I am developed a model to estimating the value of a real estate asset. My model includes a bunch of property and location related features and I am planning on adding more variables for improving the model and controlling effects. Particularly, I am interested in adding economy data such as jobs, earnings etc and measures that capture market demand.

In econometrics theory, this would fixed effects model that are "fixed" for the time period and location. How would I use them in Machine Learning context? Do they behave in a similar way as controlling for average differences across locations of observable and unobservable time-invariant features?


1 Answer 1


Fixed-effects are not consistently defined.

I guess you mean with fixed-effects binary indicators (dummy variables) in a panel setting which are either time-invariant (constant across individuals) or which take the same value for all observations within a certain time period (constant across time periods).

Fixed-effects can be under mild assumptions used to control for unobserved time-invariant heterogeneity in linear models. This however does not translate to non-linear models (due to the so-called incidental parameter problem), see for example this 1948 Journal of Econometrics publication or here (or CrossValidated).

Usually, machine learning models such as tree-based methods are not linear (additively separable). This is why the do not behave as the linear fixed-effects model in econometrics (meaning that they will not automatically take, for example, time-invariant heterogeneity into account).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.