| Standard neoclassical growth models generate two long-run hypotheses: within-country common-trend convergence, and cross-country income convergence. The within-country common-trend hypothesis specifies that per-capita output, consumption, investment and government expenditures in each country contain a common trend. The econometric methods proposed to date--which employ panel and time series data--are not ideal for examining these hypotheses. Alternatively, procedures which employ field data, which are large in both time and cross-section dimensions, promise to provide clearer insights. Quah (1990) recognizes this, and develops a field data analysis to study income convergence in a univariate framework. Unfortunately, his analysis requires prespecification of the number of autoregressive and moving average coefficients in each model. In this paper, we use both univariate and multivariate approaches to investigate the long-run properties of both neoclassical and endogenous growth models and to test both of the above hypotheses. In the univariate approach, "augmented" Dickey-Fuller tests are constructed for unit roots which are based on approximations of autoregressive-moving average models of unknown orders by autoregressions. In the multivariate approach, Johansen's cointegration procedures has been used. Applied to the Summers-Heston data, these tests indicate that income disparities have no tendency to vanish over time; that disparities of growth rates seem to vanish rapidly; and that national output, consumption, investment and government expenditures in each country appear to be driven by a common trend. Hence, the tests favor endogenous growth models. A small Monte Carlo experiment suggests that the econometric procedures employed in this paper are robust. |
