B Appendix
Our test results are not sensitive to a number of changes in our sample. In order
to show this, we have run our tests using returns in local currency (instead of US
dollar), modifying the definitions of tranquil and crisis periods, replacing rolling
averages of returns with simple daily returns, and filtering the data with US
interest rates. Table 3 summarizes the results, showing the number of countries
for which interdependence is rejected under each run of the analysis. For each
definition of our sample, we carry out the Fisher’s test, as well as our test
procedure with X7 = X7' (where the constant variance ratio is estimated using
the ‘world stock market index’) and with X7 = X7' = 0.
Our conclusions are quite robust to a change in the currency of denomination
of stock prices. This is true not only for countries that maintained a fixed or
quasi-fixed exchange rate with respect to the dollar, but also for countries that
experienced a sharp devaluation of their currency in our sample period. In the
case of Thailand vs. Hong Kong, for instance, p and pɑ are equal to 0.104
and 0.013, respectively, when using returns in local currency, while they are
0.106 and 0.005 when using returns in dollars. When we run our test setting
Xj = XÇ = 0, in our benchmark sample we reject interdependence only for Italy;
using returns in local currency we also reject interdependence for the UK. When
we set Xj = XÇ, our test rejects the null for Italy, the UK, Singapore, France
and the Philippines, regardless of the currency in which we calculate returns.
By the same token, our results are robust to changes in the timing of the
tranquil and the crisis periods. When we alter the definition of tranquil period
to include the year 1996, our test rejects the null for Italy, Singapore, France
and the Philippines, but not for the UK. As on average correlation remained
quite high at the end of 1997 (see Figures 3a-3d), we have also estimated a
model including December 1997 in the crisis period. In this case, results are
unaffected relative to our benchmark estimation.
Interestingly, if we replace two-day rolling averages with simple daily re-
turns , the number of cases in which the conditional tests reject interdependence
increases visibly, both for X7 = X7' = 0 and for X7 = X7'.19 In particular, con-
ditional on Xj = ʌɑ = 0, we reject interdependence for Italy, France, the UK;
using the estimated variance ratio together with the hypothesis Xj = X7', we
reject also for Singapore, the Philippines, Germany and Russia.
Note that we have excluded test results of the United States and Thailand,
for which the estimated correlation coefficients during the tranquil and crisis
period fall to zero. In this case, tests based on Fisher z-transformation are not
appropriate (see Stuart and Ord, 1994).
Finally, we have run the same testing procedure as in Forbes and Rigobon
(1999a), consisting in a VAR model of returns using domestic and US interest
rates and oil prices as exogenous variables. We have also expanded on their test
by including oil prices as exogenous variable. The results from these procedures
confirm our conclusions.
19Here we have excluded test results of the United States and Thailand, for which the
estimated correlation coefficients during the tranquil and crisis period fall to zero. In this
case, tests based on Fisher z-transformation are not appropriate (see Stuart and Ord, 1994).
23