How to read MASE for comparing forecast accuracy

I am recently switched from using MAPE to MASE for comparing forecast accuracy after reading the benefits of MASE over MAPE. But I am not sure how to read the values in simple terms. For example, MAPE of 6.7 would mean the error between actual values and forecasted values is only 6.7%. But if we have MASE of 0.53, how do we read that in plain English to explain to non-technical team (instead of using terms like "in-sample" or "naive forecast")?

Thanks for your help!

The excellent Forecasting: Principles and Practice, 3rd Edition has a very lucid and technically complete explanation of this that I would highly recommend. 5.8 Evaluating point forecast accuracy | Forecasting: Principles and Practice (3rd ed)

The distinct advantage of MAPE is the percentage interpretation; this attractive feature is built-in. By contrast, MASE is [S for scaled] compared to a naive or seasonal naive forecast; for each individual forecast, numbers greater than one (in absolute value) imply a worse forecast than the NAIVE/SNAIVE and those less than one (in absolute value) imply a more accurate forecast. We then take the mean of those ratios with the absolute value transformation guaranteeing that each component is non-negative.

It is worth noting that one could derive a percentage for MASE by creating a table based on the frequency of ratios between -1 and 1 [or 0 and 1 post abs()] and it would answer what proportion of the forecasts are better than NAIVE/SNAIVE though sans a metric giving any idea of by how much; the how much, on average in the metric of y, is the exact quantity provided by MASE.

1 Like

Thanks @rwalker ! I am trying to find ways to explain this to non-technical team where they can look at MASE of 0.53 as 53% or 47% accuracy, which of course is not the case and at the same time it will be a challenge to explain about the comparison with naive forecast. Just trying to find simpler way to explain it.

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.