Threshold quantile regression neural network

Abstract

ABSTRACT For modelling the threshold effect in parameters of the quantile neural network models, this paper introduces a model called threshold quantile regression neural network (TQRNN), which allows a threshold effect in the quantile regression neural network (QRNN). We develop the estimation procedure of the proposed model and suggest test statistics for threshold effect and the linear setting used typically in the quantile regression framework. We conduct Monte Carlo simulations to assess the performance of estimation and testing procedures, and compare the forecasting performance of the TQRNN model relative to the QRNN model. Through Monte Carlo simulations, we show that the estimation and testing procedures work well in finite samples, and there is little loss in forecast accuracy by employing the suggested TQRNN model even when the true data is generated from a QRNN model; on the contrary, ignoring an existing threshold effect can deteriorate the forecast accuracy of the QRNN model. All the simulation results support the usefulness of the TQRNN model.