Tools such as random forests or adaboost are powerful at solving cross-sectional binary logistic problems or prediction problems where there are many weak learners. But can these tools be adapted to solve panel regression problems?

One could naively introduce a time index as an independent variable but all this does is to provide an additional degree of freedom to the fitting algorithm. What we would like is a solution that allows information from period T-1 to have bearing on period T.

If there is not a straightforward way to do this using these algorithms, is there an alternative algorithm that can perform a panel regression making use of the information in both the cross-section and time-series?

  • 2
    $\begingroup$ So much litterature on this: you might wanna start here: cs.berkeley.edu/~pliang/cs294-spring08/lectures/time/slides.pdf $\endgroup$ – user603 Jul 25 '11 at 17:30
  • $\begingroup$ This was very interesting -- particularly the change point detection and other methods for anomaly detection. It seems entirely focused on anomaly detection as opposed to panel regression using stochastic gradient boosting methods however. If you have any research along those lines I'd appreciate it! Thanks! $\endgroup$ – Ram Ahluwalia Jul 25 '11 at 17:36
  • $\begingroup$ actually, i haven't seen boosting used much in time series analysis...Except in those cases where one suspects a sharp discontinuity/anomaly, I would think that an approach based on filters (i.e. Kalmann) would be a more efficient (statistically, computationally) way to use past residuals in a TSA setting...but since this is a bit off my area of expertise, take this as a comment (not an actual answer). $\endgroup$ – user603 Jul 25 '11 at 20:35
  • $\begingroup$ For random forest/regression trees, there are version for panel/longitudinal data, see the discussions here. BTW, @user603, the link is dead now. $\endgroup$ – Randel Nov 2 '14 at 14:23
  • 1
    $\begingroup$ Possible duplicate of How can I include random effects into a randomForest $\endgroup$ – mkt - Reinstate Monica Jun 19 '19 at 6:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.