Paper: arXiv 2601.12990

Authors: Fan Zhang, Jiabin Luo, Zheng Zhang, Shuanghong Huang, Zhipeng Liu, Yu Chen

Abstract

Generative models for financial time series often create data that look realistic and even reproduce stylized facts such as fat tails or volatility clustering. However, these apparent successes break down under trading backtests: models like GANs or WGAN-GP frequently collapse, yielding extreme and unrealistic results that make the synthetic data unusable in practice. We identify the root cause in the neglect of financial asymmetry and rare tail events, which strongly affect market risk but are often overlooked by objectives focusing on distribution matching. To address this, we introduce the Stylized Facts Alignment GAN (SFAG), which converts key stylized facts into differentiable structural constraints and jointly optimizes them with adversarial loss. This multi-constraint design ensures that generated series remain aligned with market dynamics not only in plots but also in backtesting. Experiments on the Shanghai Composite Index (2004–2024) show that while baseline GANs produce unstable and implausible trading outcomes, SFAG generates synthetic data that preserve stylized facts and support robust momentum strategy performance. Our results highlight that structure-preserving objectives are essential to bridge the gap between superficial realism and practical usability in financial generative modeling.

Complexity vs Empirical Score

  • Math Complexity: 5.5/10
  • Empirical Rigor: 6.0/10 (capped from 6.5: no out-of-sample test; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper addresses a critical issue in synthetic financial time series generation by converting domain-specific stylized facts into differentiable loss functions within a WGAN-GP framework. Empirical evaluation on 20 years of Shanghai Composite Index data demonstrates clearer alignment with real market properties and downstream trading stability compared to standard generative baselines.

Research Flowchart

  flowchart TD
    A["Research Goal: Reliable Financial Time Series Generation for Backtesting"] --> B["Data Input: Shanghai Composite Index (2004–2024)"]
    B --> C["Problem Identification: Baseline GANs fail in backtests due to ignoring asymmetry & tail events"]
    C --> D["Methodology: Stylized Facts Alignment GAN (SFAG)"]
    D --> E["Computational Process: Joint Optimization of Adversarial Loss & Differentiable Structural Constraints"]
    E --> F["Outcomes: Synthetic data preserves stylized facts and enables stable momentum trading performance"]

Cited References (21)

  1. Empirical properties of asset returns: stylized facts and statistical issues (2001)
  2. Quant GANs: deep generation of financial time series (2020)
  3. The Variation of Certain Speculative Prices (1963)
  4. Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation (1982)
  5. A long memory property of stock market returns and a new model (1993)
  6. Studies of stock price volatility changes (1976)
  7. Leverage effect in financial markets: The retarded volatility model (2001)
  8. Multifractality in asset returns: theory and evidence (2002)
and 13 more
  1. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling (2018)
  2. GPT-TST: Generative pre-trained transformer for time series forecasting (2023)
  3. Temporal Fusion Transformers for interpretable multi-horizon time series forecasting (2021)
  4. FEDformer: Frequency enhanced decomposed transformer for long-term series forecasting (2022)
  5. TimeFormer: Transformer with Attention Modulation Empowered by Temporal Characteristics for Time Series Forecasting (2025)
  6. Autoregressive Denoising Diffusion Models for Multivariate Probabilistic Time Series Forecasting (2021) — link
  7. Score-Based Generative Modeling through Stochastic Differential Equations (2021)
  8. Deep generative modeling for financial time series with application in VaR: A comparative review — → 2401.10370
  9. Generative Adversarial Networks applied to synthetic financial scenarios generation (2023) — link
  10. Quantum generative modeling for financial time series with temporal correlations — → 2507.22035
  11. Shanghai Stock Exchange Composite Index and Bank Stock Prices in China: A Causality Analysis (2008) — link
  12. Improved training of wasserstein GANs (2017)
  13. Wasserstein distance guided representation learning for domain adaptation (2018)

2 of 21 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.