In fact, measurement error was already found when we consider the re-export from China to the U.S. (eastbound trade), in Table 2. As indicated there, several of the markup estimates were made while omitting some outlying observations (i.e. particular SITC or harmonized system categories). A good example is the markup for 1994, using method A. When all the harmonized system categories are included, the markup obtained is 19.4%, as compared to 29.3% (reported in Table 2) after several observations are omitted. The difference between these two estimates is almost entirely explained by a single harmonized system category – HS 26090000, which is tin ores and concentrates. Evidence that its quantity is incorrectly measured in re-exports comes from the fact that the re-export quantity is 617 times greater than the quantity that Hong Kong imported from China in 1994! This almost surely reflects measurement error in the data, and when this single observation is omitted, the estimated markup becomes much closer to that obtained in surrounding years.
Whenever the re-export quantity is greater than the import quantity, we shall interpret this as evidence of measurement error, such as different units in measuring quantities. We compute the ratio QRATIO=(quantity re-exported through Hong Kong)/(quantity imported into Hong Kong) to alert us to the measurement problem. The storage of commodities would make the reexport quantity greater than the import quantity, so some values of QATIO greater than unity might still be acceptable. We therefore consider two criterion for eliminating outliers: first, omitting all harmonized system categories for which QRATIO>2; and, second, omitting all harmonized system categories for which QRATI0>1. The second criterion is stricter than the first, in the sense that more observations are omitted. (The actual number of observation deleted is reported in the Appendix, Tables A1 and A2). long term payday loans
The results for the Hong Kong re-exports from the United States to China are shown in Table 4. When the observations with QRATIO>2 are omitted, the extremely large and negative markups are eliminated, and most of the negative values occurring after 1988 are quite small. When the additional observations with QRATIO>l are omitted, then the negative markups are eliminated almost entirely, except for 1994 and 1995. Based on this evidence, we conclude that more realistic markups are obtained when the outlying observations (as identified by QRATIO) are omitted. Generally, excluding these observation raises the markup. This follows because the observation with high values of QRATIO will necessarily have a low value for the markups on re-exports, so that excluding these erroneous observations will raise the overall markup.