In the world of financial AI, where speed meets complexity, most systems are either too slow to adapt or too brittle to interpret the nuanced messiness of real-world finance. Enter FinKario, a new system that combines event-enhanced financial knowledge graphs with a graph-aware retrieval strategy — and outperforms both specialized financial LLMs and institutional strategies in real-world backtests.
The Retail Investor’s Dilemma
While retail traders drown in information overload, professional research reports contain rich insights — but they’re long, unstructured, and hard to parse. Most LLM-based tools don’t fully exploit these reports. They either extract static attributes (e.g., stock ticker, sector, valuation) or respond to isolated queries without contextual awareness.
But the real challenge is this: financial markets are driven by evolving events — policy changes, product launches, strategic expansions — and these are often buried in text.
FinKario’s Dual-Graph Design
FinKario addresses this by creating a two-part knowledge graph from financial research reports:
| Graph Type | Content Focus | Examples |
|---|---|---|
| Attribute Graph | Static company facts | Industry, Exchange, Ticker, Risk |
| Event Graph | Dynamic, time-sensitive causal drivers | “Overseas expansion”, “Tech upgrade” |
This is not hard-coded: FinKario uses LLMs prompted with templates from CFA, J.P. Morgan, and FIBO ontology to generate schema and extract structured data. It automates what analysts do when they annotate causal chains in reports — but at scale.
The Secret Sauce: FinKario-RAG
To query this knowledge effectively, FinKario introduces FinKario-RAG, a two-stage Retrieval-Augmented Generation process:
- Coarse Retrieval locates the right company and timeframe.
- Fine Retrieval pulls surrounding contextual nodes (e.g., market cap, industry, recent events).
Instead of naively chunking a document or doing entity-only retrieval, FinKario-RAG returns a semantically aligned subgraph that preserves causality and cross-entity reasoning. This enables the final LLM to generate investment insights that are not only accurate — but actually explainable.
Performance: Not Just a Graph, But a Gains Engine
In backtesting on Chinese A-share equities (Aug 2024–Mar 2025), FinKario-RAG beat the best financial LLMs by 18.81% and outperformed leading institutional strategies by 17.85% in predictive accuracy.
| Model | ARR | Sharpe | Accuracy | Max Drawdown |
|---|---|---|---|---|
| FinKario-RAG | 2.633 | 4.926 | 58.1% | 17.2% |
| Guolian-Minsheng | 2.012 | 3.108 | 57.5% | 16.9% |
| Stock-Chain (FinLLM) | 1.177 | 0.971 | 54.6% | 19.0% |
Removing the event graph reduced Sharpe ratio by 81% — a clear signal that event causality, not just entity presence, is what drives intelligent investing.
Why This Matters
FinKario is a step toward a middle ground between black-box LLMs and human analyst workflows. Instead of just fine-tuning more billion-parameter models, it refines what matters: structure, retrieval, and causality.
At Cognaptus, we’ve argued that modern financial AI needs to move past generic sentiment and token-level signals. The edge lies in understanding economic narratives — not just headlines. FinKario shows that with the right graph design and retrieval path, LLMs can do more than chat — they can reason with cause and effect.
Cognaptus: Automate the Present, Incubate the Future.