数值示例chap5分布模型

5、分布模型

5.1累计索赔的对数正态模型

假设个体进展因子$F_{i,j}=C_{i,j}/C_{i,j-1}$服从对数正态分布,则
$${\eta }_{i,j}=\log (F_{i,j})\sim N({\xi }_j,{\sigma }_j^2)$$
$$E[F_{i,j}]=\exp \left\{{\xi }_i+\frac{1}{2}{\sigma }_j^2\right\}$$
V a r ( F i , j ) = exp ⁡ { 2 ξ j + σ j 2 } [ exp ⁡ ( σ j 2 ) − 1 ] Var(F_{i,j})=\exp\{2\xi_{j}+\sigma_{j}^2\}\left[\exp\left(\sigma_{j}^2\right)-1\right] Var(Fi,j​)=exp{2ξj​+σj2​}[exp(σj2​)−1]
链梯模型假设下
$$E[C_{i,j}\mid C_{i,j-1}]=f_{j-1}{C}_{i,j-1}$$
故CL进展因子为
$$f_{j-1}=\exp \left\{{\xi }_i+\frac{1}{2}{\sigma }_j^2\right\}$$
参数估计

$${\hat{\xi }}_j=\frac{1}{I-j+1}\sum _{i=0}^{I-j}\log \left(\frac{C_{i,j}}{C_{i,j-1}}\right)$$
$$\widehat{{\sigma }_j^2}=\frac{1}{I-j}\sum _{i=0}^{I-j}{\left(\log \left(\frac{C_{i,j}}{C_{i,j-1}}\right)-{\hat{\xi }}_j\right)}^2$$

5.2方差参数估计与准备金估计

5.2.1方差已知

无偏估计,${\hat{\sigma }}_j^2$已知
定义$$Z_{i,j}=\log (C_{i,j})$$
则
$$Z_{i,J}=Z_{i,I-i}+\sum _{j=I_I+1}^J\log \left(\frac{C_{i,j}}{C_{i,j-1}}\right)=Z_{i,I-i}+\sum _{j=I-I+1}^J{\eta }_{i,j}$$
因此
$${\hat{Z}}_{i,J}=\hat{E}[Z_{i,J}|D_I]=Z_{i,I-i}+\sum _{j=I-i+1}^J{\hat{\xi }}_j$$

$${\hat{C}}_{i,J}^{LN}=\hat{E}[C_{i,J}|D_I]=\exp \left\{{\hat{Z}}_{i,J}+\frac{1}{2}\sum _{j=I-i+1}^J{\sigma }_j^2\left(1-\frac{1}{I-j+1}\right)\right\}$$
单个事故年的MSEP
$$\hat{m}se{p}_{C_{i,J|C_{i,I-i}}}\left({\hat{c}}_{i,J}^{LN}\right)=E[C_{i,J}\mid C_{i,I-i}{]}^2\times \left(exp\left\{\sum _{j=I-i+1}^J{\sigma }_j^2\right\}-1+exp\left\{\sum _{j=I-i+1}^J\frac{{\sigma }_j^2}{I-j+1}\right\}-1\right)$$
过程方差
$$\exp \left\{\sum _{j=I-i+1}^J{\sigma }_j^2\right\}-1$$
参数估计偏差
$$\left\{\sum _{j=I-i+1}^J\frac{{\sigma }_j^2}{I-j+1}\right\}-1$$
MSEP事故年的聚合

$$\hat{m}se{p}_{\sum i{C}_{i,j}|.}\left(\sum _{i=1}^I{\hat{C}}_{i,J}^{CL}\right)=\sum _{i=1}^{I}mse{p}_{C_{i,J}\mid C_{i,I-i}}\left({\hat{C}}_{i,J}^{LN}\right)+2\sum _{1\le i<k\le I}E[C_{i,J}\mid C_{i,I-i}]\times E[C_{k,J}\mid C_{k,I-k}]\left(exp\left\{\sum _{j=I-i+1}^J\frac{{\sigma }_j^2}{I-j+1}\right\}-1\right)$$

中心预测量,${\hat{\sigma }}_j^2$已知

$${\hat{C}}_{i,J}^{cp}=\hat{E}[C_{i,J}\mid D_I]=\exp \left\{{\hat{Z}}_{i,J}+\frac{1}{2}\sum _{j=I-i+1}^J{\sigma }_j^2\left(1+\frac{1}{I-j+1}\right)\right\}$$
$$={\hat{C}}_{i,J}^{LN}\exp \left(\sum _{j=I-i+1}^J\frac{{\hat{\sigma }}_j^2}{I-j+1}\right)>{\hat{C}}_{i,J}^{LN}$$
MSEP可以由下式估计
$$\hat{m}se{p}_{C_{i,J|C_{i,I-i}}}\left({\hat{c}}_{i,J}^{cp}\right)=E[C_{i,J}\mid C_{i,I-i}{]}^2\times \left(\exp \left\{\sum _{j=I-i+1}^J\frac{3{\sigma }_j^2}{I-i+1}\right\}+\exp \left\{\sum _{j=I-i+1}^J{\sigma }_j^2\right\}-2\exp \left\{\sum _{j=I-i+1}^J\frac{{\sigma }_j^2}{I-j+1}\right\}\right)$$

5.2.2方差未知

$$t_j({\sigma }_j^2)=\frac{I-j}{{\sigma }_j^2}\left(1-\exp \left\{\frac{-{\sigma }_j^2}{I-j+1}\right\}\right)$$
$${\hat{C}}_{i,J}^{LN\sigma ,1}=\hat{E}[C_{i,J}\mid D_I]=\exp \left\{{\hat{Z}}_{i,J}+\frac{1}{2}\sum _{j=I-j+1}^J\right\}{t}_j({\sigma }_j^2){\hat{\sigma }}_j^2$$