Overfitting and Linear Extrapolation Bias
Overfitting treats patterns specific to observed data as if they will generalize. Linear extrapolation assumes a straight-line relationship continues beyond the observed range. Either can create unwarranted confidence in a forecast.
THE IDEA
Overfitting and Linear Extrapolation Bias
Overfitting occurs when a model captures noise or incidental patterns in its data and performs poorly on new observations. Linear extrapolation extends an observed straight-line trend beyond the range in which it was measured. These are different problems: a linear model can be too simple for a relationship, while a highly flexible model can overfit it.
The forecasting concern is assuming that a good fit to past data establishes what will happen next. Changing conditions, nonlinear relationships, or rare events can undermine that assumption. Ask whether the relationship is appropriate for the task and what evidence supports using it outside the observed data.
When to use it
Evaluating investment strategies that rely solely on historical performance.
Testing long-term business projections against scenarios involving market disruptions.
Examining how policy or technology changes would affect projections based on current emission trends.
FROM THEORY TO EVERYDAY LIFE
In practice
Financial Market Forecasts
An analyst looks at a stock's price over the last five years, sees a steady upward trend, and predicts it will continue rising. They ignore market volatility, economic shocks, and mean reversion. The stock may suddenly crash, defying the linear projection.
Weather Prediction
A meteorologist observes that daily temperatures have been increasing gradually over two weeks. Assuming the linear trend continues, they forecast even hotter days. However, a cold front arrives unexpectedly, breaking the pattern.
A MOMENT TO REFLECT
Put it to work.
Think of a time you assumed a trend would continue linearly. What unforeseen event disrupted that assumption?
How can you guard against overfitting when analyzing data and making predictions?
In what areas of your life do you most often rely on simple linear models, and what could go wrong?
References
- The Black Swan: The Impact of the Highly Improbable — Nassim Nicholas Taleb (2007)
CONTINUE THE CONVERSATION
Source texts
- The Black Swan The Impact of the Highly Improbable With a new section On Robustness and Fragility (Nassim Nicholas Taleb)