Comments

Norman V. Loayza: Bill Easterly, Norbert Fiess, and Daniel Lederman have written a serious and comprehensive study on convergence in North America after NAFTA. The authors approach the subject from many different perspectives, perhaps to make up for the little time available for conducting a definitive evaluation of NAFTA's aftermath. The paper's main question is the extent to which NAFTA has contributed to making Mexico's per capita income closer to that of the United States and Canada. To provide an answer that would address the various aspects of the question, the authors examine macro- and microeconomic data; use time-series, cross-sectional, and panel econometric techniques; and consider both cross-country and (Mexican) cross-state evidence. This may seem excessive, but there is a rationale for each exercise. Microeconomic (firm-level) data can resolve aggregation biases and concentrate on productivity convergence in specific industries. Macroeconomic, time-series, and cross-country evidence can control for common events taking place internationally, provide a benchmark for comparison, and thus help us understand the effects of the unique Mexican experience with NAFTA. Finally, Mexican cross-state evidence allows an evaluation of the differing effects of NAFTA on Mexico's regions, a necessary undertaking given this country's large size and diversity.

A possible objection to the paper's emphasis on income convergence could be that a proper evaluation of NAFTA should consider other more relevant or direct aspects of the agreement, such as trade volumes and prices, foreign investment flows, capital costs, and innovation trends. This objection is unwarranted, however, on considering that this paper is part of a larger research project that evaluates NAFTA more generally and draws policy implications for Mexico and other Latin American countries. The resulting papers from this project are being collected in the volume Lessons from NAFTA, edited by Daniel Lederman, William Maloney, and Luis Servén.

The authors arrive at a nuanced conclusion on NAFTA's success. NAFTA has indeed contributed to bringing Mexico's income closer to that of the U.S., but institutional and governance factors are preventing Mexico from converging to its North American partners faster. I believe this conclusion correctly reflects the achievements and limitations of NAFTA on income convergence up to this point. At the end of my comments, I offer additional evidence supporting it. My criticism of the paper resides not in its conclusions, but in some of its methodology.

Convergence Is a Dynamic Process

The authors implicitly address the issue of convergence from two different methodological standpoints. In their firm-productivity and cross-state analyses, they regard convergence as a dynamic, transitional phenomenon. To examine it, therefore, they estimate dynamic (lagged-dependent variable) models. This is the most appropriate treatment of convergence for developing countries. Conversely, when the authors turn to their cross-country analysis, they regard convergence as a steady-state phenomenon. The econometric counterpart to this perspective is the estimation of static models, based on the comparison of output levels via cointegration analysis or cross-country regressions. This is of only limited usefulness, however, for countries that are rapidly evolving.

The first consideration is whether cointegration analysis can help determine the extent of income convergence. According to Bernard and Durlauf, long-run convergence between two countries exists if the long-run forecast of their output difference is stable. ¹ The challenge for implementing this concept is how to assess the long-run stability of the income difference. Easterly, Fiess, and Lederman choose to use cointegration analysis: U.S. and Mexican income can be said to be converging if the countries' per capita output series cointegrate with a (1,­1) vector. If this is the case, the stationary difference between the two income levels provides a measure of the extent of convergence, in which a zero difference denotes absolute convergence.

The problem with this approach is that it requires that the income difference between the two countries be stable over the sample period, whereas the concept of convergence only requires that this difference be stable in the long run...