- 14 -
• Transactions costs induced hysteresis bands which are larger for shorter horizon
data [Baldwin, 1990]
Tests of UIP with survey data
Direct tests of UIP have been attempted from survey data on expectations, and the
results favor UIP, thus implying that rejection of unbiasedness in the data comes from
imposition of rational and linear expectations. [Froot and Frankel 1989, Chinn and
Frankel 1994, 2002; Chinn 2006]. The problems with survey data are that these might not
reflect true expectations, might have different working horizons than those of the
forecasters and more importantly might be based on the current forward premium
themselves. Another issue with survey data is its availability and completeness.
Unbiasedness over time
Flood and Rose (2002) and Baillie and Bollerslev (2000) provide evidence that
deviations from UIP were lower in the 1990’s than earlier. Flood and Rose (2002) use
daily data from both industrial and some emerging market economies and find that not
only are estimated coefficients positive in the 1990’s (compared to negative coefficients
found in studies with data for previous decades) but also that UIP holds better at times of
crises. Baillie and Bollerslev (2000) estimate betas from 5 year rolling regressions,
beginning March 1973 with monthly horizon data and find that in the 1990’s even
conventional estimates of betas turned positive [See Figure 4]. Chinn and Meredith