and document that omitting JIT does not cause any problems empirically (Frankel, 1993;
and for a survey Engel, 1996). Nevertheless, if JIT are found to be significant, which is more
probable for a relatively volatile economy, then the omission may cause biased β1 estimates.
As a result of recent advances in time series econometrics, a number of studies show
that forward premium bias can be treated as a statistical artifact rather than an economic
puzzle, and hence argue that UIP works better than it seems.
Recent papers report favorable results for extremely short investment horizons or long-
term maturities, suggesting that previous unfavorable results are confined to mid-horizons or
-maturities only. Chaboud and Wright (2005) report favorable results for the UIP condition
for intra-day frequencies, while Alexius (2001), Chinn and Meredith (2005) and Chinn (2006)
conclude that temporary disturbances to the UIP condition abate over maturities longer
than a year. These imply that, for extremely short investment horizons, exchange rate risks
vanish whereas for long horizons, the effects of monetary policy actions, the volatility of
risk premia, and market expectations are lessened.
Baillie and Bollerslev (2000) argue that earlier rejections of the UIP condition are mostly
due to factors such as small-sample bias, unstable β1 estimates over different sub-periods,
and high persistence in forward premium. Similarly, Maynard and Phillips (2001) demon-
strate that differences in the persistence of exchange rate changes and of forward premia
may induce forward premium bias. Specifically, OLS estimation of the UIP condition with
a stationary dependent variable and a near-unit root regressor induces a left-tailed limiting
distribution, which in turn causes β1 to converge to zero. Liu and Maynard (2005) argue
that high persistence in forward premium can provide a partial explanation of the bias.
Through a stochastic partial break model, Sakoulis and Zivot (2005) show that ignoring
structural breaks may cause spurious persistence in forward premium, which may result in
forward premium bias. Similarly, Choi and Zivot (2007) show that accounting for struc-
tural breaks significantly reduces the observed persistence in the forward premium. Hence,
the frequently observed forward premium bias may be a statistical artifact, resulting from
estimating unbalanced test regressions, or disregarding structural breaks, or both.