Heteroscedasticity-robust model screening: A useful toolkit for model averaging in big data analytics

- Propose a new heteroscedasticity-robust model screening (HRMS) method.
- Show that HRMS has good performance in simulation.
- Demonstrate that HRMS is computationally efficient.
- Show that HRMS can lead to large gains in box office prediction accuracy.

Frequentist model averaging has been demonstrated as an efficient tool to deal with model uncertainty in big data analysis. In contrast with a conventional data set, the number of regressors in a big data set is usually quite large, which leads to a exponential number of potential candidate models. In this paper, we propose a heteroscedasticity-robust model screening (HRMS) method that constructs a candidate model set through an iterative procedure. Our simulation results and empirical exercise with big data analytics demonstrate the superiority of our HRMS method over existing methods.