Non-linear mixture of asset return models

Sandrine Tobelem-Foldvari and Pauline Barrieu present a non-linear methodology that combines different asset return models in order to define the preferable portfolio allocation when the investor is averse to model ambiguity. This method offers robustness, tractability and simplicity to investors requiring flexible and easy-to-implement blending of different ambiguous models

When quantitatively determining a portfolio’s asset allocation, a blend of different models’ outputs can be used. Traditional subjective expected utility (SEU) methods do so linearly, and as such cannot capture non-linear strategies used to guard against the uncertainty over how accurately a model reflects the  real-world distribution. For instance, it may be desirable to cap possible allocations to guard against models excessively overweighting a given asset, or to penalise the outliers when