Financial Wind Tunnel: A Retrieval-Augmented Market Simulator

TL;DR

Financial Wind Tunnel (FWT) tackles the challenge of generating realistic, controllable market data for model development amid evolving market dynamics. It integrates a retrieval-augmented diffusion model that uses cross-sectional signals from similar assets to condition generation, enabling multi-frequency synthesis, cross-market transfer, and what-if scenarios. It also includes an automated strategy optimizer to leverage simulated environments for robust downstream models. Experiments on CSI300 and HKSE demonstrate high fidelity generation (multi-frequency correlations > 0.6, cross-market IC ~ 0.475) and notable improvements in downstream tasks (annualized return, Sharpe, drawdown). This approach provides a practical wind-tunnel-like tool for stress-testing and advancing quantitative finance.

Abstract

Market simulator tries to create high-quality synthetic financial data that mimics real-world market dynamics, which is crucial for model development and robust assessment. Despite continuous advancements in simulation methodologies, market fluctuations vary in terms of scale and sources, but existing frameworks often excel in only specific tasks. To address this challenge, we propose Financial Wind Tunnel (FWT), a retrieval-augmented market simulator designed to generate controllable, reasonable, and adaptable market dynamics for model testing. FWT offers a more comprehensive and systematic generative capability across different data frequencies. By leveraging a retrieval method to discover cross-sectional information as the augmented condition, our diffusion-based simulator seamlessly integrates both macro- and micro-level market patterns. Furthermore, our framework allows the simulation to be controlled with wide applicability, including causal generation through "what-if" prompts or unprecedented cross-market trend synthesis. Additionally, we develop an automated optimizer for downstream quantitative models, using stress testing of simulated scenarios via FWT to enhance returns while controlling risks. Experimental results demonstrate that our approach enables the generalizable and reliable market simulation, significantly improve the performance and adaptability of downstream models, particularly in highly complex and volatile market conditions. Our code and data sample is available at https://anonymous.4open.science/r/fwt_-E852

Figures (6)

• Figure 1: The overall procedure of Financial Wind Tunnel (FWT). There are three core modules: retrieval, generation, and application. The data sources for retrieval include inter-market stocks, cross-market stocks, and multi-frequency dynamics. With retrieved series as conditional observation, FWT generates new time series for downstream tasks like what-if analysis, cross-market analysis, and trading strategy optimization.
• Figure 2: The self-supervised training procedure of FWT. FWT synthesizes $x_{t:t+T}$ when given a history time series $x_{0:t}$, which is used to query top-$k$ most relevant stocks to construct conditional observations. FWT predicts noise from noisy target $\mathbf{x}_h^{\text{target}}$ at the diffusion step $h$ to recover $x_{t:t+T}$.
• Figure 3: Cross-market generation case study: Liquidity crisis in the Hong Kong stock market.
• Figure 4: Cases of what-if generation. The figure illustrates a comparison between actual trends (green) with the results generated with and without what-if condition (red and grey respectively) under four different prompts.
• Figure 5: Cumulative return curve of hedge portfolio training from different dataset