Manel Baucells
University of Virginia
Room: M3 3127
An Optimal Index Policy for Search under Gaussian Learning
We tackle the longstanding open problem of optimal stopping in sequential search under unknown Gaussian parameters and a conjugate prior. We consider the two focal cases of recall and no recall. The reservation price property, which holds if the variance is known, no longer holds when variance must be learned. Under no recall, the reservation property holds past a critical sample size. In the early stage of search, however, it could be advantageous for a patient decision maker to reject an arbitrarily high offer. For recall, such behavior is optimal throughout the sampling process. Finally, we obtain a universal index policy based on comparing the standardized value of the current offer with an index, and show how to efficiently compute this index.