Alpha Research.
Information Coefficient, quintile spreads, signal decay. The toolkit professional quants use before committing capital.
A signal, properly cross-examined.
Pick a universe, pick a signal, pick a lookback. Quantle ranks every name in the universe each day by signal value, bucks the ranks into quintiles, and records the daily return of each bucket. The result is the four diagnostics quants actually use: IC time series, Information Ratio, quintile spread cumulative returns, and decay.
The signal catalog starts with four research staples — 12-1 momentum, RSI z-score, price/EMA-200, short-term reversal — each warmed up cleanly and aligned across the universe. The output is the same shape professional alpha researchers stare at: a chart you can disqualify a bad idea from in fifteen seconds.
Rank, bucket, average, decay.
At each date t, every name in the universe is ranked by signal value. Ranks are computed cross-sectionally — across symbols, not across time.
Spearman rank correlation between today's signal ranks and tomorrow's return ranks. Done per day for the IC time series; mean and SD give you IR.
Bucket ranks into N equal-count groups (default 5). Each bucket is equal-weight; daily portfolio return recorded. Q5 − Q1 long-short is the headline payoff.
Re-run the Spearman test against forward returns at horizons of 1, 2, 5, 10, 20 days. Plot IC by horizon — the slope tells you how quickly the alpha dies.
Frequently asked
- What is the Information Coefficient?
- The IC is the cross-sectional Spearman rank correlation between today's signal value and tomorrow's return, computed across every name in the universe. A positive IC means high-signal names tend to outperform; a negative IC means they tend to underperform. The IC time-series is the headline diagnostic for any alpha factor.
- What is IR and how is it different from IC?
- IR is the Information Ratio: mean IC divided by the standard deviation of IC across time. Where IC tells you direction at one point in time, IR tells you whether the signal is consistent. A signal with a 0.03 mean IC and 0.04 standard error has IR ≈ 0.75 — solid. The same mean IC with 0.10 SE has IR 0.3 — noise.
- How are the quintile portfolios formed?
- At each rebalance date, every name in the universe is ranked by signal value and bucketed into N equal-count quantiles (default five). Each bucket is held equal-weight until the next rebalance and the daily portfolio return is recorded. Q5 is the highest-signal cohort; the Q5 − Q1 long-short tells you the spread the factor would have paid.
- What signals are built in?
- Four to start: 12-1 momentum (price return over 12 months excluding the most recent), 14-day RSI z-score, log(price / EMA-200), and a short-term 5-day reversal. Each loads from clean OHLC and warms up over a documented window; custom-formula authoring is on the roadmap.
- What is signal decay?
- Decay measures how quickly a signal's predictive power dies as the forecast horizon lengthens. The tool runs the same Spearman test against forward returns over horizons of 1, 2, 5, 10 and 20 days. A signal whose IC at 1d is +0.05 but drops to 0 by day 5 is a high-turnover bet; one that holds 60% of its IC out to 10 days is far easier to trade.
- What does turnover mean here?
- Average fraction of names whose quintile membership changed day-over-day. A turnover of 60% means more than half your top-decile names rotated since yesterday — a sign the signal is noisy and would chew transaction costs. Low turnover (under 20%) is one of the few free lunches in factor research.