Taming Intuitive Predictions
Intuitive forecasts can be too extreme when they treat an imperfect signal as a reliable guide. Base rates and the strength of the evidence can help determine how far to adjust a forecast toward the mean.
THE IDEA
Taming Intuitive Predictions
An exceptional observation need not justify an equally exceptional forecast. If the observation is an imperfect signal of the outcome, prediction should account for the relevant average and the strength of the relationship.
Least squares fits a model by minimizing squared errors in the fitted data. Its usefulness for prediction depends on the model, data, and objective; minimizing that criterion does not guarantee accurate forecasts in a new setting. Other loss functions can favor different predictions.
To examine an intuitive forecast, identify the relevant base rate, assess how informative the observation is, and consider how much adjustment toward the mean is warranted.
When to use it
Predicting employee performance based on a single interview or test score.
Forecasting sales for a new product after an initial successful launch.
Estimating student grades from a single exam, considering the class average.
Making investment decisions after a period of extraordinary returns.
FROM THEORY TO EVERYDAY LIFE
In practice
Job Interview Performance
An interviewer sees an exceptional interview and predicts outstanding job performance. If interview performance is an imperfect predictor of job performance, the forecast should be less extreme. The appropriate adjustment depends on evidence about that relationship.
Stock Market Forecasting
An investor sees a stock’s sharp price increase and expects continued high returns. Before extrapolating, the investor asks whether recent gains predict future returns in this setting and how much uncertainty the forecast leaves.
A MOMENT TO REFLECT
Put it to work.
Recall a situation where you made an intuitive prediction that turned out to be too extreme. How could regression to the mean have improved your prediction?
What does minimizing squared error optimize, and when might that objective differ from what matters for your decision?
Think of a domain where predictions are commonly made (e.g., sports, weather, finance). How can taming intuitive predictions improve decision-making?
Often confused with
References
- Thinking, Fast and Slow — Daniel Kahneman (2011)
CONTINUE THE CONVERSATION
Source texts
- Thinking, Fast and Slow (Daniel Kahneman)