time of day edge finder

The problem

Intraday strategies often bleed profits during specific hours due to low liquidity or erratic volatility. Relying on aggregate performance data masks these toxic windows, leading to unnecessary drawdowns.

What it does

• Partitions trade logs into hourly and session-based buckets to isolate performance variance.
• Calculates noise-to-edge ratios to distinguish between lucky streaks and statistically stable windows.
• Flags buckets with insufficient sample sizes to prevent over-optimization on sparse data.
• Generates actionable participation schedules, explicitly identifying hours to trade or avoid.

Frameworks & tools

Designed for integration with pandas and numpy. Compatible with any CSV or JSON trade log containing standard timestamps.

Why this beats prompting it yourself

Standard LLM prompts often fail to account for the noise-to-signal ratio, confusing high variance with actual edge. This skill implements a systematic filtering logic that treats low-sample or high-volatility hours as un-tradeable by default.

Use cases

• Optimizing algorithmic execution windows to avoid high-slippage lunch hours.
• Refining manual trading plans by identifying personal performance peaks during specific sessions.
• Filtering backtest results to remove sessions that exhibit non-stationary behavior.

Known limitations

Requires a minimum trade count per bucket to ensure statistical relevance. Analysis is historical and does not predict performance shifts caused by changing market regimes.