Abstract:
|
The bootstrap method introduced by Efron (1979) is applicable under the hypothesis of independence or under specific model assumptions. The general ARIMA models are highly dependent on time, and so the residuals are also highly correlated. Due to this a different bootstrap approach must be used to deal with the dependency, we call this, the non-overlapping bootstrap method. We will show how the non-overlapping block bootstrap method can be used for parameter estimation and for forecasting in a general ARIMA model, for real life data as well simulated data. The non-overlapping block bootstrap method will be compared with the Box-Jenkins methodology on parameter estimation and forecasts. We will also compare the length of the confidence intervals for the parameters and forecasted values using the traditional methods and the non-overlapping block bootstrap. All programming was completed using the statistical software package (SAS).
|