Week 6

In order to predict more time steps ahead into future, performance of the model is evaluated by time shifting the 100 lag window. Below diagram shows the Variation of root mean square error with time shifting window of 100 Lags

Now, We can forecast further ahead, But error increase as further we predict

Below diagram show the variation of Error with time shifting using boxplots

According to the above observation variance of error also increase

Model parameters of the models shifted by time is shown below.

Enlarge model parameters for two models are shown below

Only first few lags significantly affect the model.

Problem: What is optimal number of lags? ( This question is answered in Week 7)