drawn by accounting for transaction costs (Hollifield and Uppal, 1997; Verdelhan, 2006),
possible effects of central bank interventions (McCallum, 1994; Anker, 1999; Christensen,
2000; Baillie and Osterberg, 2000; Alexius, 2002; Chinn and Meredith, 2004; Mark and
Moh, 2007), existence of limits to speculation (Lyons, 2001; Villanueva, 2005; Sarno et
al. 2006), and the possibility that investors may care for real rather than nominal returns
(Engel, 1996).
In a general equilibrium setting, along with Dumas (1992), Hollifield and Uppal (1997)
show that the segmentation of international commodity markets through proportional trans-
action costs drives the β1 coefficient downward. However, they also document that this
transaction cost cannot account for the forward premium bias in its entirety. In particu-
lar, the negative estimates of β1 observed in the empirical literature cannot be replicated
by these models even in the presence of unrealistically high transaction costs and extreme
risk aversion parameters. Verdelhan (2006) uses slow-moving external habit preferences
(following Campbell and Cochrane, 1999) together with time-varying risk-free rates and
transaction costs, and demonstrates that forward premium bias can be rationalized by such
a model, regardless of the type of transaction cost.
The existence of central bank interventions may also distort the UIP condition. In
particular, as McCallum (1994) highlights, monetary authorities’ reaction to exchange rate
movements through policy rates leads to the joint determination of the expected deprecia-
tion and the interest rate differential. In the context of empirical applications of the UIP
condition, this implies a simultaneity bias, causing lower and possibly negative β1 estimates.
Christensen (2000), however, documents that negative β1 estimates cannot only be justified
by the estimated parameter values of the monetary policy reaction function suggested by
McCallum. Chinn and Meredith (2004) extend McCallum’s model by including output and
inflation in the monetary policy reaction function and document that deviations from the
UIP condition are primarily due to monetary policy reactions to temporary disturbances in
the exchange rate. In a parallel vein, Alexius (2002) constructs a model in which a central
bank minimizes an expected discounted loss function which comprises of recent inflation,