'BoxCox Transformation in auto.arima(): Does it also transform the residuals?
I am using the auto.arima() function in the forecast package in R. I performed a Box-Cox transformation (lambda = 0.02492832, if you're curious). My data are on the order of 10^9 and is exhibiting increasing variance after differencing twice, so I think B-C is appropriate. Strangely, the residuals are on the order of 10^-2. Not sure if I have discovered a crystal ball or if I'm missing something in the way residuals are calculated when using a B-C transformation in auto.arima(). Are the residuals also transformed?
Solution 1:[1]
The residuals are on the scale of the transformed data. If you want to compute data - fitted instead, use fitted() to obtain the fitted values.
Solution 2:[2]
If the variance is equal to k, the lambda component of the residual is the coefficient of k. Thus lambda(k)=10^9. There is something missing in the box-cox transformation which is generating increasing variances. It is the forecast ''k''. Yes, the residuals are transformed. We simply 10^-2 to the forecast k. This value generates a new forecast.
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|---|
| Solution 1 | Rob Hyndman |
| Solution 2 | ouflak |