I have a question about the R^2 value equation given on page 112 (Forecasting Principles and Practice) by Dr. Hyndman.

I am working with a dataset with 100 samples and the R^2 using enter image description here gives me a value more than 1 which does not make sense. I used the alternative way to calculate R^2 and that was the square of correlation coefficient between y and y_hat. I also used the third method to calculate the R^2 and that was 1- SS_{res}/SS_{total} and they all gave me different numbers. I really appreciate if someone can help.

Thank you.

  • 1
    $\begingroup$ I marked this as a duplicate of another thread dealing with negative $R^2$ because this is a similar problem and the other thread gives detail answers to it. TL;DR: $R^2$ is guaranteed to be in [0,1] and meaningful in terms of variance explained only for linear regression. $\endgroup$ – Tim Oct 14 '19 at 18:17

Browse other questions tagged or ask your own question.