The DRL agent trained on GBM asset paths shows superior performance without transaction costs but faces challenges with 3% costs, failing to outperform the binomial strategy.
In the absence of transaction costs, the DRL agent outperformed the BS Delta method, showing its capacity for effective hedging, especially in the early exercise context.
The experiments indicate that while the DRL agent exhibits robust performance, its higher standard deviation in final P&L highlights its risk profile compared to traditional methods.
Testing with stochastic volatility models reveals real-world applicability concerns, as initial results based on arbitrary parameters may not effectively represent market behaviors.
#deep-reinforcement-learning #geometric-brownian-motion #financial-modeling #transaction-costs #stochastic-volatility
Collection
[
|
...
]