Abstracts

In this paper, we study the question of model risk control and model blending when performing asset allocation, in order to provide a clear and traceable way to account for model ambiguity, blend the different models considered and compute a cash reserve. Indeed, investors often have to decide how to allocate their wealth among various assets, by mixing different (and potentially conflicting) models for the assets returns. At a time when regulators impose more constraints on banks and financial institutions in terms of liquidity ratios and capital requirements, as specified in Basel III, taking into account model risk has become increasingly crucial in the asset allocation problem. To deal with these issues, we introduce a two-step robust ambiguity adjustment offering the advantages of being tractable and easy to implement even in large dimension. This approach decomposes the ambiguity aversion into two components: a model specific absolute ambiguity aversion and a relative ambiguity aversion across the set of different prior models considered for the asset returns. The decision process first involves the transformation of the optimal allocations under each prior through a generic absolute ambiguity function. Then the adjusted allocations are mixed through an adjustment function that reflects the relative ambiguity aversion of the investor towards the different models within the set of priors considered, as well as their relative contribution to the overall performance and risk of the portfolio. We illustrate the methodology through the study of a theoretical example and perform some empirical tests on European stock data.