Discovering Bayesian market views for intelligent asset allocation

Along with the advance of opinion mining techniques, publicmood has been found to be a key element for stock market prediction.However, how market participants’ behavior is affected by public moodhas been rarely discussed. Consequently, there has been little progressin leveraging public mood for the asset allocation problem, which is pre-ferred in a trusted and interpretable way. In order to address the issueof incorporating public mood analyzed from social media, we propose toformalize public mood into market views, because market views can beintegrated into the modern portfolio theory. In our framework, the opti-mal market views will maximize returns in each period with a Bayesianasset allocation model. We train two neural models to generate the mar-ket views, and benchmark the model performance on other popular assetallocation strategies. Our experimental results suggest that the formal-ization of market views significantly increases the profitability (5% to10% annually) of the simulated portfolio at a given risk level.