Study guides / CFA L1: CFA Program Level I / Regression & big data
Regression & big data
CFA L1: CFA Program Level I · quick review
Simple linear regression to explain one variable with another, how to judge the fit, and how big data and machine learning fit into investment analysis.
The rules
- Simple regression: Y = b₀ + b₁X + ε. The slope b₁ = Cov(X, Y) ÷ Var(X) shows the change in Y for a one-unit change in X.
- Assumptions: linear relationship, homoskedastic (constant variance) residuals, independent residuals, normally distributed residuals.
- R² = explained variation ÷ total variation; the share of Y's variation explained by X. In simple regression, R² = correlation².
- Standard error of estimate measures the typical size of residuals; smaller is better.
- Test the slope with t = (b₁ − hypothesized value) ÷ standard error of b₁. The F-test tests overall significance.
- Big data has volume, velocity and variety (plus veracity). Supervised learning uses labeled data to predict; unsupervised learning finds structure in unlabeled data. Overfitting fits noise instead of signal.
Formulas
- Slope: b₁ = Cov(X,Y) ÷ Var(X)
- Intercept: b₀ = Ȳ − b₁X̄
- Coefficient of determination: R² = SSR ÷ SST
Common traps
- A high R² doesn't prove causation.
- Predicting far outside the range of the X data used is unreliable.
- An overfit model performs well in-sample but poorly out-of-sample.
How it's tested
Interpreting slope, intercept and R², testing significance, and recognizing big data and machine learning concepts.
Practise it
The full lesson, worked examples and practice questions are in Tickmark, where every answer choice is explained.
Open the full lesson