Crises Learning Under Diagnosticity
Name
zhu-jiulei-mfin-sloan-2023-file.pdf
Description
Thesis PDF
Size
5.37 MB
Format
Adobe PDF
Checksum (MD5)
c98e178c063ae86a5bbbd54cf19dec22
Author(s)
Zhu, Jiulei
Advisor(s)
Parker, Jonathan A.
Date Issued
February 2023
Publisher
Massachusetts Institute of Technology
Abstract
We analyse ways in which diagnostic expectation in macro forecasting contaminates posterior inference and interferes with Bayesian parameter learning. Conversely, we study scenarios where parameter uncertainty dampens or magnifies extrapolation. We characterise two unique implications of such an interaction with supporting evidence from the SPF: 1) State-dependence of extrapolation even after controlling for shocks - more aggressive extrapolation when surprises are mean-reverting. 2) Asymmetric extrapolation to positive versus negative surprises - reorienting the axis to control for state effects yields a unified bias pattern across macro indicators. Additionally, oblivious agents who extrapolate are found to learn parameters more slowly and consistently underestimate the persistence of the underlying process. We question the crude use of the predictability of forecast errors as quantifiers of departure from rationality and offer an alternative approach which treats extrapolative tendencies as state-dependent and which distinguishes between two sources of error: biases and parameter confusion.
MIT Department
Sloan School of Management
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
