Explain why artificial market simulations outperform in some deep hedging settings

Determine the reasons for the observed outperformance of artificial market simulations, when used as the underlying asset price path generator in deep hedging of options, in certain settings across the evaluated risk measures. Identify the mechanisms or conditions within the artificial market simulations that lead to superior hedging or pricing performance in these cases.

Background

The paper proposes using artificial market simulations as the underlying asset simulator within the deep hedging framework and compares performance against traditional simulators such as Brownian motion and the Heston model. Across multiple experiments on European and lookback options evaluated with ERM and CVaR, artificial market simulations sometimes match or outperform the traditional approaches.

While the experiments document these performance differences, the authors explicitly note that the reasons behind the outperformance in certain settings are not fully understood, motivating a focused investigation into the mechanisms or conditions within artificial market simulations that drive these results.