A panel technique for the analysis of technology convergence: The case of the Italian regions

Differences in productivity levels represent a major component of the large cross-country differences in per capita income observed in international datasets and even in some regional ones. Nowadays, few economists would dispute neither this finding, nor that differences in productivity reflects – among other things – differences in technology levels. More controversial is the question of whether such differences in technology are stationary or temporary – that is, whether technology convergence is taking place, at what speed, under what conditions. This state of affairs is the result of several different difficulties faced by the empirical analysis on cross-country differences in per capita income growth rates. Recently, things have improved on both the analytical and the empirical side. On the analytical side, simple models in which technology convergence and capital-deepening can be studied within a common framework are now available. In these models the transitional dynamics is simple enough to be useful for empirical analysis [for instance, De la Fuente (1996) and (1997)]. On the empirical side, Islam (1995) has shown that we can test for the presence of technology heterogeneity in cross-country convergence analysis by using an appropriate fixed-effect panel estimator. The contribution of the present paper is on the empirical side. We propose a method designed to test whether part of the observed economic convergence is due to technology convergence. The method is based on the contribution by Islam (1995), but it extends it as follows. Islam’s technique was originally designed – and is currently applied – to measure cross-country differences in technology levels, assuming that such (relative) differences are at their stationary values and therefore that no technology convergence is present. The extension proposed in this paper builds on the a standard implication of models of technology convergence. If such convergence is present, the cross-sectional variance of the logs of our measure of technology should decreases over time approaching its stationary value. Alternatively, if technology convergence is absent, the variance is at its stationary value and no significant time-trend should be detected in its value. We exploit this difference to test for the presence of technology convergence in the data. First, we estimate the convergence equation over several sub-periods and use the values of the individual intercepts to compute the TFP levels. Then, we obtain the cross-section variance of the logs of our measures of TFP for each sub-period, and check whether the observed pattern is consistent either with catching-up hypothesis or with the hypothesis that the current degree of technology heterogeneity is at its stationary value. In this paper we use a panel dataset of the Italian regions, 1960-95. We apply our proposed methodology to the Italian case because it is notoriously characterized by a remarkable degree of regional heterogeneity. In spite of being one of the best known cases of regional divide, no explicit analysis of technology convergence across Italian regions is available yet. We use dynamic panel techniques (LSDV and GMM) to estimates our growth regressions. We split the whole sample period in several sub-periods to check for the presence of technology convergence. Our preliminary results reveal a significant presence of technology convergence, which reached its peak between the first and the second sub-period, and stayed significant but at a slower pace in later sub-periods. The emerging picture points to the simultaneous presence of technology convergence in a context otherwise characterized by weak output per-worker convergence. This is consistent with some recent results based on international datasets (e.g. Dowrick and Rogers [OEP (2002]).