零膨胀损失次数的贝叶斯分位回归模型

A Bayesian Quantile Regression Model for Zero-Inflated Loss Count

研究目标:建立零膨胀损失次数的贝叶斯分位回归模型。研究方法:通过增加随机扰动将离散型的损失次数数据转化为连续型数据,在预测误差平方和最小的条件下,求解出分位数水平,并应用贝叶斯方法求解分位回归模型中的参数。研究发现:基于得到的分位回归模型及相应的分位数水平,实现对未来的损失频率的预测。研究创新:借助等式关系,求解分位回归的分位数水平,避免主观选择分位数水平的弊端,实现对零膨胀损失次数贝叶斯分位回归建模。研究价值:基于一组实际数据的实证分析结果表明,该模型可以显著改进现有模型的拟合效果。

Research Objectives： We propose a new method, a Bayesian quantile regression model for zero-inflated loss count. Research Methods： This paper adds a uniformly distributed noise variable to the count variable and converts it to a continuous variable, gets quantile level by minimizing squared prediction error, and estimates the parameters by using the Bayesian method. Research Findings： Achieve the predicted frequency of future losses, basing on quantile regression Model with obtained parameters and the corresponding quantile level. Research Innovations： With the help of the equation, we can solve the quantile level of the quantile regression and avoid the malpractice of subjective in selecting the quantile level. The Bayesian quantification regression model of the zero expansive loss is achieved. Research Value： The empirical analysis based on a set of actual data shows that the new method can significantly improve the prediction accuracy of the existing models.