Asia-pacific journal of financial studies vol:36 issue:1 pages:135-161
Previous works variously ascribe the forward puzzle to non-stationary forward premium or model misspecification. However, no single theory fully succeeds in explaining the puzzle. This paper tries to differentiate two problems mentioned above by specifying non-linear model and decomposing the forward premium into long-memory co-movement and short-term filtered forward premium. We find non-linear model outperforms the classical model with respect to R-square and the forward puzzle is weakened in terms of the filtered forward premium, especially with extreme values, providing a support for Huisman et al (1998). The standard deviations are calculated in Monte Carlo Method for the overlapping observations.