Continuous-Time Dynamic Decision Making with Costly Information

TL;DR

This paper develops a continuous-time control framework for decision making under partial observations with costly information, deriving optimal strategies and analyzing how uncertainty influences information acquisition.

Contribution

It introduces a novel reduction of the problem to a deterministic control for the conditional variance, providing semi-explicit solutions and qualitative insights.

Findings

Optimal cost increases with model uncertainty.

A critical threshold determines when to acquire information.

Asymptotic behavior of acquisition rate is characterized.

Abstract

We consider a continuous-time linear-quadratic Gaussian control problem with partial observations and costly information acquisition. More precisely, we assume the drift of the state process to be governed by an unobservable Ornstein--Uhlenbeck process. The decision maker can additionally acquire information on the hidden state by conducting costly tests, thereby augmenting the available information. Combining the Kalman--Bucy filter with a dynamic programming approach, we show that the problem can be reduced to a deterministic control problem for the conditional variance of the unobservable state. Optimal controls and value functions are derived in a semi-explicit form, and we present an extensive study of the qualitative properties of the model. We demonstrate that both the optimal cost and the marginal cost increase with model uncertainty. We identify a critical threshold: below this level, it is optimal not to acquire additional information, whereas above this threshold, continuous information acquisition is optimal, with the rate increasing as uncertainty grows. For quadratic information costs, we derive the precise asymptotic behavior of the acquisition rate as uncertainty approaches zero and infinity.