The well known definition of a long term forecasting model is based on the length of the forecast horizon. In case of long term, it is usually agreed to be in the order of years (ahead of the most recent time of data).

I have observed some academic forecasting papers (will share if required), particularly those relying on ML techniques such as LSTM for high temporal resolution building energy prediction quoting their model as long term forecasting while in fact their

  1. forecast horizon is just months ahead;
  2. training period is in the order of years.

Is this just a confusion among researchers or does it show a tendency of ML researchers dealing with high granular data to just pick up few months and demonstrate that their model works fine?

I do not intend any offence to any academic community. Please share your insights.

  • $\begingroup$ What is considered long term I think tends to depend on the frequency of the forecast being done. If we have to forecast a month ahead with hourly data that could be 720 points to forecast or with weakly data it could be 4 points. $\endgroup$ – user306502 Jan 22 at 7:01

On the one hand, this is rather opinion-based.

On the other hand, it depends.

Yes, a macroeconomic forecast for the next quarter would by no stretch of the imagination be "long-term". In macro, "long-term" is indeed on the order of years.

However, consider weather (not climate!) forecasting. Currently, we can forecast the weather out for about 10-14 days with any kind of skill ("skill" in the technical sense, i.e., improving on the pure climatological forecast). I would argue that a 14-day weather forecast is already "long-term" in this context. And note that weather forecasters typically do have multiple-year histories.

So: it depends on your forecasting field.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.