劉榮玄
在NA樣本下一類雙邊截斷型分布族參數(shù)的經驗Bayes估計
劉榮玄
(井岡山大學數(shù)理學院,江西,吉安 343009 )
在平方損失下,討論一類雙邊截斷型均勻分布族參數(shù)的經驗貝葉斯(EB)估計的漸近性。按照貝葉斯(Bayes)方法,導出均勻分布族參數(shù)的Bayes估計,利用歷史樣本,采用概率密度函數(shù)的核估計方法,構造出邊緣密度函數(shù)的估計,從而得到參數(shù)的EB估計,在一定的條件下,證明所得到的EB估計是漸近最優(yōu)的,而且得到了其收斂速度,最后舉例說明滿足定理條件的參數(shù)的先驗分布是存在的。
NA樣本;均勻分布;平方損失;Bayes估計;漸近性
證明 見文獻[5]引理1。
引理2 在均方損失下,有
證明 見文獻[6]的引理2.1。
證明 見文獻 [7]引理3.1。
綜合可知引理結論成立。
, (2.7)
由引理1可得
于是有
由C-R不等式和Jensen不等式以及(2.6)、(2.10)式得
則有
, (2.12)
綜合式(2.12)和(2.13)可知,(2.11)成立。
又由凸函數(shù)的Jensen不等式有
證明 由引理2有
其中
由引理3、引理4和引理5有
由引理6得到
(2.16)
(2.17)
下面的例子說明存在滿足定理條件的先驗分布。
到此可知定理中的條件全都成立,則定理的結論成立。
[1] 韋來生.刻度指數(shù)族參數(shù)的經驗貝葉斯檢驗問題:NA樣本情形[J]. 應用數(shù)學學報,2002(3):198-201.
[2] 許勇,師義民.NA樣本情形下單邊截斷型分布族位置參數(shù)的經驗Bayes估計[J]. 應用數(shù)學,2001,14(4):98-102.
[3] 康會光,趙小山.Linex損失函數(shù)下單邊截斷型分布族參數(shù)的EB估計[J]. 應用數(shù)學,2001,14(3):82-86.
[4] 劉榮玄,曹艷華.雙指數(shù)分布族參數(shù)EB估計的最優(yōu)性[J].華中師范大學學報:自然科學版,2011,45(4):542-546.
[5] Wei L S.Convergence rates of empirical Bayes estimation for arameter ofone-sided truncated distribution[J].Annals of Math,1985,6A(2):193-202.
[6] Singh R S.Empirical Bayes estimation in Lebesgue- exponential families With rates near the best possible bate[J].Ann Statist,1979,7(4):890-902.
[7] 趙林城.一類離散分布參數(shù)的經驗貝葉斯估計的收斂速度[J].中國數(shù)學研究與評論,1981(1):59-69.
[8] 劉榮玄.指數(shù)族刻度參數(shù)EB估計的漸近最優(yōu)性[J].數(shù)理統(tǒng)計與管理,2010,29(6):1018-1025.
[9] 陳希孺.高等數(shù)理統(tǒng)計[M].合肥:中國科學技術大學出版社,1999:104-121.
[10] 劉榮玄,黃 璇.幾何模型中參數(shù)的經驗貝葉斯估計的漸近性[J].井岡山大學學報,2011,32(6):12-15.
A CLASS OF PROBLEMS OF THE EMPIRICAL BAYES ESTIMATION FOR THE PARAMETER OF THE TWO-SIDE TRUNCATED DISTRIBUTION FAMILIES
LIU Rong-xuan
(School of Mathematics and Physics, Jinggangshan University,Ji’an,Jiangxi 343009,China)
Under the condition of squared loss, we discussed theasymptotic behavior for empirical Bayes (EB) estimation of the parameters of a class of two-side truncated uniform distribution families. We derived the Bayes estimation of the parameters of the family of the Uniform distribution based on theBayesianmethods and constructed the estimate of the marginal density function by historical samples and the method of Kernel estimation of probability density function. Then we get the EB estimation of the parameters. We also proved that the EB estimation was asymptotically optimal under certain conditions. Furthermore, we get the convergence rate. Finally, we get the existences of the prior distribution of the parameters satisfying the condition of the Theorem by example.
NA samples; uniform distribution; squared loss; EB estimation; asymptotic
O212.8
A
10.3969/j.issn.1674-8085.2012.04.002
1674-8085(2012)04-0006-05
2012-03-15;
2012-04-28
劉榮玄(1959-),男,江西遂川人,副教授,主要從事概率論與數(shù)理統(tǒng)計教學和研究 (E-mail: lrx1716@126.com).