Demographic Origins of the Startup Deficit

The paper addresses the multi-decade decline in the U.S. startup rate—from about 13% in 1979 to roughly 10% by 2007—despite relatively stable survival and growth patterns of incumbent firms when conditioned on age. The central hypothesis is that a demographic-driven slowdown in labor supply growth, beginning in the late 1970s, is the primary cause of the “startup deficit.” In general equilibrium with free entry and decreasing returns at the firm level, changes in labor supply growth must be absorbed at the entry margin along a balanced growth path: slower labor supply growth necessitates slower growth in the number of firms, reducing the startup rate. The authors formalize this link in a flow-balance framework and a calibrated firm dynamics model, and test it using cross-state instrumental variable strategies and a time-series imputation that extends the startup series back to the 1960s. The study’s purpose is to provide a parsimonious, quantitatively significant, and empirically validated explanation for the broad-based decline in U.S. firm entry and to reconcile it with the stability of incumbent firm dynamics by age.

The paper builds on the literature documenting declining U.S. business dynamism and startup rates (Davis et al., 2007; Haltiwanger, Jarmin, and Miranda, 2011, 2012; Decker et al., 2014; Hathaway and Litan, 2014a; Pugsley and Şahin, 2019). Prior work explored correlates (population growth, consolidation, import competition) and ruled out compositional change across sectors and geographies as the primary driver. Quantitative explanations include skill-biased technical change effects on entrepreneurship (Salgado, 2017; Kozeniauskas, 2017). Demographics beyond growth—such as aging—have been shown to depress entry through workforce age structure and consumer behavior (Liang, Wang, and Lazear, 2018; Engbom, 2019; Bornstein, 2018). The authors’ contribution is to identify and quantify the causal role of demographic-driven labor supply growth changes on the equilibrium startup rate, and to validate it using both cross-state instruments and extended time-series evidence. The study also connects to research on macro consequences of declining entry (e.g., effects on productivity growth, labor share, jobless recoveries).

Data and measurement: The authors use the U.S. Census Bureau’s Longitudinal Business Database (LBD) and public Business Dynamics Statistics (BDS) for 1976 onward, constructing firm-level measures by aggregating establishments. Startup rate is the share of new employer firms among all employers. Exit is when all prior-year establishments of a firm report zero employment and closure; exit rates are computed economy-wide and by age. Employment growth by age cohort is decomposed into survival and conditional growth. Labor supply growth is proxied by working-age population (WAP, ages 20–64) and civilian labor force (CLF) growth.

Flow-balance framework: A simple model with identically sized firms, exogenous exit x, and exogenous labor supply growth η yields an equilibrium startup rate SR = (η + x)/(1 + η). This equates the inflow of entrants per worker to outflows due to exit and labor force expansion, providing a transparent decomposition of changes in SR into contributions from η and x.

Structural model: The paper extends Hopenhayn and Rogerson (1993) to include exogenous labor supply growth. Households grow at rate η and supply labor inelastically; firms have decreasing returns production with idiosyncratic productivity (a permanent component a drawn at entry; a persistent component s_t following AR(1)), pay fixed operating costs and quadratic adjustment costs, and face exogenous exit risk δ and endogenous exit decisions. Entry is free with cost c_e; wages and interest rates are determined so that free entry holds in balanced growth. Key implication: along the balanced growth path, free entry renders aggregate labor demand infinitely elastic to maintain a constant real wage, so changes in η are absorbed at the entry margin. The model is calibrated to 2005–2007 U.S. moments (startup rate, average startup and incumbent size, exit and conditional growth for young firms across size bins) and pins down parameters governing costs and productivity.

Cross-state empirical strategy: State-year regressions relate margins of firm dynamics to labor supply growth with state and year fixed effects. Two IVs provide exogenous variation in labor supply growth: (1) a fertility instrument using state birthrates lagged 20 years; (2) an Altonji–Card style migration instrument that weights other states’ contemporaneous labor supply growth by historical birthplace shares to capture predicted migration inflows, excluding same-division neighbors. First-stage relevance is demonstrated for both IVs. Robustness checks include detailed industry fixed effects (NAICS4), alternative labor supply measure (CLF), and state-specific linear trends, as well as clustering by state.

Time-series imputation: To test implications prior to the LBD period, the authors impute an establishment startup rate back to 1965 using County Business Patterns (CBP). They recover gross entry as net change in establishments plus predicted exits. Exit rates by state and size bin are estimated from BDS (1980–2007) with linear time trends and then projected back (holding 1980 levels pre-1979 as baseline), allowing computation of an imputed establishment startup rate that can be compared to labor supply growth trends.

• The startup rate fell from about 13.0% (1979–81 average) to 10.1% (2005–07), a 2.9 percentage point decline; the aggregate exit rate also fell by about 1 percentage point.
• Flow-balance prediction: Using SR = (η + x)/(1 + η) with WAP (CLF) trends, the predicted startup rate decline from 1979–81 to 2005–07 is 1.6 (2.0) percentage points, explaining roughly 55–70% of the observed decline. Decomposed, changes in η alone explain about 46–59% and changes in x about 40–53% of the total decline (depending on WAP vs CLF).
• Structural model: Increasing η from 1.1% (2005–07) to 1.9% (WAP) or 2.5% (CLF) raises the startup rate from 9.8% to 11.0% or 11.9%, accounting for 41% (WAP) to 72% (CLF) of the historical decline. The model endogenously generates a decline in the aggregate exit rate explaining 55% (WAP) to 100% (CLF) of the observed exit decline via age-composition effects. The implied elasticity of the startup rate to labor supply growth is about 1.5; roughly 58–59% is the direct effect of η on net entry, and 41–42% is the indirect effect via endogenous exit.
• Stability of other margins: Conditional on age, incumbent exit and conditional growth show no trend; startup size distribution is stable (mean around 6 employees). The model reproduces this: changes in η leave equilibrium wages unchanged along the balanced growth path, so incumbent dynamics and entrant sizes remain stable; changes occur via the number of entrants and firm age composition.
• Cross-state IV estimates: The elasticity of the startup rate to working-age population growth is about 1.19 (using both IVs), consistent with the model. This explains roughly one-third of the aggregate startup decline when multiplied by the national decline in η. Results are robust to industry controls, CLF growth measure, and state-specific trends. OLS elasticities are smaller, consistent with attenuation from labor demand shocks affecting incumbents.
• Incumbent dynamics: IV estimates show no statistically significant effects of labor supply growth on startup size, young-firm exit, or young-firm conditional growth, aligning with the model’s prediction that adjustments occur at entry, not along incumbent margins.
• Extended time series: The imputed establishment startup rate rises in the 1960s–1970s during accelerating labor supply growth and declines thereafter, closely co-moving with HP-filtered trends in WAP and CLF growth. Overlapping years show the imputed series tracks BDS establishment entry reasonably well.
• Alternative cost-based explanations (entry, operating, adjustment costs) require large and counterfactual parameter changes and predict changes in entrant size and incumbent margins inconsistent with observed stability; adjustment costs have negligible impact on long-run entry in the model.

The findings support a general equilibrium mechanism in which demographic-driven changes in labor supply growth are absorbed at the entry margin due to free entry and decreasing returns at the firm level. Slower labor force growth since the late 1970s reduces net entry requirements, lowering the equilibrium startup rate. The induced shift in the age distribution toward older firms lowers the aggregate exit rate without altering age-conditional survival and growth, reconciling the broad-based decline in entry with the empirical stability of incumbent dynamics. Cross-state IV evidence corroborates a causal elasticity of startup activity to labor supply growth near unity, and a longer-run imputed series confirms the co-movement between startup rates and labor supply growth over both the rise (1960s–70s) and fall (post-1980) of labor force growth. Relative to alternative hypotheses based on rising costs or frictions, the demographic mechanism both fits the data quantitatively and avoids counterfactual implications for entrant size and incumbent behavior. The results highlight that much of the startup deficit is preordained by demographic trends rather than shifts in business conditions or regulations, though other factors may contribute at the margin.

The paper identifies slowing labor supply growth—largely determined by demographics—as the dominant driver of the long-run decline in the U.S. startup rate. A simple flow-balance relation and a calibrated Hopenhayn–Rogerson style model show that changes in labor force growth can explain about half to two-thirds of the observed drop in entry, with the remainder well accounted for by an endogenously declining aggregate exit rate via age-composition effects. Cross-state IV estimates and an imputed historical series extending to 1965 provide independent validation. The mechanism also explains the stability of age-conditional survival and growth, and the broad-based nature of entry declines. Given continuing low fertility and modest labor force growth, the startup rate is likely to remain subdued, with implications for the firm age distribution and macro outcomes such as productivity growth, labor share dynamics, and recovery patterns. Future research could quantify interactions between demographics and innovation, heterogeneity across industries and regions, the role of immigration and participation margins in shaping labor supply growth, and policy levers that might offset demographic headwinds to entry without distorting incumbent dynamics.

• Identification limits: Cross-state IV estimates rely on fertility and migration instruments; while strong and plausible, they may not capture full general equilibrium feedbacks and primarily exploit short-run/year-to-year variation rather than long-run transitions.
• Time-series imputation: The historical establishment startup rate is imputed using predicted exit rates (by state and size) extrapolated from later BDS trends; alternative assumptions, while tested, still imply some measurement uncertainty and apply to establishments rather than firms.
• Model structure: The structural model assumes no aggregate uncertainty, a specific production technology with decreasing returns, inelastic labor supply, and free entry; alternative frictions or technology could alter quantitative results. Adjustment costs are modeled quadratically; costs and productivity processes are calibrated rather than directly observed.
• External validity: State-level analyses assume limited interstate mobility attenuation and exclude some potential confounders; results may understate aggregate elasticities if mobility dampens local effects.
• Measurement choices: Use of HP filters for trend extraction and specific age definitions (e.g., working-age 20–64) may influence precise quantitative decompositions, though robustness checks suggest stability of main conclusions.