Robust Quantile Factor Analysis (2501.15761v1)

Abstract: We propose a factor model and an estimator of the factors and loadings that are robust to weak factors. The factors can have an arbitrarily weak influence on the mean or quantile of the outcome variable at most quantile levels; each factor only needs to have a strong impact on the outcome's quantile near one unknown quantile level. The estimator for every factor, loading, and common component is asymptotically normal at the $\sqrt{N}$ or $\sqrt{T}$ rate. It does not require the knowledge of whether the factors are weak and how weak they are. We also develop a weak-factor-robust estimator of the number of factors and a consistent selectors of factors of any desired strength of influence on the quantile or mean of the outcome variable. Monte Carlo simulations demonstrate the effectiveness of our methods.