Hyungjin Kim

Modern semiconductor production is vertical, granular, and concentrated. This paper shows that in a bilateral oligopoly, export controls force a renegotiation of upstream input prices — a channel absent from standard quantitative trade models that treat input costs as fixed. I develop a vertical production network model in which AI-chip designers negotiate with a concentrated set of High-Bandwidth Memory (HBM) suppliers via Nash-in-Nash bargaining. Using Korean customs data, I document destination-specific price divergences consistent with relationship-specific bargaining. In the calibrated model, targeted buyers' input prices fall while rivals' input prices rise — the upstream shock-absorber and cross-buyer spillover channels. The input-price adjustment is roughly one-fifth of the downstream markup response, and a frozen-input counterfactual that ignores this renegotiation overestimates the competitive reallocation caused by export controls. Network composition matters: adding a single qualified supplier dampens the targeted buyer's markup increase by approximately 46 percent.

Combinatorial entry games combine subset choice, within-firm complementarities, and strategic substitutes, but existing uniqueness results either treat entry as binary, assume constant markups, or take the active set as given. We develop a uniqueness framework for combinatorial entry with endogenous Atkeson–Burstein markups. The key object is a decomposition of the threshold-map Jacobian into density dampening, marginal-value sensitivity, and threshold location, which implies a bathtub-shaped contraction modulus in private-cost noise. When its peak is below one, the Bayesian Nash equilibrium is unique for all noise levels, and supermodularity of the single-firm problem is preserved so existing CDCP solvers remain applicable. In a quantitative application to global cloud infrastructure entry (five providers, 42 countries), the calibrated model matches 99.5% of in-sample entry decisions, and endogenous markups sustain 15 additional firm-market pairs and lower consumer surplus by 10.6% relative to a constant-markup benchmark.

Globalization is often accompanied by declining barriers to both international trade and multinational production (MP), yet their joint implications for regional economies remain insufficiently understood. This paper examines how China's reductions in trade and MP barriers affected regional inequality in Korea, focusing on welfare and population distribution across regions. We first provide reduced-form evidence that regional exposure to MP, alongside trade, is an important and independent determinant of local employment. We then develop a multi-country, multi-sector general equilibrium model with trade, MP, and internal migration to quantify how China's cost reductions over 2000–2007 reshaped regional welfare and population allocation in Korea. The aggregate welfare gain is approximately 2.0%, with regional per-worker gains ranging from 1.7% in Seoul/Capital to 2.5% in the Southeast. Trade accounts for 1.4 percentage points of this gain, internal migration for 0.55 percentage points, and the vertical MP channel—through which affiliates abroad source inputs from the domestic parent—for approximately 0.04 percentage points. A skill-stratified extension of the model shows that the gains are regressively distributed: low-skill workers gain 2.6% on average versus 1.8% for high-skill workers, with the gap widest (−1.2 percentage points) in the Southeast manufacturing cluster. By jointly analyzing trade and MP, the paper shows that ignoring the MP channel would attenuate the estimated import-competition effect and misattribute part of the employment response to trade exposure alone.

How does intermediated trade alter the response of bilateral trade flows to cost shocks? We show that when firms can reroute shipments through intermediary hubs, the bilateral trade elasticity is strictly attenuated relative to the standard gravity benchmark—under Pareto productivity, attenuated by a factor equal to the share of trade currently flowing directly between origin and destination. We establish this result theoretically and validate it using transaction-level data from a Korean petrochemical trading house with 25 offices worldwide, where we observe the complete buyer–seller–intermediary triad for every shipment. In this network, 84.5% of origin–destination pairs route all trade through hubs, so a direct-route cost shock leaves their bilateral trade volume entirely unchanged—while standard gravity would predict a large negative response for those same pairs. Aggregating across all routes, standard gravity overstates the bilateral trade-flow response to a uniform cost shock by a value-weighted factor of 6.3 within this intermediary's network (range 3.2×–13× depending on the substitution elasticity); accounting for the share of trade that bypasses intermediaries entirely, the economy-wide overstatement is approximately 1.2–1.4× under an illustrative calibration of the intermediary sector's market share. These findings imply that bilateral gravity models systematically overstate trade-flow responses to route-specific cost shocks—such as corridor-level shipping costs, logistics frictions, and spoke-side tariffs—in sectors with deep intermediation.

Swing states—third-country firms sourcing critical inputs from China (e.g., Korea, Taiwan, the Netherlands, Vietnam, Mexico)—are the target of US coercion. As China becomes more productive, US coercion of swing states grows more, not less, aggressive. As US coercion grows, the welfare loss it imposes on swing states widens. Both relationships follow from bottleneck complementarity: swing-state firms cannot substitute away from cheap Chinese inputs, so voluntary alignment becomes too expensive for the US to buy through subsidies alone—no shift in political-economy parameters required. Past a Chinese-productivity threshold, US policy switches from subsidies alone to subsidies plus coercion. An HS6 panel difference-in-differences finds a positive cross-industry interaction in bottleneck sectors identified by the production-network literature, null or sign-flipped under non-bottleneck lists, as the model predicts. Surveillance frameworks should expect structural rather than transient coercion over swing states in bottleneck sectors as Chinese productivity continues to rise—intensity scales with that productivity, not with political-economy parameters.

Export controls do more than restrict trade: they change where the target innovates. We study a three-player model in which anticipated US controls induce China to redirect fixed R&D toward domestic production of restricted, hard-to-substitute inputs. Because China supplies intermediates to third-country firms, this directed innovation shifts input-cost incidence to exposed economies. We use the May 2020 Huawei Foreign Direct Product Rule, together with contemporaneous Made in China 2025/Big Fund support, to measure how strongly R&D is redirected: Chinese-assignee USPTO patenting rises in directly exposed chip codes relative to adjacent bottleneck codes. The calibration compares two US regimes—exercising export controls versus committing to restraint—by their combined surplus across the United States, China, and exposed third parties. Holding Chinese productivity fixed, controls are favored, but only marginally (−0.006). Once China's R&D allocation responds, the ranking flips: restraint raises combined surplus by +0.120. Exposed third parties absorb most of the incidence: 0.59% of their input-cost base, or roughly 0.5–0.9% when normalized by total model surplus. The same incidence pattern creates an institutional gap: third parties could finance US restraint, but no institution both aggregates contributions and binds the United States to restraint.