Automating Factor Discovery with Walk-Forward IC Validation

Summary

The article presents an automated workflow for discovering and refreshing cross-sectional factors for crypto perpetual contracts. It selects liquid markets, classifies volatility conditions, prompts an AI system to propose factors, and evaluates candidates with walk-forward Rank IC: factor values calculated from data through the prior bar are compared with the next bar’s cross-sectional return ranks. Candidates are screened for minimum IC, redundancy, and pool capacity; surviving factors are standardized, weighted by recent IC, and used to rank long and short candidates. A faster loop separately monitors positions and applies stop-loss, take-profit, and trailing exits.

The author reports that in a two-day live observation, many factors’ recent IC declined, and the system responded by exploring new dimensions. This demonstrates the workflow’s operation, not predictive value or profitability. The article acknowledges that the observation period is too short to establish adaptation, that current-snapshot correlation can misclassify redundancy, and that trading remains risky. Costs, robust statistical significance, and longer out-of-sample performance are not established.

Key ideas

• The workflow repeatedly generates, validates, filters, and combines candidate factors for liquid perpetual markets.
• Walk-forward Rank IC compares factor ranks based on past data with subsequent cross-sectional return ranks.
• Correlation filtering and pool limits aim to remove redundant factors and retain stronger candidates.
• Recent IC determines factor weights, while negative recent IC receives zero weight in the described combination.
• The reported two-day observation shows declining factor IC and system iteration, but does not establish long-term effectiveness or profitability.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.