Journal of Banking Regulation Экономика и эконометрика

Algorithmic capital diversion and consumer protection in fintech micro-investing: a proportional automation framework for regulators and platforms

Tsaban, Shay
Springer Nature · 2026-09-09 · DOI 10.1057/s41261-026-00356-8

Аннотация

Automated savings and micro-investing platforms increasingly govern consumer capital allocation through opaque, high-frequency algorithmic sweeps. The 2022 Consumer Financial Protection Bureau enforcement action against Digit (a deception action under the Consumer Financial Protection Act, in which the platform falsely assured users that its tool would never overdraw their accounts and failed to honour related reimbursement and interest commitments) is taken here as a motivating institutional signal that opaque, autonomous sweep mechanisms create consumer-protection exposure, not as evidence for the specific welfare mechanism this paper models. Regulators in the United States, European Union, United Kingdom, and Singapore face a common supervisory challenge: developing frameworks for algorithmic capital diversion that falls unevenly across banking, payments, investment, and consumer-protection regimes. This paper develops a conceptual framework, the Proportional Automation Constraint (PAC), that adapts principal–agent theory to this setting. The stylised utility model derived in the appendix predicts that, under its structural assumptions, each user has a welfare-maximising automation intensity, $$d^{*}$$ , that decreases with income volatility and increases with financial literacy. The paper’s central theoretical contribution is to show that this prediction is conditional on population composition : extending the model to heterogeneous agents yields an explicit threshold $$\pi ^{*}$$ on the share of present-biased, commitment-seeking users above which higher volatility raises Bthe population-average optimal automation and the aggregate prediction reverses. The prediction that uniform high-intensity automation harms volatile-income and low-literacy populations therefore holds only where commitment seekers are a minority ( $$\pi <\pi ^{*}$$ ): a scope condition that self-selection into high-automation platforms may cause to fail precisely where the framework is of most regulatory interest. The paper compares the PAC against the regulatory approaches of the CFPB, PSD2/MiFID II, the UK FCA’s Consumer Duty, and Singapore’s MAS, and against the EU AI Act’s regulatory sandbox provisions (Regulation (EU) 2024/1689, Article 57). The paper’s contribution is conceptual: it derives non-obvious conditions (the composition threshold $$\pi ^{*}$$ , a critical transparency-intervention window, and corner cases in which $$d^{*}$$ is not interior), supplies a disciplined empirical agenda for testing them, and proposes a procedural standard (a forward-looking, per-sweep liquidity gate) that is verifiable through audit without access to proprietary source code. It is not an empirical demonstration that the predicted effects obtain; that should be empirically tested.

Читать у издателя →