Institutional investors optimise over feasible sets whose shape is dictated by regulation and mandate. This paper asks whether the intrinsic geometry of those sets, as distinct from the shadow cost of the constraints that define them, carries independent information for equilibrium prices. We answer the question in two steps. First, we show that the position limits and linear matching requirements most commonly invoked in this literature generate polyhedral feasible sets whose smooth strata are affine and therefore intrinsically flat, so that no curvature mechanism can be attributed to them; the correct analytical tools there are tangent and normal cones and Karush–Kuhn–Tucker multipliers. Second, we isolate a class of genuinely nonlinear constraints, of which the active boundary of a quadratic rebalancing budget is the leading case, whose transversal intersection with the wealth constraint is a smooth submanifold of dimension N-2 with non-vanishing sectional curvature. On that class we formulate a competitive equilibrium with price impact, prove existence and uniqueness under an explicit curvature domination condition, and derive a three-term decomposition of expected excess returns into a zero-beta rate, a classical covariance premium, and a curvature term that is written explicitly as a second fundamental form contracted with the multiplier vector. A fully specified four-asset example computes every object in the decomposition in closed form and shows that the curvature term vanishes when the constraint is replaced by its tangent hyperplane, which separates it from the shadow cost of a binding constraint. An empirical exercise on S P 500 index funds documents a priced constraint-proximity factor that is not spanned by the Fama–French–Carhart factors; we interpret this factor as a measure of crowding near regulatory limits rather than as an estimate of Riemannian curvature, and we state the conditions under which the two coincide.
This paper examines the strategic interactions among AI adoption, shipping outsourcing mode selection, and coastal piggybacking (CPB) regulation in a multi-level shipping supply chain consisting of a cargo agent (CA), a liner shipping company (LS), a flag-of-convenience carrier (FOC), and a transshipment port (FOT). We develop a game-theoretic model in which AI serves dual roles: it enhances demand forecasting accuracy (information-precision effect) and reduces service time (service-time reduction effect). The cargo agent chooses between consignment and turnkey contracting under different information structures. We find that AI generates opposing effects on the CA’s outsourcing preference: the information-precision effect favors turnkey, while the service-time reduction effect favors consignment, leading to a counterintuitive outcome in which advanced AI makes consignment more attractive. The LS’s AI investment is profitable only under certain conditions and depends critically on the learning efficiency of the technology, revealing an asymmetry in how AI affects the LS versus the CA. Furthermore, AI improves social welfare only when congestion is sufficiently severe; otherwise, its demand-expansion effect may worsen congestion. CPB unambiguously reduces congestion and improves welfare, but its interaction with AI is conditional: complements when AI is immature, substitutes when mature. These findings challenge conventional wisdom and provide implications for the sequencing of regulatory and technology policies.
This study examines how product return timing links configuration selection and recovery planning in circular manufacturing systems under disruption. We develop a two-stage stochastic mixed-integer linear programming model that selects a product configuration before uncertainty is revealed and adapts return timing, production, and recovery allocation to each scenario. Return timing changes prescribed inspection-grade shares and thus the quantities available to eligible recovery routes. The model incorporates a configuration sustainability requirement, production and recovery capacities, market absorption limits, and conditional value-at-risk (CVaR) of recovery-duration-weighted disruption impacts. In an illustrative LED luminaire case, adaptive timing reduces the penalized objective under a feasible projected-plan rule. Most of this difference comes from avoiding shortage penalties. Adaptive timing matches the earliest-feasible rule at the base early-return cost; a separate benefit from optimizing timing appears at higher early-return costs. The results identify when controllable withdrawal timing can preserve service and support recovery, while distinguishing shortage avoidance from operating-cost savings.
With the increasingly complex international situation, global supply chain disruptions caused by trade conflicts are becoming more severe. To mitigate the losses caused by supply disruptions, global brands have started purchasing from local companies, even from the local competitors. This emergency procurement may lead to higher production costs and delivery delays. Therefore, global brands are now adopting natural language processing (NLP) technology to predict disruptions and procure products from local competitors in advance. However, under supply disruption, the value of NLP technology on firms’ performance and their co-opetition relationships remains unclear. Therefore, based on the NLP technology decision and the presence or absence of disruptions, we consider four scenarios to explore the effects of NLP technology on firms’ performance and their co-opetition relationships. We find that the adoption of NLP technology is conducive to predicting disruptions in advance and reducing emergency production costs, promoting a co-opetition relationship between the global brand (GB) and the local brand (LB). We also find that adopting NLP technology has two primary effects: the first is the risk response effect, which enables the GB to avoid the second-mover disadvantage; the second is the cost-saving effect, which allows the LB to produce components in advance, avoiding the increase of the emergency production cost. Only if the emergency production cost is high and the NLP technology investment cost is low, the combined advantages of these effects outweigh the investment cost, making the GB adopt NLP technology. Further, the LB can also benefit from the NLP technology when the emergency production cost is sufficiently high. Therefore, there exists a win-win outcome where adopting NLP technology benefits both the GB and the LB.
Edge computing task scheduling faces critical challenges in dynamic, resource-constrained environments, where multiple edge devices must coordinate to process time-sensitive tasks. Traditional scheduling methods struggle with the exploration-exploitation trade-offs and scalability issues. This paper presents D-MAPPO-ABC, a novel hybrid approach that integrates Distributed Multi-Agent Proximal Policy Optimization (MAPPO) with the ABC algorithm for efficient edge task scheduling. Unlike centralized approaches, our method models ABC’s three bee roles (employed, onlooker, and scout) of the ABC algorithm as distinct MAPPO policies, enabling distributed decision-making while maintaining the benefits of swarm intelligence. Through extensive experiments on a 10-agent edge computing simulation, evaluated across ten independent random seeds with Welch t-tests, D-MAPPO-ABC achieves a 95 GitHub .
