A time series can be decomposed into trend, seasonal, and irregular components. Statistical methods have been developed to smooth and forecast time series. Moving averages expose the trend by averaging out other components. Exponential smoothing extracts the systematic trend from a time series by using past data and gives a forecast. Polynomial trend models use powers of the time index as explanatory variables, avoiding the need to identify a leading indicator, an explanatory variable that anticipates changes in a time series. Autoregressions are regressions that use previous values of the response, known as lagged variables, as predictors.