Optimal sequential tests for multiple hypotheses when sampling from a Bernoulli population

In this paper we deal with the problem of sequential testing of multiple hypotheses. We are interested in minimising a weighted average sample number under restrictions on the error probabilities. A computer-oriented method of construction of optimal sequential tests is proposed. For the particular case of sampling from a Bernoulli population we develop a whole set of computer algorithms for optimal design and performance evaluation of sequential tests and implement them in the form of computer code written in R programming language. The tests we obtain are exact (neither asymptotic nor approximate). Extensions to other distribution families are discussed. A numerical comparison with other known tests (of MSPRT type) is carried out.