STRATA: Strategy-Aware Causal Representation Learning for Robust Detection of Financial Misreporting

DOI

10.2991/978-94-6239-774-3_24How to use a DOI?

Keywords

Causal representation learning; Financial misreporting; Strategic classification; Foundation models; Performative prediction

Abstract

Detecting financial misreporting is a multi-trillion-dollar problem and a notorious failure mode of off-the-shelf machine learning: modern deep models reach striking in-sample accuracy yet collapse under regulatory regime shifts and adversarial managerial adaptation. We argue that this brittleness reflects a missing causal abstraction: observed disclosures are generated jointly by the firm’s real economic state and its manager’s strategic intent; the two are statistically confounded; and the detector itself influences the strategic intent through a performative feedback loop. We formalise the problem as a structural causal model and propose Strata, a strategy-aware causal foundation model that recovers the two latent factors using regulatory regimes as auxiliary signals^[7,2], augments scarce supervision via counterfactual narratives from a financial large language model under a template, and is trained against a Stackelberg adversary that anticipates the firm’s best response^[4]. On a 1993–2023 panel of 213,481 firm-years built from SEC AAERs, Compustat, and EDGAR, Strata improves out-of-regime F₁ by 13.2 and adversarial F₁ by 13.6 points over the strongest 2022+ deep baseline, and its strategic-intent latents align with FinGPT risk signals at |r| > 0.7.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Yanze Zhou
AU  - Yikun Wang
AU  - Xuanqi Liu
PY  - 2026
DA  - 2026/09/11
TI  - STRATA: Strategy-Aware Causal Representation Learning for Robust Detection of Financial Misreporting
BT  - Proceedings of the 2026 5th International Conference on Mathematical Statistics and Economic Analysis (MSEA 2026
PB  - Atlantis Press
SP  - 276
EP  - 283
SN  - 2352-5428
UR  - https://doi.org/10.2991/978-94-6239-774-3_24
DO  - 10.2991/978-94-6239-774-3_24
ID  - Zhou2026
ER  -