Volume 26, Issue 2 -

We design a discrete time arbitrage-free model under incomplete information for application to credit risk models in the spirit of Duffie and Lando (2001). We assume a fundamental value process evolving according to a complete market model and a sequence of imperfect signals conveying information on the state of the fundamental value. The market value is computed by conditioning on the available information. The model shows that if the information setting is built consistently, the market remains complete even in the face of incomplete information. If insiders are not allowed into the market, the requirement is that the equivalent martingale measure of market prices has to coincide with the martingale measure of the fundamental price in the case of perfect information ( Information Consistency ). Such martingale measure can be recovered for arbitrary dependence structures between the fundamental process and the signal. If insiders are allowed into the market, the Information Consistency requirement has to be made more stringent by imposing the H-condition . We show that in this case only a subset of signals are consistent with no arbitrage, and are those that are not-Granger caused by the fundamental process. The model is finally applied to the study of the dynamics of the expected loss of a BBB rated firm in presence of incomplete information.

We deal with the problem of the practical use of Haezendonck risk measures (see Haezendonck and Goovaerts [8], Goovaerts et al. [7], Bellini and Rosazza Gianin [4]) in portfolio optimization. We first analyze the properties of the natural estimators of Haezendonck risk measures by means of numerical simulations and point out a connection with the theory of M -functionals (see Hampel [9], Huber [11], Serfling [19]) that enables us to derive analytic results on the asymptotic distribution of Orlicz premia. We then prove that as in the CVaR case (see Rockafellar and Uryasev [17,18], Bertsimas et al. [6]) the mean/Haezendonck optimal portfolios can be obtained through the solution of a single minimization, and that the resulting efficient frontiers are convex. We conclude with a real data example in which we compare optimal portfolios generated by a mean/Haezendonck criterion with mean/variance and mean/CVaR optimal portfolios.

We introduce an adaptive algorithm to estimate the uncertain parameter of a stochastic optimization problem. The procedure estimates the one-step-ahead means, variances and covariances of a random process in a distribution-free and multidimensional framework when these means, variances and covariances are slowly varying on a given past interval. The quality of the approximate problem obtained when employing our estimation of the uncertain parameter is controlled in function of the number of components of the process and of the length of the largest past interval where the means, variances and covariances slowly vary. The procedure is finally applied to a portfolio selection model.

In recent years optimal portfolio selection strategies for sequential investment have been shown to exist. Although their asymptotical optimality is well established, finite sample properties do need the adjustment of parameters that depend on dimensionality and scale. In this paper we introduce some nearest neighbor based portfolio selectors that solve these problems, and we show that they are also log-optimal for the very general class of stationary and ergodic random processes. The newly proposed algorithm shows very good finite-horizon performance when applied to different markets with different dimensionality or scales without any change: we see it as a very robust strategy.