Optimal asset allocation: a worst scenario expectation approach (Q438769)

In this paper, the authors deal with an investment model where not only volatility and expected returns but also the knowledge and the information of the market are taken into consideration for the determination of the optimal investment strategy. In fact, in the model there is a standard complete market with an \(n\)-dimensional Brownian motion (w.r.t an objective probability \(P\)) feeding the uncertainty and an investor who must select the optimal portfolio. However, the investor is not sure about the information she is getting from the market hence, in order to model this feature of the market, a convex set \(Q\) of probability measures is given where the size of this set depends on the investor's degree of conservativeness. In this context, where concerns about misunderstanding of the economic environment are explicitly given, it is interesting to investigate what is a sensible way to solve the portfolio problem. The proposed solution is based on the concept of coherent risk measure which is used as a criteria for computing the optimal portfolio. Therefore the authors use, as risk measure, the potential loss deviation from the expectation (w.r.t. \(P\)) of the portfolio due to the incorrect judgment of assets' returns, in other words the worst case expected loss over a set of generalized scenarios \(Q\). They show, with the help of Girsanov's Theorem, that the constraint maximization problem can be transformed in a suitable form by using the relative entropy. Namely, they assume that \(KL(Q,P)\leq K \forall Q\in F\), where \(KL(Q,P)\) is the relative entropy of \(Q\) with respect to \(P\) and the parameter \(K\) depends on the investor's confidence on the available information and her opinions on the market where she makes the investment decision. By applying tools from multiobjective optimization theory they are able to solve the problem in the general case by finding the optimal investment strategy which is the main problem. The authors also discuss the incomplete case scenario. Finally, an interesting comparison of the new proposed solution and its efficient frontier is compared with the mean-variance approach. The paper also contains numerical examples.