequency movements or to transient movements. 2 This previous work has established that the response of aggregate inflation to the various frequency components of the unemployment rate gap (i.e., to the persistent, moderately persistent, and transient components) varies not only by persistence component but also by the sign of these components.…”

Out of Bounds: Do SPF Respondents Have Anchored Inflation Expectations?

“…Without a credible commitment to an inflation target, uncertainty about the recession and its consequences for inflation, disagreement about the effects of unconventional monetary policy at the zero lower bound (ZLB), and sizeable revisions to policymakers' forecasts may have contributed to weak anchoring (Alessi et al. 2014, Ashley and Verbrugge 2022).…”