What are generalized additive models (GAMs) and why are they useful in finance?

Question

AcadiFi · Accepted Answer

Generalized additive models (GAMs) extend GLMs by replacing the linear predictor Xβ with a sum of smooth, flexible functions of each feature: g(E[Y|X]) = β₀ + f₁(X₁) + f₂(X₂) + … + fₖ(Xₖ). Each fⱼ is a smooth function (typically a spline) estimated from the data rather than assumed linear. Why this matters: many financial relationships are non-linear. The effect of interest rates on bond prices is convex, the relationship between leverage and default probability is nonlinear at extreme values, the impact of macro factors on returns may have thresholds or saturation points. Forcing linearity misses these effects. ```mermaid flowchart LR X1[Feature
X1] --> F1[Smooth
f1] X2[Feature
X2] --> F2[Smooth
f2] X3[Feature
X3] --> F3[Smooth
f3] F1 --> S[Sum] F2 --> S F3 --> S S --> G[Link g] G --> Y[Response] ``` Example: Halberd Capital predicts next-year equity returns using three features — earnings yield, 10-year Treasury yield, and VIX. A GAM fits three smooth functions, revealing that earnings yield has a strong linear positive effect, Treasury yield has a non-monotonic effect (rising then falling), and VIX shows a threshold where returns only depend on VIX when it's above 25. GAMs preserve interpretability — you can plot each fⱼ to visualize the partial effect of feature Xⱼ holding others fixed. This contrasts with black-box machine learning models. The key parameter is the smoothness penalty — controlled by generalized cross-validation to prevent overfitting while allowing flexibility. GAMs combine with GLM families (logistic GAM for classification, Poisson GAM for counts). They're particularly useful when: (1) domain knowledge suggests non-linearity but the form is unclear, (2) interpretability matters (regulatory models, credit scoring), (3) the dataset is large enough to fit flexible functions reliably. The CFA curriculum treats GAMs as a bridge between classical statistics and machine learning.

What are generalized additive models (GAMs) and why are they useful in finance?

Master Level II with our CFA Course

Related Questions

Practice Questions