0

I have a set of log-return data for a commodity and am unable to identify an appropriate ARMA model. I used auto.arima() function, and the optimized model is (4,0,4) with zero mean. However, when I run the Arima model, I get a warning message that there is a convergence problem with an optim code = 1. I wrote an iterative algorithm in R to identify the best model by minimizing AIC value, which gave me a model of (12,0,7) without mean and (12,0,7) with mean. The AIC score without mean is lower. I run the program for a max of 12 in each AR and MA to identify the best model. My algorithm keeps away from selecting any model that has convergence issues or NaNs in the standard errors for any of the iterative models. With the models that my algorithm selected, I note that there is a serial correlation present in the residuals, as there are outliers in the data. I request help with the following:

  1. Should I winsorize the log-return data to reduce the impact of outliers (OR)
  2. Should I use tsclean() to transform the outliers?

My objective is to obtain a model that has zero auto-correlation in the residuals.

Richard Hardy
  • 67,272
Jyoti Nair
  • 19
  • I feel like we have discussed this before in the comments and perhaps you even concluded that you would winsorize your data, but this post looks completely fresh (no comments or answers). Did I mistake this for another post form another user? – Richard Hardy Mar 25 '23 at 12:39
  • Yes, @Richard Hardy. We have discussed this in a different post, but by me. I received your answer in that post. Thanks very much. – Jyoti Nair Mar 26 '23 at 07:15
  • Would it make sense to close this one as a duplicate of the other one? I cannot find it, though. Did you post it from another account? – Richard Hardy Mar 26 '23 at 13:37
  • No from the same account... Here's the link:https://stats.stackexchange.com/questions/609082/autocorrelation-in-residuals-of-mean-model-to-be-used-in-a-garch-model/609095?noredirect=1#comment1131466_609095 – Jyoti Nair Mar 27 '23 at 23:59
  • How do we close a thread? – Jyoti Nair Mar 27 '23 at 23:59
  • Thanks for closing. Yes my questions have been answered in another post. – Jyoti Nair Mar 28 '23 at 08:07
  • Thank you, that is good to hear. – Richard Hardy Mar 28 '23 at 08:18

0 Answers0