Confusion about metrics in a Deep AR+ forecasting model

When I read the documentation on the metrics, and we can just use MAPE as an example, your value of 1.5993 means that the absolute value of the difference between the predicted and observed is almost 1.6 times that of the observed value, which is probably not good. If you just focus on the numerator of that equation, if the observed and predicted are close, that value will be small, and when you divide by the observed, it will be even smaller, less than 1.

MAPE is a mean, so on the average, your predictions are not very close to the real/true/observed values

Reference: https://docs.aws.amazon.com/forecast/latest/dg/metrics.html

Yeah that is what I thought having read the same docs; and a machine learning friend tells me the same.

I asked Google too, since I built a similar model on Vertex and the metrics were similar, and I think they are incorrect in telling me that the errors are the lesser of the options!