Economics · Financial, Credit and Business Indicators

Business Dynamics

Every year about one US firm in twelve exits and almost as many start, and the jobs created and destroyed each year are each larger than one in ten. This page measures these flows: firm startups and exits, establishment openings and closings, and job creation and destruction, from the Census Bureau’s annual Business Dynamics Statistics (since 1978) and the quarterly Business Employment Dynamics of the Bureau of Labor Statistics (since 1992). It gives the statistics that calibrate models with firm entry and exit, such as the exit rate of firms, and how these flows move over the business cycle.

  • Annual data: a year runs from March of the previous year to March (employment on 12 March), so 2023 is March 2022 to March 2023.
  • Quarterly data: private sector, seasonally adjusted, rates per quarter.
  • Shaded bands mark US recessions (USRECQ).

Loading data…

Firms, establishments and employment

The number of employer businesses in the United States and the jobs they provide, each year in March.

Firm
a business under common ownership or control; it may have several establishments.
Establishment
a single physical location where business is conducted, such as a plant, shop or office.
Firms and establishments

Loading chart…

Employment

Loading chart…

Average firm size

Loading chart…

Employment divided by the number of firms.

Entry and exit

How many firms start and stop each year, and how much of employment new and young firms account for. The startup rate has fallen by more than a third since the late 1970s, the decline in business dynamism documented by Decker, Haltiwanger, Jarmin and Miranda (2014).

Startup
a new firm, of age 0: all its establishments are new.
Firm exit (death)
a firm all of whose establishments close during the year.
Rates
relative to the average of the current and previous year’s count, as in the Census Bureau’s own rates; the firm exit rate, a probability, relative to the firms alive a year earlier.
Firm startup and exit rates

Loading chart…

Establishment entry and exit rates

Loading chart…

Employment shares of startups and young firms

Loading chart…

Young firms: 0 to 5 years old, from 1983, the first year in which all their ages are observed.

Job creation and destruction

Jobs created at opening and expanding establishments, and destroyed at closing and contracting ones, from March to March, in percent of employment (Davis, Haltiwanger and Schuh, 1996). Their sum, job reallocation, far exceeds the net change in employment.

Job creation and destruction rates

Loading chart…

Net job creation and job reallocation

Loading chart…

Job creation by births and destruction by deaths of establishments

Loading chart…

Net job creation by firm age

Loading chart…

Startups, which only create jobs, are left out; young firms from 1983. Young firms that survive grow fast, but many exit: “up or out” (Haltiwanger, Jarmin and Miranda, 2013).

Firm size

Small firms are most firms, but large firms employ most workers. Size is the firm’s average employment in the current and previous year, which limits the bias from firms moving between classes (Haltiwanger, Jarmin and Miranda, 2013).

Employment shares by firm size

Loading chart…

Net job creation by firm size

Loading chart…

Firm size classes: averages since 1978

Loading…

Quarterly job flows and establishment births and deaths

The BLS Business Employment Dynamics measure the same flows each quarter, for private-sector establishments, seasonally adjusted, from 1992. Quarterly flows are smaller than annual ones: a job created and destroyed within a year does not appear in the annual data.

Openings and closings
establishments with positive employment this quarter and none the previous quarter, or the reverse; they include seasonal and temporary closures.
Births and deaths
openings with no employment in the previous year, and closings with no employment for the next year: permanent entry and exit.
Gross job gains and losses

Loading chart…

Net job gains

Loading chart…

Job gains at openings and losses at closings

Loading chart…

Establishment births and deaths

Loading chart…

Deaths are published three quarters after births: an establishment is counted as dead only after a year without employment.

Establishment openings and closings

Loading chart…

Statistics for calibration

Averages of the entry and exit rates that models with firm dynamics are calibrated to. An annual probability \(s\) of exit corresponds to a quarterly probability \(1 - (1 - s)^{1/4}\); the quarterly BED rates give the quarterly frequency directly. The annual and quarterly measures differ in coverage and definition (see the note on calibration).

Loading…

Firm survival: steady state and dynamics

Models with firm exit need the quarterly probability \(s\) that a firm survives, and, if survival varies over time, the persistence \(\rho_s\) and the standard deviation \(\sigma_s\) of the shocks to \(\ln s_t\), a log AR(1) process. The steady state comes from the annual firm exit rate of the Census Bureau; the dynamics from the cycle of annual log survival, converted to a quarterly process that adds up to the annual one (method A, with the Hodrick–Prescott filter, the default), checked against the quarterly establishment death rates of the BLS (B). Detrending matters: with a linear trend instead (L), survival is much more persistent. The annual deviations and their shocks under both are in the download “Survival shocks”. How these are computed.

Loading…

Loading…

Quarterly survival rate of firms, from annual exits

Loading chart…

Each year, (1 − exit rate)^(1/4): the constant quarterly survival probability that gives the year’s exits.

Quarterly survival rate of establishments (BED)

Loading chart…

100 minus the quarterly establishment death rate.

Value of entrants

In a model in which exiting firms are replaced by entrants whose equity is all new, the steady state needs the value of entrants relative to all firms, \(\varepsilon = I_0/p^A\) per quarter. Then the relative cohort value is \(\kappa = s/(1-\varepsilon)\) and the value of the average firm relative to an entrant \(m = (1-s)/\varepsilon\) (see the Calibration page). For all firms, the employment share of startups is the proxy, from the same data and years as \(s\); the IPO counterpart, for listed firms, is on Investor Sentiment and Equity Issuance. Caveats.

Loading…

Employment share of startups, per quarter

Loading chart…

Each year, 1 − (1 − x)^(1/4), with x the share of employment in firms of age 0: the quarterly rate that gives the year’s share.

Startup size relative to the average firm

Loading chart…

Employment share of startups over the startup rate: about a quarter of the size of the average firm.

Annual series (Business Dynamics Statistics)

Mean, standard deviation, range and latest value of each series, and the average growth rate of counts.

Loading…

Quarterly series (Business Employment Dynamics)

Loading…

Business dynamics over the business cycle

Cyclical components from the Hodrick–Prescott filter, with λ = 6.25 for the annual series (Ravn and Uhlig, 2002) and λ = 1600 for the quarterly ones, compared with that of real GDP. Rates are in percentage points of deviation from trend; counts, employment and GDP in percent. Startups and job creation are procyclical, exits and job destruction countercyclical (Davis, Haltiwanger and Schuh, 1996; Clementi and Palazzo, 2016).

Firm startup rate

Loading chart…

Firm exit rate

Loading chart…

Job creation and destruction rates

Loading chart…

Gross job gains and losses

Loading chart…

Establishment births and deaths

Loading chart…

Annual cycle statistics

Standard deviation of the cycle, its ratio to that of real GDP (for counts and employment), first-order autocorrelation, and the correlation of the series \(k\) years ahead, \(x_{t+k}\), with real GDP in year \(t\). A peak at negative \(k\) means the series moves before GDP; the peak is highlighted.

Loading…

Quarterly cycle statistics

As above, with \(k\) in quarters.

Loading…

Definitions and methods

The data: Business Dynamics Statistics and Business Employment Dynamics

Business Dynamics Statistics (BDS). Built by the U.S. Census Bureau from the Longitudinal Business Database, which follows every employer establishment in the private nonfarm economy (most of the economy except government, agriculture and the self-employed without employees) and links establishments to the firms that own them. Each year is measured in March, so the flows of year \(t\) are the changes from March of \(t-1\) to March of \(t\).

  • Firm age is the age of its oldest establishment when it first appears; a startup (age 0) is a firm all of whose establishments are new. Since the data start in 1977, firms with establishments older than that are “left censored” and counted here as mature; firms of age 1 to 5 are all observed from 1983.
  • Firm exit (death): all the firm’s establishments close. A firm that is acquired, or whose establishments are sold, does not die.
  • Rates: job creation and destruction, and establishment entry and exit, are those published by the Census Bureau, relative to the average of employment (or establishments) in \(t\) and \(t-1\) (Davis, Haltiwanger and Schuh, 1996). The firm startup rate is computed in the same way; the firm exit rate is the number of firm deaths over the firms alive in March of the previous year, the probability that a firm exits within the year.
  • Firm size is the firm’s average employment in \(t\) and \(t-1\). Classes: 1–19, 20–499 and 500 or more employees, from the ten classes of the Census Bureau.

Business Employment Dynamics (BED). Built by the Bureau of Labor Statistics from the Quarterly Census of Employment and Wages, the unemployment-insurance records of private-sector establishments (almost all private employment), seasonally adjusted. Gross job gains are the employment increases at opening and expanding establishments, gross job losses the decreases at closing and contracting ones, in percent of average employment. Births are openings with no employment in the previous year, deaths closings that remain without employment for a year, which is why deaths are published later.

Annual versus quarterly. BED establishments are physical locations, without the firm links of the BDS. Annual flows miss the jobs created and destroyed within the same year, which quarterly flows capture, so four quarters of gross flows add up to more than the annual flows.

Calibration: from the data to model parameters

In a quarterly model in which each firm exits with a constant probability \(s\) per quarter, the probability of surviving a year is \((1-s)^4\), so an annual exit rate \(s_A\) implies \(s = 1-(1-s_A)^{1/4}\), and a quarterly rate \(s\) an annual one of \(1-(1-s)^4\). With exit, the Euler equation of the shareholders and the steady-state value of a firm change, and with them the discount factor implied by a return target (see the Calibration page).

Which rate to use depends on what exit means in the model. The BDS firm exit rate counts the end of firms, with all their establishments; the establishment exit rate and the BED death rate count locations, and include the closures of establishments of surviving firms. Firm exits are concentrated among small and young firms, so the share of employment, or of capital, that exits is much smaller than the share of firms: the job destruction by establishment deaths, in percent of employment, measures it.

Value of entrants. The BDS do not measure firm values, so the share of employment in startups stands for the share of value in entrants. This is a proxy: startups may be worth less per worker than older firms (less capital, lower productivity), which would make it an upper bound, or more, if their value lies in expected growth. A startup (age 0 in March) is a firm born in the previous twelve months, so its employment share is a flow over a year, converted to a quarter with \(1-(1-x)^{1/4}\). Entry and exit are not equal in the data: since 1979 the startup rate (about 10.0 percent a year) has exceeded the exit rate (8.9 percent), as the number of firms grew, while the model replaces each exiting firm by one entrant. Startups are about a quarter of the size of the average firm, so the employment share of entrants (2.5 percent a year) is much smaller than their share of firms.

Firm survival: steady state and dynamics

Steady state. With \(D_t\) the firms that exit between March of \(t-1\) and March of \(t\) and \(N_{t-1}\) the firms alive in March of \(t-1\), the annual exit probability is \(x_t = D_t/N_{t-1}\). A constant quarterly survival probability that gives the same exits satisfies \(s_t^4 = 1 - x_t\), so \(s_t = (1 - x_t)^{1/4}\); the steady state \(s\) is its average. Dividing the annual rate by four is a close but biased approximation: with \(x = 9\) percent, \(x/4 = 2.25\) percent against \(1 - (1-x)^{1/4} = 2.33\) percent a quarter.

Dynamics. In a model, \(\ln s_t = (1-\rho_s)\ln s + \rho_s \ln s_{t-1} + \varepsilon_t\), with \(\varepsilon_t\) in percent, as for the other shocks of the site’s Dynare models. Annual survival is the product of four quarterly survival rates, so annual log survival is the sum of four quarterly ones: \(\ln S_T = \sum_{j=0}^{3} \ln s_{4T-j}\). If quarterly log survival (in deviation from trend) is an AR(1) with persistence \(\rho\) and variance \(\sigma_x^2\), its four-quarter sums have

\[ \operatorname{Var}(X_T) = \sigma_x^2 \sum_{i=0}^{3}\sum_{j=0}^{3} \rho^{|i-j|}, \qquad \operatorname{Cov}(X_T, X_{T-1}) = \sigma_x^2 \sum_{i=0}^{3}\sum_{j=0}^{3} \rho^{4+j-i}. \]

Method A estimates the first-order autocorrelation and the variance of the cycle of \(100 \ln S_T\) (HP filter, λ = 6.25), finds the \(\rho\) whose ratio of covariance to variance equals that autocorrelation, then \(\sigma_x\) from the variance, and \(\sigma_s = \sigma_x\sqrt{1-\rho^2}\). This treats the annual cycle as the sum of quarterly AR(1) deviations; with about 45 annual observations, both estimates are imprecise. Method B estimates the AR(1) directly on the quarterly cycle (λ = 1600) of \(100 \ln(1 - \text{death rate})\) of establishments, a different unit (locations, including those of surviving firms) over a shorter period. Method C, interpolating the annual data to quarters before filtering, gives a smooth quarterly path that the data do not contain: its persistence is close to one by construction and its shocks are too small.

Detrending. The Hodrick–Prescott filter is the default, as for the other series of the site. With λ = 6.25 it removes movements longer than about eight years, and the annual cycle of log survival that remains is close to serially uncorrelated, so \(\rho_s\) is close to zero. Method L applies the same time aggregation to the deviations from an OLS linear trend over the whole sample: these keep the slow decline in exit rates since the 1980s and the long episodes around recessions, and are far more persistent (an annual autocorrelation of about 0.67, \(\rho_s\) about 0.86 per quarter). Which to use depends on whether the model’s survival shocks should include these slow movements. The download “Survival shocks” has, by year, \(100\ln S_T\), its HP cycle and its linear-trend deviation, and the innovations of the annual AR(1) of each.

In a Dynare model with log variables: s = (1-rhos)*log(sss) + rhos*s(-1) + eps_s; with sss the steady-state survival rate as a fraction and stderr of eps_s equal to \(\sigma_s\) in percent.

Cycles and timing

The Hodrick–Prescott filter is applied to \(100 \times \ln x\) for counts, employment and GDP, and to the rates themselves. For annual data, λ = 6.25 matches the quarterly λ = 1600 (Ravn and Uhlig, 2002). Annual real GDP is the average of the four quarters from April to March, the period a BDS year covers; the GDP cycle is filtered on its full history, from 1947. The quarterly GDP cycle is that of the Real Business Cycles page. Statistics use the years, or quarters, that each series covers.

Updates and revisions

The Census Bureau releases one BDS vintage a year, in September, about two and a half years after the March it ends with, and each vintage revises the whole history back to 1978. This page uses the latest vintage available at each update: the 2023 vintage. The BLS publishes Business Employment Dynamics every quarter, about seven months after the quarter ends, with revisions to the recent quarters.

References
  • Clementi, G. L. and B. Palazzo (2016). “Entry, Exit, Firm Dynamics, and Aggregate Fluctuations.” American Economic Journal: Macroeconomics 8(3): 1–41.
  • Davis, S. J., J. C. Haltiwanger and S. Schuh (1996). Job Creation and Destruction. MIT Press.
  • Decker, R., J. Haltiwanger, R. Jarmin and J. Miranda (2014). “The Role of Entrepreneurship in US Job Creation and Economic Dynamism.” Journal of Economic Perspectives 28(3): 3–24.
  • Haltiwanger, J., R. S. Jarmin and J. Miranda (2013). “Who Creates Jobs? Small versus Large versus Young.” Review of Economics and Statistics 95(2): 347–361.
  • Hodrick, R. J. and E. C. Prescott (1997). “Postwar U.S. Business Cycles: An Empirical Investigation.” Journal of Money, Credit and Banking 29(1): 1–16.
  • Pugsley, B. W. and A. Şahin (2019). “Grown-up Business Cycles.” Review of Financial Studies 32(3): 1102–1147.
  • Ravn, M. O. and H. Uhlig (2002). “On Adjusting the Hodrick-Prescott Filter for the Frequency of Observations.” Review of Economics and Statistics 84(2): 371–376.
  • U.S. Bureau of Labor Statistics. Business Employment Dynamics. bls.gov/bdm
  • U.S. Census Bureau. Business Dynamics Statistics. census.gov/programs-surveys/bds

Page created on 20/11/2026.

© Arthur Galichère.

  • Errors: Every effort has been made to ensure the data and calculations are accurate; any remaining errors are mine. If you believe you have found a mistake, or have a comment or suggestion, please let me know.
  • Reuse and citation: The data are those of the U.S. Census Bureau (Business Dynamics Statistics), the U.S. Bureau of Labor Statistics (Business Employment Dynamics) and the U.S. Bureau of Economic Analysis (real GDP, via FRED). The rates computed here, the statistics, cycles, charts and text are the author’s. When reusing these figures, please cite the original data providers and this website: Arthur Galichère, Business Dynamics.