GLOBAL FACTOR TRADE: Statistical Tests on Technology and Absorption 2
Nonetheless, both because the principal concern of trade economists here is in measures of net factor trade, and also for direct comparability with prior studies, in Section V we will go on to implement each of the models of technology and absorption. In doing so, we will gain a rich view of the role played by each change in improving the working of the HOV model. For reference, we will indicate the production specification associated with the distinct models of technology. www.easyloans-now.com
Our first model of technology (P1) is the standard starting point in all investigations of HOV: it postulates that all countries use identical production techniques in all sectors. This can be tested directly using our data. For any countries c and c’, it should be the case that Bc = Bc’ . We reject this restriction by inspection.
One possible reason for cross-country differences in measured production techniques is simple measurement error (P2). The Italian aircraft industry is four times as capital intensive as the US industry. While this may indicate different production techniques, the fact that net output in US aircraft is approximately 200 times larger than in Italy raises the question of whether the same set of activities are being captured in the Italian data. This raises a more general point that is readily visible in the data. Namely extreme outliers in measured Bc tend to be inversely related to sector size. In tests of trade and production theory this is likely to produce problems when applying one country’s technology matrix to another country. If sectors that are large in the US tend to be small abroad, then evaluating the factor content of foreign production using the US matrix is likely to magnify measurement error. Large foreign sectors are going to be precisely the ones that are measured with greatest error in the US.
A simple solution to this problem is to postulate that all countries use identical technologies but each measured Bc is drawn from a random distribution centered on a common B. If we postulate that
Bc = B,c
where we assume that ,c is distributed log normally. This can relationship can be estimated by running the following regression:
here $fi are parameters to be estimated corresponding to the log of common factor input requirement for factor f in sector i. We can contemplate two sources of heteroskedasticity. The first arises because larger sectors tend to be measured more accurately than smaller sectors. The second arises because percentage errors are likely to be larger in sectors that use less of a factor than sectors that use more of a factor. In order to correct for this heteroskedasticity, in all regressions we weighted all observations by the square root of the log of value added multiplied by Bfi / Bf where Bfi corresponds to the average factor intensity in sector i and B^ corresponds to the average factor intensity across all sectors.