ERC-ODTÜ Uluslararası Ekonomi Kongresi-IV, Ankara, Turkey, 13 - 16 September 2000, pp.1-41
A large body of empirical work on growth and convergence has used cross-sectional
analysis, using long-term averages of relevant variables. A negative relationship between average growth rates and initial incomes in these studies is interpreted as evidence for convergence and support for neoclassical growth models. Such approaches, however, have been criticized, among others, by Quah and Rauch [1990], Quah [1993], Evans [1996] and Lee, Peseran and Smith [1995]. Using standard time-series techniques, such as cointegration, on the other hand, suffers from low power unless the time dimension of the data is sufficiently long. Pooling data across countries, I test the convergence hypothesis by investigating the unit
root properties of income differences for three different samples. Findings suggest that income differences are persistent even after allowing for country specific effects. Next, I modify the null hypothesis of no convergence to test the implications of endogenous growth models by allowing non-zero drifts under the null. Findings suggest unambiguous convergence only within the OECD. Tests for the sensitivity of results to the sample period reveal that income differences have widened in the post-1973 period. Finally, I illustrate that the use of log-income differences, rather than levels, biases the results toward convergence.
JEL Classification: C22, C23, 040.