0

I am not an expert in statistics, and I would appreciate your help with the following.

I have a list of actual prices at different dates in the past:

Date1, Price1

Date2, Price2,

Date3, Price3,

...

Date100, Price100

Then I use two methods/models to estimate/predict prices for those different dates, and thus for each of those 100 dates, I have 100 prices that model 1 predicts and 100 prices that model 2 predicts. I would like to see whether model 1 is better than model 2. So, I calculate 100 absolute errors for model 1 and 100 absolute errors for model 2. I would like to test if absolute errors for model 1 are better (lower) than absolute errors for model 2. Since those absolute errors are not normally distributed, I cannot use the T test. I guess I can use the wilcox test: https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/wilcox.test

Would it be correct to write:

wilcox.test(AbsErrorsModel1, AbsErrorsModel2, alternative = "greater")

And if yes, how can I see if the answer is that model 1 is significantly better? Also, what is paired? Should it be set to True or False in my case?

Thank you! Have a great day.

Avocado
  • 70
  • 8
  • 1
    People would me more inclined to answer if you provide a small reproducible example with the function call and its output – Bill O'Brien Mar 19 '22 at 19:58
  • Appears to be a paired study design (so you would probably need `paired=TRUE` in the parameters) and therefore the assumption that the distribution of the abs-errors being non-normal may not invalidate the use of an ordinary t-test. Only if the paired differences were non-normal would there be a concern and even then the pair t.test might still be valid. – IRTFM Mar 19 '22 at 20:46

0 Answers0