No Contagion, Only Interdependence
2243
results is reported near the top of Table V. The base case is reported in bold
italics in the first row of the table.
For a second set of sensitivity tests, we examine how altering the defined
source of contagion can impact results. As discussed above, one difficulty in
testing for contagion during the East Asian crisis is that there is no single
event acting as a clear catalyst driving this turmoil. We begin by testing for
contagion from single East Asian markets ~other than Hong Kong! after a
significant downturn in those markets: from Thailand after its June decline;
from Thailand or Indonesia after their August losses; or from Korea for the
two-months after its crash that started in late October. Next, since conta-
gion may occur from the combined impact of movements in several East
2
0
Asian markets, we construct several East Asian indices. We test for con-
tagion from Thailand and Indonesia after the August crashes in both of
these markets; from Hong Kong and Korea during several tumultuous pe-
riods in these markets; and from Hong Kong, Indonesia, Korea, Malaysia,
and Thailand ~a five-country index! during several different periods. Sum-
mary results for a selection of these tests are reported in the middle of
Table V.
As a third series of robustness tests, we adjust the frequency of returns
and0or lag structure. In our base analysis, we focus on rolling-average, two-
day returns in order to control for the fact that different stock markets are
open during different hours. We also include five lags of the cross-market
correlations ~X ! and the vector of interest rates ~I !. We repeat this analysis
t
t
using daily returns and weekly returns. We also combine each of these re-
2
1
turn calculations with zero, one, or five lags ~as possible! of X and I .
A
t
t
selection of results is reported near the bottom of Table V.
For a final series of sensitivity tests, we vary the interest rate controls
and currency denomination. First, we include no interest rates in the model
or only include interest rates in the United States or the crisis country. Then
we repeat the analysis with local-currency returns with a variety of lag and
return structures. A sample of these results is reported at the bottom of
Table V. It is worth noting that the results reported in the last row of the
table—with daily returns, no lags, and no interest rate controls—is a test for
contagion in the simple theoretical model ~developed in Section II! before
adding the additional controls and extensions in Section III.
This series of sensitivity tests reported in Table V suggests that a wide
range of modifications to our base model do not affect the central results.
When tests for a significant change in cross-market relationships are based
on the conditional correlation coefficients, there is evidence of contagion in
2
0
To calculate each index, we weight each country by total market capitalization at the end
of 1996, as reported in Table II.
2
1
Note that in each case, estimates are only consistent if there are at least as many lags
minus one! as the number of days averaged to calculate the returns. Lags are required because
~
we utilize a moving average to measure returns. Therefore, by construction, observations at
time t are correlated with those at t Ϫ 1, t Ϫ 2, and so forth. Also note that we do not use more
than five lags due to the short length of the turmoil period.
Baroni
