Aligning with Whom in Automated Trading? A Belnap–Dunn Four-Valued Decision Layer for Financial Forecasting Pipelines
Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 18-24
DOI: 10.1609/aaaiss.v10i1.43386 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Lorenzo Cevallos-Torres und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Automated financial decision systems, like many AI systems discussed under value alignment, must turn an uncertain estimate into one committed action for stakeholders whose risk appetite and loss tolerance differ. Conventional pipelines resolve this plurality silently: one threshold, chosen by one developer on one backtest, is applied as if it were the only legitimate value position. We take a documented LSTM pipeline for one-minute S&P 500, Gold, and Euro futures data (about 19.20 million OHLCV records) as a low-dimensional case. In its documented run, a 0.02% signal threshold mapped all 23,489 evaluated observations to Hold, with a reported directional accuracy of 68.22%. We treat this outcome as a likely instrumentation or calibration artifact, not as a finding: the single configured rule returned the same action for every record, and nothing in the output signaled it. We propose a Belnap–Dunn four-valued decision layer that labels each candidate action True (supported by some stakeholder and opposed by none), False, Both (committed stakeholders disagree: a value conflict), or Neither (no stakeholder commits: an epistemic gap), instead of collapsing these cases onto one action. We state when Both and Neither are reachable, show both in one synthetic two-stakeholder population, and note that the documented single-stakeholder run reduces to the classical output. The paper is a mechanism proposal, not an empirical validation.
Zitieren
APA
Cevallos-Torres, L., Caicedo-Quiroz, R., Barzola-Monteses, J., & Tolozano-Benites, R. (2026). Aligning with Whom in Automated Trading? A Belnap–Dunn Four-Valued Decision Layer for Financial Forecasting Pipelines. Proceedings of the AAAI Symposium Series, 10(1), 18-24. https://doi.org/10.1609/aaaiss.v10i1.43386
Vancouver
Cevallos-Torres L, Caicedo-Quiroz R, Barzola-Monteses J, Tolozano-Benites R. Aligning with Whom in Automated Trading? A Belnap–Dunn Four-Valued Decision Layer for Financial Forecasting Pipelines. Proceedings of the AAAI Symposium Series. 2026;10(1):18-24. doi:10.1609/aaaiss.v10i1.43386
BibTeX
@article{CevallosTorres2026,
author = {Cevallos-Torres, Lorenzo and Caicedo-Quiroz, Rosangela and Barzola-Monteses, Julio and Tolozano-Benites, Roberto},
title = {Aligning with Whom in Automated Trading? A Belnap–Dunn Four-Valued Decision Layer for Financial Forecasting Pipelines},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {10},
number = {1},
pages = {18--24},
doi = {10.1609/aaaiss.v10i1.43386}
}
EUVIMED