Paper: arXiv 2601.17773

Authors: Jeonggyu Huh, Seungwon Jeong, Hyun-Gyoon Kim, Hyeng Keun Koo, Byung Hwa Lim

Abstract

This paper introduces MarketGAN, a factor-based generative framework for high-dimensional asset return generation under severe data scarcity. We embed an explicit asset-pricing factor structure as an economic inductive bias and generate returns as a single joint vector, thereby preserving cross-sectional dependence and tail co-movement alongside inter-temporal dynamics. MarketGAN employs generative adversarial learning with a temporal convolutional network (TCN) backbone, which models stochastic, time-varying factor loadings and volatilities and captures long-range temporal dependence. Using daily returns of large U.S. equities, we find that MarketGAN more closely matches empirical stylized facts of asset returns, including heavy-tailed marginal distributions, volatility clustering, leverage effects, and, most notably, high-dimensional cross-sectional correlation structures and tail co-movement across assets, than conventional factor-model-based bootstrap approaches. In portfolio applications, covariance estimates derived from MarketGAN-generated samples outperform those derived from other methods when factor information is at least weakly informative, demonstrating tangible economic value.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: MarketGAN effectively integrates economic inductive bias (factor models) with deep generative architectures (TCN-WGAN-GP) to tackle high-dimensional asset return simulation under data scarcity. The paper demonstrates strong empirical validation across distributional stylized facts and practical downstream portfolio optimization tasks.

Research Flowchart

  flowchart TD
    A[Research Goal: Generate realistic high-dimensional financial returns under data scarcity] --> B[Data Input: Daily returns of large U.S. equities]
    B --> C[Methodology: MarketGAN Architecture]
    
    subgraph C [Methodology: MarketGAN Architecture]
        C1[Asset-Pricing Factor Structure<br/>Economic Inductive Bias]
        C2[Temporal Convolutional Network<br/>TCN Backbone]
        C1 --- C2
    end

    C --> D[Computational Process: Joint Vector Adversarial Training]
    
    D --> E[Outcome 1: Captures stylized facts<br/>Heavy tails, volatility clustering, tail co-movement]
    D --> F[Outcome 2: Portfolio Optimization<br/>Superior covariance estimation and economic value]

Cited References (68)

  1. Quant GANs: deep generation of financial time series (2020)
  2. Generating synthetic data in finance: opportunities, challenges and pitfalls (2020)
  3. Equity portfolios generated by functions of ranked market weights (2001)
  4. Wavenet: A generative model for raw audio (2016) — link
  5. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling (2018) — link
  6. Long short-term memory (1997)
  7. Learning phrase representations using RNN encoder-decoder for statistical machine translation (2014) — link
  8. Fully convolutional networks for semantic segmentation (2015)
and 60 more
  1. Deep residual learning for image recognition (2016)
  2. Generative adversarial nets (2014)
  3. Wasserstein generative adversarial networks (2017)
  4. Improved training of wasserstein gans (2017)
  5. Unsupervised representation learning with deep convolutional generative adversarial networks (2015) — link
  6. Self-attention generative adversarial networks (2019)
  7. High-resolution image synthesis and semantic manipulation with conditional gans (2018)
  8. Empirical properties of asset returns: stylized facts and statistical issues (2001)
  9. Econophysics review: I. Empirical facts (2011)
  10. Modeling financial time-series with generative adversarial networks (2019)
  11. Time-series generative adversarial networks (2019)
  12. C-RNN-GAN: Continuous recurrent neural networks with adversarial training (2016) — link
  13. Sliced and radon wasserstein barycenters of measures (2015)
  14. Generative Adversarial Networks applied to synthetic financial scenarios generation (2023)
  15. More stylized facts of financial markets: leverage effect and downside correlations (2001)
  16. A data-driven market simulator for small data environments (2020) — link
  17. Conditional sig-wasserstein gans for time series generation (2020) — link
  18. The Variation of Certain Speculative Prices (1963)
  19. Tail-gan: Learning to simulate tail risk scenarios (2022) — link
  20. A comprehensive look at the empirical performance of equity premium prediction (2008)
  21. Empirical asset pricing via machine learning (2020)
  22. Factor-GAN: Enhancing stock price prediction and factor investment with Generative Adversarial Networks (2024)
  23. Conditional generative adversarial nets (2014) — link
  24. Predictive regressions with time-varying coefficients (2012)
  25. CAPM over the long run: 1926–2001 (2007)
  26. Bayesian analysis of stochastic betas (2005)
  27. Beta risk in the cross-section of equities (2020)
  28. Factor-loading uncertainty and expected returns (2013)
  29. Multivariate regression models for panel data (1982)
  30. Common risk factors in the returns on stocks and bonds (1993)
  31. A five-factor asset pricing model (2015)
  32. Gans trained by a two time-scale update rule converge to a local nash equilibrium (2017)
  33. The mahalanobis distance (2000)
  34. Dynamic time warping (2007)
  35. A well-conditioned estimator for large-dimensional covariance matrices (2004)
  36. Beta as a random coefficient (1978)
  37. Factorvae: A probabilistic dynamic factor model based on variational autoencoder for predicting cross-sectional stock returns (2022)
  38. Improved estimates of higher-order comoments and implications for portfolio selection (2010)
  39. Auto-encoding variational bayes (2013)
  40. Variational inference with normalizing flows (2015)
  41. Denoising diffusion probabilistic models (2020)
  42. Deep learning (2016)
  43. Autoencoder asset pricing models (2021)
  44. Deep learning with long short-term memory networks for financial market predictions (2018)
  45. Attention is all you need (2017)
  46. Finding structure in time (1990)
  47. Generative machine learning for multivariate equity returns — → 2311.14735
  48. Forest through the Trees: Building cross-sections of stock returns (2025)
  49. Missing financial data (2025)
  50. Missing data in asset pricing panels (2025)
  51. Missing values handling for machine learning portfolios (2024)
  52. Enhancing return forecasting using LSTM with agent-based synthetic data (2025)
  53. Stock market prediction based on Generative Adversarial Network (2019)
  54. Deep stochastic optimization in finance (2023)
  55. Deep empirical risk minimization in finance: Looking into the future (2023)
  56. Synthetic data for portfolio: a through of the dice will never abolish chance (2025)
  57. Deep learning in asset pricing (2024)
  58. Market-GAN: Adding control to financial market data generation with semantic context (2024)
  59. Market Generators: A Paradigm Shift in Financial Modeling (2026)
  60. Diffolio: A Diffusion Model for Multivariate Probabilistic Financial Time-Series Forecasting and Portfolio Construction — → 2511.07014

2 of 68 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.