Dynamic programming and mean-variance hedging (Q1297911): Difference between revisions

The paper deals with a market consisting of \(n+1\) primitive assets, one bond of the price process \(S_{t}^{0}=\exp(\int_{0}^{t}r_{u}du)\) and \(n\) risky assets driven by Itô processes. A contingent claim is an \({\mathcal F}_{T}\)-measurable square integrable random variable \(H\). In this context the mean-variance hedging means solving the optimization problem: \[ \min_{\theta\in \Theta}E[H-V_{T}^{x,\theta}]^{2},\quad\text{where}\quad V_{T}^{x,\theta}=S_{T}^{0}\left(x+\int_{0}^{T} \theta'_{t}d(S/S^{0})_{t}\right) \] is the terminal value of a self-financed portfolio in the primitive assets with the initial capital \(x\) and the quantity \(\theta\) invested in the risky assets. The hedging numéraire and the variance-optimal martingale measure appear to be a key tool for characterizing the optimal hedging strategy. Using the dynamic programming methods the authors provide an explicit description of the hedging numéraire and the variance-optimal martingale measure in terms of a value function for a suitable stochastic control problem. The authors study a typical incomplete market framework, a stochastic volatility model. They prove regularity of the value function and derive explicit forms of the hedging numéraire and the variance-optimal martingale measure.