Optimal portfolio policies under bounded expected loss and partial information (Q5962146)

The authors investigate a financial market with \(n\) stocks and return vector \[ R_t= \int^t_0 \mu_s ds+ \int^t_0 \sigma_s dW_s,\quad t\leq T, \] where \(T\) is the terminal trading time, \(W\) is an \(n\)-dimensional standard Brownian motion, \(\sigma\) the volatility matrix process and \(\mu\) the stochastic drift process, actually independent of \(W\). Only the returns are supposed to be observed, thus the information is incomplete (partial observation). Anyway, as usual, the process \(\sigma\) can be supposed observable since the matrix \(\sigma\sigma^T\) is nothing else than the cross variation of the observed returns \(R\). The observed filtration is denoted by \({\mathcal F}^R_t\), generated by observations \(\{R_s, s\leq t\}\). Then, the authors' aim is to optimize the expected utility \(U\) of the terminal wealth \(X^\pi_T\) with respect to the portfolios \(\pi\), restrained to satisfy a risk constraint \(E_Q[\gamma(X^\pi_T- q)^-]\leq\varepsilon\) (so-called expected loss criterium) with shortfall \(q\) and a discounting factor \(\gamma\). In a first case, \(Q\) is a risk probability measure, and in a second case it is the historical probability measure. Because of partial information, a filtering problem is to be solved -- the authors consider the innovation process \(dV_t= \sigma^{-1}_t(dR_t- E[\mu_t/{\mathcal F}^R_t]\,dt).\) The Kalman-Bucy filter and the hidden Markov model filter are used before solving the constrained optimization problem. In both cases, under sufficient assumptions, optimal terminal wealth and optimal trading strategies \(\pi\) are given. Some numerical experiments illustrate these results.