The RATS Software Forum

LASSO.RPF is an example of estimation by the LASSO (Least Absolute Shrinkage and Selection Operator) which does least squares subject to a constraint on the sum of absolute values of the coefficients. This type of constraint can (and often does) force some coefficients to zero, hence the "Selection Operator" in the title.

This requires version 11 of RATS.

Detailed Description

Thanks for adding the Lasso capability with the PLS command. As I understand the documentation, the output from PLS does not seem to yield a list of the regressors selected as belonging in the regression (and having non-zero coefficients)? So one needs to check the %beta obtained from PLS to directly construct a list of the regressors with non-zero coefficients? Is this correct? I ask because there are applications (e.g., some papers in the local projections literature) that use LASSO to determine a reduced regressor set and then rerun an OLS regression with just the reduced regressor set. (I suppose one way to implement this would be to run OLS on the full set but using RESTRICT to impose the 0's from the PLS result.) Thanks much.

Todd

The 11.00c build has a SELECT option on PLS which cuts all non-zero coefficients out of the output model. (Both the output and any derived statistics).

Terrific, thanks very much for adding that!