What is “forecasting”? Many people will have a ready answer to this question, but I would argue that not many have spent enough time thinking about it. Should we spend a couple of minutes of our time today to do that?

Straight to the point: my answer to the question comes to the following definition: A forecast is a scientifically justified assertion about possible states of an object in the future (details here. This definition was formulated by my father, Sergey Svetunkov, who spent some time of his academic carrier working on theory of forecasting). Forecasting then is just a process of making forecasts. Note that the word “scientific” here is used in a wide sense, i.e. something that is either based on real science or has some good logical rationale behind it.

The main reason why I like this definition is because it separates forecasting from foretelling, or fortune telling, or guessing, or any other non-scientific activity. In the latter, we would be using a magic ball to say what would be the demand for our product next week. This is apparently not scientific and cannot be considered as a proper “forecasting”. It is just guessing with some equipment.

A person who has done forecasting in practice might wonder how the judgmental forecasting aligns with this definition. After all, demand planners often either amend existing forecasts or override them introducing numbers purely based on their judgment. Well, this is actually the situation where the proper definition becomes useful. If you ask a demand planner why they did the adjustment and cannot get a reasonable answer (e.g., “the line was not wobbly enough”) then this is just guessing, there is no scientific reasoning behind it. If the answer is motivated by some knowledge about the forecasting process (e.g. there was some external information that was not taken into account by the model) then this is forecasting. As you see, the definition allows separating the two activities and can potentially help in improving the whole demand planning process.

One other important aspect of this definition is the absence of the word “probability”, which you can find in some other definitions in the literature. This is because there are some methods that sidestep probability all together. For example, Delphi method can be used for long term forecasting, but typically focuses on forecasting of general tendencies (structure) in the data, ignoring the uncertainty around it. So the definition above is wide enough to encompass both objective (statistical/ML) and subjective methods.

Finally, you might ask me, why bother? My answer to this is that clear definitions allow making more adequate decisions. And having correct definitions is also important if we want to understand each other. If we do not have them then we might be talking about different things, never being able to find common ground.

But what about you? Do you have a definition of “forecasting” that you like the most and why?

And here is an image of a forecaster tired of coming up with definitions.