A preprint study of South Korea's education competition finds that replacing a score contest with a capacity-preserving lottery for preferred destinations raises completed fertility, the model's children-per-family measure, from 1.8463 to 2.0862. That is a modeled increase of 0.240 children. The comparison is a counterfactual, not a direct test of an actual assignment reform.
The paper asks whether that education race raises the cost of children and reduces family size. Its proposed mechanism is positional spending: money spent to compete for rank in a score-based system rather than to add productive education. The paper's model allows preferred destinations to carry value beyond measured current wages.
What the data show
The paper's evidence includes two KLIPS groups: a benchmark sample of 770 households and a comparable younger-cohort sample of 1,129 households. A separate KEEP cohort contained 2,000 students; 574 reported positive earnings in 2015 and entered the wage regression.
The data do not show a clear measured return to private education after the strongest controls. With school fixed effects, the coefficient on the study's CSAT score measure was 0.285, with a standard error of 0.323, and was not statistically distinguishable from zero. The controlled estimate for entering a preferred destination was 0.027, with a standard error of 0.059. The estimated wage elasticity was 0.004 before school effects and 0.002 with them; the latter estimate had a standard error of 0.013 and a 95 percent interval from -0.023 to 0.026. These are conditional associations within the reported samples, so they do not establish that spending causes the outcomes.
To connect those patterns to fertility, the paper embeds the assignment mechanism in a stationary overlapping-generations model. In plain terms, it follows successive generations of households in a steady setting, while letting heterogeneous parents choose fertility, labor supply, productive education and positional effort. Access to preferred destinations is set by a fixed-share score contest. Joint calibration fits all ten point targets closely: the root mean squared relative error is 0.64 percent, and no error exceeds 2 percent.
The cost of the contest
When the contest is replaced by the lottery, positional spending disappears in the model. Productive education rises from 1.5 percent of income to 1.7 percent, while output per unit of labor falls by 0.7 percent. The simulated fertility gain therefore comes with a modeled decline in output.
The explanation depends heavily on what families value about the destination. In the benchmark, contest outlays equal 5.0 percent of income. When nonwage destination value is set to zero, outlays fall to 1.5 percent, and the lottery contrast is -0.006 under one current-wage-premium case and -0.007 under a large-firm-mover case. When the nonwage value is retained but the measured wage premium is removed, outlays remain 5.1 percent and the lottery contrast is +0.233. The paper's conclusion is that the race is sustained mainly by destination value beyond measured current wages, but that nonwage component is jointly calibrated and not separately observed.
Where policy changes differ
Transfers and taxes look weaker in this comparison. A transfer equal to 5 percent produces a modeled fertility increase of 0.031 children and a 2.5 percent welfare gain. A 20 percent broad tax and a 20 percent oracle tax each change fertility by -0.001 and raise welfare by 0.6 percent. A revenue-funded tax-transfer package changes fertility by +0.003 and welfare by +0.9 percent. These are balanced-budget model results, and the paper does not present them as direct evidence of what would happen outside the calibrated equilibrium.
Design also matters. In a qualified lottery, a pool containing half the population delivers 15 percent of the full lottery fertility effect. A 70 percent qualification pool delivers 46 percent, and an 85 percent pool delivers 75 percent. Universal qualification delivers the full effect. At a fixed admission share, random assignment carries a reported welfare gain of 16.3 percent of consumption, while output per labor unit is 0.7 percent lower. That welfare baseline excludes the child-time term and depends on keeping the admission share fixed.
A different reading of the cohort decline
The cohort exercise points to a second explanation for the decline. Holding the contest fixed, recalibrating preferences and heterogeneity produces completed fertility of 1.780 against a target of 1.786 and reproduces 76 percent of the cohort decline. The retained-target fit has an RMSE of 0.21 percent. Contest outlays remain close to 5 percent, the positional share near 0.78, and the younger-cohort lottery contrast is 0.275. The author interprets this as evidence that changing family-size preferences and heterogeneity, rather than a fiercer contest, are the main explanation for the modeled cohort shift.
Why the result is conditional
The headline policy findings are counterfactuals from a calibrated model, not estimates from a randomized assignment reform. The split between productive and positional education is inferred through calibration, and the nonwage destination component is not separately observed. Younger-cohort education histories are right-truncated, so that education comparison is descriptive. The welfare calculation excludes child time and assumes a fixed admission share. The study consequently offers a conditional policy ranking, not a measured causal effect of lotteries, transfers or taxes.
Paper data and sources
Original title: The Race for Elite Destinations: Education Competition and Low Fertility in Korea
Authors: Dongwoo Kim
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-28
DOI: Not available
Original paper · Full text