Coefficient of Variation Filter

In insider-trading and quant signal platforms, the Coefficient of Variation Filter addresses a core challenge: signals with high absolute returns may harbour outsized volatility, rendering them unreliable for portfolio integration. By computing CV = σ / μ for each signal's historical return stream, analysts identify strategies whose returns are stable relative to their dispersion. Lower CV scores indicate tighter clustering around the mean, whereas high CV flags unpredictable or regime-dependent performance. This becomes critical when vetting signals derived from Form 4 filings, insider concentration metrics, or market-abuse detection routines, where false positives or erratic signal quality can corrupt downstream trading decisions.

Implementation typically involves computing rolling or expanding windows of CV over trailing periods (e.g., 60 to 252 trading days), then applying hard thresholds or percentile ranks to admit only signals meeting minimum consistency standards. When combined with other filters such as information coefficient, signal decay, and rolling hit-rate metrics, the CV Filter forms part of a multi-layered conviction architecture. This prevents overweighting of erratic insider-activity signals that spike occasionally but lack sustained predictive power, ensuring that portfolio allocation favours signals exhibiting both statistical significance and operational stability.