Narrative-based robust stochastic optimization

Abstract

Many portfolio optimization techniques rely heavily on past data and modeling assumptions. In an uncertain and ambiguous world, these techniques are prone to amplify model misspecification and therefore have poor out of sample results. Robust optimization explicitly recognizes uncertainty in model specification and performs better out of sample. The Achilles’ heel of the method is the selection of the uncertainty set. In this paper we focus on the construction of the uncertainty set around the stochastic model specification. We propose to use narratives to select the elements in the uncertainty set to avoid using a logically inconsistent or too large uncertainty set. The narratives provide useful tools in a qualitative sense to the portfolio management process.