
The (post)pandemic period was marked by numerous port disruptions, leading to delays and rising costs for shippers. We build a fluid model in transient states to investigate how a disruption at one port can affect the disrupted port and its counterpart in another country in a circulatory container shipping system, where a stream of fleets moves goods back and forth between the two ports. Port disruption leads to two types of congestion: the inbound backlogs, which occur when ships are unable to enter the port, and the outbound backlogs, which arise when goods are unable to be loaded onto ships for transport to other ports. We analytically characterize the system’s recovery time, measured from the onset of the port closure until normal operations are restored, and the dynamics of inbound and outbound backlogs during recovery. We identify a whiplash effect in which outbound backlogs alternate between ports, creating oscillatory congestion patterns because of the disruption-induced uneven spatial distribution of ships. Our analysis reveals that simply increasing ship capacity may not expedite system recovery, highlighting the importance of coordination between port authorities and shipping companies when jointly planning capacity expansions. Furthermore, we examine several extensions of the model. We use the two-port system as a building block to analyze bipartite and cyclic network structures, and we also consider scenarios in which a port experiences multiple disruptions. In all cases, the whiplash effect persists, thereby demonstrating the robustness of our findings. Finally, we use our theory model prediction as a feature and apply machine learning techniques to predict the time that vessels spend in Shanghai Port after a typhoon disruption and show that using our model predictor, alongside all relevant vessel and voyage attributes as raw features, significantly reduces prediction errors by 10%–30% on average compared with not using it. This paper was accepted by Elena Katok, operations management. Funding: This work was supported by the Natural Sciences and Engineering Research Council of Canada [Grant RGPIN-2021-04295]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.05427 .
Changing flood insurance demand shifts climate risk across homeowners, lenders, and governments. We show that flood insurance take-up follows rising and falling home equity over the housing boom and bust across markets with different price dynamics. To isolate the causal effect of risk shifting on insurance demand, we study flood insurance policy renewals over the housing bust, exploiting granular differences in purchase timing that led policyholders to experience differential changes in skin-in-the-game exposure to flood risk for the same change in home equity. When homeowners have less equity exposure to flood losses, they are less likely to renew their policies. Lenders, in turn, have little incentive to require flood insurance when they can sell their mortgages to federally backed mortgage purchasers. This paper was accepted by Tomasz Piskorski, finance. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2025.01302 .
Different distribution shifts require different interventions, and algorithms must be grounded in the specific shifts they address. However, methodological development for robust algorithms typically relies on structural assumptions that lack empirical validation. Advocating for an empirically grounded data-driven approach to algorithm development, we build an empirical testbed comprising natural shifts across 8 tabular data sets, 172 distribution pairs over 45 methods, and 90,000 method configurations encompassing empirical risk minimization and distributionally robust optimization (DRO) methods. We find that Y|X-shifts are most prevalent in our testbed, in stark contrast to the heavy focus on X (covariate)-shifts in the machine learning literature, and that the performance of robust algorithms is no better than that of vanilla methods. To understand why, we conduct an in-depth empirical analysis of DRO methods and find that underlooked implementation details—such as the choice of underlying model class (e.g., LightGBM) and hyperparameter selection—have a bigger impact on performance than the ambiguity set or its radius. We illustrate via case studies how a data-driven understanding of distribution shifts can provide a new approach to algorithm development. This paper was accepted by Wolfram Wiesemann, data science. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.08298 .
We compare factor pricing models using the “test-asset irrelevance” result but under regime switching. We show that the unconditional squared Sharpe ratio measure is no longer appropriate for model comparison when returns are subject to regime shifts. However, conditional on a given regime, model comparison can be performed using the conditional squared Sharpe ratio metric. We find that the Fama-French six-factor model that employs a cash-based profitability factor, the Barillas-Shanken six-factor model, and the Stambaugh-Yuan four-factor model are jointly the best-performing models under the bull regime. However, no model performs significantly better than others under the bear regime. Importantly, these findings are robust to regime uncertainty. We also find that unconditional model rankings closely mirror those in bull markets, overstating the superiority of the models identified as best performing in expansions by ignoring the performance convergence across models that emerges in downturns. This underscores the importance of evaluating factor models conditional on market regimes. This paper was accepted by Lukas Schmid, finance. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2025.04942 .
As the spread of waste imposes substantial environmental and economic burdens, a newer operational strategy, waste valorization, is gaining attention. This approach transforms production-generated byproduct waste into valuable, distinct products, creating intricate production and demand interdependencies that traditional strategies like remanufacturing and coproduction cannot fully capture. In this study, we examine the impacts of these interdependencies within a two-stage stochastic production decision model to determine optimal production quantities under waste valorization. Our findings reveal that a higher waste valorization rate does not always benefit the manufacturer, especially if the waste incurs disposal costs. When product demands are uncorrelated, the manufacturer’s optimal profit can exhibit an inverse U-shaped relationship with the valorization rate, attaining its maximum at a moderate level that maintains a perfect balance in production. Conversely, when product demands are correlated, the manufacturer employs a “pool-to-center” strategy, using different pooling policies to smooth production quantities and minimize disparities with respect to the valorization rate. Surprisingly, as demand randomness or correlation intensifies, the optimal profit becomes M-shaped, with profit maximum shifting toward either low or high valorization rates, a phenomenon we term “go-to-extreme.” Although waste valorization can yield both economic and environmental benefits, we caution that a more efficient waste valorization process may lead to negative outcomes for both the manufacturer and the environment. This analysis not only sheds light on how waste valorization contributes to waste reduction and economic value, but also provides insights for production decision making. This paper was accepted by Caroline Flammer, sustainability. Funding: Financial support from the National Science Foundation of China [Grants 72301055, 72322017, 72471244, and 72101102] is gratefully acknowledged. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.08500 .