Timeline for Interpretation of mean absolute scaled error (MASE)
Current License: CC BY-SA 3.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 7, 2016 at 17:23 | history | edited | Stephan Kolassa |
edited tags
|
|
| Jun 9, 2015 at 16:17 | answer | added | denis | timeline score: 5 | |
| Nov 21, 2014 at 20:44 | comment | added | Stephan Kolassa | Related: stats.stackexchange.com/questions/124955/… | |
| Nov 17, 2014 at 13:33 | vote | accept | Richard Hardy | ||
| Nov 17, 2014 at 13:25 | history | edited | Stephan Kolassa |
added time-series tag
|
|
| Nov 17, 2014 at 13:24 | answer | added | Stephan Kolassa | timeline score: 23 | |
| Nov 17, 2014 at 12:57 | comment | added | Richard Hardy | @StephanKolassa: I skimmed through the paper and did not find a good explanation. Perhaps I should read it more carefully. But my questions is intended to be more general than that. I am not particularly interested in that instance, I just presented it as an example. I am seeking intuition about $MASE$. | |
| Nov 17, 2014 at 12:54 | comment | added | Richard Hardy | @AlecosPapadopoulos: you are right. However, I meant the out-of-sample data being quite different from the in-sample-data as a necessary, but not a sufficient condition for expecting $MASE>>1$. Perhaps I did not express myself correctly. | |
| Nov 17, 2014 at 12:22 | comment | added | Alecos Papadopoulos | I am a bit puzzled by "your guess": a structural change would mean that the sophisticated forecast would be based on partly irrelevant past data, indeed. But how a structural break would affect a "no-change" forecast depends on the break. If for example we are looking at a random walk with drift, and the structural break means that the drift, the constant term, just got lower, then the "no-change" forecast will perform better after the break, than before it. | |
| Nov 17, 2014 at 11:32 | history | edited | Stephan Kolassa | CC BY-SA 3.0 |
added a link for a cited paper
|
| Nov 17, 2014 at 11:28 | comment | added | Stephan Kolassa | In his blog post, Rob notes where this benchmark comes from: "These thresholds are the best performing methods in the analysis of these data described in Athanasopoulos et al (2010)." Have you looked at the Athanosopoulos paper? | |
| Nov 17, 2014 at 11:24 | history | asked | Richard Hardy | CC BY-SA 3.0 |