Grey-Box Methods in Forecasting Financial Markets

Abstract

We first present methods of data analysis in defining stochastic mathematical models suitable for use in forecasting financial markets. With the purpose of multi-period portfolio selection via model predictive control, we focus on input-output model structures. By capturing cause-and-effect dynamic behaviors these models exhibit improved fidelity in simulation. Second we present a probabilistic approach for augmenting the identified models with auxiliary speculative/subjective information derived from analyst and regulatory reports. The technique is an application of the Kalman filter and can be interpreted as a logical extension — to a multi-period framework — of the well-known single-period Black-Litterman approach from portfolio optimization.