Próxima Página

Página Anterior

Índice

2.3 Causality in returns

The six graphs below depict the results of the Granger test on the series of returns from MSEMF-ASIA and RIBOV-ON. All the graphs in this and the next section follow the same format. For both hypotheses, they show the p-value of the F statistic from the Granger test, as a function of the number of lags included in the regression. The number of lags are shown in the horizontal axis. Graphs are constructed for all six samples A to F. However, sample F has been extended by one month (18 observations), because too small a sample could severely diminish the power of the test.

The results are interpreted as follows. H₀⁽¹⁾ will stand for the null hypothesis "Ibovespa overnight returns do not cause MSEMF-ASIA daily returns" and H₀⁽²⁾ will stand for the null "MSEMF-ASIA daily returns do not cause Ibovespa overnight returns". Correspondingly, a small p-value for H₀⁽¹⁾ will confer credit upon the claim "Ibovespa overnight returns cause MSEMF-ASIA daily returns"; and a small p-value for H₀⁽²⁾ will strenghten the claim "MSEMF-ASIA daily returns do not cause Ibovespa overnight returns".

For sample A, the full year of 1997 (Figure 7), causality from Asian toward Brazilian returns is definitely supported. All p-values for H₀⁽²⁾ are below the 5% confidence level, with the exception of the first lag. P-values for H₀⁽¹⁾ have an unstable pattern.

Use of sample B, which corresponds to the first half of the year before the Thailand event (Figure 8) results in high p-values for both hypotheses. No causality is therefore detected. Again, the p-values for H₀⁽²⁾ are unstable over different lag structures.

On the other hand, the period after the Thailand event represented by sample C (Figure 9) reaffirms causality from Asian to Brazilian returns. P-values for H₀⁽²⁾ are consistently lower than 5% for the majority of lag structures.

The picture is unchanged if EVENT 2 is subtracted from the post-Thailand period, which is done in sample D (Figure 10). The returns on the days of the Hong Kong crash do not seem relevant in the causality effect detected before, as p-values for H₀⁽²⁾ are still below 5% (except for lags 1 and 2). Also, no feedback is supported because of the consistently high p-values of H₀⁽¹⁾.

The first half of the year represented by sample E, before the Hong Kong crash, EVENT 2 (Figure 11), again shows high p-values for both hypotheses. No causality is therefore detected.

The last sample represents the period after the Hong Kong crash (Figure 12) and shows high p-values for both hypotheses, but this time those for H₀⁽²⁾ appear much lower than those for H₀⁽¹⁾, which can be an indication of some residual causality from Asian returns, particularly considering the reduced number of observations in this sample. Anyway, the results seem much less clearcut than those obtained from splitting the full sample at the Thailand event.

All these results point to a recognizable spillover effect from the Asian markets to the Brazilian market. The study of different samples also suggests that the Thailand event in July was a change in regime and probably the starting point of the spillover effect, defined as a statistically significant precedence of East Asian markets moves over Brazilian ones. On the other hand, perhaps surprisingly, the Hong Kong crash’s import on the spillover is not clear. Sample F’s p-values, which are high and unstable for both hypotheses, indicate that no causality can be observed immediately after the crash. We interpret this evidence as a sign that the Thailand event was in fact the real trigger of what came to be called the "spillover effect" in Brazil: a causal relationship between returns generated in both regions. The Hong Kong crash appears here as the climax of the spillover effect, not its origin. The origin seems to be situated near the Thailand event. After the Hong Kong crash the causality relationship seems to vanish.

Próxima Página

Página Anterior

Índice

Fale com a CVM