political and geographic factors were also taken into consideration. This paper argues for the
inclusion of CDS premiums alongside economic fundamental factors to improve the predictability of
the crises.
4. Definition of Crises and Model Specification
To test whether sovereign credit default swaps are a leading indicator of financial crises, we
construct three models in stock markets and currency markets respectively. The first model is referred
to as the “base” model and includes variables usually used in the literature to predict crises. The
second model adds to the base model changes in CDS premiums, and will be indicative of the ability
of the sovereign CDS to improve the forecasts of the existing models. It does so by checking the
significance of the factor in presence of other factors currently used. The third model has changes in
CDS premiums as the only independent variable. The dependent variable used is a qualitative variable
while the independent variables are exogenous quantitative variables. Therefore, non-linear models
are used to link the crisis prediction indicators as the dependent variables to the changes in CDS
premiums and other quantitative variables as the independent variables.
Periods of crises are identified by constructing an indicator (Crisisi,t), which is used as the
dependent variable in the model. The indicator takes the value of 1 if there was a crisis within past 6
months and a value of 0 otherwise. In the regression models that follow, we estimate the probability
that the crisis indicator is equal to 1 in a six-month horizon in both currency market and stock market
of the emerging countries under consideration. Next, we define crises in currency and stock markets
and then explain the regression methodology.
4.1. Currency Market Crisis Definition
Following Sachs, Tornell and Velasco 1996, Kaminski, Lizondo and Reinhart (1998) and
Corsetti, Pesenti and Roubini (1999), our models specify a currency crisis when there is a
simultaneous increase in currency depreciation and foreign exchange reserves losses. As a convention
in literature, a currency pressure index is constructed by the following formula:
cpindexi,t = ∆Ei,t -(σEi,t /σRi,t ) ∆Ri,t
(3)
where ∆Ei,t measures the devaluation of the nominal exchange rate in terms of dollars, ∆Ri,t measures
the change in the country’s foreign reserves, and (σEi,t /σRi,t ) is the ratio of standard deviations. The
index has an advantage of being able to analyze speculative attacks on currencies under both fixed
and flexible exchange rate systems. An increase in the reserves reflects foreign currency inflows and
lowers the pressure on depreciation of the local currency because of the negative sign in the equation.
So, cpindexi,t measures the depreciation pressure of a currency.
We define a currency crisis when the pressure of a currency goes beyond a certain threshold. In
empirical studies, the threshold used falls between one and three standard deviations above the mean
of the index. This paper uses the following formula to identify the crisis periods:
where crisisi t is the crisis indicator of country i at time t, μcpindexi t is the sample mean of the pressure
index and σcpindexi,t is the standard deviation of the pressure index. With a threshold of two standard
deviations above the mean, a total number of 24 crises are identified. Table 1 below shows
frequencies of crises and tranquil periods in the respective countries.
crisisi,t
J1,
= 1
10,
if cpindex. ,≥ μ . , + 2.0σ . ,
i,t cpindexi,t cpindexi,t
otherwise
(4)
128