Reliability Comparison of TimeGAN Scenario Replacement versus Supplementation in Rachev-Ratio Portfolios

01 ABSTRACT

The study uses TimeGAN scenarios for Rachev-ratio portfolio allocation under a fixed training protocol, comparing them with historical scenarios and hybrid policies. The author reports that synthetic scenarios understate volatility and tail losses, distort cross-asset dependence, and that replacement raises turnover and is sensitive to the modified objective's denominator floor. The paper concludes that, within the tested implementation, return differences between generated and historical scenarios have intervals including zero, so robust improvement from generated scenarios cannot be confirmed.

02 KEY FINDINGS

1. Synthetic probability mass varies independently of draw count under a fixed training protocol, and optimal generator fidelity is not assumed.
2. TimeGAN scenarios understate volatility and tail losses and distort dependence; longer training improves reconstruction loss without improving cross-asset correlation accuracy.
3. Replacement raises turnover and is sensitive to the modified objective's denominator floor; all conditional generated-versus-historical return intervals include zero at the reference block length.