fpp2 and fpp3: Prediction intervals from bootstrapped residuals, simple forecasting methods

[amarjitsinghchandhial]

Hi,

Regarding the 4 simple methods (either via the specific instruction or ARIMA counterparts):

• Mean
• Naïve
• Seasonal naïve
• Drift

they can be applied to ANY time series!

Can bootstrapped PI's be calculated for all 4 methods?
https://otexts.com/fpp2/prediction-intervals.html
https://otexts.com/fpp3/prediction-intervals.html

If so, as benchmark models to more complex, should they be on the raw or transformed (power/Box-Cox) time series?

Amarjit

[robjhyndman]

Mean, Naive and Drift can be applied to any time series. Seasonal naive requires a frequency greater than 1.

Yes, bootstrapped PIs can be calculated for all methods.

Whether transformations should be used or not depends on the time series. See https://otexts.com/fpp3/transformations.html#mathematical-transformations

[amarjitsinghchandhial]

Thanks Rob!