Enhanced Scenario Analysis

Abstract

Investors have long relied on scenario analysis as an alternative to mean-variance analysis to help them construct portfolios. Even though mean-variance analysis accounts for all potential scenarios, many investors find it difficult to implement because it requires them to specify statistical features of asset classes which are often unintuitive and difficult to estimate. Scenario analysis, by contrast, requires only that investors specify a small set of potential outcomes as projections of economic variables and assign probabilities to their occurrence. It is, therefore, more intuitive than mean-variance analysis, but it is highly subjective. In this article, the authors propose to replace the subjective elements of scenario analysis with a robust statistical process. They use a multivariate measure of statistical distance to estimate probabilities of prospective scenarios. Next, they construct portfolios that maximize utility for investors with different risk preferences. Lastly, the authors introduce a procedure for minimally modifying scenarios in order to render them consistent with one’s pre-specified views about their probabilities of occurrence.

Keywords: Covariance matrix, Economic scenarios, Euclidean distance, Financial turbulence, Gradient descent, Mahalanobis distance, Mean reversion, Mean-variance analysis, Multivariate normal distribution, Persistence, Scale independent, Scenario analysis

JEL Classification: C01, C02, C13, C18, C4, C6, G10, G17