Machine Learning Based Policy to Ease Information Asymmetry in Non-Point Pollution Management
Date
2017-03
Journal Title
Journal ISSN
Volume Title
Publisher
Department of Applied Economics and Statistics, University of Delaware, Newark, DE.
Abstract
This research examines how an artificial neural network incorporating high-frequency monitoring data and natural system dynamics can inform policies that regulate an environmental externality with inherent information asymmetry. Using an experiment with both students and agricultural producers we study strategic behavior under various policies and measure participants’ relative values for different levels of information accuracy under such policies. First, we show that a neural-network-based recursive filter can be applied to monitoring data to estimate an individual polluter’s contribution to the ambient level of pollution, in essence, turning nonpoint sources into estimated point sources. We then test the implications of this result using an economic experiment
that explores the effects of spatial relationships and the information structure of policies on behavior and preferences. The results of the experiments show that participants change their emissions in response to both policy and information treatments and that there are no significant differences in behavior between professional and student participants. However, we find that the agricultural producers are more willing than student participants to pay for policies that more accurately target the individual sources of pollution. This latter result suggests a strong preference for polluter-pay policies instead of ambient-based policies amongst producers, even if they do not necessarily lead to higher total profits.
Description
Keywords
Nonpoint source pollution, experimental economics, neural network