Incomplete learning from endogenous data in dynamic allocation.

The writers examine the problem of incomplete learning from endogenous data in dynamic allocation. They note that this problem is commonly referred to as the “discounted multi-armed bandit problem” and that the optimal solution has been shown to be the “index rule” that selects at each stage the ac...

Full description

Bibliographic Details
Published in:Econometrica Vol. 68; no. 6; pp. 1511 - 1517
Main Authors: Brezzi, Monica, Lai, Tze Leung
Format: Article
Published: Wiley-Blackwell November 2000
Subjects:
Online Access:View this record in EBSCOhost
Description
Summary:The writers examine the problem of incomplete learning from endogenous data in dynamic allocation. They note that this problem is commonly referred to as the “discounted multi-armed bandit problem” and that the optimal solution has been shown to be the “index rule” that selects at each stage the action with the largest “dynamic allocation index.” They provide a simple proof of the incompleteness of optimal learning from endogenous data in the discounted multi-armed bandit problem. They offer a comprehensive theory about the limits of beliefs and actions that solves completely, in the context of discounted multi-armed bandits, the basic problem regarding the extent of experimentation and the long-run beliefs and actions of optimizing agents who learn by doing.