High-dimensional mixed data sampling models with a covariate-dependent threshold
提出一种高维混合数据抽样模型，允许因变量和自变量在不同频率下存在阈值效应，且阈值随协变量变化；基于MCMC和线性近似开发稀疏组LASSO估计，理论证明预测和估计的渐近一致性，模拟和GDP现在预测实证显示模型能提升预测精度。
.This article introduces high-dimensional mixed data sampling models with a covariate-dependent threshold, which allows for a threshold effect in the relationship between dependent and independent variables sampled at different frequencies and allows for threshold regimes depending on a time-varying threshold being modeled as a linear function of informative covariates. Based on the MCMC technique and linear approximation, we develop a sparse group LASSO (sg-LASSO) estimator of model parameters. We also establish non asymptotic oracle inequalities for the prediction risk, the l1 and l∞ bounds for the parameter estimator and show that these bounds can be translated easily into asymptotic consistency for prediction, estimation, variable selection, and threshold detection. Monte Carlo simulations are conducted to examine the estimation and predictive performance. The simulation results point out that the estimation, prediction, and selection procedures work well in finite samples. The model is illustrated with an empirical application to nowcasting US GDP growth. Our empirical results demonstrate that the proposed model achieves statistically significant gains in GDP nowcasting accuracy by incorporating a threshold effect to capture non linear effects and performing a regime-specific variable selection.