What it means
A cross-sectional IC compares predictions for several securities at one forecast date with their later returns over a specified horizon, while a time-series IC compares predictions and outcomes for one security across dates. The measurement method must be clear.
Pearson correlation examines linear association between numerical values, while rank correlation assesses whether the ordering of predictions aligns with the ordering of outcomes. A value near positive one indicates strong positive association under the selected method, and a value near negative one indicates opposite association, not simply a count of wrong forecasts.
A value near zero indicates little measured association in the sample. It does not prove there is no useful relationship of any kind, and small samples can produce unstable estimates.
Forecast horizon matters, since a signal intended to predict next month's return should not be judged against a different horizon without explaining the change in the question. The investment universe and return definition also matter.
Including different securities, using total rather than price return, or adjusting for market and other exposures can change the coefficient. Avoid look-ahead bias: predictions should use only information available when the forecast would have been made, with later outcomes kept separate during model development and testing.
ICs can vary over time, including changes in sign. One favourable period does not establish a durable skill, so analysts often examine repeated measurements and uncertainty rather than a single figure.
A good coefficient does not automatically create a good portfolio, because turnover, costs, risk limits, position sizing, capacity, and execution can turn a useful forecast ranking into disappointing realised returns. For non-finance managers reviewing a model, ask what the coefficient actually compares and whether the testing resembles live use.
Distinguish forecast quality from the information ratio, which concerns portfolio excess returns relative to their variability under a stated benchmark.
In practice
Real-world examples.
Example
An analyst ranks five shares by predicted monthly return and calculates a rank IC after the month ends. The report states the horizon, universe, and rank method so readers do not confuse the result with exact price prediction accuracy.
Example
A model has a positive average IC but high turnover. The investment team estimates trading costs and risk before claiming that the predictive relationship generates a useful net portfolio return.
Example
A research model uses revised historical data that was unavailable at the forecast date. Its attractive coefficient is rechecked using point-in-time inputs, because later information can make historical predictions appear artificially skilled.
Formula
Calculation
For Pearson IC, calculate the covariance between predictions and realised returns divided by the product of their standard deviations. Both standard deviations must be nonzero, and the sample and horizon must match.
Suppose three hypothetical predicted values are 1, 2, and 3, while realised values are 2, 4, and 6. Pearson correlation is positive one because the relationship is perfectly linear, even though the predicted levels differ from actual values.
A rank method can instead correlate the ranks. The expression two times the share of correct directional predictions minus one is a hit-rate transformation, not the general correlation definition of IC used in quantitative investment analysis.Case study
Seen in the real world.
This fictional case follows a company reviewing an investment-selection model for its pension advisers. The sales presentation describes an IC of 0.1 as an accuracy score without showing the underlying test. The review team asks for the securities, forecast dates, return horizon, and correlation method. It discovers that results vary substantially across periods and that several forecasts use data later revised by the provider.
The adviser reruns the analysis with point-in-time inputs and an out-of-sample period. Portfolio simulations then include turnover costs and realistic position limits instead of translating the coefficient directly into expected profit. Management receives a more cautious assessment that separates evidence of forecast association from net investment performance. The coefficient remains useful, but its definition, uncertainty, and changing behaviour are visible rather than hidden behind a single favourable number.
Watch out
Common mistakes.
- Calling IC a general percentage of correct predictions or using a directional hit-rate formula as the universal correlation definition.
- Ignoring horizon, universe, point-in-time data, and sample uncertainty when comparing model coefficients.
- Treating a positive IC as proof of profitable net trading or confusing forecast association with portfolio information ratio.
Questions
People also ask.
Does an IC of one mean every predicted return is exact?
No. Correlation can be perfect when predictions and outcomes have a consistent linear relationship but different levels or scales. It measures association under the selected method.
Can a useful model have a small IC?
Yes, but usefulness must be demonstrated with uncertainty, repeated tests, costs, and portfolio construction. A small positive estimate can also arise from noise and may not persist.
What should be disclosed with the coefficient?
Disclose the correlation method, universe, forecast horizon, return definition, sample period, data timing, and variation over time. Show out-of-sample evidence and separate forecasting results from realised net returns.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%