CAN CREDIT DEFAULT SWAPS PREDICT FINANCIAL CRISES? EMPIRICAL STUDY ON EMERGING MARKETS

Panel data unit root test as suggested by Im et al. (2003) is applied to check whether we have
stationary variables. The test rejected the null hypothesis that all series are non-stationary against an
alternative that all series are stationary at a 10 percent level of significance. Hence, the variables used
in the regressions are trend stationary.

6. Empirical Results

Models specified in the methodology were regressed to estimate the variables that affect the
probability of stock market crises and currency market crises in emerging markets. We used panel
data with one month lagged independent variables. Because of the short time period of the data we
possessed, we did only in-sample predictions. Table 4 below shows the coefficients and marginal
effects or slope coefficients of the factors supposed to affect the probability of the stock market crises
whereas Table 5 below presents the same regression results for the factors assumed to affect the
probability of the currency crises. First, we discuss the results in both markets. Then, we explore the
predictive ability of our models and finally, we compare the predictive power of models in stock and
currency markets.

In the stock market models, all three models proved to be significant. In the first model, the
regression on the currently used factors in literature produced a pseudo squared-R of 0.059 and the
model proved significant at a one percent confidence level. Unexpectedly, P/E ratios and real interest
rates are insignificant in predicting stock market crises. In the second model, we improve the pseudo
squared-R to 0.077 after adding the one month changes in the CDS premiums in the base model. One
month changes in CDS premiums are statistically significant at one percent and improve forecasts of
stock market crashes a month in advance in emerging markets. The factor has the expected positive
sign, which reflects the assumption that ceteris paribus, an increase in premiums signals a higher
probability of default in emerging stock markets. Also, changes in CDS premiums are significant also
in the third model, when they are the only factor predicting crises. So, an increase in the default
probability of bond payments by a country, which is derived from the credit default swap premium,
can be interpreted as a factor signaling an increase in the probability of a stock market crisis in the
same country.

Table 4. Logit regression models predicting stock market crises¹

Model (1)                           Model (2)                            Model (3)

Dependent variable: Crisis Indicator                moue^                         _mo∞_w                          ,,,°,e, _m

Observations

Log Likelihood
Pseudo squared-R
Chi-square

P-Value

1521
-163.73714
0.0592
20.62
0.0000

1521
-160.6004
0.0773
26.89
0.0000

1543
-169.41479
0.0296

10.34

0.0013

¹Standard errors are in brackets, * significant at 10%, ** significant at 5%, *** significant at 1%, marginal
effects give the estimated slope coefficient, coefficients are estimated for the original logistic model

132

<<   <   5   6   7   8   9   10   11   >   >>