Openness, similarity in export composition, and income dynamics

Abstract

A relevant share of the theoretical and empirical analysis on economic growth has been devoted to finding a specific role for international trade in reinforcing countries' growth rates. Not as much attention has been dedicated to the role of sectoral composition of export in influencing the effect of trade on income convergence. In this paper we look at this issue along the line of research on multiple regimes and convergence clubs, considering how openness and similarity in export composition among countries can induce convergence in income levels among the same countries. We apply our analysis to the catching-up of income levels of Central and Eastern Europe Countries to the EU benchmark. We explicitly consider the sectoral export patterns of the CEECs by comparing them to those of the 15 old members of the EU, focusing on countries' specialization as suppliers for the EU market. Our main result is that similarity in export composition has a positive, significant and nonlinear impact on catching-up. Results are robust to controlling for openness and country-size and for investment, schooling, and the quality of institutions.

Key words:

• CEECs
• transition
• growth
• semiparametrics

Acknowledgments

The authors wish to thank Fabio Sdogati, Lelio Iapadre and the participants at the conferences Economic Growth and Distribution held in Lucca, June 2004, and ETSG 2004 in Nottingham, as well as two anonymous referees for comments and suggestions. Financial support from MIUR, PRIN prot. 2002138555 is gratefully acknowledged.

Notes

To a large extent, integration though trade flows was achieved well before the formal entry of the CEECs into the EU (Landesmann, 2003).

The Bray-Curtis semi-metric – largely used in the natural sciences – is a bounded measure, 0 ≤ d _xy  ≤ 1; it has the advantage of not increasing in the number of sectors considered, n; of being invariant to proportional sub-classifications of the n sectors considered; it is not subject to the double-zeros paradox; it lessens the effect of the largest differences since difference in high sectoral export shares contribute the same as differences between small sectoral export shares; and is appropriate in the presence of skewed distributions. Therefore d _xy has been selected as our preferred choice. We also make use of the linear correlation index r _xy in the multivariate regression in order to verify the robustness of the empirical evidence to different measures of similarity.

A standard approach would be ‘to cut’ the bivariate kernels at equally spaced heights regardless of the ‘depth’ of the observations; in the present case, the contours have been carefully drawn in order to contain specific proportions of the dataset. The contour labeled ‘75’ contains the 75 per cent of the observations, and similarly for the contours labeled ‘50’ and ‘25’.

The Sachs & Warner – Wacziarg & Welch index (SW-WW) classifies countries as open if they satisfy all the following five criteria: (1) the average of unweighted tariffs in the 1990 – 1999 period is lower than 40 per cent; (2) the average of core non-tariff barriers on capital goods and intermediates is lower than 40 per cent; (3) the average black market premium over the period is lower than 20 per cent; (4) the country does not have an export marketing board; and (5) the country is not socialist.

A possible alternative is to use country fixed or random effects, but since our emphasis (as will be shown) is on the use of a semi-parametric regression, in order to give robust evidence of the nonlinear effect of EU-similarity on the catch-up of CEECs, and since, as pointed out by a referee, the literature on fixed and random effect semi-parametric regression is not thoroughly developed and proper tests to discriminate between fixed and random effects alternatives are still unavailable, we opted for a simpler functional form with a country categorical predictor, so as to allow for comparison of parametric and semi-parametric regressions.

The exclusion of potentially collinear variables (not shown) increases the significance of the country dummy parameter, increasing the F value of the regression.

Since there is a presumption of a serial correlation in openness we also used the first difference of the variable as an instrument. In this case the magnitude of the coefficients changes, but the sign and the significance remain stable.

The parameter estimate for EU-correlation is substantially lower than the one for EU-similarity; the explanation lies on the insensitiveness of the correlation index to specific relative changes in the export structure of countries. The disadvantages in the use of the Pearson correlation index as a measure of similarity in trade structures are discussed in De Benedictis and Tajoli (2007).

Diagnostics have been applied to the functional form of regression (12) with positive results.

Splines are piece-wise polynomial functions that fit together at ‘knots’ (Hastie & Tibshirani, 1999: 22); for cubic splines – as in our case – the first and second derivatives are also continuous at the knots.

Smoothing splines arise as the solution to the following simple-regression problem: find the function [fcirc](x) with two continuous derivatives that minimizes the penalized sum of squares, where h is a smoothing parameter. The first term in the previous equation is the residual sum of squares. The second term is a roughness penalty, which is large when the integrated second derivative of the regression function f ^″(x) is large – that is, when f(x) is rapidly changing slope. The endpoints of the integral enclose the data.