In NLP, the task of "predicting the next word" is an example of self-supervised learning. An essential part is that the label can be computed programmaticaly and does not require explicit human effort. Typically, this task is not an end in itself, but is used to learn a representation and is succeeded by another task (e.g., sentiment analysis or text classification).
Similarly, in forecasting it is very common to use information of a univariate time series $x_{t}$ up to time $t$, in order to predict some quantity in the future (e.g., $x_{t+1}$). The labeling can also be achieved programmatically and does not require human effort. A difference with NLP is that this is usually the final objective and an end in itself. An example would be using sales data up to time $t$ to predict future sales.
Given the obvious similarity with "predict the next word" type of tasks from NLP, should we also categorize this prototypical forecasting problem as self-supervised learning?
