Minimal entropy preserves the Lévy property: how and why (Q2485828): Difference between revisions

Using a Lévy process \(L\) for modelling in finance results in incomplete market with no unique martingale measure (mm) \(Q\). The authors suggest to choose one particular mm \(Q^E\) via minimization of the relative entropy \[ I_t(Q \mid P) := E_Q\left[\log \left.\frac{dQ}{dP} \;\right|\, {\mathcal F}_t\right] \in [0,\infty] \] of \(Q\) w.r.t. the original measure \(P\). The resulting minimal entropy mm is automatically equivalent to \(P\). The authors' main results are Theorem A and Theorem B. The first one claims that if \(L\) is an \(R^d\)-valued Lévy process under \(P\) and if \(Q^E\) minimizes the relative entropy over all \(Q\) under which \(L\) is a local martingale, then \(L\) is again a Lévy process under \(Q_E\). Theorem B shows how the Lévy property of \(L\) is preserved by passing from \(P\) to \(Q^E\), namely by using an Esscher transform of \(P\). By Girsanov's theorem any \(Q\overset{\text{loc}}{\ll}P\) can be described via two parameters \(\beta\) and \(Y\), which are in general stochastic processes. The authors' crucial idea is that one can reduce relative entropy while preserving the martingale property by a suitable averaging procedure over \(\beta\) and \(Y\). This reduces the problem from a minimization over probability measures to a minimization over non-random functions.