The Benchmark You Cannot Observe: Stars, Natural Rates, and Potential Output
This chapter defines the hidden benchmark behind claims about full employment: the star, the natural or noncyclical unemployment rate, and potential output. The goal is not to choose a magic replacement for 4 percent, but to show how economists construct a reference point from observed data and why that construction carries assumptions. Once those objects are clear, a gap becomes an interpretable relationship rather than a distance from an arbitrary line.
What a star measures
A star is an equilibrium reference point. In the Federal Reserve paper, starred variables represent the state toward which output and unemployment would be expected to converge once cyclical fluctuations had played out and the economy was expanding at its trend rate. The phrase “would be expected” matters. A star is not a physical limit that can be read from a gauge. It is a statement about what remains after the model has separated temporary disturbances from the forces that sustain the economy.
For unemployment, the related concept is the natural rate. It is the rate that would remain after the effects of a temporary demand cycle had faded, given the labor market’s underlying conditions. Those conditions include the way workers search for jobs, the way firms recruit and set wages, the composition of the labor force, and the amount of labor demand the economy can sustain. A natural rate is therefore not the lowest unemployment rate anyone could imagine. It is a model’s estimate of the unemployment compatible with a particular conception of stable, non-cyclical expansion.
The CBO uses a closely related but institutionally distinct term: the noncyclical rate of unemployment. It describes unemployment arising from all sources except changes in aggregate demand. That wording is useful because it shows what the estimate is trying to remove. A fall in demand can temporarily leave people without work; changes in demographics, participation, job matching, or the structure of production may affect unemployment even when demand is not moving in the same way. The CBO’s noncyclical rate is intended to capture the latter component.
These terms should not be treated as interchangeable labels for a single universally agreed statistic. The Federal Reserve paper is about a model of starred variables and history dependence. The CBO’s outlook is a conditional projection built from its own estimates of labor supply, potential output, and other economic relationships. Their shared lesson is narrower and more defensible: the observed unemployment rate requires a benchmark before it can be interpreted as a gap.
The same logic applies to output. Real potential GDP is the CBO’s estimate of the amount of inflation-adjusted output that could be produced if labor and capital were employed at their maximum sustainable rates. Potential labor force is its estimate of how large the labor force would be if output and other key variables were at their maximum sustainable amounts. Potential labor-force productivity is then the amount of potential output per person in that potential labor force.
The word “potential” can mislead if it is heard as “the most output technically possible.” A factory could run around the clock for a short period, but that would not necessarily be sustainable. Potential output is an estimate of capacity consistent with the economy’s continuing operation, not a promise that all resources can be used at every instant without changing prices, incentives, or future production. Likewise, potential labor supply is not a count of every person who could work under extraordinary pressure. It is a model-based estimate of sustainable participation and employment conditions.
A gap is the distance between an observed variable and one of these reference points. An output gap compares actual output with a measure of potential output. An unemployment gap compares the actual unemployment rate with a natural or noncyclical benchmark; because unemployment rises when labor-market conditions weaken, the sign convention often runs opposite to the intuitive meaning of “more slack.” The important point is not the sign but the denominator. A gap cannot be interpreted until we know which capacity estimate it uses.
That is why a 4 percent unemployment rate has no standalone economic meaning as “full employment.” It could sit above, below, or near the benchmark. The benchmark could also change while the observed rate remains 4 percent. A consistent statistic does not guarantee a constant reference frame.
The relationships are more important than the labels. The observed economy contains movements that a model may call cyclical. A supply-driven star provides one reference for judging sustainable capacity, while a broader capacity that includes the effects of earlier cycles may provide another reference for explaining inflation. The gap is not a property of the observed rate alone; it is produced by the comparison.
How the benchmark is constructed
Because stars cannot be observed directly, economists infer them from several series at once. An unobserved-components model treats the data as the result of hidden trends and more temporary cycles. Output, unemployment, inflation, wages, and expectations are not simply placed on a chart and read off. The model estimates the hidden paths that best account for their joint movements under assumptions about how each component evolves.
The simplest picture is a trend plus a cycle. The trend moves slowly and represents the underlying capacity; the cycle moves around it and represents temporary fluctuations. If output falls below its trend, the model may call the difference a negative output gap. If unemployment rises above its trend, it may call the difference a positive unemployment gap. The model’s trend is then doing the work of the star.
This decomposition is useful because it turns a vague question—“Is the economy weak because demand is weak, or because capacity has changed?”—into an estimable one. It also creates a hidden choice. How slowly must a trend move before it counts as structural? How long can a cyclical effect last before it becomes part of the trend? What other data should constrain the answer? Different responses can yield different stars even when they use the same headline unemployment series.
The conventional unobserved-components benchmark makes one especially important assumption: trend factors are independent of cyclical components. In plain language, the supply-driven path has its own shocks, and demand disturbances stay in the temporary cycle. A recession may reduce output for several years in this model, but it does not, by that fact alone, alter the supply-driven path. Persistence is assigned to the cycle rather than to the star.
Independence is not an irrational assumption. It reflects a familiar economic distinction between supply and demand. It also keeps the model identifiable and easier to interpret. If every cyclical movement could freely rewrite the trend, “potential” could become a description of whatever happened rather than a meaningful counterfactual. A benchmark needs discipline.
But discipline can become rigidity. The Federal Reserve paper points to evidence that demand shocks may have persistent effects on real activity and that long recoveries are difficult to describe as purely transitory fluctuations. A deep downturn can change firms’ investment, workers’ skills, job-finding prospects, and the organization of production. Whether those channels are large enough to alter potential capacity is an empirical question, but an independence assumption prevents the model from representing them directly.
This is the source of the disagreement hidden inside many gap estimates. Suppose output remains below its earlier path for a long time while inflation changes only modestly. A model that protects the trend from demand shocks must explain the persistent output weakness through a long-lived cycle or through other shocks. A model that permits history dependence can assign some of the persistence to capacity. The observed path is the same; the implied star and the remaining gap are not.
The distinction also clarifies why the natural rate should not be imagined as the unemployment rate that would prevail if every friction disappeared. Job search takes time, firms need to find suitable workers, and the composition of the labor force changes. Some unemployment is associated with movement between jobs or mismatch rather than with an economy-wide lack of demand. A natural or noncyclical rate is an estimate of the aggregate result of those processes under sustainable conditions, not a moral judgment about workers’ effort and not a target that can be reached by decree.
CBO’s forecast shows how the benchmark is tied to a wider production system. It defines potential output using both labor and capital. Its potential labor force depends on the working-age population and participation. Its potential labor-force productivity depends on capital accumulation and total-factor productivity. A change in any of those elements can change the amount of output the economy can sustain and the unemployment rate consistent with it.
The connection runs in both directions. If potential output grows more slowly, firms may face weaker demand for labor. That can affect the noncyclical component of unemployment. If labor supply grows more slowly, potential output may be lower, but the labor market can also become tighter in sectors where workers are scarce. The same demographic change can therefore push different observed variables in different directions. A benchmark is not moved by one isolated knob.
CBO’s definitions also make clear that the time horizon matters. In the short run, output can depart from potential because demand changes, fiscal policy shifts, or temporary disruptions affect spending and hiring. Over a longer horizon, businesses can adjust capital and production, while population and productivity trends become more important. A rate that is consistent with stable conditions after those adjustments need not be the rate that best describes the labor market during the adjustment itself.
This conditional language is important because “full employment” is often used as though it meant the disappearance of all unused labor. Neither the natural-rate idea nor the CBO’s noncyclical-rate idea requires unemployment to be zero. Workers leave jobs, enter the labor force, search for a better match, or face skills and location mismatches. Firms open and close positions at different speeds. Some of the resulting unemployment can exist in an economy whose demand is broadly consistent with its capacity. The analytical question is not whether anyone is between jobs; it is whether aggregate demand is creating more or less unemployment than the economy’s underlying labor-market structure would imply.
Nor does potential output mean that every worker and machine must be operating at maximum intensity. Capacity is sustainable only relative to the incentives and constraints that govern continued production. If firms try to keep employment above what labor supply, productivity, and demand can support, they may bid up wages and prices, shift investment, or make future production less stable. If they operate below that capacity, the observed weakness may reflect insufficient demand—or it may reflect a supply disruption that has lowered what can be produced for a time. The benchmark helps organize those possibilities; it does not eliminate the need to distinguish them.
A useful way to think about the estimate is as a counterfactual question: what would output and unemployment look like if the temporary disturbance were removed while the rest of the economy’s conditions were held in the model’s stated way? The answer depends on which disturbance is considered temporary and which conditions are held fixed. A demand shortfall can make actual unemployment high relative to the noncyclical rate. A change in participation can alter the labor force and the unemployment rate without representing the same kind of demand shortfall. A productivity change can raise potential output while changing the labor demand associated with it. These are different channels even if they eventually appear together in a single headline statistic.
The model therefore has to use relationships across data, not just the level of one series. Output says something about production, unemployment says something about the use of labor, and inflation and wages provide information about pressure relative to capacity. Long-run inflation expectations can help anchor the inflation trend. Labor productivity can help connect wage growth to the sustainable path of production. None of these signals is decisive in isolation. Their value comes from the way the model requires them to fit together.
That requirement is also the source of uncertainty. A trend estimated near the end of a sample can be revised when later observations arrive. A shock initially treated as temporary may prove persistent, or a persistent-looking movement may fade as firms and workers adjust. Estimates can differ because they use different samples, data vintages, equations, priors, or definitions of sustainable capacity. Precision in the final decimal does not remove those choices. It only records the result of applying them.
This is why an unemployment gap is never just a subtraction. The arithmetic may be simple, but the benchmark is the result of a model connecting labor supply, labor demand, production, prices, and expectations. A numerical gap can look precise while resting on uncertain judgments about how quickly trends move and which shocks have lasting effects.
The baseline model is still necessary. Without first defining the supply-driven trend and the cyclical component, it is impossible to see what changes when history dependence is introduced. The next question is not whether trends or cycles are “real.” Both describe parts of the observed economy. The question is whether a cycle can leave a mark on the trend it was supposed to be measured against—and, if so, how that mark should be separated from the underlying star.