Institutional Equity Holdings Prediction Using Node Affinities of Dynamic Graphs

2026-07-13 · Source: Machine Learning · Field: Finance & Economics — Capital Markets & Investment Management, FinTech & Digital Financial Services · Depth: Expert, quick

Summary

A new benchmark for institutional equity holdings prediction utilizes temporal graph machine learning, framing the task as node affinity prediction on a discrete-time temporal bipartite graph of managers and securities. Using SEC Form 13F filings from 2013-2025, the Node Affinity prediction model using Virtual State (NAVIS) achieved a test Normalized Discounted Cumulative Gain (NDCG) of 0.9127 with features on a dataset of 99 managers and 503 S&P 500 securities. This significantly outperforms all dynamic graph representation learning competitors. Notably, a simple Exponential Moving Average baseline reached 0.8882, highlighting strong persistence in institutional portfolios. Domain-specific node features provided only marginal gains, less than 1.2%, suggesting temporal and structural signals are highly predictive.

Key takeaway

For machine learning engineers developing financial forecasting models, particularly for institutional equity holdings, you should explore temporal graph machine learning. The NAVIS model demonstrates superior performance (NDCG 0.9127) by leveraging structural and temporal signals from SEC Form 13F data, often outperforming complex feature engineering. Consider adopting this approach to enhance prediction accuracy and streamline feature development for portfolio allocation tasks.

Key insights

Forecasting portfolio weights can be effectively modeled as node affinity prediction on dynamic bipartite graphs.

Principles

• Institutional portfolios exhibit strong persistence and smoothness.
• Temporal and structural graph signals are highly predictive.
• Domain-specific features add minimal predictive power.

Method

Predict future portfolio weights by framing holdings prediction as node affinity on a discrete-time temporal bipartite graph of managers and securities.

In practice

• Implement NAVIS for superior equity holdings forecasts.
• Prioritize temporal graph structures over complex features.

Topics

• Institutional Equity Holdings
• Temporal Graph Machine Learning
• Node Affinity Prediction
• NAVIS
• SEC Form 13F
• Portfolio Allocation

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• Data-Driven Duration Management -- Term Structure Forecasting Using Machine Learning — Neural Networks significantly enhance zero-rate curve forecasting and bond portfolio performance compared to traditional models.
• Interpretable Factor Decomposition for Decision Intelligence in Large-Scale Financial Markets: Evidence from China's A-Share Market — An interpretable machine learning pipeline quantifies auditable factor contributions to equity returns, highlighting behavioral signals' predictive power.
• Macroeconomic Message Passing for Anticipating Foreign Exchange Regime Changes: A Deep Logical Learning Approach using Graph Tsetlin Machines — Graph Tsetlin Machines with macroeconomic message passing effectively predict FX market regimes, offering interpretable, robust, and computationally efficient forecasting.
• A Three-Phase Foundation Model for Tax-Aware Personalized Portfolio Management — A multi-phase deep RL system integrates foundation models, MoE, and personalized LoRA for tax-aware portfolio management.
• Evaluating Financial Intelligence in Large Language Models: Benchmarking SuperInvesting AI with LLM Engines — Evaluating financial intelligence in LLMs requires a multi-dimensional framework beyond simple factual accuracy.
• Deep Learning That Respects the Theorems: Arbitrage-Free Neural Networks for Commodity Futures — CARRY-Net embeds economic theory into neural network architecture to achieve arbitrage-free, interpretable, and accurate commodity futures forecasting.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.