|
|
IDL Demo Library |
|
The Forecast Demo illustrates the fundamental principles of time-series forecasting. You can specify forecasting parameters to apply towards generating data.
A time-series is a sequential collection of data observations, indexed over time. Modeling a time-series as a combination of past values and residual white noise allows the extrapolation of data for future points of time.
This process is known as `Forecasting' and uses an autoregressive forecasting model of Order P, where P represents the number of past time-series values used to compute the forecast. In general, the accuracy of the forecast improves as the value of P increases.
The sample autocorrelation function is a commonly used tool to determine the accuracy of a forecasting model. The autocorrelation of a time-series measures the dependence between observations as a function of their time differences or lag. An N-element time-series with approximately 95% of its values in the interval,
[-1.96/sqrt(N), 1.96/sqrt(N)]
is said to be stationary and is the prerequisite to an accurate forecast. This interval is displayed with dashed lines on the plot of the sample autocorrelation. See the Mathematics section of the on-line help for more information.
Select "Quit" to exit the Forecast Demo and return to the IDL Demo main screen.
Select "About forecasting" for information about the Forecasting Demo.
Select the order, P, of the forecasting model.
Select the number of data points to forecast. Each new data point is represented by a red triangle.
Generate a new set of random values according to set of forecasting parameters specified with the above sliders.
IDL Demo Online Help (October 11, 2006)