metrics

核心思想:如果我们知道 的渐近分布,想求 的渐近分布,Delta Method 就是对 做一阶 Taylor 展开。

假设我们已经有:

But the key assumption here is that we need the sample is big enough, thus we could use LLN to approach.

想知道 的渐近分布。在 处 Taylor 展开:

其中 是 Jacobian 矩阵。所以:

渐近方差就是 ,实际估计时用 代入即可。