The information coefficient, or IC, is the correlation between a signal's scores and the returns that follow. It measures how well a signal ranks stocks: a higher IC means the stocks it scored highly tended to do better.
How it is calculated
On each date, compare each stock's signal score with its return over the following period, and compute the correlation. The rank IC uses the Spearman rank correlation, which is less affected by extreme values. Repeating this over many dates gives a series of ICs whose average and consistency summarise the signal.
An illustrative example
| Stock | Signal rank (1 = best) | Next-month return rank (1 = best) |
|---|---|---|
| A | 1 | 2 |
| B | 2 | 1 |
| C | 3 | 3 |
| D | 4 | 5 |
| E | 5 | 4 |
The ranks are close, so the rank correlation is high, 0.8 for this tiny example. Real signals have much lower values; an average IC of a few hundredths, for example 0.03 to 0.05, can still be useful across hundreds of stocks. These figures are illustrative.
How it is used
- Judging a signal. The average IC, and the share of periods where it is positive, show whether the signal works.
- Spotting decay. A falling rolling IC is a common sign of signal decay.
- Linking skill to results. The fundamental law of active management says the information ratio is about IC times the square root of the number of independent bets.
Cautions
- Returns over overlapping periods are not independent, which overstates statistical confidence.
- A signal can have a good IC and still be hard to trade after costs. See slippage modeling.
- Sector and size effects can drive an IC. See factor neutralization.
- An IC from a backtest may be inflated by overfitting.
Frequently asked questions
What is the information coefficient?
It is the correlation between a signal's predicted ranking of stocks and their actual subsequent returns, used to measure how well the signal forecasts.
What is a good information coefficient?
Values are usually small. An average IC of a few hundredths can be useful for a signal applied across many stocks, but a good level depends on the market, the horizon and the costs.
What is the difference between IC and rank IC?
IC uses the raw correlation of scores and returns. Rank IC uses the correlation of ranks, which is less affected by outliers.
How does IC relate to the information ratio?
The fundamental law of active management says the information ratio is roughly the IC multiplied by the square root of the number of independent bets.