Dynamic Asset Allocation Under Reservoir-Driven Multivariate Regime Switching

Abstract

A quick examination of history should be enough to convince anyone that stock markets are volatile and ever-changing – in fact, the only certainty to be found may be uncertainty. These vicissitudes of the market come in many different configurations, each with their own, complex collection of underlying catalysts. Events such as the Dot-com Bubble, the Global Financial Crisis, and COVID-19 crash (and subsequent run) are a few notable examples of perturbations within the financial markets. The pronounced differences in stock market behavior across time naturally suggest that an investor who’s able to dynamically adjust their portfolio to account for market variation and changing environments, will naturally outperform their peers. Regime switching models that can capture these attributes are thus of high interest. This thesis extends and refines the classic Markov Regime Switching Model by introducing time-varying parameters via an Echo State Network. This novel approach is called the Reservoir-Driven Markov Regime Switching Model and is shown in this paper to significantly outperform static allocation methods in out-of-sample testing.