Overfitting

What it means

A model that fits the noise in its training data as well as the signal, so it performs well in-sample and poorly on new data.

The exam angle

Guarded against by testing on data the model has not seen and by keeping it no more complex than the problem needs.

Where it appears in the syllabus

• Quantitative MethodsCFA Level I · exam weight 11–14% · Returns and benchmarks, time value, distributions, portfolio maths, regression.
• Quantitative MethodsCFA Level II · exam weight 5–10% · Multiple regression, misspecification, time series, machine learning.

Related terms

• Coefficient of determinationThe fraction of the dependent variable’s variation explained by the regression.
• Linear regressionFitting a straight line through data by minimising the sum of squared residuals, to explain a dependent variable by…
• Supervised learningMachine learning that fits a model to examples with known outcomes — labelled data — so it can predict the outcome for…
• Unsupervised learningMachine learning that finds structure in data without a target variable, for example by grouping similar observations…
• AlphaReturn in excess of what the asset’s risk exposures would predict.
• AnnuityA series of equal cash flows at equal intervals.