Backtest & validation

Event study: every day an insider buys is an event. We recompute the signal using only data available that day (no look-ahead) and measure forward returns over 5 / 21 / 63 trading days. If the signal works, higher score buckets should show higher average returns and win rates.