The outline of the paper is as follows. Section 2 reviews the EWS literature and the
financial contagion studies, respectively. Section 3 describes the empirical methodology.
Section 4 describes the dataset and the empirical analysis. Section 5 concludes.
2 Background literature
As already mentioned in the Introduction, the first method used in the EWS literature is the
signal approach proposed by Kaminsky et al. (1998a) who monitor the evolution of several
indicators. If any of the macro-financial variables of a specific country tends to exceed a given
threshold during the period preceding a crisis, then this is interpreted as a warning signal
that a currency crisis in that specific country may take place within the following months.
The threshold is then adjusted to balance type I errors (that the model fails to predict crises
when they actually take place) and type II errors (that the model predicts crises which do not
occur). Kaminsky (1998b) and Goldstein et al. (2000) base their prediction of a crisis occur-
ring in a specific country by monitoring the evolution not only of a single macro-indicators,
but also on a composite leading indicator, which aggregates different macro-variables, with
weights given by inverse of the noise to signal ratio. Goldstein et al. (2000) find that, adding
information about crisis elsewhere, reduces the prediction error, even after the fundamentals
have been accounted for. The gains from incorporating information on crises elsewhere are
highest for Asia and it can be argued that they can be motivated by the existing theoret-
ical studies on financial contagion. In particular, Kodres and Pritsker (1998) also present
a model with rational agents and information asymmetries, where financial investors are
engaged in cross market hedging. Furthermore, Calvo and Mendoza (2000) present a model
where the fixed costs of gathering and processing country-specific information give rise to
herding behavior, even when investors are rational. Calvo (1999) stresses the role played by
margin calls in one market requiring that leveraged informed investors liquidate many po-
sitions, causing financial contagion. In this case, uninformed investors may mimic informed
investors even though ex post it turns out that no new information about fundamentals was
revealed.
The alternative method in the EWS literature is to use limited dependent variable re-
gression models to estimate the probability of a currency crisis. The currency crisis indicator
is modeled as a zero-one variable, as in the signal approach, and the prediction of the model
is interpreted as the probability of a crisis. More specifically, in line with the probit regres-
sion analysis put forward by Frenkel and Rose (1996), Berg et al. (1999) use this model