SVR-based method for fixed effects interval-valued panel models

This article studies a class of fixed effects interval-valued panel models in which the response variables are intervals and the explanatory variables are single data. We propose an SVR-based estimation method to estimate unknown parameters and smoothing functions. The proposed method utilizes the kernel trick to map the original data (also called input data) to a high-dimensional feature space, which is more likely to achieve linearity than in the original space. Moreover, the proposed method is resistant to the influence of outliers. Monte Carlo experiments validate that the proposed models have good fitting accuracy and prediction performance, while the empirical application shows that the proposed models work efficiently on finite samples.