Directional Exposure in Spatio-Temporal Supply-Chain Risk Prediction: Graph Networks Versus a Strong Linear Baseline
Journal of Mathematical Finance and Risk Management · 03.11.2026 · 1 (3) · 61-68
DOI: 10.54097/ahe3p726 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Mathematical Finance and Risk Management, online erschienen am 03.11.2026. Verfasst von Mina Cruz und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Directed supplier exposures may improve disruption forecasting, but whether nonlinear graph learning adds value beyond interpretable event-history models is uncertain. We implement a two-round graph embedding, aggregation and update baseline and evaluate a directional temporal exposure graph network (DTEGN) whose logit includes lagged one- and two-hop observed incidents. Experiments use exclusively synthetic node attributes, supplier graphs and incident labels from a fully specified four-tier simulator; no operational failures are presented as real. On one chronological held-out test window, DTEGN reaches mean AUPRC 0.237 ± 0.004 across three initializations, compared with 0.194 ± 0.014 for two-round message passing. In a stronger comparison, exposure-augmented logistic regression scores AUPRC 0.279 on that window. After fitting each model on the original simulated world, we evaluate unchanged parameters on 12 independently generated worlds. Mean AUPRCs are 0.369 for DTEGN, 0.321 for two-round message passing and 0.413 for exposure-augmented logistic regression. Thus explicit directed incident exposure is a useful inductive bias, whereas gate complexity is not justified by this additive-hazard simulator. The study contributes an auditable technical implementation, transfer protocol, ablations and negative as well as positive results; real-world validity remains untested.
Zitieren
APA
Cruz, M., Hall, I., & Desai, X. (2026). Directional Exposure in Spatio-Temporal Supply-Chain Risk Prediction: Graph Networks Versus a Strong Linear Baseline. Journal of Mathematical Finance and Risk Management, 1(3), 61-68. https://doi.org/10.54097/ahe3p726
Vancouver
Cruz M, Hall I, Desai X. Directional Exposure in Spatio-Temporal Supply-Chain Risk Prediction: Graph Networks Versus a Strong Linear Baseline. Journal of Mathematical Finance and Risk Management. 2026;1(3):61-68. doi:10.54097/ahe3p726
BibTeX
@article{Cruz2026,
author = {Cruz, Mina and Hall, Irene and Desai, Xinyi},
title = {Directional Exposure in Spatio-Temporal Supply-Chain Risk Prediction: Graph Networks Versus a Strong Linear Baseline},
journal = {Journal of Mathematical Finance and Risk Management},
year = {2026},
volume = {1},
number = {3},
pages = {61--68},
doi = {10.54097/ahe3p726}
}
EUVIMED