Outperforming the market portfolio with a given probability (Q453241)

The authors consider a general setup within the stochastic portfolio theory of \textit{E. R. Fernholz} and \textit{I. Karatzas}, cf. [in: A. Bensoussan (ed.) et al., Handbook of Numerical Analysis 15, 89--167 (2009; Zbl 1180.91267)], in which risky assets' dynamics \((X_t)\) are given through SDEs driven by a Brownian motion. The setup only assumes existence of a price deflator which is a local martingale. This rules out the so-called UPBR (unbounded profits with bounded risk) but is weaker that the classical no free lunch with vanishing risk and hence does not imply that an equivalent local martingale measure exists. In this setup, the authors consider the quantile hedging problem of \textit{H. Föllmer} and \textit{P. Leukert} [Finance Stoch. 3, No. 3, 251--273 (1999; Zbl 0977.91019)]: Given a certain target position \(g(X_T)\) at time \(T\), what is the smallest capital \(U(t,X_t,p)\) required at time \(t<T\) to superreplicate \(g(X_T)\) with a fixed probability \(p\in [0,1]\)? First, a probabilistic characterisation of the solution is obtained with a (generalised) Neyman-Pearson lemma. Secondly, \(U\) is characterised as the minimal viscosity supersolution of a non-linear PDE. The latter result forms the main body of the paper. To obtain it, the authors investigate the Legendre transform \(w\) of \(U\) and show that \(w\) is a continuous viscosity solution to a linear PDE. This is obtained with an elliptic regularisation argument in which \(w\) is constructed as a limit of \(w_\epsilon\) which are classical solutions of a linear PDE arising from an enlarged system, where an additional Brownian motion is added. A Legendre duality argument allows to see that \(U_\epsilon\), the Legendre transform of \(w_\epsilon\), then solves a non-linear PDE, and a stability argument for the viscosity solution allows to conclude.