Designing robust global value chains under trade-policy uncertainty

Abstract

Trade measures change within months, while the production networks they tax take years to build, so a firm commits its network before it knows which measures it will face. Duty falls on the full customs value at every border a good crosses, so where each production step sits decides how much a measure can tax.

Trade economics prices duty on the full value of goods at each border they cross, but takes the production network as given. Network design chooses locations, but treats tariffs as a known cost. This thesis asks how a firm should design a multi-stage global value chain so that it remains cost-efficient under normal trading conditions and limits its exposure to trade policy it cannot foresee.

Its main contribution is a two-stage robust optimisation model that needs no probabilities for political decisions. The firm first commits the locations where each production step can run, and once a measure is known it adjusts production and shipments within those locations. Six regimes represent future policy. Each is a documented class of measure, such as trade defence, safeguards or export controls. It has a rule for which shipments it reaches and a cap, the increment it adds in full to their tariff. The model accepts a design only if, under every regime, its worst-case cost stays within a chosen percentage of the best the firm could have secured had it known that regime. Among the accepted designs, it commits the one that is cheapest in normal times. This adapts the $p$-robustness of Snyder and Daskin to trade policy, and tightening the percentage traces what each level of protection costs. To the author's knowledge, no earlier model combines production across successive stages, uncertainty specific to each trade measure and a tariff base that follows the design.

Applied to three constructed chain architectures and to a solar photovoltaic chain serving the United States, the model yields three findings. First, a firm can protect itself without forecasting tariff rates, because what a measure reaches matters more than its rate. Defending against bilateral escalation costs up to 12 percent more than the cheapest network under normal tariffs. Defending against anti-circumvention, at the same cap, costs nothing extra. Second, a few alternative locations deliver most of the protection. A design's worst shock cost is the most that any regime adds to its cost. A second location per step achieves 91 to 94 percent of the cut in worst shock cost that allowing every location achieves. The most protected designs cost about 4 to 10 percent more in normal times, 6.6 percent in solar, and each unit of that extra cost removes about ten units of worst shock cost. Third, protection stops where the remaining exposure is cheaper to keep than to remove. The most protected solar design cuts the worst shock cost from 189.3 to 5.5 MUSD per GW. It keeps wafer and cell production in China, because leaving China would raise its extra cost in normal times from 6.6 to 9.2 percent.

A firm should hold its alternatives where its bill of materials forces goods across a border, at merges that join separately made components, at the closing step and where supplied inputs arrive. These steps account for 155 of the 164 changes the model makes as protection tightens. Because value jumps at each merge, moving these steps towards the market leaves the crossing on cheaper goods upstream. This lowers the value a tariff can reach and protects against regimes on movements into a market. Conditions outside the production network, where the firm sells and where it buys, decide how much protection it needs.