Two puzzles in the analysis of foreign exchange market efficiency

European foreign exchange market efficiency: Evidence based on crisis and noncrisis periods

“…Conversely, Pope and Peel (1991) and Ligeralde (1997) reject the existence of a risk premium. Newbold, Wohar, Rayner, Kellard, and Ennew (1998) address the efficiency puzzle by analyzing the respective time series properties of excess returns and forward premia. They find the two time series to have different degrees of autocorrelation.…”

Testing forward rate unbiasedness allowing for persistent regressors

“…This near-unit root problem has led recent work (Bekaert and Hodrick, 2001;Maynard, 2003b;Roll and Yan, 2000;Tauchen, 2001;Goodhart et al, 1997;Newbold et al, 1998) to question the validity of empirical inference procedures underlying the forward premium puzzle, suggesting potentially serious bias and/or size distortion. It seems important to note that this is a finite sample problem, the existence of which does not depend on either the presence of a unit root or the local-to-unity approximations discussed below.…”

“…2 It is natural therefore to seek more appropriate inference procedures in testing unbiasedness. 3 One approach has been to change either the regression specification, for example by first-differencing or pre-filtering prior to estimation (Roll and Yan, 2000;Newbold et al, 1998), or the estimation technique, using robust methods such as sign tests (Maynard, 2003a). However, if one wants to stay within the framework of the original unbiasedness regression/test as in (1), it is perhaps more convenient to maintain the same estimator, but modify the critical values in order to adjust for the non-standard nature of the distribution.…”

On the robustness of cointegration tests when assessing market efficiency

“…Maynard and Phillips posit an explanation by noting that inclusion of the spot return in (4) introduces unnecessary noise that may cause finite sample bias. Additionally they suggest that this finite sample bias is of particular significance because, as reported by Newbold et al (1998), the forward premium is so small 3 in magnitude that the time series properties of the forecast error can be easily dominated by those of the much larger spot return. It follows that an examination of the time series properties of the forecast error may tell us very little about the unbiasedness hypothesis.…”