A State-based Methodology for Engineering Portfolio Risk and Improving Efficiency

This dissertation joins the body of work in finance and analytics examining the methods for characterizing financial markets and improving asset allocation decisions. It explores the existing body of literature to discuss how research in the domain of finance and investment management has incorporated the use of discrete states into decision making, particularly with applications aimed at investment portfolio construction and design. An innovative approach is presented that derives a feature set incorporating return and risk characteristics from the market and segments the market into definable, discrete states. By segmenting the population into discrete states, this research can (1) define and analyze the transitional dynamics between states, (2) create state-specific asset allocation models based on a risk-return portfolio metric, (3) calculate the probability of state membership for any observation in a holdout period, and (4) create an asset allocation strategy that is able to adapt to changing market conditions. The research examines various optimization-based approaches for deriving the state-specific asset allocations and compare the results. Several techniques are presented for evaluating the quality of the state definition scheme presented, such as whether the transitional dynamics between states hold statistically across multiple time periods. The preliminary results show that this method produces promising risk-adjusted portfolio results; however, a great deal of work remains that can build upon these initial findings.