FRM Part I · Quantitative Analysis · Free Lesson

Machine Learning Methods and Prediction

Free GARP FRM Part I lesson in Quantitative Analysis. 18 min read, ~2,641 words.

Your firm hands you a credit-risk model with 800 features and 50,000 loans. Classical econometrics wants clean parameters. The trader paying your bonus wants the lowest out-of-sample loss. Machine learning bridges the two. Risk managers need to know exactly what trade is being made.

Classical econometrics asks "what is the true coefficient on X-1, and is it significant?" Machine learning asks "what function minimizes prediction error on data I have not seen?" Different questions, different methods.

Econometrics favors linear models, few parameters, and inference about a data-generating process. ML favors highly nonlinear models, large feature spaces, and out-of-sample performance over significance.

KEY: A statistically significant coefficient may have no predictive value out of sample. A model with low in-sample R-squared may have superior predictive accuracy. Pick the method that matches the question. Capital regulation needs inference, fraud detection needs accuracy.

Most ML methods are sensitive to feature scale. A feature in dollars (range $1,000 to $1,000,000) dominates a feature in fractions (range 0 to 1) unless rescaled.

Read the full lesson, free →
Worked examples and practice. Free with a free account, no card.

Common mistakes

Bottom line

Exam shortcut

When a question gives a model with high training accuracy and low validation accuracy, the answer is overfit and the fix is regularization, more data, or a simpler model. When it asks LASSO vs. ridge with many irrelevant features, LASSO. When it asks accuracy on a 1%-positive problem, the trap is "high accuracy means good model". Wrong unless precision and recall are also strong.

The full lesson (about 2,641 words, 18 min read) adds 2 worked examples, all 6 common mistakes, a self-check, free in the app.

Learning objectives

Browse all free FRM Part I lessons or jump into free FRM Part I practice questions.