Understanding how inflation responds to real economic activity over the business cycle remains a central, unresolved question in macroeconomics and monetary economics. The Phillips curve is the tool economists employ to describe this relationship. In its modern (New Keynesian) guise, the Phillips curve posits that inflation depends positively on both a measure of the intensity of economic activity and expected future inflation, and it is also affected by cost-push factors:
Inflation = κ × (Economic activity) + β × (Expected Inflation) + Cost-push shocks,
where the cost-push term captures anything that raises firms’ desired markups — oil-price shocks and tariffs, for example. A key parameter for monetary policy is κ, the slope of the Phillips curve, which measures inflation’s elasticity to real-activity fluctuations. Its magnitude matters: a steeper curve implies more aggressive tightening in expansions and sharper easing in downturns to keep inflation on target.
A large body of research highlights the challenges related to the identification of κ, especially when only aggregate data are available. Some of the best estimates available in the literature (e.g. Rotemberg and Woodford 1997 and Hazell et al. 2022) find an astonishingly small estimate of the slope (κ ≤ 0.02), which suggests a very weak relationship between inflation and economic activity. The view that ‘the Phillips curve is flat’ has become the conventional wisdom, at least for the pre-pandemic period.
In a recent paper (Gagliardone et al. 2025a), we use pre-pandemic data to revisit the economics of the Phillips curve and reach a different conclusion: fluctuations in inflation are tightly linked to fluctuations in economic conditions, as theory and common sense predict. Put differently, the Phillips curve is not flat. The crux of the issue is which measure of economic activity better captures inflation fluctuations. We show that inflation responds to variation in production costs but is quite insensitive to variation in output. Exploiting rich microdata on prices and production costs, we estimate the pass-through of cost shocks to prices — accounting for the dampening effects due to nominal and real rigidities — and uncover a markedly steeper cost-based Phillips curve. This approach reconciles theory and empirical evidence, showing that the apparent flatness of the conventional formulation of the Phillips curve arises from using output as a noisy proxy for the true driver of inflation.
Which measure of economic activity? Output gap vs. marginal cost
The conventional formulation of the Phillips curve features the output gap — or its labour-market analogue, the unemployment gap — as a measure of real activity. These gaps, defined as percentage deviations of real output or employment from their natural levels, proxy for excess demand, and the corresponding Phillips curve slope captures inflation’s sensitivity to those proxies.
Our approach returns to first principles. In the economics underlying the Phillips curve, firms set prices in response to current and anticipated movements in marginal cost. Hence, in the primitive formulation, the percent deviation of real marginal cost from trend enters the Phillips curve as the relevant real activity variable — instead of the output or unemployment gap (Galí and Gertler 1999, Sbordone 2002). The slope of the primitive (cost-based) Phillips curve captures the elasticity of inflation with respect to real marginal cost.
The economic forces captured by this slope are rooted in the microeconomics of price setting. In a frictionless, perfectly competitive world, firms would adjust prices continuously and pass any change in production costs one-for-one to prices. In reality, posted prices remain fixed (in nominal terms) for some time — even when production costs or demand shift. Firms adjust prices infrequently because doing so is costly (e.g. time, printing, and re-tagging) and recurrent changes tend to alienate consumers and erode their trust. Moreover, firms are not price takers and strategically price their products to maximise profit margins. Thus, they may wait to see if cost changes persist before adjusting and they prefer to revise prices in sync with competitors to avoid losing market share. Each of these distortions reduces the pass-through of cost shocks into prices. The greater are firms’ deviations from the frictionless benchmark, the smaller the slope.
Micro-level identification of the slope of the Phillips curve
Estimating the strength of nominal and real rigidities — and thus the slope of the Phillips curve — using only aggregate time-series data is fraught with identification challenges. To overcome this, we propose a novel empirical approach leveraging firm-level microdata. We exploit detailed Belgian microdata on prices and costs to estimate dynamic pass-through regressions that identify both the degree of nominal and real rigidities from short-run co-movements in firm-level marginal costs and prices. We then use these estimates to recover the Phillips curve slope and compute the implied aggregate pass-through of movements in costs to inflation.
Our results deliver tight, robust parameter estimates. We find a substantial degree of nominal rigidities (prices are fixed three to four quarters, on average) and a meaningful role for strategic price complementarities (reducing the pass-through of marginal cost shocks by about half). These estimates imply an economically significant slope of the marginal cost-based Phillips curve, tightly estimated in the range of 0.05 to 0.07.
Accounting for aggregate inflation
Our estimates of the Phillips curve slope imply a substantial aggregate pass-through from production costs to inflation. To demonstrate this, we construct an aggregate cost index by aggregating producer‐level changes in average variable costs (labour and intermediate inputs) and feed it into our cost‐based Phillips curve. The resulting model-implied inflation series closely tracks observed producer price inflation (PPI), as we show in Figure 1. Quantitatively, fluctuations in production costs alone explain about 70% of the variation in inflation. The correlation coefficient between the two series exceeds 0.8.
Figure 1 Aggregate inflation dynamics through the lens of the cost-based Phillips curve
Notes: The black line represents the time series of manufacturing PPI in Belgium. The red line is the model-implied manufacturing PPI obtained feeding an aggregate cost index to a cost-based Phillips curve. Through the lens of our model, the difference between the black and red lines is the component of inflation due to cost-push shocks.
Reconciling the steep cost-based Phillips curve with the flat output-based Phillips curve
While production costs feed directly into pricing decisions, deviations in output (or unemployment) may bear only a loose relationship to inflation. Indeed, the conventional Phillips curve in Equation (1) holds only if marginal cost and the output gap move proportionally — an assumption that requires, among other things, perfectly flexible wages. When these conditions fail, the output gap becomes a poor proxy for real marginal cost, biasing estimates of κ downward. Even if proportionality roughly holds, the output-based slope can be thought as the product of the cost-based slope λ scaled and the elasticity of marginal cost with respect to output:
κ^output = κ^mc × Elasticity of marginal cost to output.
In our pre-pandemic sample (1999–2019), we estimate a very low elasticity of marginal cost with respect to output: firms’ real marginal costs respond only weakly to moderate changes in output. This implies an estimate of the elasticity of inflation to output (the slope of the conventional Phillips curve) between 0.01 and 0.02 — well in line with the estimates found in existing literature. This result shows that the ‘flatness’ of the conventional Phillips curve reflects a weak link between the output gap and marginal cost, not a weak transmission of fluctuations of economic conditions to inflation.
Lessons and the post-pandemic inflation surge
Taken together, our research documents a high pass-through of marginal costs into prices, as evidenced both by our microdata estimates and the ability of the cost-based Phillips curve to replicate aggregate inflation dynamics. A low elasticity of marginal cost with respect to output reconciles the literature’s ‘flat’ output-gap Phillips curve with our finding of strong cost–price transmission. In normal times, two factors likely underlie this low elasticity. First, in the data firms’ marginal cost schedules seem to exhibit near-constant short-run returns to scale, making their costs relatively unresponsive to output fluctuations. Second, wage rigidity likely dampens any aggregate feedback effects that operate through the demand channel.
The post-pandemic inflation surge, however, demonstrates how these relationships can shift sharply under stress.
For one, large shocks — whether from tight labour markets or intermediate-goods bottlenecks — can push firms up against capacity constraints, causing a rapid rise in marginal costs and, in turn, inflation (Boehm and Pandalai-Nayar 2022, Comin et al. 2023, Benigno and Eggertsson 2024). For another, the elasticity of inflation with respect to marginal cost itself can vary in response to large aggregate shocks, as it depends on the frequency of price adjustments. Our pre-pandemic estimates show stable adjustment rates consistent with a Calvo (1981) framework. By contrast, recent data show that the price-adjustment frequency rose significantly throughout the post-pandemic inflation surge, which would lead to a jump in the elasticity of inflation with respect to marginal cost and nonlinear inflation dynamics. State-dependent pricing models, which allow for such shifts, better capture these dynamics, both at the micro and macro level (Bloom et al. 2023, Gagliardone et al. 2025b). We will elaborate on the state-dependent nature of firms’ pricing policies (and the implications for nonlinear inflation dynamics) in a forthcoming companion Vox column.
Authors’ note: The views expressed in this contribution are those of the authors and do not necessarily reflect the views of the National Bank of Belgium, the Eurosystem, or any other institution with which the authors are affiliated.
References
Benigno, P and G B Eggertsson (2024), “The Slanted-L Phillips Curve”, NBER Working Paper 31172.
Bloom, N, P Bunn, P Mizen, O Ozturk, G Thwaites and I Yotzov (2023), “Price-setting in a high-inflation environment”, VoxEU.org, 7 August.
Boehm, C E and N Pandalai-Nayar (2022), “Convex supply curves”, American Economic Review 112(12): 3941–3969.
Comin, D A, R C Johnson and C J Jones (2023), “Supply chain constraints and inflation”, NBER Working 31179.
Forbes, K, J Ha and M A Kose (2024), “Demand versus supply: Drivers of the post-pandemic inflation and interest rates”, VoxEU.org, 9 August.
Gagliardone, L, M Gertler, S Lenzu and J Tielens (2025a), “Anatomy of the Phillips Curve: Micro Evidence and Macro Implications”, American Economic Review (forthcoming).
Gagliardone, L, M Gertler, S Lenzu and J Tielens (2025b), “Micro and macro cost-price dynamics in normal times and during inflation surges”, NBER Working 33478.
Galí, J (2015), Monetary policy, inflation, and the business cycle: an introduction to the new Keynesian framework and its applications, Princeton University Press.
Galí, J and M Gertler (1999), “Inflation dynamics: A structural econometric analysis”, Journal of Monetary Economics 44(2): 195–222.
Hazell, J, J Herreno, E Nakamura and J Steinsson (2022), “The slope of the Phillips Curve: evidence from US states”, The Quarterly Journal of Economics 137(3): 1299–1344.
Rotemberg, J J and M Woodford (1997), “An optimization-based econometric framework for the evaluation of monetary policy”, NBER Macroeconomics Annual 12: 297–346.
Sbordone, A M (2002), “Prices and unit labor costs: a new test of price stickiness”, Journal of Monetary Economics 49(2): 265–292.
