Few assumptions in modern macroeconomics carry as much weight as the one Guillermo Calvo introduced in 1983. Each period, a firm receives a random signal permitting it to reset its price. The probability is fixed, independent of how long the current price has prevailed or how far it has drifted from its optimum. From this modest device flows the New Keynesian Phillips curve, the welfare-based loss functions of Rotemberg and Woodford, and much of the quantitative apparatus central banks use to evaluate policy rules.

Yet the assumption has never been defended as a realistic description of price-setting. Firms do not wait for a metaphorical lottery before repricing. They adjust when the gap between their current and desired price becomes costly enough to justify action. Micro evidence assembled since Bils and Klenow (2004) has made this increasingly difficult to ignore. The inflation surge of 2021–2023, during which the frequency of price changes rose sharply, gave the critique fresh urgency.

The persistence of Calvo pricing therefore poses a genuine methodological question. Why does a profession committed to microfoundations retain an assumption it knows to be counterfactual? The answer involves a deliberate trade-off between tractability and fidelity, and the terms of that trade-off shift with the size of the shocks under consideration. Those terms, and what state-dependent alternatives deliver in exchange for their complexity, matter for anyone designing policy in an environment where large shocks are no longer hypothetical.

The core advantage of Calvo's assumption is that it eliminates the need to track the cross-sectional distribution of prices. Adjustment opportunities arrive independently of each firm's history, so the firms that reset in any period are a random sample of the population. The aggregate price index then evolves according to a simple recursion: today's index is a weighted combination of last period's index and the common reset price chosen by all adjusters.

Log-linearizing around a zero-inflation steady state yields the familiar New Keynesian Phillips curve, in which current inflation depends on expected future inflation and real marginal cost. In the baseline case the slope coefficient is (1−θ)(1−βθ)/θ, which maps directly to the probability of non-adjustment θ. Researchers thus have a single parameter that summarizes nominal rigidity. Combined with a dynamic IS equation and a policy rule, the result is a three-equation system that can be analyzed with pencil and paper.

This tractability extends to welfare analysis. Woodford showed that a second-order approximation to household utility under Calvo pricing produces a loss function quadratic in inflation and the output gap. Inflation's weight in that function arises from the relative price dispersion it generates. The derivation underpins the theoretical case for inflation targeting and for the analysis of optimal policy under commitment versus discretion. Few modeling choices have generated so much normative structure from so little machinery.

The assumption also integrates cleanly with richer environments. Medium-scale DSGE models in the Smets–Wouters and Christiano–Eichenbaum–Evans tradition layer indexation, wage rigidity and investment frictions atop Calvo pricing without losing their recursive structure. Heterogeneous agent New Keynesian models typically confine heterogeneity to households so that the supply block remains a single equation. Rotemberg's quadratic adjustment-cost formulation delivers an identical first-order Phillips curve, which reinforces the sense that the specific microfoundation matters little near steady state.

That last observation is the crux. Calvo's dominance rests on an implicit claim: for small fluctuations, the aggregate implications of nominal rigidity are robust to how that rigidity is microfounded. Within the regime of local approximations, this claim holds up better than critics often concede. The difficulty is that the claim is conditional. The conditions under which it fails are precisely the episodes in which policy decisions carry the greatest consequence.

Takeaway

Calvo pricing persists not because economists believe it, but because it compresses an entire price distribution into one parameter. That compression is valid only as long as shocks stay small.

Because adjustment probabilities are fixed, Calvo firms cannot respond to the magnitude of a shock by repricing more often. A monetary expansion of fifty percent moves the same fraction of firms as one of half a percent, and the output response simply scales linearly. This is a strong and testable restriction. The evidence rejects it whenever shocks become large enough to make the extensive margin of adjustment visible.

Episodes of high inflation are instructive. Gagnon's study of Mexico during the 1990s found that the frequency of price changes was fairly stable at low inflation but rose substantially once inflation exceeded roughly ten to fifteen percent. Alvarez, Beraja, Gonzalez-Rozada and Neumeyer documented a similar pattern in Argentina's passage from hyperinflation to stability, with adjustment frequency rising toward the point where prices behave almost flexibly. In Calvo, by contrast, frequency is a structural constant.

Trend inflation exposes further problems. Ascari and Sbordone showed that once steady-state inflation is positive, the Calvo Phillips curve acquires additional forward-looking terms and its slope flattens. The determinacy region of standard Taylor rules also shrinks considerably. Firms holding prices set long ago suffer severe relative price erosion, so price dispersion and its implied welfare costs rise steeply with trend inflation. These results feed directly into debates about the optimal inflation target.

The predicted dispersion is itself contested. Nakamura, Steinsson, Sun and Villar extended US micro price data back to the late 1970s. They found that the absolute size of price changes did not rise during the Great Inflation, contrary to the Calvo prediction that inefficient dispersion would increase sharply. If firms responded by adjusting more often, the welfare cost of moderate inflation may be considerably smaller than Calvo-based loss functions imply.

The 2021–2023 inflation episode offered a contemporary test. Analyses of US CPI microdata, including work by Montag and Villar, show that the frequency of price changes rose markedly as inflation climbed, while the average size of changes moved comparatively little. Inflation rose faster than a flat, Calvo-style Phillips curve would have predicted, which is consistent with the steepening that state-dependent models anticipate. A model calibrated to quiet times systematically misreads periods of turbulence.

Takeaway

A model whose adjustment frequency is a constant cannot see the margin that matters most in a crisis: how many firms choose to move at all.

State-dependent pricing models begin from a different primitive. Firms pay a fixed cost to change prices and therefore adjust only when the benefit exceeds that cost. The Ss framework of Sheshinski and Weiss formalizes this as an inaction band around the optimal price. Firms let their relative price drift until it hits a boundary, then reset discretely. Adjustment timing becomes endogenous, and the composition of adjusters is no longer random.

That non-randomness is the source of the selection effect. The firms whose prices are furthest from optimal are the ones that adjust, and they adjust by large amounts. Caplin and Spulber showed that in an extreme case, money becomes fully neutral in aggregate despite substantial individual rigidity. Golosov and Lucas calibrated a menu cost model to US microdata and found real effects of monetary shocks far smaller than in a Calvo model with matched frequency.

Subsequent work recovered meaningful non-neutrality by matching more features of the micro evidence. Klenow and Kryvtsov documented that price changes are large on average yet include many small changes, producing a distribution with fat tails. Midrigan showed that multiproduct firms with economies of scope in repricing, together with leptokurtic idiosyncratic shocks, weaken selection considerably. Nakamura and Steinsson's Calvo-plus model similarly mixes occasional free adjustments with costly ones.

Alvarez, Le Bihan and Lippi sharpened this into a sufficient statistic. Across a broad class of models, the cumulative output response to a small monetary shock is proportional to the ratio of the kurtosis of price changes to their frequency. For a given frequency, Calvo sits at one end and delivers maximal non-neutrality, while Golosov–Lucas sits at the other. The modeling choice becomes an empirical question, disciplined by observable moments rather than convenience.

Recent work also clarifies when the choice matters. Auclert, Rigato, Rognlie and Straub show that for small shocks, menu cost models generate Phillips curves closely resembling those of time-dependent models with suitably chosen adjustment hazards, which vindicates Calvo-style reasoning locally. Blanco, Boar, Jones and Midrigan emphasize the nonlinearity: large shocks raise adjustment frequency and steepen the Phillips curve. State dependence is therefore essential for large shocks and only modestly consequential for small ones.

Takeaway

The real distinction is not between realistic and unrealistic models but between models whose implications depend on shock size and those that assume they do not. Kurtosis divided by frequency is a useful lens for telling them apart.

Calvo pricing became the workhorse because it delivers a remarkable amount of analytical structure from a single parameter: a linear Phillips curve, a quadratic welfare criterion and a recursive aggregate. Its known flaws are real, but they bind most tightly outside the local regime for which it was designed.

For policy design, the implication is a division of labor rather than a replacement. Calvo-based models remain defensible for routine analysis of small disturbances and for normative questions where tractability enables insight. When shocks are large, when trend inflation is elevated, or when the cost of inflation itself is at issue, state-dependent frameworks offer more reliable guidance.

The more lasting lesson concerns model choice itself. Tractability is a legitimate virtue, but it carries an implicit domain of validity. Central banks that make that domain explicit, for instance by monitoring the frequency of price adjustment as a real-time indicator of regime change, will be better positioned when the next large shock arrives.