Reproduce “Fast Deep Hedging with Second-Order Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Konrad Mueller, Amira Akkari, Lukas Gonon, Ben Wood
A second-order optimization scheme, DH-KFAC, for training deep hedging neural network policies significantly reduces the number of optimization steps required for complex exotic options. The method achieved a fourfold speedup compared to Adam and showed that hedging with vanilla options can reduce terminal PnL standard deviation by an additional 50% while increasing skewness.
Key insights
- Second-order optimization methods can substantially accelerate the training of deep hedging models by more effectively navigating high-curvature regions of the loss landscape.
- Leveraging the differentiability of the deep hedging environment allows for the construction of a specific GGN-type preconditioner that is more suitable than directly applying the Fisher Information Matrix.
- Strategic hedging with vanilla options, beyond simple delta hedging, provides substantial additional risk reduction (reducing PnL standard deviation and increasing skewness) for path-dependent derivatives.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.