B0684
Title: High conditional quantiles for panel data
Authors: Xuan Leng - Xiamen University (China) [presenting]
Abstract: Panel quantile regression models play an essential role in real finance, econometrics, insurance, and risk management applications. However, direct estimates of the extreme conditional quantiles may lead to unstable results due to data sparsity on the far tail. Moreover, the presence of individual effects in panel quantile regressions complicates the inference for high quantiles. A two-stage method is proposed to estimate/predict the high conditional quantiles. The intermediate quantiles are first predicted according to panel quantile regressions, and the extreme quantiles are obtained by extrapolating the intermediate ones in the second stage. The asymptotic properties of the prediction method rely on a set of second-order conditions for heteroscedastic extremes. We use a metric called Average Absolute Deviation Error to evaluate the prediction performance of high conditional quantiles over different cross-sections. The asymptotic distributions of the metric for both intermediate and extreme quantiles are studied. We demonstrate the finite sample performance of the two-stage prediction, which is compared to the direct prediction for extreme conditional quantiles. Finally, we apply the two-stage method to the macroeconomic and housing price data and find strong evidence of housing bubbles and common economic factors.
