Quantitative research system / 01
Quant Research Agent & Knowledge Base
A quantitative research agent working with a layered knowledge base of 1,800 financial-engineering reports and materials from quant communities. The agent retrieves methods and evidence for research questions, checks the original text and figures, and uses them to construct and test candidate factors. Topic overviews, research notes and source materials remain linked across three layers.
Results & evidence
All 1,800 reports have been archived into 30 topics, 1,193 method families, 18,093 concepts and 89,799 references to source locations. In a debt-based factor study on 28 October 2026, the research agent QR retrieved findings from four method families, consulted the sources, implemented and tested four base factors, then recombined them based on economic rationale. The combined factor achieved a Sharpe ratio of 2.27 and Fitness of 2.05 in WorldQuant BRAIN testing, with an Excellent rating.
Topic overviews, research notes and source materials
Topic overviews give agents a concise guide to the library. The research-note layer connects methods and findings, while the source layer holds searchable text and figures. The figure includes examples of method families. Community discussions and open-source materials are stored separately; all 30 topics can be explored in the interactive example below.
View full figure ↗Research topics & knowledge network
QRL organises quantitative finance reports into 30 research topics, with links to discussions from quant communities. Methods and concepts connect to findings, their scope and source locations.
Options, index futures and convertibles: pricing and trading
Option pricing, implied volatility and hedging; index-futures basis and spreads; convertible value decomposition and contractual provisions. Each contract retains its own pricing logic.
The example starts with 01 / 30, “Options, index futures and convertibles: pricing and trading,” and draws on related materials from other topics to show six derivatives research areas. These panels are illustrative groupings, not formal subcategories of that topic.
Basis measures, timing signals and futures–spot hedging, with reported results, cost questions and validation limits retained.
This selection contains 490 nodes and 1,029 within-panel relationships. Select an entry to inspect its direct connections. Source entries and pointers do not provide full texts.
Building and combining leverage and cash-flow factors
The upper panel shows QRL's retrieval results: the relevant topic, four method families, and findings annotated with scope, source references and evidence levels. The lower panel shows the four base factors implemented and tested by QR, together with the combined factor D. All metrics are from the platform's 2019–2023 in-sample test; the factor has not been formally submitted or evaluated out of sample.
View full figure ↗QR's research process
Finding evidence and choosing a research direction
In the initial tests, several formulas using debt data directly fell below the platform's requirements. QR searched QRL for cross-sectional factors based on debt data, retrieved findings from four method families and consulted the source passages before selecting a direction. Each finding includes its scope, source and evidence level, distinguishing report findings, cited academic research and community experience.
Implementing and testing four base factors
QR implemented and tested four base factors: leverage ratio, a version with compressed extremes, and the time-series rank and deviation from the mean of cash flow from operations divided by enterprise value (CFO/EV). All four met the platform's in-sample thresholds with an AVERAGE rating. QR defined the formulas and budget before each test round; I reviewed the research direction and budget.
Comparing the combined factor with its components
The strongest individual factor was CFO/EV deviation from its mean, with a Sharpe ratio of 1.79, Fitness of 1.27 and an AVERAGE rating. Its daily PnL correlation with the leverage factor was 0.13. Adding their cross-sectional ranks produced a combined factor with a Sharpe ratio of 2.27, Fitness of 2.05 and an Excellent rating. Annualised return was approximately 4 percentage points above an equally weighted portfolio of the two individual strategies. No combination weights were fitted, and all comparisons are in sample.
Record & scope
The library supplies references for fitness design and mechanism analysis in factor research, with links from the experiments to the relevant literature.
Evaluation settings & scope
Knowledge-structure figures are based on the October 2026 archive. Retrieval uses the second index release, published on 16 October 2026.
All case metrics are WorldQuant BRAIN in-sample results for 2019–2023. The factor has not been formally submitted or tested out of sample. Three of the four base factors match publicly documented formulas and were reproduced in this account under fixed settings before use. Factor D combines two of them.
External materials come from WorldQuant BRAIN, Quant Stack Exchange, Kaggle, Numerai, QuantConnect and GitHub. They are publicly available or accessible with authorisation, used solely for personal research and not redistributed.