Time Series Analysis
Fit Time Series Models and Transfer Functions
The Time Series platform enables you to explore, analyze, and forecast univariate time series. A time series is a set of observations taken over a series of equally spaced time periods. Observations that are close together in time are typically correlated. Time series methodology takes advantage of this dependence between observations to better predict what the series will look like in the future.
Characteristics that are common in time series data include seasonality, trend, and autocorrelation. The Time Series platform provides options to handle these characteristics. Graphs such as variograms, autocorrelation plots, partial autocorrelation plots, and spectral density plots can be used to identify the type of model appropriate for describing and predicting (forecasting) the time series. There are also several decomposition methods in the platform that enable you to remove seasonal or general trends in the data to simplify the analysis. Alternatively, the platform can fit more sophisticated ARIMA models and State Space Smoothing models that have the ability to incorporate seasonality and long term trends all in one model. You can also perform a Box-Cox transformation and analyze and model the transformed series.
There are several methods to assess the forecasting performance of models. The Forecast on Holdback feature partitions the time series into a training portion to build models and a and holdback portion to assess forecasting performance.
Figure 18.1 Forecast Plot
