The Information Coefficient (IC) measures how good you are at predicting which stocks will outperform. It’s the correlation between your predictions and actual outcomes - the purest measure of investment skill.
Beginner
What It Means
IC answers: “When I predict a stock will do well, how often is that prediction correct?” It’s measured as a correlation, ranging from -1 to +1.
Portfolio Example
At the start of each quarter, you predict expected returns for 100 stocks. At quarter end, you compare predictions to actual returns.
| IC Value | Interpretation |
|---|
| 0.00 | No predictive ability (random guessing) |
| 0.03 | Weak but positive skill |
| 0.05 | Typical skilled manager |
| 0.10 | Strong skill (rare) |
Why It Matters
IC directly measures forecasting skill - the core of active management value. If you can’t predict which stocks will outperform, you can’t add value. IC tells you if your predictions have any merit.
Advanced
Mathematical Definition
IC = Correlation(Forecast Returns, Realized Returns)
IC = Cov(f, r) / (σf × σr)
Where:
- f = Forecasted returns
- r = Realized returns
- Range: -1.0 to +1.0
Realistic IC Values
Most investors overestimate achievable IC:
| Manager Type | Typical IC |
|---|
| Bottom Decile | -0.02 to +0.01 |
| Median Manager | 0.02 - 0.04 |
| Top Quartile | 0.05 - 0.08 |
| Top Decile | 0.08 - 0.12 |
An IC of 0.10 is exceptional. Claims of IC above 0.15 should be viewed with extreme skepticism.
The Fundamental Law of Active Management
IC connects to expected performance through:
E(IR) = IC × √BR
Where:
- IR = Information Ratio
- IC = Information Coefficient
- BR = Breadth (independent bets per year)
- IC = 0.05, BR = 100 independent bets
- E(IR) = 0.05 × √100 = 0.05 × 10 = 0.50
Why Small IC Matters
Even tiny IC creates value with enough breadth:
| IC | Breadth | Expected IR |
|---|
| 0.02 | 100 | 0.20 |
| 0.05 | 100 | 0.50 |
| 0.05 | 400 | 1.00 |
| 0.10 | 100 | 1.00 |
The law shows two paths to high IR: better skill (higher IC) or more independent bets (higher breadth). Most quant strategies focus on breadth since IC is hard to improve.
IC Stability
IC is not constant:
| Factor | Effect on IC |
|---|
| Market Volatility | Higher dispersion = higher achievable IC |
| Regime Changes | IC varies across bull/bear markets |
| Strategy Crowding | More users = lower IC |
| Information Decay | Signals lose power over time |
Measuring IC
| Approach | Description |
|---|
| Cross-Sectional | Rank correlation each period across all stocks |
| Time-Series | Track individual stock forecast accuracy over time |
| Quintile Spreads | Top quintile return minus bottom quintile |
Data Requirements
| Requirement | Details |
|---|
| Observations | 100+ independent forecasts minimum |
| Preferred | 500+ forecasts for stable IC estimate |
| Time Coverage | Multiple periods to confirm consistency |
| Cross-Sectional | 50-100 stocks per period typical |
Limitations
| Limitation | Description |
|---|
| Hard to Measure | Requires detailed forecast data |
| Time-Varying | IC changes across market conditions |
| Implementation Gap | Forecast IC differs from realized portfolio IC |
| Correlation Effects | High correlations reduce effective breadth |
IC vs. Hit Ratio
| Metric | Measures | Considers Magnitude? |
|---|
| Hit Ratio | Frequency of correct direction | No |
| IC | Correlation with outcomes | Yes |
