Title: Intelligent asset management and NLP
Authors: Frank Xing - Nanyang Technological University (Singapore) [presenting]
Abstract: Asset allocation models consider financial variables that are traditionally computed from numerical format data. Whereas the amount of unstructured text data has surged in the past decades, there is no place to accommodate such data in asset allocation models. We propose a possible solution to integrate such data, thus make asset management more intelligent. The solution links the text data to the old variables using natural language processing techniques. We elaborate on two examples: one is to approach asset return distribution via Bayesian revision using asset-specific news sentiment, the other is to model dependence using asset correlations estimated from the semantic information of company descriptions. In various simulations and experiments, introducing text data appears to be beneficial. Specifically, market sentiment views increase major performance indicators, such as CAGR and Sharpe ratio, by 5\% to 20\%; the semantic vine (used for covariance matrix estimation) also has a 98\% chance to outperform an arbitrary vine for portfolio optimization. Moreover, the results could potentially improve as more sophisticated NLP methods are adapted. Similar opportunities might exist for other financial variables needed for asset management.