wordpress-themes.org wordpress themes wordpress themes

GLOBAL FACTOR TRADE: Implications for Net Factor Trade 7

In Table 8 we repeat our trade results obtained above and also present our results when recast in Trefler units. The switch to Trefler units matters little. Now the coefficient on predicted factor trade actually rises from 0.82 to 0.88. Our variance ratio test statistic falls a little but overall the same basic picture emerges.. Clearly our results are robust to this specification.

Xavier Gabaix (1997) has suggested a second weighting scheme for evaluating factor content studies. If one deflates both sides of the HOV trade equation by the country’s share of absorption, one eliminates all size-based variation from the data. This adjustment is tantamount to projecting each country’s endowment point on to the same iso-income line. The results also appear in Table 8. Once again we see a steady rise in the slope coefficient as we move from T3 to T7. The final specification has a slope coefficient of 0.83, again quite similar to our primary specification.

We conclude that our results are robust to a wide variety of weighting schemes. It appears that relaxing neutral technical differences, FPE, and allowing for non-traded goods results in dramatic improvements in the HOV model regardless of the units chosen. Furthermore, accounting for the influence of trade costs on bilateral trade volumes results in further strong improvements. click here

An additional remaining question regarding our results is why in T7 we obtain a coefficient of only 0.82 when theory says it should be unity. There are three basic reasons. The first is attenuation bias due to measurement error. By conducting the reverse regression of predicted factor trade on measured, we can obtain maximum likelihood bounds for the effects of measurement error. Under specification T7, the high R2 leaves little room for measurement error to matter, with an upper bound for the coefficient of 0.84. Under specification T7′, measurement error places an upper bound on the coefficient of 0.89.

The second reason is that our adjustments apply only to the impact that country capital to labor ratios have on average technology matrices, not export technology matrices. Theory suggests that this will still under-measure factor service trade because exported goods within an industry use more extreme factor proportions than goods which in equilibrium are non-traded in the same industry.