Time Averaging Meets Heckman, Lochner, and Taber and Ben-Porath
What this paper finds — and why it matters
Layer 1: Overview
Research question and motivation: How does endogenizing retirement (career-length) choice change the labor-supply and human-capital implications of the canonical Heckman, Lochner, and Taber (1998a, HLT) life-cycle general-equilibrium model, and what does this imply for social-security reform, labor-income taxation, aggregate labor-supply elasticities, and inequality? HLT already contains two ingredients of Ljungqvist-Sargent (2006) “time-averaging” models — credit markets and within-period labor-supply indivisibilities — but shuts time-averaging down by assuming inelastic labor supply until a mandatory retirement age of 65. The authors “activate” time-averaging by letting workers choose when to retire and by adding a pay-as-you-go social security system. This matters because the micro-foundation of the high aggregate labor-supply elasticity that Prescott invoked (switching from Rogerson’s employment lotteries to time-averaging) hinges on whether workers sit at corner solutions for career length.
Model setup: A perfect-foresight OLG model in discrete annual time; agents live from age 18 to 80. Eight agent types index four innate ability levels (theta in {1,2,3,4}) crossed with two education levels (high school S=1, college S=2). Each type has a Ben-Porath (1967) human-capital technology. An aggregate CES/Cobb-Douglas production function combines physical capital and two human-capital aggregates. Within-period labor is indivisible (work full time omega=1 or not omega=0). Utility is time-separable with intertemporal elasticity 1/gamma and a fixed disutility B of working. The baseline social security program has payroll tax rate tau_p=0.10, eligibility age eta_p=65, and benefit P=8 (about 40% of average earnings), paid only to retirees; collecting nothing while working after 65 creates an implicit tax that pins all workers to a corner at age 65.
Calibration: Most parameters are borrowed or backed out from HLT (delta=0.96, gamma rounded from 0.9 to 1, tau_l=tau_k=0.15, tuition zeta=1.02 thousand 1992 dollars). New parameters: disutility B=0.8, fraction of capital held by in-model agents kappa=0.388, efficiency-decline logistic parameters phi1=0.2, phi2=75. The model targets a capital-output ratio of 4 and an after-tax interest rate of 0.05; the calibrated model reproduces HLT’s baseline and post-skill-biased-technological-change (SBTC) steady states closely (e.g., baseline interest rate 0.0588 matched; aggregate human capital H1≈274/249, H2≈280/287 in HLT/our model).
Main quantitative findings (with scope conditions): (1) Social security reform that pays benefits from 65 regardless of work removes the implicit tax wedge. At fixed prices all workers extend careers (high school +2.4 years on average; college +7.6 years to age 72.6); in general equilibrium effects are attenuated — high school workers actually retire ~1 year early (average 63.9) while college workers retire later (average 70.8). (2) Tax experiment along Prescott (2002) lines: raising tau_l with revenue rebated lump-sum produces a Laffer curve peaking at tau_l=0.54; without rebates the Laffer curve peaks at tau_l=0.73 (general equilibrium) and the small-open-economy version is nearly linear. (3) The aggregate labor-supply elasticity is zero at low tax rates (corner at 65), then rises above 1 and levels around 1.2 over a wide middle range before rising again past tau_l=0.7. (4) Ben-Porath nonconvexities create “tipping points”: e.g., high school ability-3 workers are indifferent between two starkly different career strategies over tax range 0.42-0.52, and at high tax rates workers jump discretely from long careers with high human capital to much shorter careers with little/no on-the-job investment.
Implications: College-educated (steeper-earnings-profile) workers’ labor supplies are more resilient to tax and social-security reforms than high school workers’. High tax rates with lump-sum rebates can produce a “dual labor market” / bifurcation, raising lifetime earnings inequality (Gini) while welfare conditioned on schooling converges, all at a growing efficiency cost.
Layer 2: Deep Dive
What is the core methodological contribution relative to HLT?
The authors retain HLT’s primitives (credit markets, indivisible within-period labor, Ben-Porath human capital, aggregate production) but replace HLT’s exogenous mandatory retirement at 65 with endogenous career-length choice, and add a pay-as-you-go social security system. The social security system with an implicit tax on working past 65 puts all workers at a corner solution at age 65, so the model reproduces HLT’s outcomes. This provides a choice-theoretic rationalization for retirement behavior that HLT hard-wired. They state HLT could have used this time-averaging model with endogenous retirement to obtain the same quantitative findings.
Why is there no separate identification/empirical strategy in the usual sense?
This is a calibrated/quantitative general-equilibrium model, not a reduced-form causal study. Parameters are borrowed or ‘backed out’ from HLT (who estimated human-capital technologies via nonlinear least squares on NLSY 1979-1993 earnings profiles for white male civilians, plus CPS 1963-1993 and NIPA aggregates). New parameters are calibrated to be compatible with HLT: B and the efficiency-decline parameters (phi1, phi2) are jointly set so all agents retire at 65 in baseline; kappa=0.388 is set to match HLT’s interest rate given a capital-output ratio of 4; sigma (dispersion of nonpecuniary college cost) is calibrated to match the 8% rise in the relative college skill price between HLT’s two steady states; ability-specific means mu_theta target college enrollment rates from Taber (2002, Table 1).
What are the three forces that make high school workers retire earlier than college workers under the social security reform?
First, the social security system redistributes from high-ability to low-ability agents (equal benefit, proportional payroll tax), and the income effect on low-ability (mostly high school) workers reduces their labor supply; removing social security entirely (recalibrating kappa from 0.388 to 0.767) shows lowest-ability high school workers extend careers most. Second, per Ljungqvist-Sargent (2014), the more elastic an earnings profile to accumulated work, the longer the career; giving high school workers college workers’ more productive human-capital technology lengthens their careers. Third, a time-averaging ‘apprenticeship’ effect: college is treated as a fixed pre-work requirement Z tacked onto an optimal working span, so at an interior solution optimal career length = baseline length + Z; this accounts for roughly a 4-year career-length difference between high school and college workers in the relevant perturbed economy.
How do the effects of a labor tax increase depend on how revenue is spent, and what is the mechanism?
Following Prescott (2002): if revenue is rebated lump-sum (a good substitute for private consumption), the income effect of the tax is suppressed and the substitution effect dominates, sharply reducing labor supply (Laffer peak at tau_l=0.54). If revenue is squandered or spent on poor substitutes, income and substitution effects roughly cancel under balanced-growth preferences, so labor supply is little affected (Laffer peak at tau_l=0.73 in GE; nearly linear/flat in the small-open-economy version where capital inflows hold the interest rate constant at 0.059). With lump-sum rebates the equilibrium interest rate is U-shaped in the tax rate and the Laffer curve eventually approaches zero (output collapses); without rebates the interest rate rises monotonically to offset what would otherwise be capital inflows.
What are the Ben-Porath nonconvexities and the ’tipping points’?
Returns to on-the-job human-capital investment can only be harvested over a long enough career, so the value function over retirement ages can become non-concave with two local maxima: a long career with high end-of-life human capital versus a short career with little/no investment. As a determinant (tax rate, disutility, technology productivity) changes incrementally, the optimal response can be discontinuous — a discrete jump to a much shorter career and much less human-capital accumulation. Example: at tau_l=0.45 high school ability-3 workers have two optima, retirement at 65 (high human capital) and early retirement at age 50 (low human capital); they are indifferent over tax range 0.42-0.52. The nonconvexity is intrinsic to the Ben-Porath technology and arises even in a laissez-faire economy with interior career-length solutions, not only because of the social-security corner.
How is the indifference between career strategies handled in equilibrium (heterogeneity and computation)?
When otherwise-identical agents become indifferent between two career strategies, the regularity condition of a unique solution fails. The authors extend the equilibrium definition to allow equilibrium fractions of identical agents choosing different strategies; market clearing pins down these fractions (a ‘convexification’). Computationally they identify the ‘most indifferent’ worker type (smallest gap between the two local maxima; threshold 0.05%) and vary the fraction retiring at each age until GE conditions are satisfied. They also introduce continuous retirement ages via cubic-spline interpolation of the value function, validated against a closed-form analytical formula for agents who do not accumulate human capital (largest deviation only about half a month at tau_l=0.61).
What heterogeneity is documented across the eight worker types?
College enrollment rises with ability in baseline (about 0.11, 0.34, 0.56, 0.86 for ability groups 1-4 in the authors’ model). Group 4 has the second-highest average disutility of attending college, so 14% of group 4 become high school workers despite large advantages, and group 4’s enrollment falls most sharply with higher taxes. Group 1 has the highest disutility and lowest college human capital, so only ~11% attend college, falling below 1% above tau_l=0.45. End-of-life human capital of lower ability groups (1,2) falls monotonically with taxes, while higher ability groups (3,4) initially raise human capital as the interest rate falls. High school ability-1 workers eventually stop working entirely at the highest tax rates, with lifetime labor earnings falling to zero, relying on lump-sum transfers and social security.
What does the paper find for aggregate labor-supply elasticity, and why is ~1.2 notable?
With lump-sum rebates, after an initial range of zero elasticity (all at the corner of retiring at 65), the elasticity quickly rises above 1 and levels around 1.2 over a substantial middle range, then rises again after tau_l=0.7 (as physical capital gets scarce and the interest rate rises steeply). The ~1.2 is notable because in the Ljungqvist-Sargent (2014) framework with the same utility, the analytical aggregate elasticity is exactly one regardless of the learning-by-doing wage exponent; the model obtains ~1.2 despite college workers being stuck at the corner until tau_l≈0.6, because falling college enrollment shifts would-be college workers into earlier-retiring high school careers. Without rebates the elasticity is suppressed.
What are the inequality findings?
Two measures: present value of lifetime labor earnings and lifetime utility. The pre-tax earnings Gini is roughly flat for the first five percentage points above baseline (all still retiring at 65), then rises nearly one-to-one with the tax rate until tau_l=0.65, flattens as college ability groups 2 and 3 switch to short careers, drops when group 4 (highest earners) switches, then rises again as college workers’ relative earnings surge (driven by the rising college skill premium compensating for tuition and nonpecuniary costs). Using the Holter-Ljungqvist-Sargent-Stepanchuk (2025) ex post-ex ante welfare measure, higher taxes with lump-sum transfers shrink welfare inequality conditional on schooling even as income inequality grows, at an efficiency cost that accelerates above tau_l=0.4.
How do taxation results differ under the social security reform versus the baseline social security system?
Laffer curves under the reform (Figure 12a) closely resemble the baseline (Figure 2a). The key difference is that under the reform workers are at interior career-length solutions, so high school workers’ average retirement age falls with the very first tax increments (rather than staying stuck at 65), and college workers raise average retirement ages over a mid-range of taxes. At sufficiently high taxes the two economies become identical (above tau_l=0.74 with, 0.72 without rebates), because the implicit post-65 tax wedge becomes irrelevant once everyone retires early. Under the reform, college workers’ careers are ‘anchored’ near the age where human-capital efficiency depreciates rapidly rather than by the official retirement age.
How does the paper relate to and differ from Fan, Seshadri, and Taber (2024)?
FST (2024) independently endogenize career lengths in a Ben-Porath model estimated on SIPP data for male high school graduates, with nine worker types differing in disutility B(theta), learning ability A(theta), and initial human capital H(theta). A key difference: FST impose identical Ben-Porath exponents across all workers, so the Ljungqvist-Sargent force (more elastic earnings profiles imply longer careers) is largely absent; and FST do not impose balanced-growth preferences, so income effects of higher wages do not cancel. The authors suspect the sharp declines in career length with higher productivity in FST’s first two rows reflect income effects, and that time-averaging strengthens income effects. In the authors’ own balanced-growth model, the level of wages does not affect labor supply — only the terms on which human capital can be accumulated.
What robustness/sensitivity checks and appendices are reported?
Appendix C: sensitivity analysis of disutility B and the efficiency-decline function e(n); searching over (B, phi1) that keep all agents retiring at 65 yields end-point coordinates approximately (0.59, 0.09) and (0.9, 0.31), with the baseline (B=0.8, phi1=0.2) chosen as an intermediate pair subject to no noticeable efficiency decline before the 60s. Appendix D: alternative social security reforms raising benefits — college workers keep retiring at 65 while high school workers retire ever earlier. Appendix F.1: elasticity of the aggregate human-capital composite Q. Appendix G: replacing the Ben-Porath technology with exogenous earnings-experience profiles yields less polarization (lower Gini) and a lower aggregate labor-supply elasticity. The authors also note an unresolved discrepancy: their present-value earnings are 6.9-7.0% (high school) and 7.1-7.2% (college) lower than HLT’s Table II, but college enrollment is little affected since differences are similar across schooling.
What are the main caveats and policy scope conditions?
Results depend on balanced-growth preferences (income/substitution effects of wage levels cancel), on HLT’s estimated human-capital technologies and nonpecuniary college-cost distributions, and on the auxiliary kappa device for targeting the capital-output ratio. The disutility B and efficiency-decline parameters are not pinned down by data when workers sit at the 65 corner, hence only a sensitivity analysis. Limited heterogeneity (only 8 types) means aggregate smoothness comes from convexification rather than from a continuum of switching agents. The central policy warning — that high enough tax wedges or distortions can dislodge even high-productivity workers into a ‘dual labor market’ with earlier retirement and less human-capital accumulation, risking an implosion of activity — applies within this calibrated structure.