Building Probabilistic Causal Models Using Collective Intelligence

Abstract

The purpose of this article is to show a novel approach to automatically generating probabilistic causal models (Bayesian networks [BNs]) by applying natural language processing (NLP) techniques to a corpus of millions of digitally published news articles in which different authors express views on the future states of economic and financial variables and geopolitical events. The authors will show how to derive BNs that represent the wisdom-of-the-crowds: forward-looking, point-in-time views on various variables of interest and their dependencies. These BNs are likely to be of interest to asset managers and to economists who want to gain a better understanding of the current drivers of an economy based upon a rigorous probabilistic methodology. Additionally, in an asset allocation context, the BNs the authors derive can be fed to an optimization engine to construct a forward-looking optimal portfolio given the constraints of the asset manager (e.g., budget, short constraints). The authors demonstrate various automatically derived BNs in a financial context.