Using multiple model forms to capture and forecast the components of complex time series
I recently needed to fix the fence in my back yard. It’s old, wood, and has been threatening to topple over for some time now. Between curses it really struck me what number of tools I needed to make use of to get the job done, and the way sometimes you really want a couple of tool for the job.
What does this must do with time series regression? Generally, little or no. Particularly, quite a bit: today we’ll be diving into using mixed models for time series evaluation and forecasting. Or in additional DIY terms — using more model tools to get the forecasting job done.
So, without too far more rambling, we are going to get cracking with:
- Revisiting the massive picture, and talking a bit about mixed models.
- Taking a look at some real-world data.
- Using an easy model to capture the trend in our time series.
- Seasonality 3 ways: decision trees, linear regression, and classic time series.
- Putting Humpty Dumpty back together to get a single time series prediction.
Aside: you’ll be pleased to know that my thumbs have made a full recovery following some hammer-related accidents.
As at all times, we’re attempting to construct essentially the most “accurate” model possible. On this case, we’re specializing in forecasting so we’ll prioritise models which might produce essentially the most accurate estimates of future time series values.
I’ve previously written at length about using a regression approach based on Meta’s Prophet methodology. In those articles, I explained why I selected to make use of the LASSO model: non-stationarity made using classic methods difficult, the necessity for sensible extrapolation ruled out tree-based approaches, and the will for simplicity and explainability excluded any type of neural network.
But I also hinted at using mixed model forms — that’s, modelling each of the time series components using a unique model form, and mixing the output of the…