Ridge regression
Least squares with a charge on the size of the coefficients. It pulls every coefficient towards zero, most of all along directions where the data say little.
01The formula
- the penalty: how much you charge for large coefficients (0 gives ordinary least squares)
- the sum of squared coefficients
- the signals and the outcomes, as in least squares
- the identity matrix: adds λ to each signal's own variance
02See it move
03Where it is used
- Quant researcherFits forecasts on many overlapping signals without the coefficients exploding in opposite directions.
- Portfolio managerUses the same idea to shrink noisy covariance estimates before building a portfolio.
- Risk managerStabilises factor exposures that would otherwise swing from one estimation window to the next.
04Why traders use it
When signals are correlated, plain least squares can give huge offsetting coefficients that fit the past and fail out of sample. Ridge trades a little bias for a big cut in variance.
05Worked example
Here the OLS estimates are 1.5 and 0.5 and the signals have correlation 0.60. With λ = 1.00 (per observation) ridge gives 0.758 and 0.473: a coefficient vector 57% as long, for an extra squared error of only 0.575.
06How it is built
- Add the penalty to the squared error: .
- Set the gradient to zero: .
- Collect terms: .
- Adding to the diagonal makes the matrix easy to invert even when signals are almost identical, which is why the answer is stable.
07The common mistake
Penalising signals measured in different units. Ridge charges every coefficient the same, so standardise each signal first or the penalty falls mostly on whichever one happens to have small units.
08In Python and LaTeX
import numpy as np rho, lam = 0.6, 1.0 S = np.array([[1, rho], [rho, 1]]) # X'X / n b_ols = np.array([1.5, 0.5]) b = np.linalg.solve(S + lam * np.eye(2), S @ b_ols) print(b.round(3)) # [0.758 0.473]
\hat{\boldsymbol\beta}_{\text{ridge}} = \arg\min_{\boldsymbol\beta} \|\mathbf y - X\boldsymbol\beta\|^2 + \lambda\|\boldsymbol\beta\|^2 = (X^\top X + \lambda I)^{-1} X^\top \mathbf y09Where it fits
Rather not code the maths yourself? Describe your trading idea and let AI build the bot.
Build a bot with AgenKitThe worked example and the model run the same code, so every number matches. Models simplify markets on purpose. This is education, not financial advice.