2

Specifically, I would like to estimate a model that includes, on the RHS:

  • Individual and time fixed effects
  • A set of variables that may or may not be included

I would like to keep the coefficients on the fixed effects 'unrestricted', so basically as estimated with OLS ($\lambda = 0$).

I am not aware of a formalised procedure for this, let alone implementation in any software package.

My intuition would be to regress the full set of FEs on my dependent variable, and then use the resulting residuals as the dependent variable in the LASSO.

The main reason I am trying to do this is to establish some comparability with earlier literature that relied on OLS completely. The fixed effects merely have the function of absorbing some differences in measurement units, and I am only interested in the variables that are selected beyond that.

whuber
  • 281,159
  • 54
  • 637
  • 1,101
ljrog
  • 31
  • 3
  • 1
    Your intuition is almost right: you also need to regress all the remaining independent variables against the FEs and replace them with their residuals. See https://stats.stackexchange.com/a/28493/919 or https://stats.stackexchange.com/a/46508/919. – whuber Nov 28 '17 at 19:48

0 Answers0