Title: Estimating bond risk premia via sequential learning
Authors: Tomasz Dubiel-Teleszynski - London School of Economics (United Kingdom) [presenting]
Konstantinos Kalogeropoulos - London School of Economics (United Kingdom)
Nikolaos Karouzakis - University of Sussex (United Kingdom)
Abstract: A Bayesian learning framework for the estimation and predictability of bond risk premia under a dynamic term structure model is implemented. We develop a sequential process for investors who learn about parameters, state variables and model uncertainty, when new information arrives. We account for model uncertainty by implementing an analysis of the time-varying parameters, in particular those driving the market price of risk specification and assess the efficiency and economic importance of risk restrictions from a forecasting perspective. The methodology improves the numerical behavior of estimation and addresses the issues of the instability of parameters and the ill-behaved likelihood functions. It allows for statistical and economically plausible parameter estimation when it comes to out-of-sample bond return predictability. The estimates are capable of capturing the stylized facts of the yield curve behavior, such as the violation of the expectation hypothesis, through the predictability of excess returns, and the persistence of interest rates and provide better forecasts of bond excess returns, improving investor utility.