5 Empirical evidence
We now present an application of our methodology to the international effects
of the October 1997 stock market crisis in Hong Kong.10 Using data from
Thomson Financial Datastream, we analyze correlation between stock market
returns of Hong Kong with ten emerging economies (Indonesia, Korea, Malaysia,
the Philippines, Singapore, Thailand, Russia, Argentina, Brazil and Mexico)
and the G7 countries. In our benchmark estimation we calculate two-day rolling
averages of daily returns in US dollars and we define tranquil and turbulent
periods as starting from January 1 1997 to October 17 1997 and from October
20 1997 to November 30 1997 respectively.16 This definition of the crisis period
follows the crash recorded by the stock market index in Hong Kong, which lost
25 per cent of its value in just four days starting on October 20 1997. Hong
Kong stock prices declined until the end of November, apparently influencing
returns in several other markets.
Although our test procedure is symmetrical, we adopt the common practice
of testing for contagion as a phenomenon in which correlation is significantly
higher during the crisis period. Hence, our test hypotheses are:
Hq : pɑ ≤ φ interdependence
Hi : pc > φ contagion .
Looking at the definition (2) of the coefficient of interdependence φ, note that p
and S, as well as the coefficient pɑ, can be easily estimated from the data. The
main challenge in carrying out this test is to find good estimates of λj and ʌɑ.
We will proceed as follows. First, we set up a conditional test fixing the value
of variance ratios parametrically; namely, we calculate minimum thresholds for
λj and λ6' at which the difference between pc and φ becomes statistically sig-
nificant. Second, we compare these threshold with empirical estimates of these
variance ratios, obtained using different methods.
5.1 Conditional test: identification of threshold values for
A and Ac
In this section, we identify critical thresholds for λj and λ6' at which the null
hypothesis is rejected at a given confidence level. To clarify the meaning of these
thresholds, consider first the case in which ʌɑ = λ1∙. By inspecting equation
(3), we see that φ is monotonically decreasing in λj∙, for given p and S. Sup-
pose we find pɑ significantly larger than φ for a given λj = λ' ; it follows that
pɑ is significantly larger than φ also for any λj = λw > λz. Therefore, we can
look for the minimum value of λj — denoted with λ — at which the hypothe-
sis of interdependence would be rejected at some prespecified confidence level.
Analogously, in the case ʌɑ = λ1∙, equation (2) shows that φ is monotonically
decreasing in λ6'. Hence, for any given λj we can look for the minimum value of
λj , λ , at which the hypothesis of interdependence would be rejected. In the
first case, the result of the conditional test will be a threshold A; in the second
laFor a comparison, see Forbes and Rigobon (1999a).
10We use US dollar returns because they represent profits of investors with international
portfolios. As stock markets in different countries are not simultaneously open, two-day rolling
averages of returns have been preferred to simple returns.
15