Dissertation Defenses

Doctoral Dissertation Announcement

---

Candidate: Sagarika Mishra

Degree of: Doctor of Philosophy

Department: Economics

Title: Essays on Asymmetric Loss and Learning in Macroeconomic Forecasting

Committee:
Dr. Matthew L. Higgins, Chair
Dr. C. James Hueng
Dr. J. Kevin Corder

Date: Thursday, July 10, 2008 1:00 p.m. – 3:00 p.m.
5302 Friedman Hall

Abstract:
This dissertation consists of three essays that answer the questions whether agents have asymmetric loss, why agents have asymmetric loss, and whether agents engage in least squares learning.
In the first essay, I test the rationality of inflation forecasts from the Livingston survey using the Mincer-Zarnowitz (MZ) regression when agents have asymmetric loss. I show that the MZ regression is inappropriate when agents have asymmetric loss. I demonstrate how the MZ regression can be suitably modified to test forecast rationality when agents have asymmetric loss. When I augment the MZ regression with higher order moments of the forecasts, the rationality of the inflation forecasts can not be rejected for linex and linlin loss.
In the second essay, I explain why agents have asymmetric loss using GDP growth rate forecasts from the SPF. Under asymmetric loss the bias can be explained by a time-varying asymmetry parameter or by time-varying higher order moments. However, in the absence of time-varying second order moments the bias can only be explained by the time-varying asymmetry parameter. Using linex and linlin loss, I estimate the time-varying asymmetry parameter and the bias by maximum likelihood estimation. I find that the factors which agents knowingly use to bias GDP growth rate forecasts are the lag growth rate of GDP, the duration of business cycle in the presence of recession, a Republican government in the presence of recession, and uncertainty in the presence of recession.
In the third essay, I test whether agents learn monetary policy by least squares when there are shifts in monetary policy, using the three month T-bill forecasts from the SPF. I derive the conditional mean, variance and covariance of the forecast errors when agents learn by least squares in the presence of structural shifts. I identify the structural break dates in the policy rule using the Bai and Perron (1998, 2001, and 2003) test. Using those dates, I estimate the mean and variance of the forecast error within each regime. When I correct the bias from the survey forecast error using the estimated mean, I find survey forecasts are consistent with least squares learning.