Before the Covid-19 pandemic, a large body of research suggested that the Phillips curve – which plots the inverse relationship between unemployment and inflation – exhibited a relatively flat slope. Consequently, it was widely believed that a temporary tightening of labour markets would have only a modest impact on inflation.
This perspective has come under further scrutiny following the high and persistent inflation from 2021 to 2023. During this period, inflation surged alongside high job vacancies and low unemployment – reigniting a longstanding debate on the ‘nonlinearity’ of the Phillips curve.
Recent research has provided empirical evidence suggesting that the Phillips curve may exhibit strong nonlinearity, with inflation accelerating rapidly when the vacancy-to-unemployment ratio exceeds unity – that is, it is more than one (Benigno and Eggertsson, 2023; Gitti, 2024). Figure 1 replicates the typical raw US data, both at the national level and across metropolitan statistical areas (MSAs).
Figure 1: US labour market tightness and inflation, raw data, 2000-23
Source: Beaudry et al (2025)
Note: Each dot represents a quarter (Panel A) or a quarter-city (Panel B). Labour market tightness is measured as logθ, where θ=V/U. Inflation is quarter-to-quarter CPI core inflation (annualised). Light grey dots correspond to logθ<0, dark grey dots correspond to logθ≥0. The black line is the fitted cubic relation between inflation π and logθ, dotted lines delimit the 95% confidence interval. Sample is 2000Q1-2023Q4 for Panel A and 2000Q3-2024Q3 for Panel B.
This perspective suggests that labour market tightness played a crucial role in the inflationary surge observed in late 2021 and 2022, and that the subsequent easing of labour market conditions was instrumental in bringing inflation back down. More specifically, had the supply shocks induced by Covid-19 and related policies not been accompanied by labour market tightness, inflationary pressures would probably have been significantly lower, potentially reducing the necessity for an aggressive monetary policy response.
Given the importance of correctly interpreting recent inflationary dynamics, we investigate the robustness of the empirical evidence in support of a nonlinear Phillips curve (Beaudry et al, 2025). Our findings suggest that the evidence is fragile, as reasonable alternative model specifications and data choices yield markedly different conclusions.
The stakes of this debate are high. An unwarranted prior belief in a highly nonlinear Phillips curve could result in sub-optimal policy decisions.
For example, consider a scenario where there is a supply-side shock of similar magnitude to the Covid-19 crisis, but without the accompanying labour market tightness. A misinterpretation of the recent inflationary episode could lead policy-makers to underestimate the necessity of a forceful monetary response, under the erroneous assumption that inflation will revert to target levels automatically.
At the same time, if the apparent nonlinearity in recent data is instead a reflection of de-anchored short-run inflation expectations, failure to recognise this could lead to an underestimation of the risk of inflation persistence in the absence of a robust monetary response. Both cases would be dangerous for price stability.
Our work highlights some key empirical patterns that underscore both the plausibility and fragility of the nonlinear Phillips curve.
Figure 1 illustrates a pronounced nonlinear relationship between inflation and the vacancy-to-unemployment ratio. But this observation does not establish causality.
From a Phillips curve perspective, the inflation surge following the pandemic could have been driven by multiple factors, including cost-push shocks, inflation expectations and labour market tightness. Disentangling these effects is challenging, given their simultaneous movements during the post-pandemic period.
To visualise this problem, Figure 2 presents the evolution of the vacancy-to-unemployment ratio (Panel A) and inflation (both actual and expected), as measured by the Michigan Survey of Consumers (Panel B).
Figure 2: US labour market tightness, inflation and inflation expectations, 2000-23
Panel A: Labour market tightness
Panel B: Inflation and inflation expectations
Source: Beaudry et al (2025)
Note: Labour market tightness is measured as θ=V/U. Inflation π is quarter-to-quarter CPI core inflation (annualised). Expectations are one-year-ahead and obtained from the Michigan Survey of Consumers. Grey areas represent quarters with θ≥1. Sample is 2000Q1-2023Q4.
The data reveal that inflation expectations tended to be elevated precisely when labour market tightness was high, suggesting that inflation dynamics may have been driven, at least in part, by short-run de-anchoring of expectations in response to supply shocks.
To explore whether the observed nonlinearity is confounded by inflation expectations, we use a framework known as the expectations-augmented Phillips curve. The importance of inflation expectations was a central lesson from the inflationary episode of the 1970s.
In Figure 3, we re-examine the data, plotting inflation minus 0.99 times one-year-ahead inflation expectations (obtained from the Michigan Survey of Consumers) against the log of labour market tightness, as a New Keynesian Phillips curve would suggest. Results are displayed for both aggregate and MSA-level data.
Figure 3: Inflation and labour market tightness with expectations, 2000-23
Source: Beaudry et al (2025)
Notes: Each dot represents a quarter. Dark dots indicate observations with logθ≥0 and light dots observations with logθ<0. Inflation is quarter-to-quarter CPI core inflation (annualised). The measure of inflation expectations π_(t+1)^e is the national Michigan Survey of Consumers one year-ahead inflation expectation (adjusted to obtain one quarter-ahead expectation). The black line is the fitted cubic relation between the y-axis variable and logθ, dotted lines delimit the 95% confidence interval. Sample is 2000Q1-2023Q4.
Once inflation expectations are taken into account, the previously striking nonlinearity observed in Figure 1 largely disappears. In Panel A, representing aggregate data, nonlinearity also dissipates when alternative household and firm-based expectation measures are used. But the nonlinearity persists when expert-based inflation expectations are used.
In Panel B, which examines city-level cross-sectional data, the nonlinearity vanishes for any measure of inflation expectations when time-fixed effects are included.
Our empirical investigation suggests limited support for a nonlinear Phillips curve in cross-sectional data. In time-series data, controlling for inflation expectations largely accounts for the apparent nonlinearity. The only instance in which nonlinearity appears is in aggregate data when expert expectation measures – those that performed poorly in forecasting inflation post-2020 – are used.
Two competing interpretations of recent inflationary dynamics emerge from this analysis, and distinguishing between them remains challenging. Policy-makers should exercise caution and refrain from prematurely adopting either view until more conclusive data or methodological advancements resolve the debate.
On one hand, there is the view that – in addition to supply shocks – labour market tightness played a very important role in generating high inflation post-2020 because the Phillips curve is highly nonlinear, and the labour market was very tight.
On the other hand, there is the view that the Phillips curve is likely to be flat and that a de-anchoring of short-run inflation expectations following the supply shocks probably played a central role in the inflationary outcomes. Based on our previous research, we tend to favour that view (Beaudry et al, 2024).
The main difficulty is differentiating between these two views, and weighting their respective merits, relates to the difficulty of knowing how to control properly for inflation expectations. Misinterpretation could lead to significant policy mis-steps, particularly in stagflationary scenarios.
