How state street boosts investment research
- State Street wanted to help analyst get the gist of investment research.
- Extractive summarisation is a useful NLP concept, helping pluck salient details out of long texts.
- The asset manager used extractive summarisation to condense the research into shorter snippets.
Investment research volumes overwhelm analysts
State Street has estimated that top financial research teams can produce enough content to consume 12,000 sheets of paper per day. With limited time, investors may overlook or miss important research insights.
In this environment, making sense of information quickly can become a competitive edge. Additionally, owing to the European MIFID-II regulations, investment firms now need to pay for research. Understanding which research adds value helps optimize this expense.
State Street decided to develop a solution that helps investment professionals efficiently read and interpret lengthy research reports. The system was rolled out internally, with a view of expanding it as a service to clients.
ML concept: Extractive Summarsation
To decide if a long article is worth reading, it’s helpful to have a short summary featuring the content’s highlights.
One way to generate such a paragraph is to identify within the text several key sentences, which are likely to contain the most crucial information. Extractive Summarisation is the task of automatically identifying which of the sentences are the best choice.
This is done by scoring sentences using automated text-analysis criteria, which attempts to rank the sentences by importance. One such criterion is ranking sentences based on the relative rarity of the words they contain, using a term-weighting scheme. (See NLP Concept on the ING Merchant Bank case study.)
The output of an extractive summarisation task is usually the relevant sentences in the order they appear in the original text. (see image to the right.)
Often, a combination of measurements is used to decide which are the most potent sentences.
Extractive summarisation applied to investment research
In 2017, State Street launched its Quantextual Idea Lab, to improve investment research. The lab launched a research aggregation software, which reviews lengthy research reports and, in addition to tagging and classifying documents, also summarises their content.
Analysts and portfolio managers can quickly extract findings relevant to their investment strategies. The algorithm identifies words and phrases (including financial terms), and uses the keywords to classify documents into topics, asset classes and regions. (See the Topic Modelling concept on the Deutsche Bank case study.)
The system automatically tags and categorizes research, while allowing human reviewers to add and refine tags. Additionally, users can query the research in everyday language.
To help researchers analyse search results, text summarisation algorithms condense the research into shorter snippets, while preserving their nuanced academic or economic tone.