Mean-reversion in forward premium

T. Choudhry International Review of Economics and Finance 8 1999 433–453 443 Table 9 Unbiased forward rate hypothesis test Null Hypotheses—F Statistics Country H : a 1 5 1 H : a , a 1 , a 3 5 0, 1, 0 Australia 4.649 4.924 Canada 7.675 3.137 France 0.319 5.231 Germany 0.186 3.939 Hong Kong 131.770 46.462 Italy 0.595 3.224 Japan 3.720 5.199 South Africa 4.723 1.719 U.K. 0.292 107.930 Implies rejection of the null at the 1 level. Implies rejection of the null at the 5 level. Implies rejection of the null at the 10 level. of a unit coefficient on the forward rate is accepted only in France, Germany, Italy and the United Kingdom Table 9. This result is quite different from the previous semi-strong efficiency test Table 6. The null unbiased hypothesis is only found to be true for South Africa Table 9. Thus, once again South African currency is the only currency that provides evidence supportive of the semi-strong efficiency. Table 10 provides a comparison between results presented in this article and results of other research articles that apply some form of cointegration procedure. 26 As the table indicates results presented are not identical in each article. Of course along with the estimation procedure the sample period and the data frequency are also not identical. Most results presented do provide some evidence of the forward rate effi- ciency. Lack of evidence is noted during the latest floating era. This article and the Corbae et al. 1992 article do help redeem the latest floating era.

4. Mean-reversion in forward premium

As suggested above, violation of the forward rate unbiasedness may be due to the existence of risk aversion on the part of the participants of the foreign exchange market. Risk averse participants require a premium on forward contracts that exposes them to exchange rate risk. Relying heavily on Crowder 1994, p. 556, the forward rate consists of the future spot rate, a risk premium d and a rational expectations forecast error e. f t 5 s t 1 1 1 d t 1 1 1 e t 1 1 . 9 Furthermore, the future spot rate can be decomposed into current spot rate plus future depreciation. s t 1 1 5 s t 1 Ds t 1 1 . 10 444 T. Choudhry International Review of Economics and Finance 8 1999 433–453 Table 10 Summary of comparison between our results and others Article Data Method Results Barnhart and Monthly data Engle and Granger No evidence of the For- Szakmary 1991 Currencies of Canada, cointegration method ward rate hypothesis Germany, Japan, and the UK against the US dollar 1974–1988 Corbae, Lim, and Weekly data Canonical cointegra- Some evidence of the Ouliaris 1992 Currencies of Canada, tion regressions Forward rate hypothesis Germany, Swiss, France, Japan, and the UK against the US dol- lar 1976–1985 Crowder 1994 Monthly data Johansen cointegra- No evidence of the For- Currencies of Canada, tion method ward rate hypothesis Germany, and the UK against the US dollar 1974–1991 McFarland, Daily data Phillip and Hasen Some evidence of the McMahon, and Currencies of Germany, fully modified ap- Forward rate hypothesis Ngama 1994 Belgium, France, Italy, proach and the US against the UK pound 1922–1925 Phillips, McFarland, Daily data Fully modified least Some evidence of the and McMahon Currencies of Belgium, absolute deviations Forward rate hypothesis 1996 France, Italy, and the US against the UK pound 1922–1925 Phillips and Daily data Fully modified least No evidence of the For- McFarland Currency of Australia absolute deviations ward rate hypothesis 1997 against the US dollar 1984–1991 This paper Monthly data Fractional cointegra- Some evidence of the Currencies of Australia, tion and the Harris- Forward rate hypothesis Canada, France, Ger- Inder cointegration many, Honk Kong, It- method aly, Japan, South Africa, and the UK against the US dollar 1985–1996 T. Choudhry International Review of Economics and Finance 8 1999 433–453 445 Table 11 Forward premium unit root test results: KPSS test Lags Australia Canada France Germany H Kong Italy Japan S Africa U.K. 8.413 2.192 1.240 4.551 4.028 0.859 1.338 2.585 3.565 3 2.450 0.762 0.393 1.330 1.290 0.292 0.479 0.956 0.115 6 1.486 0.499 0.248 0.789 0.908 0.194 0.298 0.603 0.692 Implies rejection of the null of stationarity at the 1 level. Implies rejection of the null of stationarity at the 5 level. Implies rejection of the null of stationarity at the 10 level. KPSS critical value: 0.739 1, 0.463 5 and 0.347 10. Source Kwiatkowski et al. 1992. Substituting Eq. 10 into Eq. 9, we get f t 2 s t 5 Ds t 1 1 1 d t 1 1 1 e t 1 1 . 11 Eq. 11 shows that the forward premium f t 2 s t is the sum of the spot rate deprecia- tion, risk premium and rational expectations error. 27 Given that the spot rate deprecia- tion and the rational expectation error are stationary, 28 if the risk premium is covariance stationary, then the forward premium must be stationary and the spot and the forward rate are cointegrated with the cointegrating vector of 1, 21; consequently the unbi- ased hypothesis holds Baillie Bollerslev, 1994a. 29 We conduct unit root tests GPH and KPSS to check for the stationarity of the forward premium f t 2 s t . Table 11 presents the results from the KPSS tests with 0, 3 and 6 lags. The KPSS test is applied without a trend, thus the null hypothesis is stationarity with a drift. The null is rejected at all lags length in six cases: Australia, Canada, Germany, Hong Kong, South Africa, and the United Kingdom. The null of stationarity is accepted using 3 and 6 lags in the case of Italy and using only 6 lags for France and Japan. Given the cointegration tests results and the unbiased test results, a nonstationary forward premium for Canada and South Africa and a stationary premium for Italy is quite surprising. 30 Baillie and Bollerslev 1994a indicate that forward premium may be a long memory process; in other words, it may be a fractionally-integrated process, so that imposing a value of 0 and 1 for d may be too restrictive. Baillie and Bollerslev, applying maximum likelihood estimations methods, find evidence indicating the forward premium to be fractionally integrated for several currencies. Table 12 shows results from the GPH tests of unit root in the nine forward premiums under consideration. Tests are conducted for two different values of m, i.e., 0.55 and 0.60. The hypothesis testing regarding the value of d is based on two tests: a two-sided test where the null is d 5 1 and alternate null is d ? 1; and a one-sided test where the null is as before d 5 1 and the alternate null is d , 1 or d . 1. If the two-sided test is unable to reject the null of d 5 1 then the second test is not considered. For Australia, France, Germany, Japan, South Africa, and the United Kingdom the null of d 5 1 is not rejected using both values of m. Once again, the South African result is quite surprising given the cointegration and unbiased hypothesis results. The Canadian results show the forward 446 T. Choudhry International Review of Economics and Finance 8 1999 433–453 Table 12 Foward premium unit root test results: GPH test Country m 5 0.55 m 5 0.60 Australia d 5 0.9284 d 5 0.8066 20.3248 21.0235 Canada d 5 0.9046 d 5 0.6386 20.4326 21.9123 c,e France d 5 0.9168 d 5 0.7122 20.3772 21.5227 Germany d 5 1.0255 d 5 0.8293 0.1155 20.9033 Hong Kong d 5 0.3971 d 5 0.5622 22.7336 a,d 22.3166 b,e Italy d 5 0.5773 d 5 0.6885 21.9165 c,e 21.6482 a,e Japan d 5 0.6912 d 5 0.7742 21.4002 21.1946 South Africa d 5 1.2064 d 5 0.8240 0.9359 20.9312 U.K. d 5 0.8085 d 5 0.7960 20.8681 21.0793 a Implies rejection of the null of the one-sided test at the 1 level. b Implies rejection of the null of the one-sided test at the 5 level. c Implies rejection of the null of the one-sided test at the 10 level. d Implies rejection of the null of the two-sided test at the 1 level. e Implies rejection of the null of the two-sided test at the 5 level. t -statistics in parentheses. Two-sided test H : d 5 1 and H 1 : d ? 1. One-sided test H : d 5 1 and H 1 : d , 1. Two-sided critical values: 2.576 1, 1.960 5, and 1.645 10. One-sided critical values: 2.326 1, 1.645 5, and 1.282 10. premium to be fractionally integrated when m is equal to 0.60. In both tests, the alternate null hypotheses are accepted. Results from Hong Kong and Italy show long memory forward premium using both values of m. 31 The difference in the KPSS and the GPH tests results are quite considerable especially for Canada and Hong Kong. 32 Fig. 1 shows the autocorrelation function of the nine forward premiums and the 5 confidence interval two standard error. Interpretation from the figure is quite difficult but the forward premium from Australia, France, Germany, Japan and the United Kingdom seems to decay to zero quite slowly. Our results basically imply two points: first, forward premiums may have a long memory as indicated by Baillie and Bollerslev 1994a; and second, the nonstationary forward premium may not be the cause behind the failure of the forward rate unbiased- ness. Other factors cited earlier may be responsible. Baillie and McMahon 1989 and McFarland et al. 1994 suggest that the failure of rational expectations may also be a reason behind the lack of the unbiasedness of the forward rate. Frenkel 1980 T. Choudhry International Review of Economics and Finance 8 1999 433–453 447 Fig. 1. Autocorrelation function of forward premium. 448 T. Choudhry International Review of Economics and Finance 8 1999 433–453 provides evidence of the failure of the unbiased hypothesis during the 1920s due to the absence of rational expectations. Our results also indicate that the definition of the forward premium may very much dictate mean-reversion. Results provided in our article advocate the need for further research in this field.

5. Conclusions