Flat forecasts

About once a week someone will tell me there is a bug in my forecast package for R because it gives forecasts that are the same for all future horizons. To save answering the same question repeatedly, here is my response.

A point forecast is (usually) the mean of the distribution of a future observation in the time series, conditional on the past observations of the time series. It is possible, even likely in some circumstances, that the future observations will have the same mean and then the forecast function is flat.

  • A random walk model will return a flat forecast function (equal to the last observed value of the series).
  • An ETS(A,N,N) model will return a flat forecast function.
  • An iid model will return a flat forecast function (equal to the mean of the observed data).

This is not a bug. It is telling you something about the time series — namely that there is no trend, no seasonality, and insufficient temporal dynamics to allow the future observations to have different conditional means.

I discussed this once with another consultant and he told me that he sometimes adds some random noise to his forecasts, just to stop his clients questioning the flat forecast functions. Unfortunately, that increases the forecast error, but he thought it was better to give them what they wanted rather than what was best!

Related Posts:

  • buggyfunbunny

    There’s an olde wive’s tale (or may be old crony’s) about the consultant and the client, to wit…

    client: Mr. Consultant, please tell me the time! I don’t know what time it is!
    consultant: Client, give me your hand, there there, all will be well. [client proffers left hand, consultant pushes back sleeve, revealing watch.] Client, it is 10:07 AM
    client: Ah!!! Thank you so much. I was so worried I knew not the time.

  • ericstrom

    Rob, would a compromise be possible ? Forecast function provides a note as to why the flat forecast (e.g. note the flat forecast due to random walk model being most appropriate).

    By the way, great book and great packages. Thanks much !

  • dan_

    Rob, this is an excellent statement. In addition it is so clear and so to the point, simply. Thank you for posting it.

  • dp_planner

    I have just come across this article. I provide forecasts to clients for inventory optimization. When it comes to client, you have to sell the forecast. The two main points of sell is credibility and accuracy. For a long term out look (more than 2-3 months out), a flat forecast is not credible since it is there has to be some type of variation in the business going forward. Having said that, a flat forecast for a current month forecast, is the most credible (from the clients perspective). That’s the constant battle with the client.