极值分布下联合位置与散度模型的变量选择

Variable Selection in Joint Location and Dispersion Models of the Extreme-value Distribution

极值分布在地震、洪灾和其它自然灾害的预测中是非常有用的.在许多应用方面,很有必要对散度建模.本文推广经典极值回归模型,研究了联合位置与散度模型,并提出了一种同时对位置模型和散度模型的变量选择方法.同时证明了惩罚极大似然估计具有相合性和oracle性质,通过随机模拟研究了所提出方法的有限样本性质.

The extreme-value distribution is very useful in predicting the probability that an extreme earthquake, flood or other natural disaster will occur. In many applications, there is a great need to model the dispersion. In this paper, a unified procedure is proposed to simultaneously select signifi- cant variables in joint location and dispersion models which provide a useful extension of the general extreme-value regression model. It is further shown that the presented penalized maximum likelihood estimator enjoys the consistency and the oracle property. Numerical simulation is conducted to exam- ine the finite sample properties of the proposed method.