Some recent developments in nonlinear and linear time-series analysis are considered. Both theoretical results and empirical comparisons are used to demonstrate that substantial improvement in forecasting accuracy can be obtained via adaptive forecasting. The adaptive forecasting method has certain advantages. For example, the procedure is extremely simple to apply. In addition, the adaptive method can be readily extended to forecasting linear aggregates of future observations. The user simply estimates the parameters of an assumed model by minimizing the sum of squares of the forecast errors with respect to the linear aggregates of interest. Such an adaptive approach would, in general, stretch the usefulness of an assumed model and link more closely the problem of parameter estimation to loss function and local approximation in statistical modeling.