apparently, the man rivals any member of this forum for deserving recognition as "most likely to be a computer program"
Which is unfortunate, actually! I think we were all hoping that he'd be a slower thinker, more humanly curious, and instead say "hmm...I don't know why there are discrepancies in the results. But thank you for pointing them out! For my next research project I will do a meta-analysis of SWR research to determine how and why different sources come to different results."
Because, really, he's probably more suited to do the job than any of us!
And it seems like it's a pretty important topic that deserves more exploration. Yeah, we know that any SWR number that gets spit out needs some error bars around it, but given how often SWR numbers are reported, it would seem important to at least understand how replicable the research is, and to what extent different data sources or methodologies contribute to the size of the error bars.
FIRECalc gives the most insight into the effects of using different sources for "bonds". With a 60/40 portfolio, the safemax rate for each of the four "bond" options is:
Commercial Paper: 3.77%
Long Interest Rate: 3.67%
30 Year Treasury: 3.63%
5 Year Treasury: 3.71%
That's a smaller range than the range between the FIRECalc results and Kitces's results, so it suggests that even if "short-term government bonds" were used instead, they wouldn't fully explain the difference. So there must be more to it, and if different researches use different methodologies, there's a good chance that at least one of them are "wrong".