How NLP Is Advancing Asset Management

Although natural language processing (NLP) has been around for decades, the recent and rapid rise of deep learning algorithms together with the increasing availability of massive amounts of text data are creating new and appealing opportunities for the tech across many industry sectors, including in the investment world.

Unlike rule-based or statistical algorithms which depend on human-crafted rules or task-specific ad-hoc features, the deep learning approach trains a single end-to-end model, discovering the rules and features along the way. Such models are able to obtain high performance across a variety of NLP tasks. Recent research breakthroughs have also improved DL models’ understanding of context and made them better able to tackle large-scale problems.

The asset management industry is particularly interested in these capabilities. Leading firms like Citadel and AIG are leveraging signals from alternative data such as social network info, shopping history, shipping info, GPS and satellite data, etc. in a bid to increase active investment return — and are continuously exploring how NLP technology can improve efficiency and scalability in this practice. Automation of the ingestion and analysis of public filings and quick consumption and sentiment scoring of news and social media content are also being introduced by in-house teams and third-party providers in asset management.

UBS wealth management is using NLP to speed up the process of due diligence. The company’s innovation group developed an AI-armed investment platform which detects negative news by reading vast amounts of documents fetched from search engines. Hours of human work are thus reduced to seconds, freeing asset managers to focus on downstream tasks. The same is happening for the company’s client screening process, where profiles are scanned to identify whether features meet a client’s criteria.

Periscope Capital, a US$250 million Toronto hedge fund, built a neural network to provide a sentiment score whenever a tweet, blog posting, or comment comes in. Views are then aggregated to assist in real-time investment strategy development. The startup implemented the project in the US$11 million BUZZ US Sentiment Leaders ETF (Exchange-traded Fund).

S&P Global is using DL and NLP to generate sentiment scores on Chinese Small & Medium Enterprise (SME) filings for its clients in an attempt to increase the corpus’ interpretability and reduce the side effect of conflicting and/or missing data. With a layered neural network, the S&P R&D team extracts risk signals from millions of documents to provides early alerts.

The firm’s Kensho team has been using NLP since its inception. A text processing engine assigns topics/themes to incoming news automatically and provides a surveillance service for clients including Goldman Sachs and JP Morgan. S&P data scientists are also working on applying state-of-the-art neural network architecture to generate sentiment scores with higher accuracy when companies appear in news articles.

Of course, no algorithm is perfect. Asset management goals are generally open-ended, which machine intelligence is not very good at. Director of leading Robo-advisor firm Betterment Dan Egan told CNBC he still has reservations regarding technology’s ability to generate accurate financial advice or manage entire portfolios. Algorithms excel at specific, well-defined problems and so the key is to survey the wide range of asset management activities to identify business use cases where text data is rich.

Although AI’s incorporation into asset management remains at an early stage, it is clear the tech has much to offer in today’s complex and interconnected global investment environment.

---

Author: Jingya Xu | Editor: Michael Sarazen | Producer: Chain Zhang