Adaptive Learning in Continuous-Time: Techniques and Theory
Loading...
Date
2021-09-13
Authors
Lester, Chandler
Journal Title
Journal ISSN
Volume Title
Publisher
University of Oregon
Abstract
How we model individual’s expectations and predictions in economic models playsan essential role in economic outcomes. We can assume that individuals are well
informed and developed nuanced views on the economy, meaning they understand and
have detailed knowledge of economic parameters and economic models, or we can
suppose individuals are observant and develop perceptions of the economy and make
decisions based on available data.
One method of including this level of realistic behavior in economic models isadaptive learning. In adaptive learning models, agents use simple forecasting rules to
make predictions about future values of economic variables or the state of the economy.
The work presented in this dissertation builds a framework for examining these dynamics
in a high-frequency setting. It is important to extend these behavioral modeling
techniques to this setting because increasing data are available at higher frequencies. This
work combines existing continuous-time modeling techniques with emerging research
from economics to develop modelings in which an agent can respond to high-frequency
information.
This dissertation demonstrates that complex high-frequency learning is possibleand has potential benefits and improvements over discrete-time counterparts. The
dominant theme of this work is defining and mathematically developing a framework for
examining bounded rationality in continuous-time models. In chapter two, basic
exogenous adaptive rules are explored in a simple Ramsey Model setting. Chapter three
introduces shadow-price learning and more complicated endogenous learning rules,
including a derivation of continuous-time recursive least squares and the definition of a
continuous-time mapping between an agent’s perceptions and actuality. Chapter four
builds on the dynamics defined in chapter three by applying them to a linearized Real
Business cycle model. We find that the continuous-time learning dynamics offer some
improvements to the volatility of predictions.
Description
Keywords
continuous, learning, macroeconomics