Microsoft Office has, since 2007, used a formula layout algorithm based on the algorithm Knuth described in appendix G of the TeXbook. But there's an apparent divergence in the final output from the description. The algorithm that MSOffice uses is allegedly based on that as described in the Open Type Math Table. Precise details of this can be found in the paper Open Type Illuminated by Ulrik Vieth (obtainable from the Netherlands' TUG at http://www.ntg.nl/maps/38/03.pdf).
For a screen resolution of 96dpi and a font size of 11pt, the algorithm says to shift the superscript baseline relative to the original baseline by 5 pixels, and the subscript by 3 pixels (up and down, respectively). (For a more precise description of what the algorithm says in this case, please see http://i.stack.imgur.com/ckGL5.png.)
The figure below contains a blown up view of various renderings, with Word's view on the left, the algorithmic reconstruction in the middle, and TeX's rendering on the right:
The three horizontal lines are not from the figures, but are added to highlight the difference and parallelisms between them.
As can be seen in the image, the result obtained for the formula
$a^y_y$ following the algorithm described above is different from the result obtained from MS Word. As can also be seen, the algorithm conforms to TeX's implementation fairly well (given the inaccuracies inherent in producing these diagrams, "fairly well" could mean "precisely" here).
Is there another algorithm at work here? Has Knuth's algorithm been superseded? Or is it simply that when placing the superscript
y then MSWord chooses to place the bottom of the
y at the superscript baseline instead of the baseline of the