Working Paper: CEPR ID: DP16364
Authors: Kfir Eliaz; Daniel Fershtman; Alexander Frug
Abstract: We consider a decision-maker sequentially choosing which task to attend to when payoffs depend on both the chosen and unchosen tasks. We show that when tasks are substitutes (complements) such that the flow payoffs are a sum (product) of the tasks’ outputs, the optimal policy is an index policy, generalizing the independence of irrelevant alternatives (IIA) property known in the classic multi-armed bandit problem. We illustrate the usefulness of our model through several applications, including repeated bargaining, dynamic occupational choice, on-the-job training, R\&D, and dynamic supervision.
Keywords: allocation of attention; scheduling; time management; multitasking; task juggling; multiarmed bandits
JEL Codes: D81; D83
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| tasks are substitutes (C78) | optimal policy characterized by an index policy (C61) |
| optimal policy characterized by an index policy (C61) | DM's flow payoff from chosen and unchosen tasks (D79) |
| tasks are complementary (Y80) | optimal policy requires distinguishing between augmenting and non-augmenting states (C61) |
| optimal policy requires distinguishing between augmenting and non-augmenting states (C61) | DM's flow payoff is a product of the outputs from all tasks (C69) |
| task choice (C78) | overall payoffs (J33) |
| attention allocation (D91) | future productivity (O49) |
| task substitutability and complementarity (D10) | decision-making strategies (D87) |