- 13 -
therefore the interest differential, the more the currency appreciates within any day
during which it does not depreciate.
Meredith and Chinn (1998) and Chinn (2006) obtained panel estimates for UIP at 5
and 10 year horizons for 4 countries and obtained betas close to 1, although these were
imprecisely estimated. Lothian and Simaan (1998) used time averaged long-horizon data
to obtain evidence in favor of UIP for 1974-1994. Cheung et. al (2005) also note more
evidence of UIP at long, rather than at short horizons. Figure 3 below summarizes panel
data evidence on unbiasedness for horizons varying from 3 months to 10 years.
Several explanations have been forwarded for long-short result discrepancy. These
include:
• Weak exogeneity of long run rates in a model where monetary authorities use short
term rates as instruments.[MacCallum 1993, Anker 1999, Chinn and Meredith
2004, Moore 1994]
• Segmented Short and Long Term bond markets, because of ‘preferred habitat’ [Lim
and Ogaki 2003; Alexius and Sellin, 2001] or ‘limited participation’ [Mizrach and
Occhino, 2004; Lahiri et.al., 2003; Alvarez et. al., 2001]
• Different expectations at different horizons [Frenkel and Froot 1987, Froot and Ito
1989]
• For very high frequency data, the risk involved in taking an open position goes to
zero. If one believes that the estimates of unbiasedness at other horizons are
corrupted by the existence of a time-varying risk premium, then the very high
frequency data, by eliminating that risk premium, should yield a coefficient of 1 if
UIP in fact holds.