The Information Coefficient (IC)

The Information Coefficient (IC) is a correlation coefficient that measures how well some person's (or model's) 'prediction' matched the 'actual result' that followed.

The range of the value is from -1 to +1.

- IC = +1: the prediction matched perfectly (the ranking matches exactly). - IC = 0: prediction and reality have no relationship (coin-flip level). - IC = -1: the prediction matched exactly in reverse.

In other words, IC summarizes 'predictive power' as a single number. If you predicted the future returns of several stocks, you compute the correlation between the ranking of those predictions and the ranking of actual returns.

Interestingly, even a skilled manager's IC is actually quite small. In stock prediction, an IC of just 0.05–0.1 is considered fairly good.

This is because markets contain enormous randomness, so no matter how skilled you are, individual predictions can't be right every time. Merely 'being right a little more often' can create a large performance difference.

The idea that organizes this is Richard Grinold's 'fundamental law of active management.' Roughly, it says that the larger 'skill (IC) × √(number of independent bets)' is, the greater the excess performance relative to risk. It means that even with somewhat low predictive power, making many different bets can let skill accumulate.

Individuals rarely calculate their own IC precisely. But the concept teaches humility.

Even professionals have an IC of around 0.1—that is, a level of being right 'about six times out of ten.' If so, the conviction that 'I can read the market' is likely, in most cases, overconfidence.

The reason the memory of predicting correctly stays strong is confirmation bias and recency bias. From the IC's perspective, you can understand why 'staying in the market for a long time' is a better strategy for most individuals than showing off predictive skill through frequent trading.