2.3 Summary

• Time-varying regression is a simple approach to forecasting that allows a non-linear trend.
• The uncertainty in your forecast is determined by how much error there is between the fit an the data.
• Fit must be balanced against prediction uncertainty.
• R allows you to quickly fit models and compute the prediction intervals.

Careful thought must be given to selecting the polynomial order.

• Standard methods are available in R for order selection
• Using different orders for different data sets has prediction consequences