Exact Distributions of Bayesian Cox-Snell Bounds in Auditing
推导了审计中Cox-Snell贝叶斯模型下相对高估误差上限的精确概率分布，帮助审计师利用先验知识评估误差风险。
Consider a population of recorded values in which errors may occur, but only overstatement errors. An auditor wants to find an upper bound for the relative overstatement error A, that is, the total of the errors expressed as a fraction of the total of the recorded values. From past experience, the auditor usually has fairly good insight into the characteristics of the population at hand. Therefore, it seems natural to make use of this prior knowledge and to consider the problem from a Bayesian point of view. Up to now the most flexible and mathematically convenient class of Bayesian models was proposed by Cox and Snell [1979]. The Bayesian approach combines the prior knowledge with the observations to find an upper bound for A. Before the observations have been actually obtained, this upper bound is a random variable, say L. In the Cox-Snell model, L can be expressed by means of a simple formula. For theoretical populations, this paper shows how the exact probability distribution of L can be derived, following the lines suggested by Moors [1983]. Of course, expectation and variance of L follow at once, as well as the probability that L exceeds A1. The latter is a most important