1: \begin{abstract}

2: 

3: Hedging exotic options in presence of market frictions

4: is an important risk management task. 

5: Deep hedging can solve %provides approximate solutions

6: such hedging problems

7: by training neural network policies

8: in realistic simulated markets.

9: Training these neural networks may be delicate and suffer from slow convergence, particularly for options

10: with long maturities and complex sensitivities

11: to market parameters.  

12: To address this,

13: we propose

14: a second-order optimization scheme

15: for deep hedging.

16: We leverage pathwise differentiability

17: to construct a

18: curvature matrix,

19: which we approximate

20: as block-diagonal and Kronecker-factored

21: to efficiently precondition gradients.

22: We evaluate our method on a challenging and practically important problem: hedging a cliquet option

23: on a stock with stochastic volatility by trading in the spot and vanilla options. 

24: We find that

25: our second-order scheme

26: can

27: optimize the policy in $1/4$ of the number of steps

28: that

29: standard adaptive moment-based optimization takes.

30: \end{abstract}

31: