零膨胀损失次数回归模型及其应用

在许多情况下,损失次数数据一方面具有零膨胀特点,另一方面尾部较长,存在过离散特征。此时,通常的损失次数模型（如泊松回归）将不能很好地拟合实际损失数据。本文对零膨胀泊松回归、零膨胀负二项回归、零膨胀广义泊松回归和零膨胀泊松—逆高斯回归的模型构造和参数估计进行讨论,并用这些模型对一组实际损失数据进行了拟合。结果表明,当实际数据存在零膨胀特点时,零膨胀回归模型可以显著改善对实际损失数据的拟合效果。

Zero-inflated Loss Count Models and Their Applications

Under many circumstances,loss count data appears to be both zero-inflated and long-tailed,and this will make the usual models such as Poisson regression models unsuitable to fit the data properly.The paper discusses some important zero-inflated models,including zero-inflated Poisson regression,zero-inflated Negative Binomial regression,zero-inflated Generalized Poisson regression,and zero-inflated Poisson Inverse Gaussian regression.The paper also applies these models to a practical data set.The result show...