Signal Research

Signal Research is a quantitative skill from the CLI Trader project that teaches trading agents how to systematically discover, evaluate, and maintain alpha signals for stocks and crypto markets. Built by the CLI Trader team, this skill moves beyond discretionary trading ideas and into the domain of structured, data-driven signal engineering. It gives your agent the methodology to transform raw market data into quantitative features, test their predictive power rigorously, and combine surviving signals into multi-factor models that drive systematic trading decisions.

The skill covers the full signal research pipeline. Feature engineering transforms raw price, volume, and fundamental data into candidate signals — returns over various lookback windows, volatility metrics, z-scores of key ratios, relative strength measures, and cross-sectional rankings. Each candidate signal is evaluated using information coefficient (IC) and information ratio (IR) analysis, which measure how well the signal predicts future returns and how consistent that predictive power is over time. Signal decay analysis tracks how quickly a signal’s edge erodes, helping the agent determine appropriate holding periods and rebalance frequencies. Walk-forward cross-validation is the core testing methodology — the agent trains on historical data and tests on subsequent out-of-sample periods, sliding the window forward to simulate real deployment conditions. Signals that survive this process are combined into multi-factor models using weighting schemes that account for signal correlation, capacity, and decay characteristics. Throughout the entire pipeline, the skill enforces overfitting detection by tracking the number of hypotheses tested and applying appropriate statistical corrections.

In a trading workflow, the agent uses this skill to continuously research and validate new signal ideas, monitor the health of existing signals in production, and flag when a signal’s performance has degraded beyond acceptable thresholds. This creates a systematic research process where your trading edge is actively maintained rather than assumed to persist indefinitely, and where every signal in your portfolio of strategies has a clear statistical justification for its inclusion.

For the complete methodology — including formulas, decision trees, worked examples, and validation checklists — see the full Signal Research skill guide.