This workflow is a fully automated, AI-driven quantitative factor discovery and validation system purpose-built for cryptocurrency markets. Users simply submit a natural-language factor description via an instant messaging channel, and the system delivers a complete backtesting report within minutes — covering the full pipeline from idea to insights.

Workflow Overview

1. Factor Description Input: The system listens for user messages through an instant messaging channel, accepting natural-language inputs such as "RSI overbought expecting pullback" or "consecutive small declines with shrinking volume, anticipating a sharp drop."

2. AI Factor Code Generation: A large language model parses the user's description to identify the factor's expected direction (long/short signal), calculation logic, and data dependencies, then generates a standardized JavaScript factor function. The system handles three input paradigms — standard factors (Momentum, RSI, MACD, etc.), descriptive factors, and composite conditional factors — and automatically sets the correct sign convention: positive values for bullish expectations, negative values for bearish expectations.

3. Multi-Asset K-line Data Retrieval: The system concurrently fetches up to 365 periods of daily historical data for a pre-configured watchlist. Data is cleaned to remove anomalies (extreme amplitude, OHLC logic violations, etc.) and organized into a structured dataset indexed by symbol.

4. Factor Computation & Multi-Dimensional Validation: Factor values are computed daily across all assets, with the following validation suite applied:

   • IC / RankIC Analysis: Measures correlation between factor values and next-day returns, with t-statistics for significance testing
   • IR Analysis: Evaluates the Information Ratio and consistency of predictive signal over time
   • Monotonicity Test: Splits assets into five quantile groups by factor value and verifies whether returns increase monotonically
   • Long-Short Symmetry Analysis: Assesses the balance between long-side and short-side signal strength
   • Factor Decay Analysis: Estimates the effective half-life of the factor via autocorrelation, informing rebalancing frequency
   • Market-Cap Domain Consistency: Validates IC consistency across small-, mid-, and large-cap segments
   • Transaction Cost Simulation: Net-return backtesting inclusive of fees and slippage, assessing whether turnover erodes factor alpha

5. AI-Driven Interpretation: A second AI model professionally interprets the validation results and produces a concise structured report — including an overall score (0–100), letter grade (A+ to D), improvement suggestions, and risk warnings — formatted as standardized JSON.

6. Result Delivery: The final report is split into two messages — "Factor Evaluation Report" and "Generated Factor Code" — formatted and pushed to the user's messaging channel with second-level response latency.

Use Cases

• Rapid factor prototyping and validation for quantitative researchers
• Building and expanding factor libraries for crypto multi-factor models
• Pre-development feasibility screening for trading strategy ideas

Vedio Link：Factor Analysis Workflow