RESEARCH DOSSIER

Entry Exit and Heterogeneous Firms

Abstract. This research dossier investigates firm entry, continuation, exit, and the evolution of heterogeneous producer populations. It examines irreversible entry, informational learning, productivity transitions, dynamic stopping, stationary industry distributions, innovation and market-access selection, and the empirical distinction between technical efficiency and profitability. The analysis combines conditional derivations, transparent hypothetical examples, comparative empirical research, and official business-demography conventions. It distinguishes productive formation from administrative registration, and accounting contributions from causal and welfare effects. Its analytical synthesis establishes why demographic counts alone are insufficient to evaluate productive selection or aggregate efficiency. The study uses documented source findings and does not claim a new theory or independent estimation of confidential microdata.

Keywords: firm dynamics; market entry; exit; productivity heterogeneity; firm selection; business demography; sunk costs; survival; resource reallocation; industry equilibrium; learning; microeconomics

The economic object and its boundaries

A market is a changing population of productive organizations. Its aggregate output is produced by incumbents, by businesses that have recently begun operation, and by organizations whose operations will cease. A theory of production that describes only the optimal output of a continuing firm therefore leaves one of the determinants of industry output unexplained: the constitution of the population itself. Firm dynamics extends the microeconomic analysis of production to the decisions through which firms enter, survive, expand, contract, and leave.

A firm is an organization that coordinates inputs and exercises control over production decisions. An establishment is a geographically or operationally identifiable producing unit. An enterprise is a statistical organizational unit whose precise administrative definition depends on the data system. A firm may possess several establishments, enter an industry with a new establishment, or change its industrial composition while continuing to exist. The economically relevant entry event must consequently be specified before the incidence of entry is measured.

Entry in its strongest productive sense involves the mobilization of factors for a new productive activity. Entry to a product market can also occur through diversification by an incumbent elsewhere. Registration confers a legal status and may precede production, accompany it, or formalize an existing activity. Exit can mean abandonment of a product, permanent closure of an establishment, liquidation of an enterprise, or withdrawal of a legal unit. Ownership transfer can preserve the productive organization. These events have different implications for competitive supply, organizational knowledge, jobs, and resource reallocation.

The empirical classification matters. In their study of U.S. manufacturing, Dunne, Roberts, and Samuelson (1988) distinguish genuinely new firms from incumbents entering an industry through a new facility or a change in output mix. Their entry and exit study covers four-digit industries from 1963 through 1982. A quantity that combines these events can describe market entry, but it cannot without additional information measure the creation of new firms across the economy.

Firm heterogeneity denotes dispersion in the characteristics determining feasible production and expected returns. Relevant characteristics include physical productivity, organizational capability, product quality, demand, financing opportunities, capital vintage, location, managerial information, age, and exposure to uncertainty. Productivity is a conditional input-output relationship; profitability is a monetary residual after costs. A firm can possess a productive technology and fail because it cannot finance working capital or reach customers. It can survive with a comparatively inefficient technology because it has advantageous demand, a protected asset, or unusually cheap funding. These distinctions ground the entire investigation.

The scope of this dossier is the producer's dynamic problem and the evolution of the producer population. Strategic market power, mergers as a competition-law question, and the general distribution of wealth are adjacent objects. They enter here only insofar as they affect entry, exit, selection or measurement. The method combines conditional derivations from established economic concepts, transparent illustrative calculations, and a critical review of empirical findings. No new confidential microdata have been estimated, and the illustrative results below are analytical examples rather than estimates for a particular country.

Historical development of firm dynamics

The transition from a representative production unit toward a distribution of firms required two intellectual moves. First, it was necessary to explain why organizations facing a common price could have divergent outcomes. Second, it was necessary to make entry and departure endogenous, so that the distribution itself becomes an equilibrium object. A static cost curve alone answers neither question. It supplies the current production choice conditional on the firm's existence, while dynamic analysis supplies the conditions of existence.

Jovanovic's (1982) Selection and the Evolution of Industry gives a learning-based account of differences in growth and survival. A producer discovers information about its own efficiency through operation. Good information can support expansion, and unfavorable information can lead to withdrawal. Size and age can thus reflect an accumulated history of decisions and signals rather than immutable firm types. Its explanation depends on the informational environment and the selection process, not on the claim that every individual firm improves mechanically with age.

Hopenhayn's (1992) dynamic industry equilibrium develops an explicit competitive industry with idiosyncratic productivity uncertainty and irreversible entry expenditure. The population distribution can be stationary even while firms change size and are replaced. This provides a population-level equilibrium concept in which aggregate constancy and continuing microeconomic adjustment coexist. The methodological significance is that an equilibrium must determine the distribution of active producers, not merely the output chosen by a hypothetical typical producer.

Subsequent work linked the dynamic population to measurable production behavior. Olley and Pakes's (1992 working paper; 1996 journal article) telecommunications equipment study recognizes that firms choose inputs and liquidation in response to information about productivity. This connection makes measurement part of the same scientific problem as selection. If the sample contains only surviving firms or if an unobserved productive advantage causes greater input use, a conventional regression does not automatically recover the underlying technology.

The framework also became important for trade and innovation. Melitz (2003) embeds heterogeneous producers in a trade equilibrium, making market participation conditional on productivity and additional market-access costs. Klette and Kortum's (2004) Innovating Firms and Aggregate Innovation links firm innovation processes to industry turnover and technological progress. In the latter research program, firms can gain or lose productive opportunities through innovation rather than through an exogenous efficiency draw alone. Each mechanism illuminates a distinct source of heterogeneity. Their assumptions must be preserved when their implications are compared.

Empirical research has enlarged the question again: a dynamic industry can be studied not merely through the frequency of entry and exit but through the speed with which high-return opportunities attract resources. The outcome of transition matters alongside the transition rate. The resulting analytical agenda combines productivity measurement, organizational demography, labor flows, institutional constraints, and the distribution of economic opportunities.

Production decisions and the meaning of productivity

Consider a one-output price-taking producer with productivity a > 0, capital k > 0 fixed within the period, labor l >= 0, wage w > 0, output price p > 0, and avoidable operating cost f >= 0. Its technology is y = a k^alpha l^beta, with 0 < beta < 1 and alpha >= 0. Holding the predetermined state fixed, the interior operating choice solves p a k^alpha beta l^(beta-1) = w. The optimal labor input is

l* = (p a beta k^alpha / w)^(1/(1-beta)).

At that choice, the operating contribution before the fixed cost is (1-beta)p y*. Hence the maximized current operating profit is (1-beta)p y* - f. These formulas are elementary conditional implications of diminishing marginal returns. For common prices and capital, current operating profit increases with a. If higher a is also associated with better expected future states, a threshold-type continuation decision becomes plausible. The inference depends on both the current solution and the transition process.

Capital is held fixed only within this calculation. Once investment becomes a choice, the firm compares the acquisition cost and future productive benefits of capital; with adjustment costs, investment may be discontinuous. An inherited machine can have a low resale value but a high continuation value. Its original purchase cost is sunk. An alternative-use value or a debt payment falling due can nevertheless affect the forward decision. Economic cost, accounting depreciation, and cash outflow must remain separate.

The revenue counterpart to the production function is r = p_i y_i. If a common industry deflator P is used, log deflated revenue is log y_i + log(p_i/P). Its residual after subtracting the elasticities of measured inputs includes log a_i and a relative-price component. The residual may also absorb differences in product mix, quality, measurement error, utilization, and unmeasured inputs. Two producers with the same technology therefore need not have the same measured revenue productivity; two producers with identical observed revenue productivity need not have identical technologies.

In their productivity and profitability analysis, Foster, Haltiwanger, and Syverson (2008) use settings where quantities and prices can be observed separately. The authors report opposing price relationships for physical and revenue productivity and lower prices among their young producers. Such findings constrain the interpretation of selection regressions. A positive association between a sales-based productivity measure and survival mixes production efficiency with the conditions under which output is sold.

Consider a hypothetical example. Producer A uses ten units of a composite input, produces twenty physical units, and charges a price of one. Producer B uses ten inputs, produces fifteen units, and charges two. A has physical output per input of two; B has 1.5. Revenue per input is two for A and three for B. If the input price is one per unit and there are no other costs, profits are ten and twenty respectively. The rankings differ. If a subsequent demand shift changes B's price to 0.8 while output and inputs remain fixed for the period, its profit falls to two while its technical productivity remains 1.5. These constructed figures demonstrate the separate information carried by the three measures. They do not estimate a real firm's efficiency or infer any demand elasticity.

Even the physical ranking depends on the meaning of a unit of output. A medical service and a standardized manufactured component are not comparable through counts alone, and quality variation can matter within a narrowly defined product. Where multiple outputs are produced jointly, explicit output indexes or a multi-output production technology are needed. The empirical analyst must therefore name the productivity concept, the unit of observation, the prices and deflators, and the assumed production-function form before assigning an economic interpretation to dispersion.

Entry, continuation and the option to exit

An entrepreneur chooses entry before receiving all information about the future. An entry project requires resources that could be committed elsewhere, and the expenditure may be irreversible. Let c_e be the sunk cost of attempting entry, G the distribution of initial entrant states, V(s) the value after a state s becomes known, and beta_d in (0,1) the discount factor. In a timing convention where the attained state is realized next period, the value of an entry attempt today is beta_d E_G[V(s)] - c_e. Immediate realization would instead omit that initial discount factor. Timing must be stated because it changes the numerical entry condition.

If free entry is available to a continuum of identical prospective entrants and competition adjusts prices or other equilibrium objects, positive entry is consistent with beta_d E_G[V] = c_e. If entry attempts are zero, the corresponding inequality is beta_d E_G[V] <= c_e. These conditions assume an unrestricted extensive margin and a stable entry prospect. When licenses are rationed, financing is scarce, or entrepreneurial opportunities differ, some prospective firms can face binding constraints while others earn positive expected surplus. Free entry is a particular institutional organization of opportunity rather than a property of all markets.

For an incumbent, define V(s) at the beginning of an operating period and d(s) as the realizable payoff from immediate exit, net of closure costs. Let pi(s,u) be current profit under feasible action u and P(ds'|s,u) a transition probability on future states. A conventional stopping formulation is

V(s) = max{d(s), sup_u [pi(s,u) + beta_d E[V(s')|s,u]]}.

Existence and boundedness require additional regularity, such as bounded returns or suitable growth conditions on state and rewards. Decisions are conditional on information actually available. When the continuation term strictly exceeds d(s), continued operation is optimal; when it is strictly smaller, immediate exit is optimal. Equality permits indifference. Default, legal restrictions, or irreversible capacity loss may constrain the available actions and must be represented in the feasible set.

Suppose current avoidable profit is negative, but a future recovery remains possible. If immediate exit pays zero, current profit is minus two, next period yields nine with probability one half and zero otherwise, and beta_d = 0.9, a two-period continuation calculation gives -2 + 0.9(4.5) = 2.05. Continuation is valuable despite the current loss. At a recovery probability of one fifth, continuation is worth -0.38, and closure is preferred. Its break-even recovery probability is 2/(0.9*9), approximately 0.247. This is a finite-horizon pedagogical calculation with no intermediate options, no further payoff after the second period, and the stipulated terminal return. It isolates the role of expectations. A liquidity-constrained firm may still close in the first scenario even if its continuation value is positive.

Under one-dimensional ordered states, increasing current returns, and stochastically monotone transitions, continuation value can be monotone. With constant exit value, a suitable single-crossing property then yields an upper continuation region and a lower exit region. If liquidation proceeds increase rapidly with productivity, if firms face nonmonotone financing terms, or if productive assets have unusually high alternative values, continuation may no longer have this simple ordering. A scalar threshold should be derived from the assumptions rather than imposed on all real firms.

The sunk cost of original entry and the avoidable cost of continued operation enter at different decision points. The incumbent can rationally remain at a price that makes entry unattractive for an otherwise identical outsider because the outsider still must incur the initial expenditure. The difference is an ordinary consequence of irreversibility. An industry may therefore display low entry without high exit, or delayed exit after a temporary reduction in demand. A reversible-entry approximation removes this asymmetry and can substantially alter the predicted population response.

Learning, growth and selection

Learning introduces an information state. An entrepreneur may know the technology available in principle but be uncertain about her own cost parameter, the reliability of a process, or the demand for a product. Operation generates observations that update expected returns. For a simplified Bayesian illustration, let an unknown permanent parameter theta have a normal prior with mean m_0 and variance v_0 > 0. Assume observations x_j = theta + e_j with independent normal noise of known variance sigma_e^2 > 0. After n observations, posterior precision is 1/v_0 + n/sigma_e^2 and posterior mean is (m_0/v_0 + sum_j x_j/sigma_e^2)/(1/v_0 + n/sigma_e^2).

This illustrative updating rule describes information accumulation under stated distributional assumptions. It is not a transcription of Jovanovic's specific technology or theorem. A firm with a favorable posterior can choose to expand; one with an unfavorable posterior can stop. Conditional on previously surviving, the population of older firms will have a different distribution of inferred prospects from the original cohort. Expected parameter values and posterior uncertainty are distinct state variables in such an analysis.

Jovanovic (1982) explains the link between incomplete information and selection. The implication for evidence is precise: estimates of age-related growth or productivity taken only from continuing firms reflect the firms selected into observation. Consider two equally weighted entrants with constant physical productivity of 1 and 3. If the first closes and the second survives, average observed productivity rises from 2 to 3 even though neither technology changed. The fifty percent rise in the observed mean must not be called technological progress inside the surviving firm.

Now suppose both entrants remain, but the stronger firm grows relative to the other. A size-weighted mean can rise even if the unweighted mean does not. These are separate sources of aggregate improvement: observed productivity may rise through selection on participation, reallocation of weights among continuers, and genuine changes inside individual firms. Different statistical constructions distinguish these channels with different precision.

Passive learning is only one mechanism. Active investment can change the state itself, while organizational training can alter both the mean and uncertainty of future outcomes. An omitted factor, such as founder skill, can generate persistence without ongoing learning. A favorable demand shock can produce growth without improved technology. Consequently, a low hazard among old firms is consistent with selection, accumulated capabilities, persistence in demand, access to cheap finance, or combinations of these. Learning is supported more directly by evidence that firms' beliefs or responses to information evolve systematically with operating experience.

Age and size are not mutually substitutable controls. Size is partly an outcome of age, historical productivity, and market opportunity; controlling for size may condition on a mediator. Age is similarly conditioned on survival. A cohort-based analysis asks how the original population fares with time, including closures. A cross-sectional analysis asks how firms of different ages compare at a date. Neither answers the other question without assumptions about cohort effects, common period shocks, and selection. A balanced panel may be appropriate for a particular within-survivor estimand but cannot describe the complete entrant cohort.

Industry equilibrium and population distributions

The distribution of firms evolves through a stock-flow identity. Let mu_t be a finite measure of incumbent firms at the start of period t, q_t(s) in [0,1] the probability of continuing to the next period, and P_t(B|s) the conditional probability that a continuer's next state belongs to measurable set B. Let M_t denote attempts at entry and G_t(B) the subprobability measure of entrants reaching the next period in B. This permits unsuccessful entry attempts; G_t(S) is the share that becomes active. Then

mu_(t+1)(B) = integral q_t(s) P_t(B|s) mu_t(ds) + M_t G_t(B).

The formula follows from counting continuing and newly active units on a consistent boundary. It excludes mergers and changes in statistical coverage or assumes that they are already represented in the transition. If these events matter, further terms are necessary. Let N_t = mu_t(S). Then N_(t+1) - N_t equals the active entrants less departing incumbents. In a stationary finite population with a common unit and timing, entry and exit masses balance, while states of individual firms can continue to change.

For a finite-state pedagogical version, let T be the substochastic transition matrix among surviving states, g a row vector of entrant state probabilities, and m a constant entrant mass. With row-vector convention, stationarity solves mu = mu T + m g. If the spectral radius of T is less than one, the inverse of I-T exists and

mu = m g (I-T)^(-1) = m g (I + T + T^2 + ...).

Each term corresponds to a different surviving cohort. The finite-state formula shows how the current distribution depends on both the entry mix and the duration of participation. It is a standard linear representation of an absorbing transition process and is used here to explain established firm-dynamics reasoning, not to introduce a new economic model.

For two types L and H, suppose T is diagonal with survival probabilities 0.5 and 0.9, and ten new active firms of each type enter every period. Stationary masses are 10/(1-0.5) = 20 and 10/(1-0.9) = 100. The entrant mix is fifty percent high type; the active population is approximately 83.33 percent high type. Ten exits of each type are offset by ten entrants of each type. With physical productivities 1 and 3, the average among active firms is (20+300)/120 = 8/3, above the entrant mean of 2. This is duration weighting without any within-firm technological change.

Changing the low type's survival probability to 0.8 while preserving the entrant flow raises its stationary mass to 50; the high type remains at 100. The population average becomes (50+300)/150 = 7/3. The mass expands and average productivity declines. If each firm of either type uses one unit of input per period in this example, total output rises from 320 to 350 because the population's total inputs increase. The result demonstrates why mean productivity, total output, and business survival can point in different directions. In a general equilibrium, input scarcity and market demand would also adjust; the partial example cannot be interpreted as a welfare result.

Hopenhayn (1992) provides the economic foundations for this distinction between stationary industry aggregates and changing firm-level states. An equilibrium must combine optimal firm decisions, a consistent distribution, an entry condition, and market clearing. For example, at a clearing output price, demand equals the integral of firm output over the active population. Input markets also clear at their relevant prices. Solving the entry condition at a fixed price and then increasing entrant mass indefinitely without changing prices generally violates these requirements.

Stationarity can fail if firms accumulate indefinitely, entry flows vary persistently, transition probabilities change, or expected durations are not finite. A stationary distribution of normalized shares need not be a stationary distribution of firm counts. A growing economy may instead possess a stable distribution of relative sizes while the scale of production changes. The empirical decision to use a stationary model should be matched to the horizon and environment under examination.

Innovation, heterogeneous opportunity and international selection

In competitive industry dynamics, productivity changes can be modeled as shocks, as the accumulation of knowledge, or as outcomes of directed investment. Each treatment carries a different economic interpretation. If a firm's innovation effort affects the arrival of successful projects, its transition law is endogenous. Entry can occur because an innovation creates a feasible project. Exit can result from the loss of all viable projects rather than from an isolated negative shock. Klette and Kortum (2004) connect this innovation perspective to aggregate technological change and firm dynamics. The important boundary is that a productivity transition estimated under a fixed law may not describe the response to a policy that changes innovative effort.

International market access creates additional participation decisions. In a standard constant-elasticity differentiated-product setting, assume elasticity sigma > 1, common wage w, labor productivity phi > 0, and the markup sigma/(sigma-1). Domestic marginal cost is w/phi. If the market demand index is fixed and domestic revenue is A_d phi^(sigma-1), domestic current profit under fixed cost w f_d equals A_d phi^(sigma-1)/sigma - w f_d. The domestic zero-profit cutoff in this simplified comparison is (sigma w f_d/A_d)^(1/(sigma-1)). For destination x with iceberg cost tau >= 1 and revenue coefficient A_x, export revenue is A_x (phi/tau)^(sigma-1), and the export cutoff is tau (sigma w f_x/A_x)^(1/(sigma-1)).

These expressions are elementary CES profit conditions for a stipulated environment, not a full solution of the Melitz equilibrium. A distinct export cutoff above the domestic cutoff requires relative market-access costs and demand parameters that make exporting more demanding at the margin. If foreign demand is sufficiently favorable or domestic fixed costs sufficiently high, that particular ordering is not implied by the algebra alone. Multidestination costs produce several participation margins.

Melitz's (2003) heterogeneous-firm trade model derives selected steady-state effects of opening to trade and changing trade costs under its general equilibrium conditions. More productive firms can access export markets, resource shares shift, and some firms cease production. The interpretation here is limited to the mechanism and model's conditions. The observed disappearance of a domestic producer after liberalization could, in a different setting, reflect exchange-rate changes, imported-input costs, liquidity shocks, or local demand movements. Empirical evidence must distinguish these channels.

Pavcnik's (2002) Chilean plant-level trade study explicitly deals with productivity estimation and nonrandom plant exit. The study reports improvements inside import-competing plants as well as output shifts among producers after trade reform. Thus changes in average industry performance can include both selection and changed performance among continuers. A comparison that holds physical technology fixed is analytically useful, but a real episode can contain both processes.

For a multi-establishment firm, the response to international opportunity may be a change of destination mix or closure of one site while headquarters and other sites persist. A country-level export entry series can consequently rise without a corresponding increase in domestic firm births. The proper economic unit is the firm-destination, product-destination, establishment, or enterprise, according to the margin under study. Such measurement choices precede causal analysis.

Accounting for reallocation and productivity

Let p_it denote a consistently measured productivity index for firm i at date t, and s_it its nonnegative aggregation weight with sum_i s_it = 1. Define aggregate P_t = sum_i s_it p_it. The arithmetic applies to this specified index; a weighted mean of log productivity is not generally identical to aggregate output divided by a composite quantity of input. A researcher must justify the index and weights for the question.

For a fixed population of n firms, define mean pbar = sum_i p_i/n and mean share 1/n. Then

P = pbar + sum_i (s_i - 1/n)(p_i - pbar).

The second term records how weights covary with measured productivity, with the normalization made explicit: it is n times the covariance computed with divisor n. An increase in the term can result from high-productivity firms expanding or from changes in productivity among firms already holding large shares. The covariance is descriptive, because high sales shares may reflect prices or demand and because productive capacity responds to input-market conditions.

For two periods t=0,1, distinguish continuing firms C, entrants E, and exits X; denote changes by Delta. A base-period centered exact decomposition is

Delta P = sum_C s_i0 Delta p_i

  • sum_C Delta s_i (p_i0 - P_0)
  • sum_C Delta s_i Delta p_i
  • sum_E s_i1 (p_i1 - P_0)
  • sum_X s_i0 (p_i0 - P_0).

To establish the identity, expand the product s_i1 p_i1 for continuers, add entrants, and subtract the old products of exiting firms. Centering the share changes and demographic contributions on P_0 introduces terms proportional to the difference between total period-one and period-zero shares. Both sums equal one, so those terms cancel. The decomposition is defined for the chosen endpoints; it does not estimate the hypothetical productivity of an exited producer had it survived. Entry and exit terms are relative to initial aggregate productivity, not necessarily relative to each other.

In a constructed example, the initial population comprises A with share 0.6 and productivity 2, and X with share 0.4 and productivity 1. Then P_0=1.6. In the next period A has share 0.7 and productivity 2.2, and entrant E has share 0.3 and productivity 1.8. P_1=2.08 and the difference is 0.48. The continuing within term is 0.12, the between term 0.04, the interaction term 0.02, the entry term 0.06, and the exit term 0.24. Their sum is 0.48. The example shows that an entrant less productive than a continuing producer can have a positive entry contribution because its comparator is the original aggregate. It also shows how an accounting convention can assign a positive exit contribution without establishing the welfare value of a particular closure.

Changing the time window changes the classification. A new firm that enters and exits between endpoints is absent from both populations, although it may have incurred considerable resource costs. A short interval can show initially weak entrants; a longer interval captures their surviving growth and selection. Changing weights from employment to sales or value added changes the scientific estimand. A productivity decomposition is therefore inseparable from the study's demographic boundary, index and horizon.

Hsieh and Klenow's (2009) manufacturing misallocation analysis uses a model-based comparison of manufacturing plants in China, India and the United States. Their counterfactual gains under a reallocation assumption are not observed realized gains from closing firms. Calculated wedges can combine genuine constraints, model misspecification, and measurement error. A large gap between measured marginal revenue products is a reason to investigate the allocation mechanism; it does not by itself establish that a particular industrial subsidy or closure decision will increase aggregate welfare.

The authorial assessment is that population replacement, resource reweighting and technological change should be reported as analytically distinct observations. A high entry rate can accompany small entrants and weak aggregate contribution, while a small number of highly successful firms can drive substantial growth. A scalar count of births or deaths discards these dimensions. Comparing demographic activity with aggregate productivity requires both firm histories and the economic weights of the units involved.

Business demography and measurement conventions

Administrative sources permit the observation of large populations but define their own units and coverage. The Eurostat business demography metadata defines activity using economic indicators and applies continuity rules to births, deaths, survival and takeovers. Employer demography employs a one-employee activity threshold. A firm can cross that threshold without originating or dissolving the underlying organization. A permanent death normally awaits the absence of reactivation within two years, so a newly released death count can be preliminary. A legal cancellation and an economic death can occur at different dates.

The U.S. Business Dynamics Statistics provides annual establishment births and deaths, firm startups and shutdowns, and employment-flow measures. Its referenced 2023 vintage covers 1978–2023; this statement describes the vintage checked, not a claim that newer releases cannot exist. Establishment-level employment changes can be allocated across firm characteristics. The denominator, employment reference period and age assignment should be obtained from the specific table's documentation before comparing rates.

The World Bank new business density has the identifier IC.BUS.NDNS.ZS. It measures formal limited-liability registrations during a year per thousand persons aged 15–64. It has a population denominator, not a stock of active firms. It excludes informal activities from its operational definition, and the underlying source systems differ across participating economies. The series can inform the study of formalization and entrepreneurial registration intensity. It cannot be converted into an enterprise birth rate by shifting the decimal point.

A reproducible longitudinal data dictionary should contain a permanent unit identifier; organization and establishment links; founding or first-observed date; legal registration date if available; activity status and status reason; industry at each reference date; geography; employment, hours, wage bill and other inputs; nominal quantities and prices where observable; capital and its valuation method; output measure; calendar period; and source vintage. Discontinuity flags should identify changes in ownership, mergers, reorganizations, industry codes and coverage. Weights, missing values and disclosure suppression require their own fields. A data dictionary describes available inputs; no variable is assumed present solely because it is desirable.

For annual active-population denominator A_t, enterprise birth rate b_t = B_t/A_t and death rate d_t = D_t/A_t. For employer units, the numerator changes as well as the population. In a simplified stable-boundary study with N_0 beginning stock, 100 active entrants, 80 departing incumbents and no other transitions, a net stock change of 20 follows. If N_0=1,000, rates normalized by beginning stock are 10% and 8%, whereas rates normalized by all 1,100 distinct units active at some point in the period would be about 9.09% and 7.27%. The same underlying transitions yield different reported percentages. The second denominator in this example is an explicitly constructed period-activity count, not a claim about any particular national statistical agency.

A cohort survival rate at age a is the fraction of an original entrant cohort still active under the stipulated continuity rule. A discrete hazard at age a is the conditional probability of exit over the interval among units at risk at its beginning. For consistent observation, survival through age a is the product over prior intervals of one minus each interval hazard. A firm disappearing from a survey may be an exit, an unmatched identifier, a refusal, or a unit outside the current sampling frame. Treating all forms of missingness as closure biases the estimated hazard. Right censoring requires explicit treatment where the subsequent history has not yet been observed.

Businesses can move to the informal sector without terminating production, and informal activity can enter a registration database without creating a new productive unit. Harmonization across countries should therefore be attempted at the level of definitions and risk populations. The same printed term, such as startup, frequently describes a different event.

Identification, research design and statistical diagnostics

An empirical analysis of firm productivity is subject to at least three related problems: unobserved heterogeneity, simultaneity in input choices, and endogenous observation. Consider log output y_it = alpha k_it + beta l_it + omega_it + epsilon_it, where omega_it is an efficiency state known to the firm when it makes some decisions and epsilon_it is an unanticipated disturbance. If higher omega induces higher labor use, the regressor l is correlated with the composite error. Ordinary least squares need not recover beta. If continued observation depends on omega, estimates on surviving firms may also differ systematically from those in the original cohort.

Olley and Pakes's semiparametric production-function approach makes the economic decisions generating these problems explicit. A proxy-variable strategy uses an observed decision informative about otherwise latent productivity, under assumptions concerning timing and invertibility. The chosen input must contain distinct identifying information. Zero investment, adjustment constraints and measurement problems can limit the scope for proxy inversion. A valid regression specification must distinguish flexible inputs from predetermined inputs and explain what information firms observe before choosing each.

An empirical hazard specification might be Pr(exit next period | currently active, z_it) = Lambda(alpha_age + beta'z_it + period effects + industry effects), with Lambda the logistic link. A positive coefficient on a debt ratio is an association conditional on the specified variables. Debt may reflect the financing of expansion or prior adverse shocks. A production shock can simultaneously change both the leverage ratio and closure probability. Causal interpretation requires credible exogenous variation or additional identifying restrictions. Age-period-cohort effects also have an exact linear dependency; imposing unrestricted additive effects for all three does not give uniquely identified coefficients without further restrictions.

A study of registration reform must identify what treatment changes. A licensing intervention may reduce an administrative delay, eliminate a fee, alter enforcement, or change the legal status of preexisting firms. A staggered reform can offer useful variation, but comparisons across early and late adopters need credible pretrends and attention to heterogeneous treatment effects. If entrants draw workers and customers from untreated neighbors, the no-interference assumption is violated. A reform estimate can describe the total local change in a specific outcome without demonstrating the national general equilibrium effect.

Bruhn's (2008) Mexican registration reform study exploits implementation at different times across municipalities and reports a five percent increase in registered businesses in eligible industries in the working-paper abstract. The reported additional businesses were associated with previous wage earners becoming owners, rather than greater registration propensity among existing informal owners. This is evidence for a particular institutional episode and the outcome measured there. It does not establish a universal response to all licensing reforms. The later journal edition is bibliographically identified; the quoted estimate comes from the 2008 working-paper record.

A robust empirical workflow begins with continuity checks and a demographic event table; documents coverage changes; reports raw distributions and age-size composition; checks extreme values and missingness; estimates technology or a clearly labeled alternative performance proxy; examines cohort hazards; and tests sensitivity to different measurement and aggregation choices. A causal design should document treatment timing, comparison populations, anticipation, spillovers, and whether the outcomes are levels, rates or changes. Confidence intervals concern estimation uncertainty under the sampling and modeling assumptions. They do not automatically reflect uncertainty about the definition of a firm or a future revision in source data.

A fully identified structural model additionally needs assumptions about the stochastic process of productivity, the information set, adjustment technology, entrant opportunity distribution, costs and equilibrium closure. Different combinations of entry costs and entrant quality can produce the same observed count. Different combinations of productivity shocks and financial constraints can produce the same exit rate. The data must supply moments or independent restrictions that separate these mechanisms. Matching a stationary size distribution alone cannot establish their unique identification.

Comparative empirical record

The United States provides several complementary examples. The telecommunications equipment research by Olley and Pakes documents industry restructuring and an allocation channel in which capital moves toward productive plants. The study's finding concerns that industry and period of technological and regulatory change. It is a concrete reason to distinguish an aggregate increase in productivity from an increase in average plant-level growth.

Haltiwanger, Jarmin, and Miranda's (2013) firm age and job creation study reports that the apparent size-growth relationship is sensitive to specification, while conditioning on firm age removes the systematic size relationship in their analysis. Startups and young businesses are central to the employment dynamics they measure. A small incumbent and a recently founded firm should therefore not be placed in a common policy category solely because their employee counts are identical.

The later business dynamism study by Decker and coauthors (2020) reports weakening employment adjustment to productivity despite greater dispersion of shocks in U.S. data. The distinction is analytically important. Low job reallocation may reflect either fewer opportunities to move resources or reduced responsiveness to opportunities that remain. The authors provide suggestive productivity evidence; the abstract does not establish the precise cause or a universally valid policy magnitude. An increase in friction is one interpretation requiring additional institutional and empirical evidence.

A multi-country perspective comes from the OECD DynEmp study by Criscuolo, Gal, and Menon (2014), which examines 17 OECD countries and Brazil and emphasizes young firms' contribution to net employment creation among small and medium enterprises. The cross-country results vary in young firms' expansion prospects. A high share of jobs attributable to entrants does not imply that most individual entrants expand rapidly; the aggregate can be concentrated in a small upper tail. Age, size, industry and observation window remain relevant.

The EU illustrates the importance of vintage and denominators. Eurostat's October 2025 2023 enterprise demography bulletin reports enterprise birth and preliminary death rates of 10.5% and 8.5%, respectively, for the EU business population. The two-percentage-point difference is an arithmetic comparison of rates with a common active-unit denominator. It is not an exact measure of growth in the beginning-of-year enterprise stock, nor an estimate of improved productivity. The death figure is preliminary in that publication; the source vintage is retained rather than being presented as the most recent estimate.

For developing economies, Roberts and Tybout's (1997) producer turnover comparison draws on manufacturing data from Chile, Colombia and Morocco. It discusses initially modest productivity effects of plant replacement and stronger effects over longer horizons as surviving entrant cohorts develop. This provides a substantive rationale for following cohorts after entry instead of attributing an entire long-run effect to the first-period difference between entrant and exit means.

The comparative cases answer different questions. Chile's trade episode bears on changes in exposure to competition and import opportunities, Mexico's registration reform on formal access costs, the U.S. evidence on job dynamics and producer allocation, and the multi-country OECD work on age and growth structures. No numerical meta-estimate is appropriate because the studies use different units, samples, treatments and outcomes. The cases are complementary demonstrations of mechanisms rather than repeated measurements of a single universal parameter.

A complete global research program would additionally require consistent treatment of informal activities, subsistence enterprises, nonemployer units, public enterprises and family ownership. The literature and official metadata considered here permit concrete conclusions for the named samples. They do not supply a harmonized worldwide microdata panel. Accordingly, the scientific claims are international in scope while their quantitative generalization remains conditional on definitions and evidence.

Finance, institutions and the quality of turnover

Financial arrangements modify the feasible set in the dynamic firm problem. A productive firm with a positive expected continuation surplus can face a borrowing constraint at the moment it must pay for inputs. The constraint may be temporary and reflect a shortfall in pledgeable assets or a collapse in cash collections. An entry project can also have positive social value while the entrepreneur lacks collateral. In these situations, selection by realized survival is partly selection by financing access. Conversely, continued access to subsidized resources can preserve a project whose expected return is below the opportunity cost of those resources.

In the two-period example developed earlier, suppose the original loss of two must be financed immediately, future borrowing is unavailable, and the firm's available cash is one. The option-valued continuation may be profitable but infeasible. A transfer of one or an equivalent loan with a suitable repayment schedule can restore feasibility; the actual value after financing depends on its terms. A firm with negative unconstrained continuation value would require a different justification. The arithmetic distinguishes the inability to finance positive expected value from the preservation of negative expected value, a distinction that cannot be recovered from observed insolvency alone.

Institutions affect the costs of forming, operating, transferring and terminating an enterprise. A change in entry requirements can influence the mass of potential experiments and the composition of prospective founders. A change in insolvency law can alter expected loss, recovery, creditor behavior and managers' willingness to abandon a project. The combined equilibrium response can be ambiguous: reducing the cost of failure may stimulate experimentation, but altered recovery probabilities may also change loan prices and credit allocation.

Labor markets mediate the welfare consequences of exit. If workers can immediately move to comparable jobs, closure can release inputs for higher-valued activities with modest transition costs. If local opportunities are scarce or skills specialized, closure may impose large earnings and relocation losses. Some valuable knowledge is embodied in a team, a supplier relation, or an organizational process and cannot be recovered through the resale price of machines alone. An enterprise's sale as a going concern can preserve this knowledge while changing ownership. The statistical coding of such an event matters for the attribution of productivity and employment outcomes.

The social appraisal of turnover therefore requires a counterfactual. It asks what would happen to the productive inputs, customers, organizational knowledge and public resources if the firm entered, continued, transferred its assets, or disappeared. Private value maximization weighs the owner's feasible alternatives. Social value can additionally account for worker transitions, consumer surplus, fiscal costs, environmental effects and innovation spillovers. These are standard welfare considerations; they do not imply a uniform policy preference for either preserving firms or increasing closure rates.

The evidence warrants separate consideration of entry conditions, expansion conditions and exit procedures. Simplifying registration may make projects easier to start without solving the constraints on their subsequent scale. An expansion policy may raise output of survivors without increasing the number of firms. Predictable closure procedures may facilitate resource reuse without changing the initial quality of entrants. Policies should be evaluated against the stated mechanism and measured outcome. Demographic totals provide a useful monitoring signal but cannot substitute for specifying opportunity costs and the population at risk.

Original analytical synthesis and robustness

The findings can be integrated around four separable empirical questions. What is the distribution of opportunities available to potential entrants? What productive and financial states determine continued participation? How do resources move among producers in response to those states? How do the observed demographic events correspond to the economic events of interest? These questions organize the synthesis of established research; they do not constitute a new theory, doctrine, or named model.

The first implication is that the observed firm population is an outcome of selection and duration. An entrant distribution contains potential or newly realized projects. An incumbent distribution weights past entrant cohorts by their conditional survival and growth. Even with fixed technologies and a constant entrant mix, different rates of continuation can produce substantially different incumbent distributions, as the two-type calculation demonstrates. Comparing an incumbent productivity distribution across economies without taking age and entrant composition into account can therefore confound production differences with population history.

The second implication concerns the information content of survival. The theoretically relevant selection criterion is expected feasible continuation value relative to exit value. Physical productivity contributes to continuation value, but the relationship can be altered by price, financing, exit options and market access. The price counterexample and the liquidity example establish separate routes through which a performance ranking can differ from a survival ranking. They identify conditions under which the inference from survival to technical efficiency fails. Testing whether selection is productive requires direct information about the omitted states.

The third implication links microeconomic transitions to aggregate measurement. The reallocation identity separates a continuing firm's performance change from the effects of changes in weights and population membership. Its exit contribution compares an observed predecessor with a chosen benchmark. A causal evaluation instead compares realized outcomes with an unobserved path under an alternative decision or policy. It requires a design supplying that counterfactual. The same principle applies to structural simulations: the model gives conditional counterfactuals, while a decomposition gives an accounting identity. The distinction prevents numerical precision from being mistaken for causal identification.

The fourth implication concerns the ambiguous sign of changes in turnover. Let the variance of idiosyncratic productivity shocks increase while the ability to adjust falls. A measurable decline in resource reallocation can coexist with greater dispersion of opportunities. Conversely, a rapid churn of low-capital firms can increase counts of both starts and closures while hardly moving major productive resources. The relevant comparison must examine shock distributions, adjustment responses, entry composition, exit composition and the weights of the units. The U.S. evidence on responsiveness reinforces the scientific relevance of this separation.

The fifth implication is the non-equivalence of demography and development. Consider two hypothetical economies with identical annual registrations. In one, formal registrations primarily create previously absent activities; in the other, they primarily reorganize existing informal production into legal entities. Both can gain from improved legal access, but their immediate implications for productive capacity differ. Likewise, equal death rates can describe efficient transfer of resources or the destruction of valuable organizational capital. The type of transition, rather than the count alone, determines the causal question.

Robustness follows from these distinctions. Recompute descriptive results with employment and output weights; retain original cohorts including exits; separate establishment from firm transitions; distinguish new productive units from registry events; test measurement with physical quantities where feasible; examine alternative age and entry definitions; preserve source vintages and preliminary-death flags. Where the substantive interpretation changes across conventions, publish the change instead of selecting the convenient convention. Where an estimated effect is driven by the upper tail or by one sector, report this concentration.

The provisional hypotheses set in the scientific passport are resolved at the level supported here. H1, that outcomes depend on explicit assumptions about selection and entry, is supported by the conditional derivations and counterexamples. H2, that heterogeneous environmental conditions can modify observed outcomes, is supported by the comparisons across distinct study populations and institutional episodes, without claiming a pooled causal effect. H3, that different measurement conventions alter interpretation, follows from the explicit denominator and productivity counterexamples and official metadata. None of these resolutions claims original econometric estimation or priority for the underlying mathematical identities.

Limits and future research

Several scientific problems remain open at the level of empirical identification. Information about entrepreneurs who never attempt entry is often incomplete. This limits inference about the opportunity distribution G and the consequences of entry barriers. Matching firm registration to actual operation, founder histories, financial applications and market opportunity could improve the measurement of the extensive margin. The sample must include unsuccessful applications and prospective founders, rather than only companies that appear in a business register.

The joint dynamics of productivity, demand and finance are also difficult to distinguish. Revenue productivity can change when either technology or prices change. A business can seek additional credit because it expects growth, or because it experiences distress. An investigation linking producer quantities and prices with credit records, input quantities, firm balance sheets and closure outcomes could make several competing explanations more discriminating. Access to sensitive administrative data and disclosure control would be essential.

International evidence requires common event definitions. A productive unit in one jurisdiction can be registered as a corporation, a partnership or a sole proprietorship in another. Harmonization should preserve the distinctions among enterprise births, first hiring, first sales, changes in legal identity, dormancy and final closure. Equivalent research protocols may be applied locally to confidential data, followed by aggregation of disclosure-safe results. Even then, differences in coverage and institutions require careful interpretation.

Transition costs must be measured alongside reallocation. A study of closures should follow workers, capital, suppliers and organizational capabilities to the extent that data permit. A longer horizon is needed to distinguish delayed reallocation from permanent loss. Household outcomes and local labor-market conditions can inform the welfare interpretation, but their direct incorporation involves additional assumptions and a wider scientific scope than this dossier.

Finally, the innovation channel challenges an equilibrium based on a fixed entrant opportunity distribution. If policy or technological change alters the costs and quality of new projects, the entrant distribution, survival process and responsiveness may all shift together. An investigation of those shifts needs independent measurements of entrepreneurial opportunity and innovation outcomes. This is an agenda for research under recognizable existing approaches. It is not a claim to have estimated a new structural system.

The evidence in this study consists of the accessed original theory texts, institutional metadata, statistical bulletins, and original author or publisher abstracts where full articles were unavailable. Reported empirical findings are limited to what those records support. No replication of confidential microdata, independent econometric estimation, exhaustive literature coverage, or external peer review is claimed. Illustrative arithmetic is explicitly hypothetical; the derivations state their assumptions. These boundaries preserve the distinction between a scientific synthesis and a new statistical investigation.

Conclusion

Entry and exit are integral components of microeconomic production. The output and employment of an industry are determined by the behavior of individual producers and by the evolving distribution of those producers. Firm heterogeneity makes the extensive margin consequential: different projects enter, different organizations accumulate capabilities, different firms expand, and different states lead to the continuation or abandonment of activity.

The decisive theoretical distinction is between physical productivity, monetary profitability and expected feasible continuation value. Each quantity contains distinct information. Learning, productivity transitions, market access, finance, salvage opportunities and institutional restrictions shape entry and survival. A threshold is justified when the relevant monotonicity and single-crossing conditions obtain; stationarity requires a consistent population transition with finite mass. These conditions explain the usefulness of canonical firm-dynamics models and the limits of their application.

The decisive empirical distinction is between the measured event and the productive event. Firm formation, first hiring, product entry, registration, legal cancellation, closure and transfer cannot be used interchangeably. Productivity measurement must further distinguish physical output from revenue and account for endogenous input decisions, nonrandom survival, and the aggregation weights of entering, continuing and departing units.

The comparative literature reviewed in this dossier demonstrates the importance of firm age, the possibility of allocation gains through resource movement, and differences in adjustment across countries, industries and historical episodes. It supports an assessment of business dynamism in which the quality and direction of resource reallocation matter alongside the volume of births and deaths. Profitability and productivity distributions, cohort outcomes, institutional barriers and transition costs complete the relevant evidence.

The principal authorial conclusion is that the scientific appraisal of entry and exit requires a joint reading of demographic transitions, production measurements, feasible decisions and economic counterfactuals. This conclusion synthesizes established mechanisms and identifies their evidentiary boundaries. It leaves the individual firm, its market environment, and the evolving industrial population connected within a coherent microeconomic account.

References

  • Bruhn, Miriam. 2008. License to Sell: The Effect of Business Registration Reform on Entrepreneurial Activity in Mexico. World Bank Policy Research Working Paper 4538. Journal version published 2011, Review of Economics and Statistics 93(1): 382–386. Source
  • Criscuolo, Chiara, Peter N. Gal, and Carlo Menon. 2014. The Dynamics of Employment Growth: New Evidence from 18 Countries. OECD Science, Technology and Industry Policy Papers No. 14. Paris: OECD Publishing. DOI: 10.1787/5jz417hj6hg6-en. Source
  • Decker, Ryan A., John Haltiwanger, Ron S. Jarmin, and Javier Miranda. 2020. Changing Business Dynamism and Productivity: Shocks versus Responsiveness. American Economic Review 110(12): 3952–3990. DOI: 10.1257/aer.20190680. Source
  • Dunne, Timothy, Mark J. Roberts, and Larry Samuelson. 1988. Patterns of Firm Entry and Exit in U.S. Manufacturing Industries. RAND Journal of Economics 19(4): 495–515. Source
  • Eurostat. 2025. More Businesses Opened than Dissolved in the EU in 2023. October 13. Dataset bd_size. Source
  • Eurostat. Business Demography (bd): Reference Metadata. European Commission. Accessed November 17, 2026. Source
  • Foster, Lucia, John Haltiwanger, and Chad Syverson. 2008. Reallocation, Firm Turnover, and Efficiency: Selection on Productivity or Profitability? American Economic Review 98(1): 394–425. DOI: 10.1257/aer.98.1.394. Source
  • Haltiwanger, John, Ron S. Jarmin, and Javier Miranda. 2013. Who Creates Jobs? Small versus Large versus Young. Review of Economics and Statistics 95(2): 347–361. Source
  • Hopenhayn, Hugo A. 1992. Entry, Exit, and Firm Dynamics in Long Run Equilibrium. Econometrica 60(5): 1127–1150. Source
  • Hsieh, Chang-Tai, and Peter J. Klenow. 2009. Misallocation and Manufacturing TFP in China and India. Quarterly Journal of Economics 124(4): 1403–1448. DOI: 10.1162/qjec.2009.124.4.1403. Source
  • Jovanovic, Boyan. 1982. Selection and the Evolution of Industry. Econometrica 50(3): 649–670. Source
  • Klette, Tor Jakob, and Samuel Kortum. 2004. Innovating Firms and Aggregate Innovation. Journal of Political Economy 112(5): 986–1018. DOI: 10.1086/422563. Source
  • Melitz, Marc J. 2003. The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity. Econometrica 71(6): 1695–1725. DOI: 10.1111/1468-0262.00467. Source
  • Olley, G. Steven, and Ariel Pakes. 1992. The Dynamics of Productivity in the Telecommunications Equipment Industry. U.S. Census Bureau CES Working Paper 92-02. Journal version published 1996, Econometrica 64(6): 1263–1297. Source
  • Pavcnik, Nina. 2002. Trade Liberalization, Exit, and Productivity Improvements: Evidence from Chilean Plants. Review of Economic Studies 69(1): 245–276. DOI: 10.1111/1467-937X.00205. Source
  • Roberts, Mark J., and James R. Tybout. 1997. Producer Turnover and Productivity Growth in Developing Countries. World Bank Research Observer 12(1): 1–18. DOI: 10.1093/wbro/12.1.1. Source
  • U.S. Census Bureau. Business Dynamics Statistics (BDS). Revised December 17, 2025. Accessed November 17, 2026. Source
  • World Bank. World Development Indicators: New Business Density (IC.BUS.NDNS.ZS). Entrepreneurship Database. Accessed November 17, 2026. Source