The class of univariate processes considered in this chapter includes stationary and non-stationary cases, where the latter are either integrated (and so can be made stationary by an appropriate level of differencing) or are stationary around a deterministic polynomial of time. We assume in all cases that the model is the process that generated the data: in subsequent chapters, this assumption is relaxed. We define forecasts, forecast errors, forecast-error variances and prediction intervals, and make explicit the distinction between conditional and unconditional forecasts. The impact of parameter estimation uncertainty is discussed. We reconsider the ‘informativeness of forecasts’, with an illustration in terms of predicting stock-market prices, and the related notion of ‘the limit to forecastability’ (the horizon up to which forecasts are informative). We also discuss forecasting in non-linear models, the impact of ARCH on prediction intervals, and point predictions for asymmetric loss functions.