Member-only story
Time Series Chapter 7 : Autocorrelation & Partial Autocorrelation
Understanding how past values influence the future
In the last lecture, we learned how to make our first forecasts using simple baseline models like Naïve, Mean, and Drift. These gave us a sense of how predictions behave and why forecasting needs careful evaluation. Now, before we move on to more advanced models, we need to understand a very important question:
“How much does the past influence the future?”
Not all days affect tomorrow equally, yesterday might matter a lot, last week might matter a little, and last month might have almost no influence at all. To see these relationships clearly, we use two simple tools:
- ACF (Autocorrelation Function) shows how strongly a series is related to its past values
- PACF (Partial Autocorrelation Function) shows direct influence after removing effects of earlier lags
These plots help us understand the “memory” of the data. Once we can see how far back the influence goes, choosing the right forecasting model becomes much easier.
This lecture is all about making those hidden relationships visible with simple visual intuition, not heavy math.
