I personally feel that the size of the mape depends on three factors:
1, depends on the variability of data, for example, if you have two normal distribution, the mean is zero, then two variance of a big, a small, you can try the two distribution to generate some random number, your forecast is 0, but you will find that the variance of mape. 2. Mape depends on your model or prediction. Assuming that you now have only one distribution with a mean of 0, if your prediction is 0 it should be smaller than the mape with a prediction of 1. 3. Mape depends on the size of the number in the data. For example, if you have two data, one is 100 and one is 1, your prediction is 101 and 2 respectively, the error is 1, but the mape is large and small. Therefore, I think mape can only be used to evaluate the same set of data from different models. For example, for the same set of data, the mape given by model a is smaller than that given by model b, so the conclusion is that model a will be better. But if I just say Mape =10%, I can’t tell if the model is good or bad.