Auditing Algorithmic Bias in Transformer-Based Trading

2025-10-01

AI Summary

This paper identifies a frequency preference bias in Transformer-based trading models: their predictions rely heavily on low-frequency price movements while systematically neglecting volatility signals, introducing algorithmic bias risks. To address this, we propose a novel interpretability auditing framework grounded in Partial Information Decomposition (PID), which— for the first time—quantifies each asset’s information contribution across multiple frequency bands and precisely characterizes the model’s dependence on distinct frequency components. Empirical analysis demonstrates that the framework effectively uncovers latent low-frequency bias in model decisions and confirms a statistically significant negative correlation between model confidence and price movement frequency. Our work establishes a new paradigm for fairness assessment of financial time-series models and provides a practical, deployable technical pathway for algorithmic auditing and regulatory oversight in high-stakes financial applications.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Learning Preferences or Rankings
• Knowledge Representation and Reasoning: Preferences

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environments
• Responsible Web: Algorithmic accountability and transparency on the web

Abstract

Transformer models have become increasingly popular in financial applications, yet their potential risk making and biases remain under-explored. The purpose of this work is to audit the reliance of the model on volatile data for decision-making, and quantify how the frequency of price movements affects the model's prediction confidence. We employ a transformer model for prediction, and introduce a metric based on Partial Information Decomposition (PID) to measure the influence of each asset on the model's decision making. Our analysis reveals two key observations: first, the model disregards data volatility entirely, and second, it is biased toward data with lower-frequency price movements.

Problem

• Auditing algorithmic bias in transformer-based trading models
• Quantifying how price movement frequency affects prediction confidence
• Measuring asset influence on model decisions using information decomposition

Innovation

• Transformer model for financial prediction
• Partial Information Decomposition metric for auditing
• Quantifies bias toward low-frequency price movements