I found one the other day in the area of finance. The Chinese researchers claimed to have discovered a small tweak to a long established indicator which they de

I found one the other day in the area of finance. The Chinese researchers claimed to have discovered a small tweak to a long established indicator which they described as giving a remarkable increase in r-squared value across a cross section of markets. Sounds great, who wouldn't want to use this? So I implemented and find that their increase was due entirely to applying a log transform of the input variables. The resulting clusters were tighter, but it had zero predictive capability. Very disappointing but in my experience, this is not uncommon.