Solved – What’s the difference between a TSLM forecast and STL forecast

forecastingrtime series

Just want to preface that I'm relatively new to working with time series forecasts.

I'm pulling together a forecast I just want to understand a bit about the difference between an STL forecast and a TSLM forecast. I know that TSLM allows you to put in an external regressor, but I have only included the trend and seasonal in my model. Just want to know why the two forecasts look different/have such different prediction intervals.

library(forecast)
library(tidyverse)

autoplot(forecast(stl(time_series, s.window="periodic", robust=TRUE), h=123, method="ets")) +theme(legend.position="none")

STL
a_tslm <- tslm(time_series ~ trend + season)

autoplot(forecast(a_tslm, h=123)) +theme(legend.position="none")

TSLM

Best Answer

I think I got my answer.

A TSLM is literally just that. Essentially a data frame is created with 3+ columns. Column 1 is your y value. Column 2 is the numbers 1 to however_many_observations_you_have. Column 3 is a factor variable that corresponds to your seasonal. In my case, 1 to 12. Then it just compiles a regression based off of that.

STL on the other hand will find the trend (using loess). Then break out the seasonal.

Related Question