Deep Hedging the Volume-Weighted Average Price Risk in Order-Driven Markets

Problem: Buying Q units of a certain asset S over the time interval [0, T] (one trading day), where our average purchase price is compared to the market volume-weighted average price (VWAP) during the same time interval. We focus on the buy side; the sell case is symmetric.

We show how to build a trading policy (i.e., a program to buy in the market via MOs and LOs) to hedge the downside risk of our wealth (i.e., minimize potential large losses).

This paper analyzes VWAP (Volume-Weighted Average Price) risk in order books using deep hedging. Market scenarios are simulated with a model calibrated to real data, and neural networks are trained to determine how much to trade actively and passively during the session, minimizing the risk of extreme losses. The results show that optimal strategies depend on the market microstructure: in cases where spreads are narrow, more aggressive strategies are used initially, while in orders with wider spreads, passive strategies are preferred. Thus, deep hedging emerges as an effective tool for designing trading strategies to minimize VWAP risk in order books.