"... In PAGE 10: ... The absolute return is defined without considering the day-of-week effect, as that in the previous section. The summary statistics and the time plots, where it is expressed in percentage points, are depicted in Table1 and Figures 1 to 3 respectively.... In PAGE 15: ...From Table1 , one can see that the number of observations for the three markets vary in all samples, due to different holidays in different markets. For each market, the in- sample mean is more or less the same as the pre-crisis (pre-crisis post-sample) mean.... In PAGE 26: ... The average log-likelihoods of the ARCH-like model before deletion are more or less the same though. Comparing Table 3 with Table1 (summary statistics of the absolute return), one can see that the smaller the sample standard deviation, the larger the average log- likelihoods, across all markets and all periods, except for the pre-crisis NIKKEI. The z- test and the LR-test are essentially the same.... ..."

"... In PAGE 21: ... To avoid this problem, we use the ICSS method to identify breaks in variances of stock returns by using the model (19) - (21). Table7 reports the number of sudden changes in variance as identified by the ICSS algorithm for stock returns. Periods 3 and 7 have 17 break points and period 9 has only 4 change points and so on.... In PAGE 21: ... The significant changes in variance are a little bit more than those in level of absolute returns. The 3rd panel of Table7 shows the results of fitting breaks that correspond to the points of breaks in variance to the level of absolute stock returns. When breaks in variance are introduced, the evidence is somewhat mixed.... ..."

"... In PAGE 20: ...4.1 Event Study Results Table3 presents results of the event study of absolute ab normal returns and abnormal trading volume for the entire sample of 1,273 pre-listing events and 1,422 post-listing events. Given our concerns about non-normality in returns and volume, the table presents a variety of statistics.... In PAGE 20: ... Both the W-test and T-test statistics evaluate whether the Before period reaction is larger than the After period reaction. Table3 indicates statistically significantly greater return volatility and volume reactions to earnings shocks in the post-listing period. These heightened responses are particularly strong for volume, which typically doubles its size after U.... ..."

"... In PAGE 20: ... We count the days when the absolute returns of a single stock (i 6 = i0; j = j0) were over a certain threshold on two or more consecutive days. When a threshold value of u = 3:5 is used, we flnd that the maximal range of consecutive days from which the jumps in returns are over the threshold value are 3, 3, 2 days (Columns 2, 4 in Table4 ) for JPY/USD, CAD/USD, and GBP/USD respectively. The empirical estimates of tail dependence indexes can be computed by the division of counts for each range over the total number of days that exceedances occur.... In PAGE 20: ...etween 3.5 and 3.7. This phenomenon gives empirical evidence that variables in the left panel and variables in the middle panel may be tail dependent and it is appropriate to model the transformed data in an M4 process. We summarize the counts information within each range in Table4 . Table 4 suggests that the exchange returns between difierent currencies may be tail dependent, especially between (JPY/USD, CAD/USD) with an empirical tail dependence index estimate 11% and between (JPY/USD, GBP/CAD) with an empirical tail dependence index estimate 24%, and there may be lag-1 day tail dependence within each individual sequence.... In PAGE 20: ... Table 4 suggests that the exchange returns between difierent currencies may be tail dependent, especially between (JPY/USD, CAD/USD) with an empirical tail dependence index estimate 11% and between (JPY/USD, GBP/CAD) with an empirical tail dependence index estimate 24%, and there may be lag-1 day tail dependence within each individual sequence. From Table4 , since the counts for each individual return to have values above the selected threshold u more than 2 consecutive days is negligibly small, and hence it may be safe to say the maximum moving range for each return is 2, i.... ..."

"... In PAGE 21: ... In some sub-series, 7 As we increase the maximum permitted number of breaks or decrease the minimum number of observations to detect a break within that sample, we get a little bit more breaks. 8 In Table6 , we use (23) instead of (22) to get break apos;free apos; series since the absolute stock returns show some co-break in mean and variance. Because we estimate breaks by Bai apos;s method, (22) is suitable one.... ..."

"... In PAGE 24: ...ncrease dramatically after U.S. listing, although the changes are not statistically significant. Table6 presents estimates of cross-sectional regressions to explain cumulative absolute abnormal returns. Slope dummy terms are used to distinguish the pre-listing versus post-listing associations.... In PAGE 26: ...S. In the Panel B of Table6 , various specifications (e.g.... In PAGE 27: ...S. Table 7 p resents cross-sectional results to explain abnormal trading volume in a format similar to Table6 . Across the two panels, there is only weak evidence that abnormal trading volume around earnings releases is more prominent after cross listing.... In PAGE 51: ... Table6 . Cross-sectional regressions to explain cumulative absolute abnormal returns.... ..."