Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FinCARE: Financial Causal Analysis with Reasoning and Evidence

Alejandro Michel, Abhinav Arun, Bhaskarjit Sarmah, Stefano Pasquali

TL;DR

FinCARE delivers a hybrid causal-analysis framework for finance by fusing statistical causal discovery with domain-grounded knowledge from a SEC 10-K–derived knowledge graph and reasoning from LLMs. The approach enhances three canonical causal-discovery paradigms—PC, GES, and NOTEARS—via KG constraints and/or LLM-informed hypotheses, preserving DAG structure. On a synthetic 500×18 dataset, KG+LLM improvements are substantial, with F1 gains of up to +366% over baselines and NOTEARS achieving near-complete edge recovery (F1=0.759 from 0.163). Counterfactual predictions are highly accurate (MAE=0.003610) with perfect directional accuracy, supporting proactive risk management and strategic decision-making in dynamic markets. Ablation studies reveal that a focused LLM module (MissingEdgeDiscoverer) and algorithmic KG-constraints yield the best performance, while direct KG prompt injections offer limited benefit. Overall, FinCARE grounds statistical discoveries in financial-domain expertise and empirical validation, providing a robust basis for causal scenario analysis in portfolio management.

Abstract

Portfolio managers rely on correlation-based analysis and heuristic methods that fail to capture true causal relationships driving performance. We present a hybrid framework that integrates statistical causal discovery algorithms with domain knowledge from two complementary sources: a financial knowledge graph extracted from SEC 10-K filings and large language model reasoning. Our approach systematically enhances three representative causal discovery paradigms, constraint-based (PC), score-based (GES), and continuous optimization (NOTEARS), by encoding knowledge graph constraints algorithmically and leveraging LLM conceptual reasoning for hypothesis generation. Evaluated on a synthetic financial dataset of 500 firms across 18 variables, our KG+LLM-enhanced methods demonstrate consistent improvements across all three algorithms: PC (F1: 0.622 vs. 0.459 baseline, +36%), GES (F1: 0.735 vs. 0.367, +100%), and NOTEARS (F1: 0.759 vs. 0.163, +366%). The framework enables reliable scenario analysis with mean absolute error of 0.003610 for counterfactual predictions and perfect directional accuracy for intervention effects. It also addresses critical limitations of existing methods by grounding statistical discoveries in financial domain expertise while maintaining empirical validation, providing portfolio managers with the causal foundation necessary for proactive risk management and strategic decision-making in dynamic market environments.

FinCARE: Financial Causal Analysis with Reasoning and Evidence

TL;DR

FinCARE delivers a hybrid causal-analysis framework for finance by fusing statistical causal discovery with domain-grounded knowledge from a SEC 10-K–derived knowledge graph and reasoning from LLMs. The approach enhances three canonical causal-discovery paradigms—PC, GES, and NOTEARS—via KG constraints and/or LLM-informed hypotheses, preserving DAG structure. On a synthetic 500×18 dataset, KG+LLM improvements are substantial, with F1 gains of up to +366% over baselines and NOTEARS achieving near-complete edge recovery (F1=0.759 from 0.163). Counterfactual predictions are highly accurate (MAE=0.003610) with perfect directional accuracy, supporting proactive risk management and strategic decision-making in dynamic markets. Ablation studies reveal that a focused LLM module (MissingEdgeDiscoverer) and algorithmic KG-constraints yield the best performance, while direct KG prompt injections offer limited benefit. Overall, FinCARE grounds statistical discoveries in financial-domain expertise and empirical validation, providing a robust basis for causal scenario analysis in portfolio management.

Abstract

Portfolio managers rely on correlation-based analysis and heuristic methods that fail to capture true causal relationships driving performance. We present a hybrid framework that integrates statistical causal discovery algorithms with domain knowledge from two complementary sources: a financial knowledge graph extracted from SEC 10-K filings and large language model reasoning. Our approach systematically enhances three representative causal discovery paradigms, constraint-based (PC), score-based (GES), and continuous optimization (NOTEARS), by encoding knowledge graph constraints algorithmically and leveraging LLM conceptual reasoning for hypothesis generation. Evaluated on a synthetic financial dataset of 500 firms across 18 variables, our KG+LLM-enhanced methods demonstrate consistent improvements across all three algorithms: PC (F1: 0.622 vs. 0.459 baseline, +36%), GES (F1: 0.735 vs. 0.367, +100%), and NOTEARS (F1: 0.759 vs. 0.163, +366%). The framework enables reliable scenario analysis with mean absolute error of 0.003610 for counterfactual predictions and perfect directional accuracy for intervention effects. It also addresses critical limitations of existing methods by grounding statistical discoveries in financial domain expertise while maintaining empirical validation, providing portfolio managers with the causal foundation necessary for proactive risk management and strategic decision-making in dynamic market environments.
Paper Structure (17 sections, 7 equations, 5 figures, 3 tables)

This paper contains 17 sections, 7 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction & Motivation
  2. Related Work
  3. Knowledge Graph Integration
  4. Statistical Causal Discovery with Knowledge Graph Constraints
  5. Knowledge Graph Constraint Formulation
  6. PC Algorithm: Constraint-Based Discovery
  7. GES Algorithm: Score-Based Discovery
  8. NOTEARS Algorithm: Continuous Optimization
  9. Statistical Causal Discovery with LLMs
  10. Experiments
  11. Benchmark Dataset
  12. Causal Graph Recovery
  13. LLM Enhancement Analysis
  14. Knowledge Graph Prompt Integration
  15. Counterfactual Estimation
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Graph Recovery Performance Across All Methods ($\pm$1 standard deviation)
  • Figure 2: Causal Graph Recovery Comparison: Baseline vs. KG+LLM vs. Ground Truth
  • Figure 3: Leave-One-Out Ablation Analysis of LLM Reasoning Modules on Graph Recovery
  • Figure 4: Counterfactual Estimation for Regulatory Change Intervention
  • Figure 5: Counterfactual results on market risk and return