-
Comments
- Economía
- Brookings Institution Press
- Volume 5, Number 1, Fall 2004
- pp. 116-129
- 10.1353/eco.2005.0006
- Article
- Additional Information
Economía 5.1 (2004) 116-129
[Access article in PDF]
Comments
[Return to Article]
Daniel Lederman: The editors of Economía are to be commended for recruiting the perfect pair of authors to write a paper on economic development in East Asia and Latin America.1 Both authors are well-known research economists who know much more than I about economic growth and the costs of financial crises. The fact that De Gregorio is a long-time policymaker and observer of Latin American economic issues and Lee is a well-known Korean economist is the cherry on the sundae.
De Gregorio and Lee provide substantial food for thought that merits further analysis. I will not take space here to discuss the costs of a balance-of-payments crisis, although that section of the paper is quite interesting and informative. Their calculations of the high costs of the Asian crises in the 1990s compared with those suffered by Latin American economies in the same period, plus their estimates of the determinants of these costs, seem to be consistent with existing case-study evidence my coauthor and I put together in 1998.2 However, some of the econometric issues raised below are applicable to the econometrics of the costs of crises.
It is difficult to argue against the basic findings of their econometric analyses, as is often the case with empirical studies of the determinants of economic growth across countries and over time. The authors find that education, trade, institutions, and so forth are good for growth. They offer no surprises, and this is perhaps the main weakness of the paper. It does not really advance our knowledge of the fundamental factors that drive growth. Its main contribution is the comparison between East Asia and Latin America, but the reader is left with a sense that everything matters for growth and that East Asia has performed better than Latin America because it has done many things—if not everything—better.
This is simply not the case. Many of the results on the determinants of economic growth are not as robust as the authors argue. My comments [End Page 116] aim to provide insights about econometric pitfalls that might affect not only the cross-country growth regressions, but numerous other applications, as well.
Econometric Pitfalls
The empirical model estimated by the authors and by literally hundreds of other researchers working on the determinants of economic growth can be written as follows:
where y is the natural logarithm of per capita GDP. Most researchers, including De Gregorio and Lee, subtract y in the previous period (t- 1) from the left-hand side of equation 1. This is trivial, however, for the only difference is that the beta on the lagged value of y on the right-hand side minus one equals the coefficient on the initial level of per capita income that was estimated by De Gregorio and Lee in their growth regressions. That is, the economic growth models are models of the determinants of per capita income, and the explanatory variables are admittedly country characteristics that supposedly determine the steady-state level of development for each country. The conditional convergence coefficient is thus nothing more than an estimate of the persistence of GDP per capita over time across countries.
In equation 1, η i captures any country-specific characteristic that does not change over time. In the growth literature, Barro and Sala-i-Martin argue that for international comparisons, it is important to capture this cross-country heterogeneity in the steady-state level of development, because it is unlikely that all countries use the same technology and have the same economic preferences.3 Ignoring the influence of these fixed effects probably biases the estimates of ß in equation 1 or of any other additional regressor.
De Gregorio and Lee deal with this heterogeneity, as many have done in the past, by also estimating model 1 or its equivalent in differences, which eliminates the influence of the unobserved heterogeneity: [End Page 117]
