RDP 2007-10: Trade Costs and Some Puzzles in International Macroeconomics 5. Further Robustness Checks

In this section I examine a number of further checks of the robustness of my results. First I use a different measure of trade costs obtained by running the following regression:

where f_kit captures the average transport costs for shipments of good k to the US originating in country i at time t. Here, goods are disaggregated to the Standard International Trade Classification (SITC) revision 2, 3-digit level (again using Feenstra 2005). The betas are allowed to vary across countries and decades (which are indicated with the subscript s) and provide estimates of the transport costs for each country over time.

The regression attributes some of the trade cost to a commodity-specific effect (as some goods are likely to be more expensive to transport than others) while allowing there to be an effect that varies across time and across countries (represented by the betas), which captures in some sense the underlying transaction costs for the country. While this is a direct and intuitive way of attempting to calculate transportation costs, there are a number of reasons why it may be problematic. For example, it assumes a specific, though not implausible, assumption about functional form. However, Appendix B provides some evidence that this trade-cost measure may be reasonable.

The results from using estimates of betas as an alternative measure of trade costs in Equations (1) to (4) are reported in Table 9. The table provides support for trade costs explaining all three puzzles, since the interaction term is of the right sign in all specifications and similar in magnitude to estimates provided in Section 3 of the paper. Similar results are obtained if trade costs are assumed to follow a country-specific trend.

Up to this point I have been using trade-cost measures with the US as the proxy for general trade costs (largely due to data availability and reliability). There are a number of ways I attempt to assess whether this is likely to be affecting my results.

First, it is possible that by using trade costs with the US I may be able to describe well countries that are significant trading partners with the US, but poorly describe other countries. To assess this I run some of the simpler regressions using only a sub-sample of countries that trade substantially with the US – Australia, Canada, Japan and New Zealand (Table 10).^[25] Compared to the whole sample, arguably this sample suggests that trade costs provide a less convincing explanation for the puzzles; for the investment equation the interaction coefficient is insignificant and for one of the consumption equations it is negative. Alternatively, estimating the same regressions without these countries that trade highly with the US (that is, estimating for the European economies which tend to trade more amongst themselves) leads to results somewhat consistent with those of the full sample – the interaction term in the investment and the consumption (PPP) equations are positive and significant and the interaction term in the consumption (exchange rate) equation is also positive. However, the interaction term in the real exchange rate equation is negative and insignificant. This provides some evidence about the extent to which the main results are being driven by countries that trade substantially with the US.

As a second way to assess whether my results may be affected by their reliance on US import costs, I examine whether the extent of trade (as a share of GDP), in place of trade costs, effects the qualitative results. The idea behind this approach is that higher trade costs are likely to be reflected in lower openness and so openness may be a good and more broadly based indicator of trade costs. Moreover, it may also be better able to capture the extent of non-tradables, which are likely to lead to deviations between prices faced by locals and foreigners. Table 11 provides some further support to trade costs explaining part of the puzzles. The results indicate that the more open the economy the less extreme the puzzle (as the interaction term is negative).

Footnote

These countries have 20 per cent or more of their developed economy trade with the US (based on IMF DOTS for 1999). [25]