The forward premium puzzle and Markov-switching adaptive learning,

Abstract

Using Markov-switching adaptive learning, this paper builds on adaptive learning (AL) techniques pioneered in Evans and Honkapohja (2001, 2012) and presents an empirical analysis and explanation of the forward premium puzzle, first characterized in Fama (1984). Furthermore, this paper addresses the need for using mean-square stability as the criterion for stability when observing state dependent parameters rather than the traditional stability conditions presented by Blanchard and Kahn. Using these tools, I am able to parameterize a monetary exchange rate model to mimic the empirical observations which lead to the forward premium puzzle. I find that agents who possess knowledge of central bank interest rate regime changes and who use constant gain learning can replicate the negative bias present found in the forward premium regression. By using a robust parameterization aimed at weighting information in a representative agent's learning process, I am able to remove the negative bias found empirically in the forward premium regression. The central finding of this paper is that under constant gain learning, regime changes by central banks, modeled by Markov-switching liquidity preferences, explain the forward premium bias found empirically. This is different from most research, which focuses on persistent monetary model fundamentals to explain the bias. This result holds under robust parametrizations of constant gain learning, monetary fundamental persistence, and Markov-switching liquidity preferences, which govern the model. Thus, this paper offers a novel solution to the forward premium puzzle.