Semiparametric analysis of the specialization-income relationship

Abstract

In this article we investigate the empirical relationship between overall specialization and per capita income. The metric we use to measure overall specialization is the median of the sectoral distribution of the Balassa Index of Revealed Comparative Advantages applied to four-digit (SITC rev.2) sectoral export manufactured data for 39 countries over the period 1985 to 2001. Once we take into account countries' specificities using fixed effects GAM, the results of the semiparametric analysis indicate that sectoral diversification of manufactured exports increases monotonically with development, and thus the reconcentration observed at high levels of income in the recent literature may not be linked to trade-induced specialization.

Notes

¹The only noticeable exception is Feenstra and Rose (2000). They notice that ‘…there is a strong negative correlation between the number of goods that a country exports and its ranking [income level]…,because rich countries tend to be open and diversified exporters, while poor countries tend to be closed and specialized exporters’.

²Following De Benedictis and Tamberi (2004) we consider the median of the sectoral empirical density function of the Balassa (1965) Index (BI) of Revealed comparative advantages as a summary measure of overall specialization.

³We also compare the median of the BI to three other popular indexes of overall specialization: the Gini and Herfindahl indexes, and the so-called country Gini. All indexes are correlated to each other, with the only exception of the Herfindahl index that appears to be different from the others in some respects. Comparisons are available on request from the authors.

⁴Manufacturing is defined as the sum of sectors from code 5 to 9, included. The total amount of sectors included in the database is 786.

⁵The choice of total income as a basis for the selection of countries was made to avoid possible distortions due to the presence of too small economies; nevertheless relative per capita incomes range from 7 (Pakistan) to 100 (USA) in 1998.

⁶See Fox (2000a, b) for a discussion on nonparametric regression methods.

⁷GAMs were introduced by Hastie and Tibshirani (1986) and are described in detail in Hastie and Tibshirani (1990).

⁸Under additional assumptions of no bias, these upper and lower curves can be viewed as approximate 95% pointwise confidence intervals bands.

⁹For example, Imbs and Wacziarg (2003) provided empirical evidence of a U-shaped pattern using data on sector level employment and value added with sectoral diversification followed by sectoral concentration.