FRM Part I · Quantitative Analysis · Free Lesson

Regression Diagnostics

Free GARP FRM Part I lesson in Quantitative Analysis. 19 min read, ~2,813 words.

You fit a regression to your trading desk's daily P&L against three risk factors. The R-squared is 0.62 and every coefficient looks significant. A month later, your hedges drift, the standard errors flip sign on a re-estimate, and your boss asks why the model lied. Diagnostics catch what the t-stats hide.

Ordinary least squares (OLS) delivers the smallest-variance unbiased linear estimator under five conditions: linearity in parameters, zero conditional mean errors, homoskedasticity, no autocorrelation, and no perfect collinearity.

When these hold, OLS is BLUE. Drop any one and a different estimator may dominate, or the standard errors become invalid. The exam tests which assumption breaks each variation and what the consequence is. Bias in beta, bias in SEs, or loss of efficiency.

KEY: Heteroskedasticity and autocorrelation do NOT bias the coefficient estimates. They bias the standard errors. The point estimates of beta stay correct on average; the inference around them is wrong.

Heteroskedasticity is when error variance changes with the regressors. Picture residuals from a regression of P&L on position size. Small positions produce small residuals, large positions produce large residuals.

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

Common mistakes

Bottom line

Exam shortcut

When a regression diagnostic question gives you a residual pattern and asks "what's the issue?", match the visual: fanning out → heteroskedasticity (robust SEs); curvature → omitted nonlinear term; serial pattern in residuals → autocorrelation (Newey-West SEs). When the question gives high R-squared, high F-stat, but weak individual t-stats, the answer is multicollinearity, check VIFs. Memory aid: "Hetero hurts SEs not betas. Multi hurts variance not bias.

The full lesson (about 2,813 words, 19 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.