When Recessions Leave a Mark: How Scarring Changes Capacity
This chapter examines the mechanism that makes an unemployment benchmark history-dependent: a cyclical disturbance can alter the capacity against which later output and unemployment are judged. The useful question is not whether every recession causes permanent damage. It is how a past deviation can persist in the model, how long it lasts, and why the Federal Reserve paper keeps that history-dependent capacity separate from the supply-driven star. That distinction lets us discuss persistence without turning a model estimate into a claim of inevitable economic decline.
From a cyclical shock to a second capacity path
Start with the baseline picture. A cyclical shock moves output or unemployment away from the supply-driven trend. A demand shortfall, for example, can reduce production and hiring even though the economy’s underlying ability to produce has not changed immediately. In a conventional trend-cycle decomposition, the shock belongs to the cycle. The trend remains the reference point, and the distance between the observed economy and that trend is the gap.
The scarring framework adds a feedback path. Past cyclical deviations can influence productive capacity, so the trend that matters for current production is no longer necessarily the same as the purely supply-driven trend. The paper calls the broader object productive capacity. It includes current supply forces, but also a component that responds to earlier deviations of output from the supply-driven potential.
This does not mean that the model simply relabels every downturn as a supply shock. It separates two ideas that are easy to confuse. The first is the cause of the initial movement: a cyclical disturbance can begin as a demand-driven change in the cycle. The second is the persistence of the consequences: part of that disturbance can alter the path of capacity for a time. A shock can therefore be cyclical in origin and supply-relevant in its later effects.
The timing matters. The scarring term in the Federal Reserve model is linked to a lagged cyclical deviation rather than to the current observation alone. A weak period leaves information in the state of the model that affects later capacity. The current cycle still exists, but it no longer carries all of the persistent movement in output. Capacity absorbs some of the history, and the remaining cycle can be smaller and less drawn out.
The same logic applies to unemployment, though with a separate labor-market capacity. The paper defines a supply-driven unemployment trend and a broader unemployment trend influenced by past deviations of unemployment from its supply-driven path. If a cyclical downturn raises unemployment, the model can allow that history to affect later labor-market capacity. The direction and persistence of the effect are estimated rather than assumed to be a permanent increase in the natural rate.
That last qualification is essential. A higher observed unemployment rate during a recession is not itself evidence that the natural rate has risen. It could be a temporary cyclical gap. The scarring model asks whether the joint behavior of output, unemployment, wages, and prices is better explained if some of the past movement changes the capacity path. The answer is constrained by the model’s equations and data, not decided by the headline unemployment series.
The two capacity paths serve different interpretive purposes. The purely supply-driven star answers a policy-oriented counterfactual: where would the economy be if cyclical effects were removed but its underlying supply conditions were retained? The broader, history-dependent capacity helps explain current inflation and wage pressure because firms and workers respond to the capacity that actually exists after past disturbances have worked through the economy. These paths can be close, but they need not coincide.
The sequence shows why “scarring” is a mechanism rather than a metaphor. The initial shock enters the cycle. The cycle changes observed activity. A lagged deviation then feeds into a second capacity path, while the supply-driven star remains a separate reference. The later inflation and wage readings are interpreted relative to the capacity that the model says is relevant. At the end of the process, the gap depends on which path is used as the denominator.
This also explains why a scar need not move every measure in the same direction. A negative cyclical shock can lower the history-dependent productive path, while the supply-driven star follows a different trajectory. The observed economy may then look closer to the broader capacity than to the supply-driven star. For inflation, that can mean a relatively small capacity gap even when output is far below an earlier trend. For policy, the distance from the supply-driven star may still be large. The two comparisons answer different questions.
Temporary scarring is not permanent hysteresis
The paper distinguishes temporary scarring from hysteresis by asking whether the history-dependent component eventually returns toward the supply-driven path. In the temporary case, a cyclical disturbance has persistent effects, but those effects decay. Productive capacity can be away from the purely supply-driven trend for years without being displaced permanently. In the hysteresis limit, the history-dependent component follows a unit-root process: past cyclical deviations permanently alter the capacity path.
The distinction is not semantic. It changes what a later recovery means. Under temporary scarring, a return toward the supply-driven star remains possible as the scar fades. Under hysteresis, the economy’s capacity has been permanently rewritten by the earlier cycle; recovery is measured against a new path. The observed unemployment rate could return to its previous level while the underlying productive benchmark has changed, or it could remain different because the benchmark itself has moved.
The Federal Reserve paper’s theoretical framework makes the two cases a continuum. When the scarring parameters are absent, the productive trend is independent of demand factors. When both the level and speed terms contribute and the history-dependence coefficient is below the hysteresis boundary, demand shocks have persistent but ultimately temporary effects. When the coefficient reaches the boundary, the effect becomes permanent. The model can therefore represent more than the binary choice between “nothing changed” and “everything changed forever.”
The preferred empirical specification points toward temporary scarring. Its posterior mean history-dependence coefficient is about 0.58 for productive capacity, with a 68 percent credible interval from 0.35 to 0.79. For labor-market capacity, the corresponding mean is about 0.47, with an interval from 0.21 to 0.71. The authors describe these magnitudes as persistent but not extremely persistent. They are evidence that history matters within the estimated model, not a schedule that says exactly how much any future recession will change the natural rate.
The paper is more cautious about unemployment permanence than a casual use of “hysteresis” might suggest. In its unemployment block, the coefficient on the level of the unemployment gap is constrained to be strictly negative. That assumption rules out unemployment hysteresis in the model. A negative coefficient on the speed of the unemployment gap can still produce temporary unemployment scarring, but the preferred framework does not let a cyclical unemployment shock create a unit-root, permanently higher unemployment trend.
Output is treated differently. The paper estimates a scarring model for productive capacity and separately examines the limiting output-hysteresis specification. The hysteresis case is not the preferred conclusion simply because it is more dramatic. It is a boundary case in which the history-dependent component never returns to the supply-driven path. The paper compares it with temporary scarring and finds that the scarring specification is preferred in its model evaluation, while the forecasts of the two history-dependent alternatives can be closer for some variables.
This model-specific boundary prevents two opposite mistakes. The first is to assume that a recession cannot affect potential output or the natural rate because those are “structural.” The scarring framework explicitly allows past cyclical conditions to matter. The second is to assume that any persistence proves permanent damage. The preferred estimate is temporary scarring, and the unemployment component does not permit hysteresis. Evidence that a past shock changes the path for a while is not evidence that the old path is unrecoverable.
Why persistence moves out of the cycle
Allowing scarring changes the division of labor between the cycle and the trend. In the benchmark specification, the cycle must explain the persistent, hump-shaped behavior of output and unemployment. That makes the cycle highly persistent: it carries the memory of the recession so that the trend can remain insulated from it. In the scarring specification, part of that memory is stored in capacity. The cycle can react and decay more quickly because it no longer has to account for every medium-run movement in real activity.
This reallocation affects the estimated relationship between slack and inflation. In both specifications, the cyclical component enters the inflation equations. But the scarring specification estimates a smaller cycle. To account for comparable inflation movements with a smaller cyclical gap, it estimates steeper price and wage Phillips curves. In the paper’s results, the price slope rises from about 0.09 in the benchmark to about 0.48 in the scarring specification; the wage slope changes from about −0.15 to about −0.67.
Those numbers should be read as part of a decomposition, not as a claim that the real economy suddenly became five times more sensitive to unemployment. If the denominator used to measure the inflation-relevant gap becomes smaller, the estimated slope attached to that gap can become larger. The model is changing both the measured gap and the relationship used to explain inflation. The comparison tells us that trend-cycle assumptions affect Phillips-curve estimates; it does not provide a universal multiplier for policy analysis.
The reallocation also changes which shocks are asked to explain the data. Under the benchmark, a flat Phillips curve and a highly persistent cycle make it difficult to match rapid inflation movements without implying very large or long-lasting output and unemployment responses. The model therefore relies more heavily on cost-push shocks. Under scarring, a smaller and more responsive cycle can account for more of the inflation movement, while supply-driven trend shocks account for more of the persistent real-side variation.
The trade-off is important. A steeper Phillips curve makes cyclical innovations more powerful for inflation, but that same power limits how much the model can use them to explain output and unemployment. If it used large cyclical shocks to reproduce all of the real-side persistence, the resulting inflation movements would be too large relative to the data. The model instead shifts more of the variance toward supply-driven stars and the history-dependent capacity component.
This is why a low unemployment reading can be misleading in two directions. A model that puts too much persistence in the cycle may infer that the economy is far beyond its supply-driven capacity and that policy must lean against it aggressively. A model that allows a history-dependent capacity may infer a smaller inflation-relevant gap, while still finding a meaningful policy-relevant distance from the supply-driven star. Neither conclusion follows from the unemployment rate alone; each depends on how the model distributes persistence.
The Federal Reserve paper’s estimates are persuasive within their stated design, but the design sets boundaries on the conclusion. The authors use U.S. data from 1987:Q1 through 2024:Q4 and restrict attention to a period they regard as more homogeneous than earlier samples. Their model assumes time-invariant parameters, so it is not designed to capture changes in the degree of scarring across regimes. The paper is also a preliminary working paper, and its results depend on the chosen equations, priors, treatment of volatility, and use of inflation and wage data.
Those limits do not make the mechanism unhelpful. They tell us how to use it. When someone says a 4 percent unemployment rate is below, above, or equal to full employment, ask whether the claim assumes that trends and cycles are independent. If the claim allows history dependence, ask whether the proposed effect is temporary or permanent and whether the evidence supports unemployment scarring, output scarring, or both. Then ask which gap is being discussed: the gap relative to capacity that explains inflation, or the gap relative to the supply-driven star that informs a policy judgment.
The remaining problem is identification. A second capacity path cannot be defended merely because it is plausible. The model needs evidence that distinguishes a movement in the cycle from a movement in capacity. That is where the level and speed of inflation and wage relationships become central: they provide the cross-checks that determine whether persistence belongs in the cycle, in the trend, or in both.