Title: Machine learning solution of dynamic models with rational inattention
Authors: Jozef Barunik - UTIA AV CR vvi (Czech Republic)
Lukas Vacha - Institute of Information Theory and Automation of the CAS (Czech Republic) [presenting]
Abstract: An asset pricing model is proposed under uncertainty with agents having quantile preferences and limited information processing capacity. In contrast to the standard asset pricing that relies on expected utility, we introduce a dynamic quantile model for asset pricing, in which the agent maximizes stream of future quantile utilities. The agent has a limited amount of attention since the information she obtains is costly. In our model, the agent maximizes stream of her future quantile utilities according to her quantile utility preferences subject to information costs constraints. We solve this high-dimensional problem using machine learning tools leading to a sequence of supervised learning problems. This approach makes this task computationally feasible. Results indicate that there is a significant benefit when a standard expected utility is expanded into quantile preference utilities.