When in 1976, Robert Merton introduced jumps to the modelling of stock prices, he attributed the occurrence of such to „the arrival of an important piece of information on the stock“. The relationship between news and asset price movements has, however, so far not been addressed, due to mathematicians focusing specifically on numbers and linguists on text. At riskl.io we bridge this gap. This presentation is based on ongoing research. A scientific paper will be available by the time of the data festival. The presentation itself will focus more on data and visual examples and will avoid scientific complexity.
The presentation formulates the problem statement based on a set of news stories (i.e. Dieselgate, Cambridge Analytica, Donald Trump / Elon Musk tweets) and their impact on stock prices. It is shown that in order to correctly classify relevant news, it is better to start with locating jumps in the asset prices and use these to label the unsorted news data. This ‘market sentiment’ approach is a new alternative to a purely text-based sentiment analysis that generalises this known concept. Using NLP techniques and Bayesian methods allows to bootstrap common market-moving themes from data. The presentation discusses the advantages of applying actuarial mathematics and machine learning to identify relevant news. Once identified, quantitative jump statistics can then be aggregated across themes and levels (company, market, sector, but also portfolio) and offers a new family of financial information, especially suitable for retail investors. The presentation concludes with discussing the challenges we were facing with data visualisation. A way to use the data was to automatically create annotations in stock charts, thus visually combining price with text data. Interestingly, there was no technical solution that could do that so we set out to develop our own stock chart based on D3.js.
Dr. Stephan Werner, Founder / CEO, riskl.io