2.1 Four empirical regularities
2.1.1 Sharp falls in stock prices tend to concentrate in periods of
international financial turmoil.
In the literature, many authors have observed that financial crises are not ran-
domly distributed. For instance, Eichengreen et al. (1996) noted that clusters of
speculative attacks on the exchange rate of several countries in different periods
are separated by long phases of tranquillity.
The stock market crises in our sample follow patterns that are consistent with
this observation. While referring to IMF (1998 and 1999) for a comprehensive
analysis of financial markets in our sample, here we note that, in the second half
of the nineties, many Asian stock indexes started to decline some time before
the eruption of the crisis in 1997 and together stayed on a descending path until
the end of 1998 (see Corsetti et al. (2000) and Corsetti (1998)). Stock markets
in Asia were not affected by the Mexican crisis, with the exception of Hong
Kong. In Latin America, stock markets partially survived the Asian crisis, but
were all greatly affected by the Mexican crisis and by the Russian/Brazilian
crisis. The impact of the last episode was especially strong, bringing about a
drop of over 50 per cent in stock indices. At the end of March 2000, stock prices
of most emerging market economies had not recovered relative to their historical
level. In the G7 countries, the impact of the Mexican and the Asian crises was
negligible, while the effect of the Russian/Brazilian crisis was much deeper. Yet
stock markets recovered quickly also after this crisis.
2.1.2 Volatility of stock prices increases during crisis periods.
Volatility of stock market returns is shown in Figures 1a and 1b. In Asia , stock
market volatility increases everywhere in 1997-99 relative to 1990-96, with the
sole exception of the Philippines.4 In 1997 and 1998 volatility records two
peaks, corresponding to the Asian and the Russian/Brazilian crises.0 By con-
trast, Hong Kong is the sole country in the region that is significantly affected
by the Mexican crisis. Overall, average volatility in 1997-99 is almost twice than
in 1990-96. As regards Latin American countries, in the second half of the 1990s
stock market volatility either decreases relative to previous record-high levels,
as in the case of Argentina, or it remains constant, as in the cases of Brazil and
Mexico. Volatility in these two countries is subject to large swings in correspon-
dence with the crisis episodes — yet, it is around its sample average by the end
of the decade. In Russia, volatility increases in 1997, peaking (dramatically) in
1We compute ‘instantaneous’ volatility of returns for country i at time t as an exponential-
weighted moving average given by σgt = ʌ/(l — ι?) ∑⅛Zo ^h(ri,t-h ~ ri)2, where ι? is the
decay factor (that we set equal to 0.96), T is the length of the moving window (that we pose
equal to 3 months), r⅛ ⅛ = log(a⅛ ιl∣xi ħ-l) where Xi ⅛ is the value of the country Ts stock
market index at time h and r⅛ is the average of the variable ri h in the period [t — T + 1,⅛].
Analogously, instantaneous covariance is σ⅛∙jt = (1 — г?) ∑t=Q1 ^h(rt,t-ħ — — r j)
and the instantaneous correlation coefficient is ρij t = σ⅛j∙jt∕(σιttσj,t)∙ Volatility has also been
estimated with a simple GARCH(I1I) model, which yields essentially the same results and,
hence, it is not shown. Variables computed using daily, weekly and monthly returns give very
similar results.
υ Only in Argentina, Indonesia and Thailand volatility reached higher levels in 1990 than
in 1997-99.