Trade-creating regime-wide rules of origin: a quantitative analysis

Abstract

Regime-wide rules of origin (ROO), such as diagonal cumulation, de minimis and self-certification requirement, can be applied to reduce additional administrative and compliance costs for verifying restrictive ROO. However, empirical evidence related to the trade effect of various regime-wide ROOs is very few. We quantitatively investigate the trade effect of regime-wide ROOs by estimating the modified gravity equation with panel data on 36 238 country pairs covering 151 countries for 16 years from 1990 to 2005 at 5 year intervals. From our empirical experiments, we find that implementation of regime-wide ROOs such as diagonal cumulation and de minimis generate more trade between members of free trade agreements (FTAs). However, we also find that certification requirement does not produce positive trade effects. In addition, we confirm the effectiveness of the Poisson pseudo-maximum likelihood (PPML) estimator dealing with the zero trade issue and the presence of heteroscedasticity compared to the traditional log-linearized model estimation.

Keywords:

• free trade agreements
• rules of origin
• zero trade
• Poisson pseudo-maximum likelihood

JEL Classification:

• C23
• F15

Notes

¹ PPML estimation technique introduced by Santos Silva and Tenreyo (2006) has been applied to solve the zero trade issue and the presence of heteroscedasticity. In contrast, Head et al. (2010) and Martin and Pham (2008) criticize the biased estimates by the PPML when there is a large number of zero trade (17.6% in our case), and they drop observations that are recorded as zero bilateral trade instead of adopting the PPML. However, Santos Silva and Tenreyo (2011) empirically prove that the PPML estimation is well behaved, especially for constant elasticity models such as the gravity model, even when the proportion of zero trade in the data set is very large.

² CU is included as an independent dummy to avoid the problem of selection bias and because CU is not subject to ROO.

³ See Baldwin and Taglioni (2006), Baier and Bergstrand (2007), and Magee (2008). For the PPML estimation, we applied exporter and importer fixed effect which is not time-varying because of the failure of convergence, but the time-varying effect is controlled by including the year dummy.

⁴ We only report results from the dyad fixed effect estimation. We conducted the Hausman test (Hausman, 1978) and found that the null hypothesis, where the individual effects are not correlated with other regressors in the model, has been rejected. However, from the random effect estimation, we found that the conventional gravity variables (X’ in Equation 1) representing the transaction costs behave the way the model predicts and the estimated coefficients are statistically significant but not reported.

⁵ Confirms the findings from Park and Park (2009) and supports the better trade-creating effect of full cumulation to other alternative methods.

⁶ Confirms the findings from Santos Silva and Tenreyo (2006) that criticize biased estimates by log-linearization in the presence of heteroscedasticity.

⁷ It is because zero trade pairs would be estimated to trade close to zero and it results in residuals close to zero (Santos Silva and Tenreyo, 2006).