The recent high-inflation episode generated renewed interest among monetary policymakers and academics about the slope of the Phillips curve and potential non-linearities in price-setting behaviour (Forbes et al. 2021, Gagliardone et al. 2025, Benigno and Eggertsson 2023, Powell 2023, Mann 2022, Schnabel 2024). However, there are multiple empirical challenges with the identification of the Phillips curve using aggregate data (e.g. Beaudry et al. 2025, Mavroeidis et al. 2014, Tenreyro and McLeay 2018). These include accurately controlling for supply-side shocks, inflation expectations, monetary policy responses, and broader measurement challenges. In a new research paper (Bunn et al. 2025), we use data from large surveys of UK and US firms to study how prices respond to firm-level demand shocks and, separately, cost shocks. We develop a general equilibrium model with menu costs to rationalise the micro-level estimates and analyse the implications for the aggregate Phillips curve.
Data
We use data from the Decision Maker Panel (DMP) survey for the UK, and data from the Survey of Business Uncertainty (SBU) in the US. The DMP is a large and representative survey of CFOs in UK businesses with ten or more employees. The survey is carried out online and receives close to 2,500 responses each month. In the survey, firms are asked how the average price that they charge has changed over the last year. Firms are also asked how they expect their own prices to change over the next year. 1 The US SBU is organised in a similar manner, run by the Federal Reserve Bank of Atlanta and collecting monthly data from around 1,000 firms.
Firm price responses to demand shocks
We use three distinct empirical approaches to study how firm prices respond to demand shocks (e.g. due to changes in consumer preferences), and to test for non-linearities. First, we run a randomised survey experiment that asks firms how they would adjust their prices in response to a series of hypothetical sales volume shocks. Firms are randomised into one of four shock scenarios – ±5%, ±10%, ±15%, and ±20% – and asked to respond to both positive and negative shock outturns. 2 The identical question is asked to UK firms in the Decision Maker Panel and US firms in the Survey of Business Uncertainty.
Figure 1 shows the average price response to these hypothetical demand shocks, for UK firms (Panel A) and US firms (Panel B). Across both countries, we find strong evidence of a convexity, with firms raising prices in response to positive shocks, whereas the response to negative shocks is around zero. Further tests confirm that the non-linearity is statistically significant in both of these samples.
Figure 1 Nonlinear price responses to hypothetical sales volume shocks
Notes: This figure reports responses to the question ”Suppose that your business’s sales volume over the next 12 months is X per cent higher/lower than you currently expect. How would that affect the average price you charge, relative to what you currently expect?” Panel A is based on 6,394 observations from 2,486 UK firms in the Decision Maker Panel. Panel B is based on 787 US firms which responded to the June 2024 survey wave of the Survey of Business Uncertainty.
The randomised survey approach provides strong causal evidence that firm prices respond more to positive demand shocks compared with negative shocks. However, one limitation is that they are based on hypothetical scenarios, which do not necessarily reflect how firms would respond in real life. To address this concern, in the second empirical exercise we calculate firm-level price growth and sales growth forecast errors using actual data on prices, sales, and expectations of firms. This leverages the strong panel dimension of the DMP survey, comparing firm realisations for price growth and sales growth to year-ahead expectations reported a year earlier. The results again suggest a significant asymmetry, with positive sales growth forecast errors associated with more than two times larger price growth forecast errors compared with negative sales growth forecast errors.
Finally, in the third empirical exercise we analyse the response of firm prices to the Covid-19 shock. The impact of Covid-19 on demand is estimated using specific questions about the impact of the pandemic on firm sales asked over 2020-2022. Firm sales declined sharply in 2020Q2 by over 30%, with substantial heterogeneity across firms and sectors. Once again, we find that the effect on inflation was asymmetric: positive demand shocks from Covid raised inflation at the firm level by five times more than negative Covid demand shocks reduced it. This demand asymmetry explains why inflation did not fall by more during the first year of the pandemic.
Overall, multiple empirical approaches using firm data from both the US and the UK point to a significant non-linear response of prices to demand shocks. We extend these results in two ways. First, we show that the convex response to demand shocks is pronounced when inflation is high. Firms with average price growth above 4% exhibit a clear non-linearity in price responses across all three exercises above; firms with price growth below 4% have a linear response. Second, we show that the non-linearities are a short-term phenomenon: over longer time horizons (more than three years), the responses to positive and negative sales growth converge.
Firm price responses to cost shocks
Going beyond demand shocks, we also analyse how firm prices respond to cost shocks. These could be changes in firm input costs, like energy or labour. This is important for at least two reasons. First, the standard New Keynesian Phillips curve models prices at the firm level as a mark-up over marginal cost. Marginal costs are notoriously difficult to measure, and the literature therefore usually focuses on average or unit costs, or estimates derived from assumed production functions. Second, the inflation episode over 2022-2025 has been driven in large part by cost increases.
In Figure 2, we present evidence from two separate empirical exercises on the response of firm prices to costs. Panel A shows evidence from a randomised survey experiment on the response of firm prices to hypothetical unit cost shocks (using the same format as the sales volume shock questions). The pass-through of positive cost shocks is close to three times as strong as that to negative shocks. In Panel B, we consider price growth and unit cost growth forecast errors; again, we find that the response to positive cost errors is around twice as strong as the response to negative errors. Quantitatively, the impact of cost shocks on prices is also much stronger than the impact of demand shocks.
Figure 2 Nonlinear price responses to unit cost
Notes: Panel A reports responses to the question, “Suppose that your business’s unit costs over the next 12 months are X per cent higher/lower than you currently expect. How would that affect the average price you charge, relative to what you currently expect?” The scatter plot is based on 3,728 observations from 1,864 UK firms in the Decision Maker Panel. Panel B shows the relationship between unit cost growth forecast errors and annual price growth forecast errors. The scatter plot is based on 2,260 observations for 1,364 UK firms in the Decision Maker Panel.
We use the responses to hypothetical demand and cost shocks along with data on real sales growth and unit cost growth to calculate the importance of nonlinearities for the increase in inflation over 2022 (Table 1). Between 2018 and 2022, price growth among firms increased by 4.4 percentage points on average. This quantification yields two important insights. First, cost shocks were a much more important driver of the rise in firm price growth (by 2.7-3.8 percentage points in the linear vs. non-linear specifications) in 2022 compared to demand shocks (0 to 0.1 percentage points). Second, taking non-linearities into account helps account for an additional 1.2 percentage points (or around one-fourth) of the total increase in price growth. Thus, non-linearities in price-setting are not only significant in a statistical sense, but also quantitatively for understanding recent inflation dynamics.
Table 1 Quantifying the contribution of nonlinearities for the 2022 rise in inflation
A model of state-dependent price setting
To rationalise these findings, we build a model of price-setting behaviour based closely on Nakamura and Steinsson (2010). The model features menu costs, trend inflation, and diminishing returns at the firm level. Menu costs create an inaction region for firms in which prices do not change in response to cost or demand shocks. Positive trend inflation generates an asymmetry in the distribution of firms within the inaction zone. The non-linearity emerges from trend inflation pushing firms closer to their price increase thresholds. Figure 3 presents firm-level Phillips curves using simulated data from the model, for different rates of trend inflation. With positive trend inflation (right panel), the model is able to reproduce the convex relationship between demand shocks and prices; for zero or negative inflation, the convexity is absent, or even reverses. Furthermore, in the paper we show that the model generates a convex aggregate Phillips curve in response to demand shocks.
Figure 3 Simulated firm-level Phillips curves at low and high inflation rates
Notes: This figure presents simulated firm-level data from the model outlined in Section 6 of Bunn et al. (2025). The left panel shows a simulation with -2% trend inflation in the model, the middle panel shows a simulation with 0% trend inflation in the model, and the right panel shows a simulation with 2% trend inflation in the model (our baseline setup). In each panel, the x-axis is the one-month change in idiosyncratic demand and the y-axis is the annualised one-month inflation rate.
Conclusion
In this column, we present new evidence from UK and US firms that prices respond asymmetrically to both demand and cost shocks, rising by more following positive shocks. This non-linearity can help explain around one-fourth of the rise in price growth in 2022 compared to pre-pandemic and is rationalised in a state-dependent model of price-setting. Our results highlight the importance of taking asymmetries in pricing behaviour into consideration, both in empirical and theoretical work.
Source: cepr.org
