This paper proposes a novel approach to improve accuracy in long-termforecasts based on nonlinear models, and provides an empirical analysis oflong-run electricity demand forecast for Korea. Our approach is based on thenotions of catch up and convergence, which exploits the fact that variouseconomic characteristics of developing countries evolve following closely withsome time lags the historial patterns of corresponding characteristics ofdeveloped countries. To forecast electricity demands in developing countries,for instance, we may therefore utilize the historical patterns of thedeterminants of electricity demands in developed countries. In particular, wefind that the time series of income elasticity of electricity demand in Koreaevolves in the same pattern as that of electricity demand in Japan. Thepatterns are extracted using cointegrating regressions with time-varyingcoefficients, and the actual gap and catch-up periods are determined tomaximize the performance of our forecasting model. Our procedure is shownto perform substantially better than other models that are commonly used inlong-term forecasts. In particular the optimal forecasts are obtained under theassumption that Korea’s catch-up period is shorter than the gap periodbetween Korea and Japan. Our empirical results are also consistent with thosedone by the studies of the convergence of productivity.