Please use this identifier to cite or link to this item: https://hdl.handle.net/10419/208887 
Authors: 
Year of Publication: 
2014
Series/Report no.: 
PhD Series No. 9.2014
Publisher: 
Copenhagen Business School (CBS), Frederiksberg
Abstract: 
This thesis consists of three essays of which two are about return predictability while the last essay covers term structure models. Return predictability is still a heavily debated issue among nancial economists as well as practitioners in the nancial industry. The ability to predict stock returns out-of-sample, that is, by relying on information available at time t, is still controversial. In a recent paper, Goyal and Welch (2008) comprehensively reexamine the performance of 14 predictor variables that have been suggested by the academic literature to be powerful predictors of the U.S. equity premium, that is, the S&P 500 index return minus the short-term interest rate. The authors conclude that none of these predictor variables led to robust predictions across di erent forecast horizons and sample periods which consistently beat benchmark models such as the historical mean. In a response to Goyal and Welch (2008) Campbell and Thompson (2008) nd evidence of out-of-sample predictability by putting some economically meaningful restrictions on the coe cients of the predictive regressions. However, the out-of-sample explanatory power is nil, but nonetheless it is economically signi cant for investors with mean-variance preferences. The predictability literature argues that the out-of-sample predictability deteriorates due to structural breaks such as macroeconomic instability, changes in monetary policy, new regulations etc. Thus, not only the predictor model changes over time, but also its coef- cients. Goyal and Welch (2008) explain that more sophisticated models accounting for structural breaks might be able to consistently beat historical mean predictions. Additionally, predictability su ers from model uncertainty, meaning that there is only little consensus about the correct predictor variables and hence, the correct speci cation of the predictor model is unknown. The Bayesian framework accounts for model uncertainty by computing posterior model probabilities for all possible predictor models. Thus, Bayesian forecasts condition on the whole information set as opposed to conditioning on a single predictor variable and lead to more accurate forecasts.
Persistent Identifier of the first edition: 
ISBN: 
9788793155190
Creative Commons License: 
cc-by-nc-nd Logo
Document Type: 
Doctoral Thesis

Files in This Item:
File
Size





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.