On \(L^2\)-projections on a space of stochastic integrals (Q1381569): Difference between revisions

Let \(X\) be a vector-valued semimartingale, \(\Theta\) be the space of all predictable processes \(\vartheta\) such that the stochastic integral \(\vartheta\cdot X\) is a square-integrable martingale. A recent paper by \textit{Delbaen} et al. [Finance Stochastics 1, No. 3 (1997)], gives necessary and sufficient conditions that the set \(G_T(\Theta)\) of the terminal values of this martingales is a closed subspace in \(L^2\). The authors describe the structure of the projection on \(G_T(\Theta)\) of an arbitrary \({\mathcal F}_T\)-measurable random variable \(H\in L^2\) and prove that under a certain condition the resulting integrand \(\vartheta ^H\in \Theta\) can be written in a feedback form. The proofs use the variance-optimal martingale measure and weighted norm inequalities. The results have applications in the theory of mean-variance hedging for incomplete markets.