0

I am analysing data on hourly electricity prices to try to do some forecasting, in my dataset I have mostly positive values, but some negative, around 20 per region for the two year duration. I want to do autoregressive (AR) forecast of this data, that is, trying to explain future prices based on previous prices.

I might want to add other variables such as weather forecasts later but for now I will concentrate on what I described above. Descriptive statistics

Here is a example of the distribution: Histogram

I want to transform this data to be able to do forecasts but I can't do log since I have negative values. I have read that i could add a constant (x+1) -> min. x is close to 0, or min. x is 1 and then do the log, I have also read that I could do a fisher transformation to get a better distribution. I am turning to you for some suggestions on what would be best practise in this case.

If you need some other measures I just ask and I will provide.

All the best

Simon Rydstedt
  • 35
  • this is not really a statistical question. Please think about what is the relationship between prices and the other variables that you are interested in. (you don't explain whether price is an input or output) [ and update the question with those details]. There is no need to log data just because it is skewed. – seanv507 Sep 25 '22 at 09:48
  • 1
    I did edit my post now, I do not have other variables right now but I want to use a AR model to predict future prices based on previous prices. – Simon Rydstedt Sep 25 '22 at 10:09
  • the comments of the duplicate post should also be helpful. but essentially you should aim to get the right functional form. a quick google : https://mpra.ub.uni-muenchen.de/29958/ ..."Moreover, this transformation might be in general more appropriate for power prices than the log transformation, considering fundamentals of power price formation. Eventually, a thorough treatment of negative prices is indispensable since they significantly affect business" – seanv507 Sep 25 '22 at 13:26

0 Answers0