Bayesian Robust Financial Trading with Adversarial Synthetic Market Data

TL;DR

This work tackles the robustness gap in algorithmic trading under regime shifts driven by macroeconomics by integrating macro-conditioned data generation with a Bayesian adversarial training framework. It models trading as a two-player zero-sum Bayesian Markov game, where an adversary perturbs macro indicators while the trader maintains a belief over hidden market states and learns a robust policy via Bayesian neural fictitious self-play, achieving a Robust Perfect Bayesian Equilibrium. A macro-conditioned TimeGAN-inspired generator produces diverse, realistic market trajectories conditioned on macro indicators, while a Quantile Belief Network captures uncertain market states to inform decision-making. Empirical results across nine ETFs show the proposed method outperforms nine baselines in profitability and risk management, including during extreme events like the COVID-19 shock, and data realism analyses confirm superior fidelity of generated sequences. The approach offers a scalable, macro-aware solution for trading under uncertain and shifting market dynamics, with practical implications for stress testing and risk-aware deployment.

Abstract

Algorithmic trading relies on machine learning models to make trading decisions. Despite strong in-sample performance, these models often degrade when confronted with evolving real-world market regimes, which can shift dramatically due to macroeconomic changes-e.g., monetary policy updates or unanticipated fluctuations in participant behavior. We identify two challenges that perpetuate this mismatch: (1) insufficient robustness in existing policy against uncertainties in high-level market fluctuations, and (2) the absence of a realistic and diverse simulation environment for training, leading to policy overfitting. To address these issues, we propose a Bayesian Robust Framework that systematically integrates a macro-conditioned generative model with robust policy learning. On the data side, to generate realistic and diverse data, we propose a macro-conditioned GAN-based generator that leverages macroeconomic indicators as primary control variables, synthesizing data with faithful temporal, cross-instrument, and macro correlations. On the policy side, to learn robust policy against market fluctuations, we cast the trading process as a two-player zero-sum Bayesian Markov game, wherein an adversarial agent simulates shifting regimes by perturbing macroeconomic indicators in the macro-conditioned generator, while the trading agent-guided by a quantile belief network-maintains and updates its belief over hidden market states. The trading agent seeks a Robust Perfect Bayesian Equilibrium via Bayesian neural fictitious self-play, stabilizing learning under adversarial market perturbations. Extensive experiments on 9 financial instruments demonstrate that our framework outperforms 9 state-of-the-art baselines. In extreme events like the COVID, our method shows improved profitability and risk management, offering a reliable solution for trading under uncertain and shifting market dynamics.

Figures (4)

• Figure 1: State representation (x-axis) and reward (y-axis) reduced to one dimension via t-SNE (Algorithm \ref{['alg:tsnegeneration']} in Appendix G). The shift in distribution between training (blue points) and testing (orange points) highlights the out-of-distribution issue during testing.
• Figure 2: Overall architecture of our framework. An adversarial agent and a pre‐trained generator jointly form an adversarial environment, while a trading agent is trained on observations augmented by this environment.
• Figure 3: Overall architecture of the data generator. The encoder, decoder, and forecaster are pre-trained (marked with gear icons). The blue routes are only for training, while the red routes are for both training and inference.
• Figure 4: Comparison of trading actions (top) and cumulative returns (bottom) for our method, RARL, and DQN on DBB from 2021 to 2024. Our method adapts to both high-volatility pandemic phase (between the blue lines) and calmer phases (otherwise).