Moving from Hunches to Confident Decisions

Finance practitioners and their clients love graphs, and it’s easy to ‘see’ why.

These days, there is another sort of graph driving 21st century finance: Knowledge Graphs were born out of the early semantic networks in the 1960s. The concept gained prominence with the Semantic Web initiative in the 2000s, spearheaded by Tim Berners-Lee (who like Al Gore, invented the Internet) which aimed to make web data machine-readable through linked data. Google’s Knowledge Graph, launched in 2012, popularized the term by enhancing search results with contextual information, integrating data from sources like Freebase and Wikipedia. Since then, companies like Amazon, Microsoft, and financial institutions have adopted knowledge graphs for purposes ranging from recommendation systems to risk analysis. Their purpose is to provide a flexible, scalable framework for organizing and querying complex data, enabling deeper insights, improved decision-making, and enhanced automation.

Knowledge graphs are structured data models that represent information as a network of entities (nodes) and their relationships (edges), forming a graph that captures complex, interconnected knowledge. Nodes might represent real-world entities like companies, people, or concepts, while edges define relationships such as "owns," "influences," or "is part of." This structure, often stored in graph databases like Neo4j, enables efficient querying and reasoning over relationships, making knowledge graphs ideal for applications requiring contextual insights. They are typically built using ontologies or schemas to ensure consistency and can incorporate diverse data sources, from structured databases to unstructured text, often enriched with semantic technologies.

A picture is worth a thousand words, but a Knowledge Graph is worth a lot more!

Knowledge graphs are powerful tools in investment management, particularly for portfolio management and building explainable, audit-friendly models. They organize complex data into interconnected nodes (entities like companies, assets, or markets) and edges (relationships like ownership, correlation, or influence).

This exoteric, intelligible depiction means that these tools are extremely useful for portfolio management, while simultaneously providing for the intelligibility and oversight that AI has struggled with. 

Boosting portfolio management

Knowledge graphs enhance portfolio management by integrating and analyzing diverse data sources intelligently. This allows for:

• Holistic Data Integration: Knowledge graphs unify disparate data—market data, company financials, news sentiment, ESG (environmental, social, governance) factors, and macroeconomic indicators—into a single framework. For example, a graph can link a company’s stock to its supply chain, competitors, and regulatory environment, revealing hidden risks or opportunities.

• Dynamic Risk Assessment: By mapping relationships, knowledge graphs identify systemic risks, such as exposure to a specific sector or geopolitical event. For instance, if a portfolio holds multiple stocks tied to a single supplier, the graph can highlight concentration risk.

• Portfolio Optimization: Graphs enable scenario analysis by simulating how changes (e.g., interest rate hikes or commodity price shocks) propagate through interconnected entities. This helps managers rebalance portfolios to maximize returns or minimize volatility.

• Personalized Investment Strategies: For client portfolios, knowledge graphs can incorporate investor preferences (e.g., ESG priorities or sector exclusions) and match them to suitable assets, ensuring alignment with goals.

This, in turn, allows for much more explainable models

In investment management, explainability is critical for building trust and ensuring decisions are justified. Knowledge graphs support explainable models with:

• Transparent Decision Logic: Unlike black-box models (e.g., deep neural networks), knowledge graphs explicitly represent relationships and dependencies. For example, if a model recommends divesting from a stock, the graph can trace the decision to specific nodes (e.g., poor earnings) and edges (e.g., high debt correlation with market downturns).

• Contextual Insights: Graphs provide context by linking decisions to external factors, such as news events or industry trends. This makes it easier to explain why a model prioritizes certain assets over others.

• Human-Readable Outputs: Portfolio managers can query the graph (e.g., “Why is this stock underperforming?”) and receive intuitive answers based on connected data points, improving communication with clients or stakeholders.

Explainable models are more compliant and audit-friendly models

Regulatory compliance and internal audits require models to be traceable and reproducible. Knowledge graphs excel here through:

• Data Lineage and Provenance: Graphs track the source and transformation of data, ensuring auditors can verify inputs (e.g., market data feeds or underwriting criteria) and their impact on decisions. This is crucial for compliance with regulators like the SEC or FINRA.

• Reproducible Decisions: The structured nature of graphs allows auditors to reconstruct the logic behind investment choices. For example, a graph can show how a risk model flagged a stock due to its exposure to an out of favor sector.

• Version Control and Updates: Knowledge graphs can log changes in relationships or data over time, enabling auditors to review historical decisions and ensure consistency.

‘Drawing’ conclusions

Knowledge graphs serve as both an intermediary step and a stand-alone solution in investment management, particularly for portfolio managers. As an intermediary step, they integrate diverse data into a structured network of entities (nodes) and relationships (edges), generating enriched features contextual embeddings for downstream models. This role streamlines data preprocessing and ensures relationship-aware inputs, improving model performance in tasks like return forecasting or stress testing.

As a stand-alone solution, knowledge graphs enable direct querying for insights, such as identifying portfolio stocks exposed to regulatory risks, with transparent, human-readable explanations (e.g., tracing a stock’s underlying exposures). Their traceability supports auditability by logging data provenance and decision logic, meeting regulatory requirements for the SEC or FINRA. By facilitating real-time risk monitoring and compliance accessibility without additional models, knowledge graphs provide a flexible, queryable framework for decision-making. Their dual utility—feeding models while independently supporting transparent, defensible decisions—makes them invaluable for managing portfolios in regulated environments. It’s easy to see what all the buzz is about. 

Photo by Alina Grubnyak on Unsplash