Optimal model averaging for divergent-dimensional Poisson regressions
提出一种新的模型平均方法，处理协变量维数随样本量增长的泊松回归模型不确定性，通过无偏估计KL散度选择权重，在模型误设或存在正确模型时均具有优良性质，并用于预测企业专利数量。
This paper proposes a new model averaging method to address model uncertainty in Poisson regressions, allowing the dimension of covariates to increase with the sample size. We derive an unbiased estimator of the Kullback–Leibler (KL) divergence to choose averaging weights. We show that when all candidate models are misspecified, the proposed estimate is asymptotically optimal by achieving the least KL divergence among all possible averaging estimators. In another situation where correct models exist in the model space, our method can produce consistent coefficient estimates. We apply the proposed techniques to study the determinants and predict corporate innovation outcomes measured by the number of patents.