Time Series 101: MASE Forecast Accuracy

In this Time Series 101 video, we learn about forecast accuracy by examining the MASE, mean absolute scaled error, using visuals. MASE is a little bit different from the other forecast accuracy measures we have learned about in this playlist. It uses the MSE of the naïve forecast as a scaling factor. Thank you for watching, and enjoy!
#timeseries #timeseriesanalysis #statistics #datascience

[@joseparedes3765]

As i'm working on my thesis in energy demand forecasting, i decided to use the mase as a performance metric. Thing that comes to my mind now is: ok i'm comparing my forecasting performance to the performance of the naive forecast. And also great, i got something for example like 0.5 for mase on my forecast. So that only means that im better than the naive. What about the naive being quite bad. Being a little bit better than the naive isn't a great help in this case.

[@Wrapped01]

Do I understand this correctly that the number of records would just cancel out and we are basically just summing the error of our forecast divided by the error of the naive forecast?

[@stopworrying8850]

Hallo my teacher,
I study psychology and i need your help.
We want to see how does the mental health condition of refugees change over a period of one year. We collected the data and this over 3 . Time period: baseline , after 6 months, after 12 months.
Now i want to see if there is any difference in health condition over this period i means 3 times, what kind of tests is the best to choose for in SPSS??
And what kind of tests is better for only 2 period? I mean if i just want to see the difference between baseline and 12 months later or 6 months with 12 months i meant only 2 period of times?
Is anova good or GLN general linear model?
And how can i see the difference in every person in itself?
Also how cann I filter and choose only those persons, who i want to compare i mean only those, who participated in the whole experiment and not those, who participated only in baseline or 6 months?
The last question is how can i have a demographic characteristics table in SPSS?
I will be happy for your help . You have Videos on that will be even better 😍

[@Chrysaries]

Is the naïve approach feasible when predicting far into the future? Right now I'm forecasting 24 h into the future, so I don't have the timesteps necessary for this naive prediction. This makes me doubt that MASE>1 is by definition useless

[@michaelclerkin2972]

It really depends on what you are predicting. All forecasting methods can only forecast one time step ahead (This is arbitrary as a time step can be 1 day, 1 week, 1 year, 1 decade etc etc) into the future based on observational data otherwise you will be predicting off of other predictions.

[@chrisw4525]

@@michaelclerkin2972 not true there are sequence to sequence forecasting methods e.g. an encoder-decoder RNN