Linear estimation of the UIP condition may also cause a forward premium bias when
the true data generating process reveals strong nonlinearity, which is confirmed by smooth
transitive regression (STR)estimations of Sarno et al. (2006) and Baillie and Kilic (2006).
Using excess currency return as the transition variable, Sarno et al. show that forward pre-
mium bias becomes persistent yet economically small as the expected exchange rate change
is closer to the UIP-implied level. Baillie and Kilic (2006) use lagged/risk-adjusted forward
premium as the transition variable and conclude that forward premium bias occurs when
the premium is small and/or negative and UIP is less likely to be rejected when the premium
is larger. These results favoring nonlinearity/asymmetricity within the UIP context are also
in line with the earlier literature documenting that positive versus negative (Bansal, 1997;
Bansal and Dahlquist, 2000), and normal versus extreme (Huisman et al., 1998) forward
premia/interest differential deserve different treatments. In particular, through empirical
analyses, Bansal, and Bansal and Dahlquist show that the UIP condition is more likely
to hold when the interest differential is negative, whereas Huisman et al. report that for
extreme and positive forward premia, the UIP condition cannot be rejected. Hence, the
stylized substantial departures from the UIP condition may not mean foreign exchange
market inefficiency when the data shows strong nonlinearity.
Albuquerque (2006) estimates equation (3b), treating forward premium as endogenous,
and using lagged values of forward premia and interest rates as instruments. The results
of two and three-stage least squares with time-varying fixed effects reveal positive and
significant slope coefficients, weakening the forward premium bias.
In addition, Maynard (2006) questions the conventional regression-based analyses and
directly tests the forward rate unbiasedness hypothesis using sign and covariance-based
tests, and reports that the previously documented forward premium puzzle may in fact be
a statistical artifact.
To sum up, the existence of persistent deviations from the UIP condition has remained
a highly debated topic in international finance. In particular, earlier literature documents
that the forward premium/interest differential predicts the movements in the exchange rate