数理统计讲义第二章课后答案8

2.18设$(X,Y)=((X_1,Y_1),\cdots ,(X_n,Y_n))$为取自两维正态总体的简单随机样本。求总体均值${\mu }_X,{\mu }_Y$，总体方差${\sigma }_X,{\sigma }_Y$和总体相关系数${\rho }_{XY}$的最大似然估计(对于两维正态总体参数的最大似然估计，除了利用一般的求多维正态总体参数最大似然估计的方法外，也可将两维分布密度写为一维的边缘分布和条件分布的乘积，并利用密度的这一表示式求出各参数的最大似然估计)。

参考答案：

方法一：(多维正态总体参数)设$x_1,\cdots ,x_n$是对应$X_1,\cdots ,X_n$的样本值,$y_1,\cdots ,y_n$是对应$Y_1,\cdots ,Y_n$的样本值。

总体$Z\stackrel{d}{=}N(\mu ,\Sigma )$

其中$Z=\left(\begin{array}{c}X\\ Y\end{array}\right),\mu =\left(\begin{array}{c}{\mu }_X\\ {\mu }_Y\end{array}\right),$$\Sigma =\left(\begin{array}{cc}{\sigma }_X^2& {\rho }_{XY}{\sigma }_X{\sigma }_Y\\ {\rho }_{XY}{\sigma }_X{\sigma }_Y& {\sigma }_Y^2\end{array}\right)$

总体$Z$的密度函数为
p ( z ) = 1 ( 2 π ) 2 ∣ Σ ∣ exp ⁡ { − 1 2 ( z − μ ) ′ Σ − 1 ( z − μ ) } p(z)=\frac{1}{\sqrt{(2\pi)^2|\Sigma|}}\exp\{-\frac{1}{2}(z-\mu)'\Sigma^{-1}(z-\mu)\} p(z)=(2π)2∣Σ∣ ​1​exp{−21​(z−μ)′Σ−1(z−μ)}

似然函数为

L ( μ , Σ ) = ∏ i = 1 n p ( z i ) = 1 ( 2 π ) n ∣ Σ ∣ n 2 exp ⁡ { − 1 2 ∑ i = 1 n ( z i − μ ) ′ Σ − 1 ( z i − μ ) } \begin{array}{rl}&L(\mu,\Sigma)=\prod_{i=1}^{n}\limits p(z_i)\\&=\frac{1}{(2\pi)^n|\Sigma|^{\frac{n}{2}}}\exp\{-\frac{1}{2}\sum_{i=1}^{n}\limits(z_i-\mu)'\Sigma^{-1}(z_i-\mu)\}\end{array} ​L(μ,Σ)=i=1∏n​p(zi​)=(2π)n∣Σ∣2n​1​exp{−21​i=1∑n​(zi​−μ)′Σ−1(zi​−μ)}​

对数似然函数为

$\begin{array}{rl}l(\mu ,\Sigma )& =\ln L(\mu ,\Sigma )\\ & =-n\ln (2\pi )-\frac{n}{2}\ln |\Sigma |\\ & -\frac{1}{2}\sum _{i=1}^n(z_i-\mu {)}^{\prime }{\Sigma }^{-1}(z_i-\mu )\end{array}$

由$\{\begin{array}{rl}\frac{dl(\mu ,\Sigma )}{d\mu }& =\frac{1}{2}\sum _{i=1}^n{\Sigma }^{-1}(z_i-\mu )=0\\ \frac{dl(\mu ,\Sigma )}{d\Sigma }& =-\frac{n}{2}{\Sigma }^{-1}\\ & +\frac{1}{2}\sum _{i=1}^n(z_i-\mu )(z_i-\mu {)}^{\prime }({\Sigma }^{-1}{)}^2=0\end{array}$

得$\{\begin{array}{rl}\mu & =\frac{1}{n}\sum _{i=1}^n{z}_i=\bar{z}\\ \Sigma & =\frac{1}{n}\sum _{i=1}^n(z_i-\mu )(z_i-\mu {)}^{\prime }\\ & =\frac{1}{n}\sum _{i=1}^n(z_i-\bar{z})(z_i-\bar{z}{)}^{\prime }\end{array}$

故${\mu }_X,{\mu }_Y,{\sigma }_X,{\sigma }_Y,{\rho }_{XY}$的最大似然估计量为
$${\hat{\mu }}_X=\bar{X},{\hat{\mu }}_Y=\bar{Y},$$$${\hat{\sigma }}_X=\sqrt{\frac{1}{n}\sum _{i=1}^n(X_i-\bar{X}{)}^2},$$$${\hat{\sigma }}_Y=\sqrt{\frac{1}{n}\sum _{i=1}^n(Y_i-\bar{Y}{)}^2},$$$${\hat{\rho }}_{XY}=\frac{\sum _{i=1}^n(X_i-\bar{X})(Y_i-\bar{Y})}{\sqrt{\sum _{i=1}^n(X_i-\bar{X}{)}^2}\sqrt{\sum _{i=1}^n(Y_i-\bar{Y}{)}^2}}.$$

方法二：(将两维分布密度写为一维的边缘分布和条件分布的乘积)设$x_1,\cdots ,x_n$是对应$X_1,\cdots ,X_n$的样本值,$y_1,\cdots ,y_n$是对应$Y_1,\cdots ,Y_n$的样本值。

总体$Z\stackrel{d}{=}N(\mu ,\Sigma )$
其中$Z=\left(\begin{array}{c}X\\ Y\end{array}\right),\mu =\left(\begin{array}{c}{\mu }_X\\ {\mu }_Y\end{array}\right),$$\Sigma =\left(\begin{array}{cc}{\sigma }_X^2& {\rho }_{XY}{\sigma }_X{\sigma }_Y\\ {\rho }_{XY}{\sigma }_X{\sigma }_Y& {\sigma }_Y^2\end{array}\right)$

总体$Z$的密度函数为 p ( z ) = 1 ( 2 π ) 2 ∣ Σ ∣ exp ⁡ { − 1 2 ( z − μ ) ′ Σ − 1 ( z − μ ) } p(z)=\frac{1}{\sqrt{(2\pi)^2|\Sigma|}}\exp\{-\frac{1}{2}(z-\mu)'\Sigma^{-1}(z-\mu)\} p(z)=(2π)2∣Σ∣ ​1​exp{−21​(z−μ)′Σ−1(z−μ)}

$X$的边缘密度为 p ( x ) = ∫ p ( z ) d y = 1 2 π σ X exp ⁡ { − ( x − μ X ) 2 2 σ X 2 } p(x)=\int p(z) dy=\frac{1}{\sqrt{2\pi}\sigma_X}\exp\{-\frac{(x-\mu_X)^2}{2\sigma_X^2}\} p(x)=∫p(z)dy=2π ​σX​1​exp{−2σX2​(x−μX​)2​}

$Y$的边缘密度为 p ( y ) = ∫ p ( z ) d y = 1 2 π σ Y exp ⁡ { − ( y − μ Y ) 2 2 σ Y 2 } p(y)=\int p(z) dy=\frac{1}{\sqrt{2\pi}\sigma_Y}\exp\{-\frac{(y-\mu_Y)^2}{2\sigma_Y^2}\} p(y)=∫p(z)dy=2π ​σY​1​exp{−2σY2​(y−μY​)2​}

$X$在$Y=y$下的条件分布 p ( x ∣ Y = y ) = p ( z ) p ( y ) = 1 2 π 1 − ρ X Y 2 σ X exp ⁡ { − 1 2 ( 1 − ρ X Y 2 ) [ ( x − μ X ) 2 σ X 2 − 2 ρ X Y ( x − μ X ) ( y − μ Y ) σ X σ Y + ( y − μ Y ) 2 σ Y 2 ] + ( y − μ Y ) 2 2 σ Y 2 } \begin{array}{rl} p(x|Y=y)&=\frac{p(z)}{p(y)}\\&=\frac{1}{\sqrt{2\pi}\sqrt{1-\rho_{XY}^2}\sigma_X}\exp\{-\frac{1}{2(1-\rho_{XY}^2)}\\&[\frac{(x-\mu_X)^2}{\sigma_X^2}-\frac{2\rho_{XY}(x-\mu_X)(y-\mu_Y)}{\sigma_X\sigma_Y}\\&+\frac{(y-\mu_Y)^2}{\sigma_Y^2}]+\frac{(y-\mu_Y)^2}{2\sigma_Y^2}\} \end{array} p(x∣Y=y)​=p(y)p(z)​=2π ​1−ρXY2​ ​σX​1​exp{−2(1−ρXY2​)1​[σX2​(x−μX​)2​−σX​σY​2ρXY​(x−μX​)(y−μY​)​+σY2​(y−μY​)2​]+2σY2​(y−μY​)2​}​
似然函数为 L ( μ X , Σ X ) = ∏ i = 1 n p ( x i ) = 1 ( 2 π ) n 2 σ X n exp ⁡ { − ∑ i = 1 n ( x i − μ X ) 2 2 σ X 2 } L ( μ Y , Σ Y ) = ∏ i = 1 n p ( y i ) = 1 ( 2 π ) n 2 σ Y n exp ⁡ { − ∑ i = 1 n ( y i − μ Y ) 2 2 σ Y 2 } L ( ρ X Y ) = 1 ( 2 π ) n 2 ( 1 − ρ X Y 2 ) n 2 σ X n ⋅ exp ⁡ { − 1 2 ( 1 − ρ X Y 2 ) ∑ i = 1 n [ ( x i − μ X ) 2 σ X 2 − 2 ρ X Y ( x i − μ X ) ( y i − μ Y ) σ X σ Y + ( y i − μ Y ) 2 σ Y 2 ] + ∑ i = 1 n ( y i − μ Y ) 2 2 σ Y 2 } \begin{array}{rl} L(\mu_X,\Sigma_X)&=\prod_{i=1}^{n}\limits p(x_i)\\&=\frac{1}{(2\pi)^{\frac{n}{2}}\sigma_X^n}\exp\{-\frac{\sum_{i=1}^{n}\limits(x_i-\mu_X)^2}{2\sigma_X^2}\}\\ L(\mu_Y,\Sigma_Y)&=\prod_{i=1}^{n}\limits p(y_i)\\&=\frac{1}{(2\pi)^{\frac{n}{2}}\sigma_Y^n}\exp\{-\frac{\sum_{i=1}^{n}\limits(y_i-\mu_Y)^2}{2\sigma_Y^2}\}\\ L(\rho_{XY})&=\frac{1}{(2\pi)^{\frac{n}{2}}(1-\rho_{XY}^2)^{\frac{n}{2}}\sigma_X^n}\cdot\\&\exp\{-\frac{1}{2(1-\rho_{XY}^2)}\sum_{i=1}^{n}\limits[\frac{(x_i-\mu_X)^2}{\sigma_X^2}-\\&\frac{2\rho_{XY}(x_i-\mu_X)(y_i-\mu_Y)}{\sigma_X\sigma_Y}+\frac{(y_i-\mu_Y)^2}{\sigma_Y^2}]\\&+\sum_{i=1}^{n}\limits\frac{(y_i-\mu_Y)^2}{2\sigma_Y^2}\} \end{array} L(μX​,ΣX​)L(μY​,ΣY​)L(ρXY​)​=i=1∏n​p(xi​)=(2π)2n​σXn​1​exp{−2σX2​i=1∑n​(xi​−μX​)2​}=i=1∏n​p(yi​)=(2π)2n​σYn​1​exp{−2σY2​i=1∑n​(yi​−μY​)2​}=(2π)2n​(1−ρXY2​)2n​σXn​1​⋅exp{−2(1−ρXY2​)1​i=1∑n​[σX2​(xi​−μX​)2​−σX​σY​2ρXY​(xi​−μX​)(yi​−μY​)​+σY2​(yi​−μY​)2​]+i=1∑n​2σY2​(yi​−μY​)2​}​

对数似然函数为$\begin{array}{rl}l({\mu }_X,{\Sigma }_X)& =\ln L({\mu }_X,{\Sigma }_X)=-\frac{n}{2}\ln 2\pi \\ & -n\ln {\sigma }_X-\frac{\sum _{i=1}^n(x_i-{\mu }_X{)}^2}{2{\sigma }_X^2}\\ l({\mu }_Y,{\Sigma }_Y)& =\ln L({\mu }_Y,{\Sigma }_Y)=-\frac{n}{2}\ln 2\pi \\ & -n\ln {\sigma }_Y-\frac{\sum _{i=1}^n(y_i-{\mu }_Y{)}^2}{2{\sigma }_Y^2}\\ l({\rho }_{XY})& =\ln L({\rho }_{XY})=-\frac{n}{2}\ln 2\pi \\ & -\frac{n}{2}\ln (1-{\rho }_{XY}^2)-n\ln {\sigma }_Y\\ & -\frac{1}{2(1-{\rho }_{XY}^2)}\sum _{i=1}^n[\frac{(x_i-{\mu }_X{)}^2}{{\sigma }_X^2}\\ & -\frac{2{\rho }_{XY}(x_i-{\mu }_X)(y_i-{\mu }_Y)}{{\sigma }_X{\sigma }_Y}+\frac{(y_i-{\mu }_Y{)}^2}{{\sigma }_Y^2}]\\ & +\sum _{i=1}^n\frac{(y_i-{\mu }_Y{)}^2}{2{\sigma }_Y^2}\end{array}$
由

$\{\begin{array}{rl}\frac{dl({\mu }_X,{\Sigma }_X)}{d{\mu }_X}& =-\frac{2n{\mu }_X-2n\bar{X}}{2{\sigma }_X^2}=0\\ \frac{dl({\mu }_Y,{\Sigma }_Y)}{d{\mu }_Y}& =-\frac{2n{\mu }_Y-2n\bar{Y}}{2{\sigma }_Y^2}=0\\ \frac{dl({\mu }_X,{\Sigma }_X)}{d{\sigma }_X}& =-\frac{n}{{\sigma }_X}+\frac{\sum _{i=1}^n(x_i-{\mu }_X{)}^2}{{\sigma }_X^3}=0\\ \frac{dl({\mu }_Y,{\Sigma }_Y)}{d{\sigma }_Y}& =-\frac{n}{{\sigma }_Y}+\frac{\sum _{i=1}^n(y_i-{\mu }_Y{)}^2}{{\sigma }_Y^3}=0\\ \frac{dl({\rho }_{XY})}{d{\rho }_{XY}}& =0\end{array}$

得

$\{\begin{array}{rl}& {\mu }_X=\bar{x}\\ & {\mu }_Y=\bar{y}\\ & {\sigma }_X=\sqrt{\frac{1}{n}\sum _{i=1}^n(x_i-\bar{x}{)}^2}\\ & {\sigma }_Y=\sqrt{\frac{1}{n}\sum _{i=1}^n(y_i-\bar{y}{)}^2}\\ & {\rho }_{XY}=\frac{\sum _{i=1}^n(x_i-\bar{x})(y_i-\bar{y})}{\sqrt{\sum _{i=1}^n(x_i-\bar{x}{)}^2}\sqrt{\sum _{i=1}^n(y_i-\bar{y}{)}^2}}\end{array}$

故${\mu }_X,{\mu }_Y,{\sigma }_X,{\sigma }_Y,{\rho }_{XY}$的最大似然估计量为
$${\hat{\mu }}_X=\bar{X},{\hat{\mu }}_Y=\bar{Y},$$$${\hat{\sigma }}_X=\sqrt{\frac{1}{n}\sum _{i=1}^n(X_i-\bar{X}{)}^2},$$$${\hat{\sigma }}_Y=\sqrt{\frac{1}{n}\sum _{i=1}^n(Y_i-\bar{Y}{)}^2},$$$${\hat{\rho }}_{XY}=\frac{\sum _{i=1}^n(X_i-\bar{X})(Y_i-\bar{Y})}{\sqrt{\sum _{i=1}^n(X_i-\bar{X}{)}^2}\sqrt{\sum _{i=1}^n(Y_i-\bar{Y}{)}^2}}.$$

---

2.19设$X_1,\cdots ,X_n$为取自具有指数分布族 { p ( x ; λ ) = λ − 1 e − x / λ I ( 0 , ∞ ) ( x ) , λ > 0 } \{p(x;\lambda)=\lambda^{-1}e^{-x/\lambda}I_{(0,\infty)}(x),\lambda>0\} {p(x;λ)=λ−1e−x/λI(0,∞)​(x),λ>0}的简单随机样本。验证$\lambda$的无偏估计$\hat{\lambda }=\bar{X}$既是$\lambda$的矩法估计，又是$\lambda$的最大似然估计。说明在形为$c\bar{X}$的$\lambda$的估计量中，在均方误差准则下存在优于$\hat{\lambda }$的估计。

参考答案：

总体$X$的密度函数为
$$p(x;\lambda )={\lambda }^{-1}{e}^{-x/\lambda }{I}_{(0,\infty )}(x),\lambda >0.$$
总体矩：$${\mu }_1=EX={\int }_0^{+\infty }x{\lambda }^{-1}{e}^{-x/\lambda }dx=\lambda ,$$
样本矩：$$A_1=\frac{1}{n}\sum _{i=1}^n{X}_i=\bar{X}.$$
令$${\mu }_1=A_1,$$则$$\lambda =\bar{X}.$$
即$\lambda$的矩法估计量$$\hat{\lambda }=\bar{X}.$$

设$x_1,\cdots ,x_n$是对应$X_1,\cdots ,X_n$的样本值。

总体$X$的密度函数为

$$p(x;\lambda )={\lambda }^{-1}{e}^{-x/\lambda }{I}_{(0,\infty )}(x),\lambda >0.$$
似然函数为
$$L(\lambda )=\{\begin{array}{r}\frac{1}{{\lambda }^n}{e}^{-\frac{{\sum }_{i=1}^n{x}_i}{\lambda }},x_i>0\\ 0,x_i\le 0\end{array}(i=1,\cdots ,n)$$
对数似然函数为
$$l(\lambda )=\ln L(\lambda )=-n\ln \lambda -\frac{{\sum }_{i=1}^n{x}_i}{\lambda },$$
$x_i>0(i=1,\cdots ,n)$

令

$$\frac{dl(\lambda )}{d\lambda }=-\frac{n}{\lambda }+\frac{{\sum }_{i=1}^n{x}_i}{{\lambda }^2}=0,$$
得
$$\lambda =\frac{{\sum }_{i=1}^n{x}_i}{n}=\bar{x}.$$
则$\lambda$最大似然估计量为$${\hat{\lambda }}_{MLE}=\bar{X}.$$

设${\hat{\lambda }}_1=c\bar{X},$有
$$\begin{array}{rl}& E({\hat{\lambda }}_1)=E(c\bar{X})=cE(\bar{X})=c\lambda ,\\ & Var({\hat{\lambda }}_1)=Var(c\bar{X})=c^{2}Var(\bar{X})=\frac{c^2{\lambda }^2}{n}\end{array}$$

则
$$\begin{array}{rl}MSE(\bar{X})& =Var(\bar{X})+(E(\bar{X})-\lambda {)}^2\\ & =\frac{{\lambda }^2}{n},\\ MSE({\hat{\lambda }}_1)& =Var({\hat{\lambda }}_1)+(E({\hat{\lambda }}_1)-\lambda {)}^2\\ & =\frac{c^2{\lambda }^2}{n}+(c\lambda -\lambda {)}^2\\ & =[\frac{n+1}{n}(c-\frac{n}{n+1}{)}^2+\frac{1}{n+1}]{\lambda }^2.\end{array}$$
当$c=\frac{n}{n+1}$时，${\hat{\lambda }}_1=\frac{n}{n+1}\bar{X}$的均方误差$MSE({\hat{\lambda }}_1)=\frac{1}{n+1}{\lambda }^2$小于$\bar{X}$的均方误差$MSE(\bar{X})=\frac{{\lambda }^2}{n},$

故在形为$c\bar{X}$的$\lambda$的估计量中，在均方误差准则下存在优于$\hat{\lambda }$的估计。

---

如果您在看完以后，无论是题目还是参考解答过程有什么问题，希望不吝指出！如果有更好的解题思路或者好的问题分享，欢迎通过后台发送给我们！