It doesn't get to the history of the term. If the regression has just one continuous predictor, then the r-squared is algebraically identical to the squared Pearson correlation coefficient, though the interpretation (causation vs association) is different.
I don't know what Excel is really doing ... but ... no one should be using Excel for statistical calculations. There have been paper after paper showing flaws in Excel's algorithm. Sometimes, it isn't able to compute variances properly.
From "On the accuracy of statistical procedures in Microsoft Excel 2007", B.D. McCullough and and David A. Heiser, Computational Statistics & Data Analysis
Volume 52, Issue 10, 15 June 2008, Pages 4570-4578.
"The statistical literature has regularly identified flaws in Excel’s statistical procedures at least since Sawitzki (1994), and Microsoft has repeatedly proved itself incapable of providing reliable statistical functionality. It is little wonder that introductory texts on statistics warn students not to use Excel when the results matter (e.g., Keller (2001) and Levine et al. (2002))."
Bad enough so that "introductory texts on statistics warn students not to use Excel when the results matter". In other words, we're not talking about advanced methods in bioinformatics that involve millions of data points ... we are talking about every day statistics.
Do your results matter?
Message was edited by: Paige