Exchange Rate Economics: Theories and Evidence

342 Market microstructure approach
the 3-month horizon and for all groups,except banks and import industries,
at the 6-month horizon. In sum Ito’s,unbiasedness tests reveal some evidence
of heterogenity in the sense that the rejection of unbiasedness is not universal
across company groups and forecast horizons.
MacDonald (1992) uses a disaggregate survey data base supplied by Consensus
Forecasts of London,to test the unbiasedness of individual forecasters (sample
period October 1989–March 1991). This survey data base is particularly valuable
since it contains disaggregate survey responses for three key currencies (dollar–
sterling,dollar–mark and dollar–yen) conducted simultaneously in seven financial
centres. Although these results tend to confirm unbiasedness tests using aggregate
data (which give a strong rejection of unbiasedness – see Chapter 15),in the
sense that the vast majority of forecasters do not have unbiased forecasts,there is a
significant minority that do produce unbiased forecasts. An interesting aspect of this
study is that German forecasters have almost a 100% record in producing unbiased
forecasts of the German mark,but produce as biased forecasts as other country
forecasters for the non mark currencies. MacDonald and Marsh (1996a),Chionis
and MacDonald (2002) have updated the Consensus unbiasedness results (period
October 1989–March 1995) and essentially confirm the findings of MacDonald.
Survey-based error orthogonality tests are based on the following equation:
s e t+k − s t+k = α + βI t + ε t+k ,(14.2)
where I t is a publicly available information set and the null hypothesis is α = 0
and β = 0 (for further issue relating to this kind of equation see Chapter 15).
Ito (1990) conducts his error orthogonality tests using the forward premium,past
forecast errors and the past exchange rate change as informational variables. At
the individual level Ito finds that about three-quarters of the individuals in his
survey fail the orthogonality test. MacDonald (1992) finds that the forecasters who
produce biased forecasts also failed the error orthogonality test (14.2) when the
information set consisted of the fourth lagged survey forecast error (the fourth lag
was used to avoid potential misalignments which may have produced spurious
correlations). MacDonald and Marsh (1996a) and Chionis and MacDonald (2002)
have extended and updated the Consensus results of MacDonald using the forward
premium as the informational variable; again the null hypothesis of orthogonality
was rejected in the vast majority of cases.
14.1.2 Expectational mechanisms
The following expression combines three expectational mechanisms,namely,
bandwagon,adaptive and regressive (see Chapter 15 for a discussion):
s e t+1 = ω + γ(s t − s t−1 ) + τ(s e t − s t ) + ν(¯s t − s t ). (14.3)
In the context of this equation the null hypothesis of static expectations may be
tested as γ = ν = τ = 0. In MacDonald (1992),a comprehensive examination of
the expectations formation mechanisms of all of the respondents to the Consensus