•
Bühler, Gonon, Teichmann & Wood (2018) train neural nets to hedge derivatives under frictions instead of relying on Black–Scholes deltas. Plain-language intro. Not investment advice.
Takeaway. Deep hedging replaces textbook delta with a neural policy: the net chooses hedge trades to minimise a risk/cost objective when markets have transaction costs, liquidity limits, and incomplete dynamics.
Deep Hedging (Hans Bühler, Lukas Gonon, Josef Teichmann, Ben Wood, 2018; arXiv:1802.03042) is the after-work bridge from Black–Scholes notebooks to friction-aware policies.
Fit for learners: RL-style hedging under frictions — teach yourself to replace Black-Scholes hedges with neural policies in PyTorch.
This Hub note is a plain-language intro. Not investment advice.
Black–Scholes delta assumes continuous, frictionless rebalancing in a complete market. Real books pay spreads, cannot trade continuously, and face unhedgeable risk. A “perfect delta” can be a bad policy once costs exist.
Research pattern—not a live options desk.
You already know “delta = N(d1).” This paper asks: what if the loss is P&L risk under costs? That is a PyTorch computational-graph homework once you can simulate geometric Brownian motion in NumPy.
Tiny evening exercise: simulate 1,000 toy GBM paths; hedge a call with a fixed 0.5 share vs a pathwise BS delta (ignore costs first); print the distribution of terminal P&L—then add a 10 bp cost and watch the “perfect” hedge worsen. Learning only.
Landing: arXiv abs
Not investment advice. This Hub note is educational only. Deep Hedging is academic research—not a recommendation to automate option hedges with real capital.
Single course CTA: SoloQuant Academy’s Introduction to Python Programming—loops, functions, and arrays first; then simulated P&L paths become readable code instead of mystery.