A1428
Title: On desparsified focused information criterion for high-dimensional linear models: with an application to PM2.5 data
Authors: Cy Sin - National Tsing Hua University (Taiwan) [presenting]
Abstract: A modification of the covariate selection criterion (CSC) is proposed to minimize the mean squared errors (MSE) of the focused parameter in causal methods such as difference-in-differences (DiD). Following a previous approach, a debiased Lasso is first applied to a high-dimensional linear model. All regularized models then proceed through the focused information criterion (FIC), and the MSE of the average treatment effect (ATE) parameter is minimized. The method is applied to PM2.5 data comprising 365 city-level observations. Apart from socioeconomic and meteorological covariates, the analysis includes 365 cross-section-specific dummies and covariates.
