The variance-optimal martingale measure for continuous processes (Q1915163)

Let \(S\) be a multidimensional semimartingale on \((\Omega, {\mathcal F}, P)\) and consider the set \({\mathcal M}^s (P)\) of all signed measures \(Q\) with \(P\)-square-integrable density and such that each elementary stochastic integral of \(X\) has \(Q\)-expectation 0. These measures come up in applications in financial mathematics. \(Q^{\text{opt}} \in {\mathcal M}^s (P)\) is called variance-optimal if its density has minimal \({\mathcal L}^2 (P)\)-norm. The main result is that \(Q^{\text{opt}}\) is automatically equivalent to \(P\) if \(S\) is continuous and if \({\mathcal M}^s (P)\) contains at least one element which is equivalent to \(P\). [Note that \({\mathcal M}^s\) should read \({\mathcal M}^e\) in Theorem 1.3.] For discontinuous \(S\), this conclusion fails in general. An approximation problem is discussed as an application.