Hi all! I'm using Power BI to produce a forecast for the next 5 years, based on monthly volume data since 2007. My chart of the forecast looks like this (green line is actual monthly data, dotted black line is trend line and solid black line is forecast):
As you can see, a trendline on the data shows a gradual upward trend. However the 5 year forecast shows the exact same values each month for each year - e.g. Jan-19, 20, 21, 22, 23 all show the same forecast value. All the Februarys show the same forecast value, etc. It's just the same pattern repeated 5 times. Is this what is meant to be shown? Unless I'm missing something obvious, I would think that Power BI would show the forecast data increasing each year based on the existing trend?
PS I have set seasonality to '12' as my data is monthly.
I am having the same issue even with 4 months of data and only forecasting one month of data. If I use the Anamolies option istead of forecase, it gives a bit more accurate results when it analyses the sales histroical data. The forecase option is far from reailty, I think the problem lies with the type of data we are feeding into Power BI, it might less fluctuations in the data to be able to tell the real forecast.
Please leave the "Seasonality" option empty for a test.
Hi Yuliana. Thanks for your reply. If I leave the Seasonality option blank (so it resorts to the 'auto' setting), I get the following chart:
This forecast seems to more closely match the trend of the data, but I now lose any variation in the data and I just get a straight line!
Hi All. Just a small update on this issue. I've now tried experimenting with a date slicer on the dataset and I've noticed that if I exclude a portion of the early data, at a certain point Power BI will give me a forecast which looks more sensible. For example if I only look at monthly volumes from Sep-2011 onwards, I get the following chart:
At least now, my forecast is showing variation and is following the trendline (dotted line). My theory is that Power BI will only apply a similar trend to the forecast data if the R-squared value is above a certain value? Or if the correlation coefficient is above a certain value?
Just for comparison, here again is what the forecast looks like if I reset the slicer (using the entire dataset):
Does this seem consistent with anyone elses experience with forecasting in Power BI please?
Hmm, any chance you could share some data, would be interesting to play around with this.
Hi Greg. Thanks for your reply. Unfortunately I'm unable to share that data although it's just based on a simple table of month names, e.g. Jan-07, Feb-07, etc and a figure for each month. Are you able to replicate something similar to see if you get the same strange forecast, i.e, is the forecast the same pattern for each projected year?
Seems like a lot of typing...
Vote for your favorite t-shirt design now through March 28.
Find out more about the March 2023 update.
Find out more about the online and in person events happening in March!