McKenzie Marshall: NLP in Asset Management (Barings)

2019-07-06 · Source: Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai · Field: Finance & Economics — Capital Markets & Investment Management, FinTech & Digital Financial Services · Depth: Intermediate, long

Summary

McKenzie Marshall of Barings discusses the application of Natural Language Processing (NLP) in active asset management, emphasizing its role in augmenting investment research to generate "alpha" by efficiently processing extensive qualitative data like news and regulatory documents. Barings' approach focuses on assisting analysts, not automating their roles. The productionization of NLP solutions involves three sequential subtasks: identifying companies in documents (a Named Entity Recognition problem), rectifying these entities to internal IDs using algorithmic solutions, and providing an additive ranking metric, initially a sentiment score. Key challenges in NER included training a custom "company" label to filter irrelevant entities like regulatory bodies or platform products used as verbs, and distinguishing company names from general acronyms or product modifiers. Sentiment analysis presented difficulties due to the prevalence of neutral text, the need for polarity conventions, and the variability of journalistic versus regulatory language. Focused annotation management was crucial for specialized, high-value models.

Key takeaway

For Machine Learning Engineers building NLP solutions in finance, prioritize augmenting human analysts over full automation. Your success hinges on meticulously training custom Named Entity Recognition models with focused annotation management. This ensures precise identification of domain-specific entities. Combine NLP outputs with robust rules-based entity rectification. Simplify sentiment scores into clear buckets (good/bad/neutral) to deliver pragmatic, high-value tools that directly support investment research processes.

Key insights

Effective NLP in asset management augments human analysis by precisely identifying and contextualizing company-specific information from diverse text sources.

Principles

• Augment, don't automate, human analytical processes.
• Specialized annotation management drives business value.
• Algorithmic cleaning enhances NLP pipeline pragmatism.

Method

Productionizing NLP for text consumption involves identifying companies via custom NER, rectifying entities to internal IDs with rules-based fuzzy matching, and deriving relevance metrics like sentiment from polar sentences.

In practice

• Train custom NER labels for domain-specific entities.
• Combine NLP with rules-based algorithms for entity resolution.
• Bucket sentiment scores (good/bad/neutral) for clarity.

Topics

• Natural Language Processing
• Asset Management
• Named Entity Recognition
• Sentiment Analysis
• Data Annotation
• Entity Resolution

Best for: NLP Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• NLP-based Page Classification for Efficient LLM Extraction from Brazilian Public Tender Documents — Combining NLP page classification with LLM extraction significantly reduces token count and cost for document processing.
• Predicting Overnight Stock Changes Post-Earnings: Tomorrow’s Price Hidden in Today’s Language — Earnings call language, processed with domain-specific NLP, predicts overnight stock movements.
• Luxury and nationalities, my first data science project — The project applies sentiment analysis and NLP to hotel reviews to understand luxury tourism demand by nationality.
• Exploring NLP Applications in Food Research: my ATRIUM TNA visit to the GATE group — NLP tools significantly enhance social media data analysis for consumer behavior and food research.
• NLP: From Prototype to Production — Production NLP prioritizes specific business utility and maintainability over raw academic accuracy.
• Evaluating Reference-Free Summarization Quality Metrics for Portuguese: A Study with Human Judgments in Financial News — Decomposing summarization evaluation into atomic QA pairs surpasses holistic LLM-as-a-Judge methods for Portuguese financial news.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai.