Polynomial and Regularized Regression Models for Financial Data
Summary
The article introduces extensions to ordinary linear regression for financial modeling. Polynomial regression adds transformed predictor terms to represent curved relationships while remaining linear in the coefficients. The article describes forward selection and backward elimination for choosing polynomial order, and cautions that high orders can overfit and that outliers can affect estimates.
It also explains ridge regression as an L2 penalty that shrinks coefficients to address instability from correlated predictors, while retaining all variables. Predictor scaling is needed, and the penalty strength must be selected, commonly with cross-validation. The introduction lists Lasso, Elastic Net, and Least Angle Regression as further approaches for feature selection and high-dimensional data, though the supplied text is incomplete and does not develop them fully. These are modeling techniques, not demonstrated trading strategies: the document provides no empirical results showing predictive accuracy or profitability, and fitting patterns in financial data does not by itself establish reliable forecasts.
Key ideas
- Polynomial regression represents curved relationships by adding powers or other transformed terms of predictors.
- Forward selection adds polynomial terms incrementally, while backward elimination starts with a larger model and removes terms.
- High polynomial order can overfit, and polynomial regression is sensitive to outliers.
- Ridge regression penalizes large coefficients to stabilize estimates when predictors are correlated.
- Ridge regression requires scaled variables and shrinks coefficients without setting them exactly to zero.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.