Advanced Econometrics

Download the full article

Methods to deal with high-dimensional problems are of interest in micro-econometrics mostly as a way to perform model selection, whether it is in a context of a non-parametric model estimated by sieve approximation, selection of control variables or instruments. We reviewed the particular case of fixed effect panel data models. These models do not conform to a straightforward application of the regular Lasso since the assumption of approximate sparsity in the individual-specific heterogeneity appears unrealistic and that temporal correlation must be taken into account especially when a Within transformation of the model is considered. A convincing estimator called Cluster-Lasso has been proposed by Belloni et al. (2014): it has desirable theoretical properties based on assumptions that are usual in the high-dimensional literature and is also computationally efficient. In this report, we discuss its properties.