No perfect general formula. You need to know the business very well before you can make an informed selection.

For example, if you know that your company changed case reserve rule, you must adjust accordingly.

For example, if you know that your company take actions on settling on large claims, you must adjust accordingly.

Again, you need to know the business, either internal change or external change (such as reinsurance, regulation, etc.).

顶起来，大家也来发表一下意见，P&C的准备金评估真的很需要经验交流啊

这个问题我曾经想过。目前行业许多公司在KPI考核中，有保单终极赔付率的指标。这样，可以通过大量数据的GLM计算本期风险暴露（可以简单理解为已赚保费）所有保单的终极赔款，然后减掉当期的已决和已报案未决。
此种方法的缺点：1、需要改变事故年的评估方式，改为保单年。2、目前精算在评估准备金中，多多少少会有一定的保守因素，而新方法评估相对客观，有INBR不足的风险。
其他可以继续聊一下，以上仅供参考。

GLM method is better for predictive modeling work but not for reserving, which the purpose is to reflect historical performance. Also, GLM is not good at dealing with interactions. And, you need to develop loss to ultimate first in order to obtain expected loss cost. Sounds like the question of which one comes first, "chicken or eggs".

You can consider other reserving methods, such as Munich Chain ladder, which would generate LDFs for each accident year based on paid vs. incurred loss triangles.

Anyway, you still need to make adjustments in line with business changes rather than using formulae directly.

I want to know if there are some book and material about the estimating unpaid claim,except the book offerred by CAS,because I have read the book ,and I can't find any idea for choosing method