Search:
Match:
1 results

Analysis

This paper addresses the critical problem of data scarcity and confidentiality in finance by proposing a unified framework for evaluating synthetic financial data generation. It compares three generative models (ARIMA-GARCH, VAEs, and TimeGAN) using a multi-criteria evaluation, including fidelity, temporal structure, and downstream task performance. The research is significant because it provides a standardized benchmarking approach and practical guidelines for selecting generative models, which can accelerate model development and testing in the financial domain.
Reference

TimeGAN achieved the best trade-off between realism and temporal coherence (e.g., TimeGAN attained the lowest MMD: 1.84e-3, average over 5 seeds).