InsidersTradesSigma
Quantitative Signals & Scoring
A normalized metric quantifying the incremental predictive power of a signal relative to a random or baseline model, expressed as the ratio of signal-driven outperformance to the baseline prediction error.
The Prediction Lift Coefficient is a diagnostic statistic employed in insider-trading and quantitative scoring platforms to isolate and measure the genuine forecasting contribution of a candidate signal beyond what a naive or historical baseline would achieve. It addresses the critical question of whether observed signal correlations with future price movements reflect true predictive edge or mere statistical artifacts. In quant signal construction, lift coefficients are computed by comparing the ex-post hit rate, directional accuracy, or information ratio of a signal-conditioned portfolio versus an equally-weighted or market-cap-weighted control portfolio. High lift values indicate robust signal discriminatory power, while values near unity suggest the signal adds negligible value beyond market mechanics or momentum carryover.
A lift of 1.0 means the signal does no better than the baseline; the further above 1.0, the more genuine edge it carries. The hard part is making the comparison honest. The lift must come from walk-forward, out-of-sample testing with point-in-time data, or look-ahead bias and curve-fitting will inflate it and a signal that looks predictive in a backtest will collapse in live trading.