Closed brimoran closed 2 years ago
Thanks for finding this issue. Investigation in progress.
Great :)
After investigating this issue, I have found that the package behaviour is consistent with the assumptions of the distribution forecasts as twopiece normals. That is, the skewness parameter should be in (-1, 1). So, the only change that I plan to make to the package is to raise an appropriate warning when parameters out of range are provided. I wrote a brief explanation about this on my blog. Thanks again for highlighting this to me.
Interesting! Great blog post. Thanks for looking into this and for publishing your insight into what the out of range skewness implies.
After some communication with BoE, they confirmed that what is published in their recent spreadsheet as Skewness is not the actual $\gamma$ parameter. Instead, they publish Skewness = difference between the mean and the median of the distribution. This quantity indeed can be outside the interval (-1, 1). So, what I did was to change my package twopiece (release 1.3.1) to accept this quantity and then calculate the parameter $\gamma$. This will in turn solve the issue with the fancharts. Besides, I updated fanchart (release 1.0.0) to require twopiece>=1.3.1. So, you can just update fanchart and the correct twopiece will be installed. Once again, thanks for highlighting this issue! It was very helpful.
I love what you've done here.
I wanted to see if I could reproduce the latest BoE fanchart (August 2022) but I am encountering the attached error message.
I have played around a bit with the data to try to see what is causing the issue and I think it occurs when Skewness >= 1 as it is in the latest BoE projections for two periods