Adversarial Machine Learning Attacks on Financial Reporting via Maximum Violated Multi-Objective Attack

TL;DR

This paper introduces MVMO attacks that significantly improve the ability of malicious actors to manipulate financial reports by optimizing multiple objectives simultaneously, revealing vulnerabilities in current financial modeling defenses.

Contribution

The paper presents Maximum Violated Multi-Objective (MVMO) attacks that outperform standard methods by 20 times, demonstrating realistic fraud scenarios and potential for substantial financial report manipulation.

Findings

MVMO attacks achieve 20× more satisfying attacks than standard methods.

In about 50% of cases, firms could inflate earnings by 100-200%.

Firms could reduce fraud scores by 15% using these attacks.

Abstract

Bad actors, primarily distressed firms, have the incentive and desire to manipulate their financial reports to hide their distress and derive personal gains. As attackers, these firms are motivated by potentially millions of dollars and the availability of many publicly disclosed and used financial modeling frameworks. Existing attack methods do not work on this data due to anti-correlated objectives that must both be satisfied for the attacker to succeed. We introduce Maximum Violated Multi-Objective (MVMO) attacks that adapt the attacker's search direction to find $20\times$ more satisfying attacks compared to standard attacks. The result is that in $\approx50\%$ of cases, a company could inflate their earnings by 100-200%, while simultaneously reducing their fraud scores by 15%. By working with lawyers and professional accountants, we ensure our threat model is realistic to how such frauds are performed in practice.