For a family of uniform distribution on the interval [8 - (1/2), e - (1/2)]' the informationinequality for the bayes risk of any estimator of e is given under the quadratic Joss and the uniform priordistribution on an interval [-c,c]. The lower bound for the Bayes risk is shown to be sharp. And also thelower bound for the limit inferior of Bayes risk as c -jo 00 is seen to be attained by the mid-range estimator.
Let X be an observable random vector and Y a random variable to be observedin future. Assume that the joint distribution of X and Y depends on an unknownparameter. In this paper we consider a way of the construction of a predictioninterval for Y based on X for a discrete exponential fq.mily of distributions. Inparticular we asymptotically construct the prediction interval in the binomial andPoisson cases, and give practical applications to the prediction of the number ofwins of the Japanese professional baseball teams and that of home runs of the playersin the major league of the United States.