Summary
Two concerns observed when using fect for estimation:
1. Slow Convergence Speed
Iterative convergence is noticeably slow when dealing with large sample sizes or high panel dimensions. This is especially pronounced when using cross-validation for parameter selection, leading to long overall runtimes that hinder practical usage.
2. Results Lack Robustness
Estimation results show non-trivial variation under different initial value settings, subsample splits, or minor parameter adjustments. It would be helpful to improve robustness at the algorithm level or through better default parameter choices to make results more stable and reliable.
Source repo: xuyiqing/fect
Summary
Two concerns observed when using
fectfor estimation:1. Slow Convergence Speed
Iterative convergence is noticeably slow when dealing with large sample sizes or high panel dimensions. This is especially pronounced when using cross-validation for parameter selection, leading to long overall runtimes that hinder practical usage.
2. Results Lack Robustness
Estimation results show non-trivial variation under different initial value settings, subsample splits, or minor parameter adjustments. It would be helpful to improve robustness at the algorithm level or through better default parameter choices to make results more stable and reliable.
Source repo: xuyiqing/fect