Explore the natural rate of unemployment with U.S. data, job flows, and examples that show why two 4% labor markets feel different.

Natural Rate of Unemployment Explained With Job Flows

Natural Rate of Unemployment Explained With Job Flows

In two towns, exactly 4% of the labor force is looking for work. Someone starting a search in the first town can expect to find a job in five months. Across town, the expected wait is only two and a half months. How can both figures be right? The natural rate of unemployment describes a balance between people losing and finding work. That balance tells you how many people you might see searching on a given day. To learn how long they wait, you need to know how quickly people move.

What is the natural rate of unemployment?

The natural rate of unemployment is the unemployment rate an economy tends toward once temporary business-cycle pressures fade, given its workers, employers and labor-market institutions. Even in a healthy economy, people graduate, contracts end and firms take time to fill vacancies. Meanwhile, a worker with the right skills may live hundreds of miles from the employer who needs them.

Economists use several related benchmarks. A simple job-flow model asks where unemployment would settle if transition probabilities stayed constant. By comparison, the Congressional Budget Office (CBO) estimates a noncyclical rate for the United States. An inflation-based measure, often called the NAIRU, asks what unemployment rate would be consistent with stable inflation under specified conditions. These concepts overlap, but they answer different questions. CBO series and methodology notes Congressional Research Service on unemployment and inflation

Why can actual unemployment fall below the natural rate?

During a hiring boom, employers can fill jobs more quickly or retain people they might otherwise let go. As a result, actual unemployment can fall below a long-run estimate. A recession can reverse those movements. Moreover, statisticians can revise the estimate as new evidence arrives. Treat the difference as a clue about economic conditions, with uncertainty on both sides of the comparison.

How to calculate the natural rate of unemployment

Start with 10,000 people who either hold a job or actively seek one. For this exercise, nobody enters or leaves the labor force. Let E represent employed people, U unemployed people and L their combined total. Each month, a fraction s of the initially employed move into unemployment. Meanwhile, a fraction f of the initially unemployed move into employment. Both probabilities use the same time interval, but each uses a different denominator.

Labor force and unemployment share
L = E + U   ·   u = U / L
Monthly change in the number unemployed
Ut+1 − Ut = sEt − fUt

At the steady state, the two flows balance. Therefore, sE = fU. Substitute L − U for E, expand the left side and collect terms to obtain sL = (s + f)U. Finally, divide by L(s + f), assuming at least one transition probability is positive.

Steady-state unemployment share
u* = s / (s + f)

For example, a 1% monthly separation probability and a 24% monthly job-finding probability imply 1 / (1 + 24) = 4%. Among 9,600 employed people, 96 enter unemployment in an average model month. Similarly, 24% of 400 unemployed people, or 96, find jobs. The count stays at 400 even as individuals move in both directions.

Which flow should you measure?

Suppose an analyst divides all hires during a month by the number unemployed. Some hires come straight from another employer. Others come from outside the labor force. Consequently, the result can misrepresent the chance that an unemployed person finds work. To estimate f, follow people who began the month unemployed and count how many hold a job at the next observation. Likewise, estimating s requires employed people who move specifically into unemployment. BLS labor-force status flows and downloadable series

Two economies with 4% unemployment and different risks

Imagine two labor markets with the same population and the same 400 unemployed workers. In Market B, both transition probabilities are exactly twice those in Market A. Consequently, each market has the same ratio of separations to job finding. Workers experience that equilibrium very differently.

What doubling both monthly probabilities changes
MeasureMarket AMarket B
Monthly chance of moving from work to unemployment0.8333%1.6667%
Monthly chance of finding work if unemployed20%40%
Steady-state unemployment rate4%4%
Expected transitions into unemployment per month80160
Expected transitions into work per month80160
Expected unemployment spell5 months2.5 months
Chance a new spell lasts beyond 6 monthly transitions26.2%4.7%

The table uses exact separation probabilities of 1/120 and 1/60. For example, 9,600 ÷ 120 gives 80 monthly transitions. With a constant monthly job-finding probability, the expected spell lasts 1/f periods. Similarly, the chance of remaining unemployed after six transitions equals (1 − f) to the sixth power. These calculations describe hypothetical workers with identical probabilities throughout a spell.

Which market would a worker choose?

A person already looking for a job faces better near-term odds in Market B. However, a person who has a job faces twice the monthly risk of entering unemployment there. For a household with little cash, a single long interruption could matter more than several short ones. By contrast, frequent interruptions could prove especially costly for a worker who loses benefits each time a job ends. The headline rate alone cannot rank the two markets for either person.

How fast does unemployment return to its steady state?

Keep the labor force fixed and divide the change in unemployment by L. Because employed people represent the fraction 1 − u, the unemployment rate follows a simple recurrence. Subtracting its steady-state value then reveals how much of the initial gap survives each period.

One-period update
ut+1 = s + (1 − s − f)ut
Gap after t periods
ut − u* = (1 − s − f)t(u0 − u*)

For example, take s = 0.01 and f = 0.24 again. Start the economy at 10% unemployment. In the first month, 9,000 employed people generate 90 entries, while 1,000 unemployed people generate 240 exits. Thus, the unemployment count falls to 850 and the rate reaches 8.50%. As hiring draws down the pool of job seekers, the absolute decline gets smaller.

Unemployment after a one-time shock
Illustrative rates with an unchanged 4% steady state. Bars start at zero and span a scale of 0–10%.
Start 10.00%
Month 1 8.50%
Month 2 7.38%
Month 3 6.53%
Month 6 5.07%
Month 12 4.19%

After one year, unemployment reaches about 4.19%. In this example, the gap shrinks by 25% each month. That percentage applies to the distance from 4%, so a forecast that cuts the whole unemployment rate by a quarter each month will soon undershoot. For a spreadsheet, put 0.10 in cell A2 and enter =0.01+0.75*A2 in A3. Then copy the formula down and display the cells as percentages.

When does the model overshoot?

Usually the textbook example assumes s + f is below one. In that case, the gap keeps its sign and shrinks. However, two individually valid probabilities can sum to more than one. With s = 0.40 and f = 0.80, the gap changes sign and shrinks by a factor of 0.20 each period. At the extreme, s = f = 1 makes everyone switch status every period. Therefore, balancing the flows identifies a steady state, while the size of the probabilities determines how the economy approaches it.

What U.S. unemployment spells actually look like

The U.S. Bureau of Labor Statistics (BLS) asks unemployed respondents how long their search or layoff has lasted so far. Consequently, its duration table gives a snapshot of spells in progress. A worker who has searched for nine weeks at the interview may find a job tomorrow or remain unemployed for months. BLS survey definitions and duration measures

U.S. unemployment duration, annual averages for 2023 and 2024
Time unemployed so far2023 share2024 share2024 people
Under 5 weeks34.7%33.1%2,235,000
5 to 14 weeks30.7%29.8%2,012,000
15 to 26 weeks15.2%15.7%1,064,000
27 weeks or more19.4%21.4%1,449,000

In 2024, the median elapsed duration was 9.6 weeks, while the mean reached 21.6 weeks. Moreover, 825,000 people had already spent at least 52 weeks unemployed in the average survey month. Long spells pull the mean upward. Because the survey takes repeated snapshots, adding up the monthly counts would count some people several times. Rounding can also keep the categories from adding exactly to the total. BLS annual averages, Table 30

Why the majority of spells can account for a minority of weeks

Consider 100 completed spells. Suppose 90 people each spend two weeks unemployed and ten people each spend 40 weeks unemployed. Together, the short spells generate 180 person-weeks and the long spells generate 400. Therefore, just 10% of the spells account for 400 ÷ 580, or 69%, of all unemployed weeks.

Choose a spell at random and you will probably pick a short one. Instead, choose an unemployed person-week at random and you will probably pick a long one. This distinction also explains why taking the reciprocal of the BLS median duration will not recover the monthly probability of finding a job. The units differ, the survey observes unfinished spells and workers have different chances of leaving unemployment.

Which age group had higher unemployment but shorter searches?

Two BLS tables let us check an apparent paradox. Among U.S. teenagers aged 16–19, the 2024 unemployment rate was almost four times the rate for workers aged 35–44. Yet the median unemployed teenager had spent about four fewer weeks out of work. To see how both statements can hold, align the age groups and survey year before comparing the denominators.

U.S. unemployment rates and elapsed search times by age, 2024
Age groupUnemployment rateMedian durationMean duration
16–1912.7%6.2 weeks12.2 weeks
20–247.3%9.2 weeks19.0 weeks
25–344.3%9.6 weeks21.0 weeks
35–443.2%10.5 weeks24.9 weeks

The rate counts unemployed people relative to everyone in that age group’s labor force. By comparison, the duration measures cover only people currently unemployed. Frequent entries into unemployment can keep the first number high even when spells tend to be brief. However, these two tables alone cannot identify how much of the age gap comes from job loss, labor-force entry or differences in hiring. BLS rates by age, Table 3 BLS duration by age, Table 31

Who enters unemployment without losing a job?

A graduate can enter unemployment while looking for a first job. Similarly, a parent returning to paid work can enter unemployment after a period of caregiving. In 2024, new entrants and reentrants together accounted for roughly four in ten unemployed Americans. The calculation uses the reason for unemployment reported in the household survey. BLS annual averages, Table 29

Reason for unemployment in the United States, 2024 annual averages
GroupPeople, thousandsShare of all unemployed¹Within group at 27+ weeks
Lost a job or finished temporary work3,25348.1%18.5%
Left a job voluntarily81012.0%19.0%
Returned to the labor force2,05230.4%26.5%
Entered for the first time6479.6%23.1%

¹ Shares divide the published group counts by 6.761 million and round to one decimal. For example, the roughly 39.9% combined entrant share uses (2,052 + 647) ÷ 6,761 before rounding. By comparison, the final column divides by each group’s own total. It answers how common long searches are within that group.

What happens when someone stops searching?

Imagine a town with 950 employed people and 50 unemployed people. If ten job seekers stop searching and move outside the labor force, the rate falls from 5% to 40 ÷ 990, or 4.04%. Meanwhile, the town has gained zero jobs. This arithmetic explains why analysts check participation and employment alongside unemployment.

Under U.S. survey rules, active search generally means taking a concrete step in the previous four weeks, alongside availability for work. People on temporary layoff who expect recall can qualify without searching. Moreover, discouraged workers form a specific subset of people outside the labor force who want a job and have stopped searching for labor-market reasons. Someone who leaves work to study full time does not automatically meet that definition. BLS labor-force classifications

Frictional, structural and cyclical unemployment

A hiring manager needs time to read applications, and applicants need time to compare offers. Economists call the unemployment that arises during this matching process frictional. For example, a bilingual teacher might spend several weeks finding a school with a suitable opening even when local districts are hiring.

Structural unemployment can persist when the jobs on offer and the available workforce fit poorly. A factory closure may leave skilled employees far from employers who value their experience. Alternatively, a credential requirement may delay entry into a growing occupation. In a wage-rigidity model, a real wage above the market-clearing level also leaves more people seeking work than firms choose to hire.

Cyclical unemployment follows changes in overall demand. For instance, a company with a suitable opening and a qualified applicant might postpone hiring after customers cancel orders. In practice, one worker can face all three forces during the same search. Consequently, the length of a spell cannot identify its cause by itself.

Why might an employer keep wages high?

Training a replacement takes time, and a vacancy can disrupt a shift. Therefore, an employer may profit from paying an efficiency wage if better retention or higher productivity offsets the extra payroll expense. Collective bargaining offers another mechanism. Existing employees, sometimes called insiders, can negotiate conditions that matter to them, while applicants outside the firm have less influence. A binding minimum wage provides a third mechanism in the competitive model below.

A worked wage-floor example with employment and payroll

Suppose a competitive economy produces output using capital K and employment N. Set capital at 1,000 units and fix labor supply at 1,000 workers. Each worker supplies one unit of labor. Firms sell output at a normalized price of one and hire until an additional worker’s output equals the real wage w. Fractional employment represents divisible units of labor.

Production and marginal product
Y = K1/3N2/3   ·   w = (2/3)(K/N)1/3
Labor demand at a given wage
Nd = K[2/(3w)]3

With all 1,000 workers employed, the real wage equals two-thirds of a unit of output. Now impose a floor of one full unit per worker. Because the floor exceeds the clearing wage, firms move along their labor-demand curve. Substituting w = 1 gives 1,000 × (2/3)³, or about 296.30 units of employment.

Two wage regimes in the same hypothetical economy
OutcomeFlexible wageBinding floor
Real wage per worker0.6671.000
Employment1,000.00296.30
Unemployment rate0.00%70.37%
Output1,000.00444.44
Total worker earnings, w × N666.67296.30

A worker who retains a job gains 50% in pay. However, total labor earnings fall by about 55.6% because firms employ far fewer people. Why does the example produce such a large response? Its labor demand is proportional to w raised to the power −3. The model therefore builds a large employment response into the production technology, alongside fixed capital and a 50% wage increase.

How does that result compare with actual wage changes?

Cengiz and colleagues studied 138 prominent state minimum-wage changes in the United States from 1979 through 2016. They counted jobs that disappeared below the new threshold and jobs that appeared at or just above it. On average, the authors found little change in the overall number of low-wage jobs over the following five years. Consequently, applying the numerical result above to a particular city would require separate evidence on the size of its wage change and the response of local employers. Quarterly Journal of Economics study

How would a productivity shock change the answer?

Multiply the production function by a productivity factor A. Then labor demand becomes N = K[2A/(3w)]³. Suppose productivity falls by 10%, starting from full employment and a real wage of two-thirds. With flexible pay and fixed labor supply, employment stays at 1,000 and the real wage falls to 0.60. However, if a contract holds the wage at two-thirds, firms demand 1,000 × 0.9³ = 729 workers. In that case, unemployment rises to 27.1%. The same loss of productivity reaches households through different combinations of pay and employment.

Which policies can change the natural rate of unemployment?

Job-search assistance can help an applicant find a suitable opening sooner. Training can address a skill gap if employers actually need the new skill. Meanwhile, a policy that protects jobs may change the chances of both dismissal and hiring. The two-state formula lets us identify exactly when those changes offset each other.

What happens if a policy cuts the monthly separation probability?
Illustrative scenarioSeparation sFinding fSteady-state rate
Before policy1.00%24.00%4.00%
Separations fall by one quarter0.75%24.00%3.03%
Both probabilities fall by one quarter0.75%18.00%4.00%
Hiring falls by half0.75%12.00%5.88%

Thus, a 25% reduction in separations buys room for a reduction of up to 25% in the job-finding probability before steady-state unemployment rises above its starting value. Even at the break-even point, expected jobless spells increase from 4.17 to 5.56 months. That distinction matters when deciding whether a policy helped workers who were already unemployed.

Why unemployment insurance complicates the policy test

Benefits can give a household time to pay bills while someone looks for work. They can also change how urgently a worker accepts an offer. Economist Raj Chetty estimated that access to cash accounted for about 60% of the benefit-induced increase in unemployment duration in his study. Therefore, observing a longer search alone tells a policymaker little about whether the additional time harmed the worker. A fuller evaluation would track earnings after reemployment, financial hardship and how long the next job lasts. Chetty’s research on liquidity and unemployment insurance

Why the U.S. natural rate changes over time

Selected first-quarter CBO estimates trace an increase between 1960 and 1980 and a decline over later decades. For example, the early-1980 estimate exceeds the early-2024 estimate by about 1.80 percentage points. These observations belong to one published series. Earlier analysts working with information available at the time could have reached different estimates. CBO quarterly observations via FRED

U.S. noncyclical unemployment rate in selected first quarters
CBO estimates in percent, rounded to two decimals. Each bar starts at zero on a 0–7% scale.
1960 Q1 5.46%
1980 Q1 6.22%
1990 Q1 5.76%
2000 Q1 5.12%
2024 Q1 4.43%

How workforce composition can move the average

Imagine two groups with unemployment rates of 10% and 3%. If the higher-rate group’s labor-force share rises from 20% to 30%, aggregate unemployment rises from 4.4% to 5.1%. Neither group experienced any change in its own unemployment rate. Consequently, changes in the size of younger cohorts can move the national benchmark even when individual hiring risks hold steady.

Historically, economists have examined changes in workforce age, productivity growth and the industries that hire or release workers. For example, when productivity slows unexpectedly, wage expectations can take time to adjust. Conversely, faster productivity growth can let firms raise pay with less pressure on unit labor costs. These mechanisms give possible explanations for the rise and later decline in the estimated benchmark. Four decades of aggregate observations cannot assign a precise share to each mechanism. Congressional Research Service review of historical explanations

Why a shorter workweek tells a different story

In 2024, unemployment averaged 5.9% of the EU labor force aged 15–74. Among labor-force participants aged 15–24, the rate reached 14.9%. Meanwhile, employed people aged 20–64 averaged 36.0 actual weekly hours in their main job, including both full-time and part-time workers. For context, the U.S. annual unemployment rate was 4.0% among people aged 16 and over in 2024. Notice the change of denominator and age range before comparing the figures. BLS U.S. annual rate Eurostat unemployment figures Eurostat working-hours figures

Greece averaged 39.8 hours on that measure, compared with 32.1 in the Netherlands. Thus, a single European average hides a 7.7-hour difference between two member countries. An international comparison also needs information on part-time work, leave, industry mix and household constraints. For a related comparison with the United States, see European unemployment explained by jobs, hours and pay.

How does higher pay affect hours?

An extra hour of leisure costs more in forgone earnings when the hourly wage rises. Therefore, the substitution effect encourages more work. At the same time, a higher wage lets someone afford the same purchases with fewer hours. If they value extra leisure as income rises, the income effect encourages fewer hours. A shorter observed week cannot, by itself, tell us which effect dominated or whether workers could freely choose their schedules.

Three short problems to test your intuition

1. Can you recover both transition rates from a 6% unemployment rate?

Only the ratio follows from the headline. Rearranging the formula gives s/f = 0.06/0.94, or about 0.0638. For example, if the monthly finding probability equals 30%, the monthly separation probability must be about 1.915% in steady state. Alternatively, a 15% finding probability implies a separation probability of about 0.957%. To choose between the two, collect a transition count or a credible measure of completed spell length.

2. What does a four-week search imply for someone with short contracts?

Suppose comparable workers spend four weeks unemployed on average and 36 weeks employed before their next spell. Under constant weekly transition probabilities, f = 1/4 and s = 1/36. Therefore, the steady-state rate equals 10%. You can check the result another way. Over repeated cycles, four out of every 40 weeks are spent unemployed. Using one person’s most recent search and longest-ever job would mix incompatible observations and give a misleading estimate.

3. How much office space can stay vacant in a stable market?

Suppose 2% of occupied offices become vacant each month and 18% of vacant offices find tenants. By the same flow arithmetic, the steady-state vacancy share equals 2/(2 + 18), or 10%. Some vacancy can make it easier for firms to relocate. However, an empty space with the wrong layout or in a declining district can remain vacant for a different reason. As with unemployment, the duration and location of vacancies help distinguish routine turnover from a persistent mismatch.

How to read the next unemployment report

Start with the number of unemployed people and the size of the labor force. Next, ask how many people moved into work, lost work or stopped searching. Compare the median time out of work with the share already searching for six months or more. Finally, label each figure as a survey observation, a model estimate or an assumption in your own calculation. Guide to the unemployment rate with worked examples

Return to the two towns with 4% unemployment. A worker starting a search in the slower market has about a one-in-four chance of still looking after six monthly transitions. In the faster market, the chance is closer to one in twenty. Both towns balance entries and exits. What changes a worker’s experience is how long that balance leaves them waiting.

Comments

No comments yet. Why don’t you start the discussion?

Leave a Reply

Your email address will not be published. Required fields are marked *