How the reduction of Temporary Foreign Workers led to a rise in vacancy rates in South Korea
(Journal of Human Capital, Fall 2026 (vol. 20, no. 3))
This study investigates the causal relationship between the reduction of low-skilled temporary foreign workers (TFWs) and job vacancies in South Korea's manufacturing sectors, utilizing the COVID-19 quarantine policy as a natural experiment. Employing a Difference-in-Differences methodology, the research reveals that sectors with high dependence on TFWs, particularly for permanent positions, experienced significantly elevated vacancy rates for a two-year period following the onset of the pandemic. The inability of native workers to fill these positions highlights the critical role of foreign labor in mitigating labor shortages. Notably, vacancy rates began to decline only after the government relaxed quarantine restrictions, facilitating the re-entry of TFWs into the country. These findings are corroborated by local projection methods.
The Causal Effects of Tariff-Rate Quota Policies on Agricultural Product Retail Prices
with Youngmi Kim (Under Review, Agricultural Economics)
(한국어 버전: 할당관세 정책이 농산물 소매가격에 미치는 인과적 영향 (Unpublished Institutional Working Paper, Not peer-reviewed)
Tariff-Rate Quotas (TRQs) can be administered either to protect domestic industry or to promote imports aimed at alleviating supply shortages and stabilizing consumer prices. Prior empirical studies have predominantly focused on protectionist TRQs; empirical evidence regarding trade-promoting TRQs remains scarce. This paper addresses this gap by examining Korea's voluntary TRQs implemented outside WTO/FTA commitments. We estimate the causal effect on retail prices for 40 agricultural products employing Local Projection Difference-in-Differences, which exploits staggered TRQ introductions and heterogeneous tariff reduction intensities. Averaging across treated products, TRQs do not significantly alter retail prices. However, effects exhibit substantial heterogeneity: for leafy and root vegetables (Group 1), retail prices do not decline, whereas for fruits (Group 2), a 1%p reduction in the tariff rate lowers retail prices by approximately 0.9%, implying a pass-through rate of roughly 90%. Mechanism analysis reveal no significant response in import volumes for either group, thereby ruling out a import quantity channel. Tariff-exclusive import prices remain unchanged, implying that tariff-inclusive import prices decreased and suggesting approximately 100% pass-through at the import stage. Wholesale pass-through diverges between groups: characterized by low import dependence, Group 1 exhibits elevated wholesale prices, indicating that wholesalers capture the tariff reduction gains; in contrast, characterized by high import dependence, Group 2 demonstrates wholesale price declines that transmit the gains from tariff reductions entirely to the retail stage.
로봇 혁신과 인간 혁신이 노동소득분배율에 미치는 영향 (Automation, Human Task Innovation, and Labor Share)
(심사중, 노동경제논집)
본 연구는 유럽연합 9개국을 대상으로 로봇 혁신과 인간 혁신이 노동소득분배율에 미치는 영향을 분석한다. 일반균형 모형을 통해 실증 분석을 위한 축약 회귀식을 도출하며, 2005 2019년 기간의 특허 기록과 인지 과업 지수를 활용한다. 내생성 문제를 해결하기 위해 미국 특허와 인지 과업 지수에 기반한 시프트-쉐어 도구변수를 사용한다. 분석 결과, 로봇 혁신의 증가는 노동소득분배율을 유의하게 감소시키는 것으로 나타나, 상당한 수준의 노동-로봇 대체가 존재함을 시사한다. 반면, 인간 혁신은 노동소득분배율과 양(+)의 관계를 보이나, 표본 기간 동안 인간 혁신 속도의 둔화로 인해 노동소득분배율을 오히려 감소시켰다.
(This study examines the impact of robotic and human innovation on labor share across nine European Union countries. Using a general equilibrium model, I derive a reduced-form specification for empirical analysis. The study employs patent records and a cognitive task index spanning 2005 2019. To address endogeneity concerns, I utilize shift-share instruments based on US patents and the cognitive task index. The findings reveal that increased robotic innovation significantly reduces labor share, indicating substantial labor-automation substitution. Conversely, human task innovation demonstrates a positive relationship with labor share, though slowing innovation rates during the sample period have led it to contribute to a decline in labor share.)
(심사중, 노동경제논집)
본 연구는 생성형 AI 확산기에 고용이 인지(cognitive) 수준에 따라 어떻게 달라졌는지를 반복성 강도(routine)와 AI 노출 강도를 사용하여 분석한다. 먼저 O*NET 과업 정보를 대규모 언어모델(LLM)로 채점해 SOC 6자리 직업분류에서 반복성 점수와 인지 점수를 세밀하게 구하였다. 그 후 반복성 강도를 연속형 처치로, 인지 점수를 표본 분할 기준으로 삼아, 2단계 이중차분으로 미국의 고용량을 분석하였다. 반복성 강도를 처치로 하면 사후 계수의 부호가 인지 수준을 경계로 갈렸다. 저인지의 음(-)의 사후계수는 AI에 의한 새로운 현상이 아니라 반복편향 기술변화(Routine-Biased Technological Change)의 지속된 패턴으로 설명되는 반면, 고인지의 양(+)은 반복편향 기술변화의 예측과 다르게 고인지 그룹에서 오히려 반복적인 직업이 떠오르고 있음을 시사한다. 한편 같은 설계에 AI 노출지수를 처치로 넣으면, 저인지 직업군에서 사후 계수가 유의하지 않았다. 반면 고인지 직업군에서는 사후 계수가 음(-)으로 추정되어 반복성을 처치군으로 썼던 경우와 부호가 반대이다. 반복성과 AI 노출지수라는 두 측정치는 상관관계가 없으므로, 두 지표는 노동시장의 서로 다른 메커니즘을 비춘다고 해석된다. 이는 반복편향 기술변화 문헌이 의지해 온 반복성 지표가 생성형 AI의 고용 영향을 설명하는 메커니즘을 대변하지 않을 수도 있음을 시사한다.
(Using routine intensity and AI exposure, this paper analyzes how employment shifted across cognitive levels during the diffusion of AI. This study first construct the data by using a large language model to score O*NET task descriptions, obtaining granular routine and cognitive scores at the 6-digit SOC occupation level. The paper then analyze U.S. employment with a two-stage Difference-in-Differences design that treats routine intensity as a continuous treatment and uses the cognitive score to split the sample. The sign of the post-treatment coefficient splits at the cognitive threshold. The negative coefficient in the low-cognitive group is not a new, AI-driven phenomenon but is explained by the continued pattern of Routine-Biased Technological Change, whereas the positive coefficient in the high-cognitive group suggests that, contrary to what Routine-Biased Technological Change predicts, routine occupations are gaining ground within the high-cognitive group. When the same design instead uses AI exposure as the treatment, the post-treatment coefficient is not statistically significant for the low-cognitive group, whereas for the high-cognitive group it is estimated to be negative —opposite in sign to the case using routine intensity as the treatment. Since these two measures, routine intensity and AI exposure, are uncorrelated, they are interpreted as offering different perspectives on the labor market. This suggests that the routine measure on which the Routine-Biased Technological Change literature has relied does not represent the mechanism through which AI affects employment.)