This vignette describes two automatic forecasting algorithms implemented in the R package forecast. These algorithms are used for forecasting univariate time series data. The first algorithm is based on innovations state space models that underlie exponential smoothing methods. The second algorithm is a step-wise approach for forecasting using ARIMA models. Both methods are applicable to both seasonal and non-seasonal data and are illustrated using four real time series. The package also includes other forecasting methods such as the Theta method, cubic splines, and simple exponential smoothing. The first algorithm, based on exponential smoothing, uses a state space model to produce point forecasts and prediction intervals. It includes models with additive and multiplicative errors, and the choice between them is based on the AIC criterion. The second algorithm, based on ARIMA models, uses unit root tests and the AIC to select the appropriate model order. It also includes a step-wise procedure for traversing the model space to find the best model. The forecast package provides functions for automatic forecasting, including ets(), auto.arima(), and forecast(). These functions can be used to produce point forecasts and prediction intervals for a wide range of time series models. The package also includes other functions such as theta(), splinef(), and meanf() for different forecasting methods. The forecast() function is a generic function that can be used with various time series models, and it provides a common format for the results. The package also includes plotting functions for visualizing forecasts and prediction intervals.This vignette describes two automatic forecasting algorithms implemented in the R package forecast. These algorithms are used for forecasting univariate time series data. The first algorithm is based on innovations state space models that underlie exponential smoothing methods. The second algorithm is a step-wise approach for forecasting using ARIMA models. Both methods are applicable to both seasonal and non-seasonal data and are illustrated using four real time series. The package also includes other forecasting methods such as the Theta method, cubic splines, and simple exponential smoothing. The first algorithm, based on exponential smoothing, uses a state space model to produce point forecasts and prediction intervals. It includes models with additive and multiplicative errors, and the choice between them is based on the AIC criterion. The second algorithm, based on ARIMA models, uses unit root tests and the AIC to select the appropriate model order. It also includes a step-wise procedure for traversing the model space to find the best model. The forecast package provides functions for automatic forecasting, including ets(), auto.arima(), and forecast(). These functions can be used to produce point forecasts and prediction intervals for a wide range of time series models. The package also includes other functions such as theta(), splinef(), and meanf() for different forecasting methods. The forecast() function is a generic function that can be used with various time series models, and it provides a common format for the results. The package also includes plotting functions for visualizing forecasts and prediction intervals.