My opinion is that using simple stupid and open formats is the better choice. The binary closed office only document format would be my last choice if I have other better choices. The binary format also has the disadvantage that even different office versions display it differently in contrast to PDF that has more consistent across different viewers and viewer versions. The pdf can be a better choice in this case too, not to mention that there are a lot of free (and platform independent/opensource) pdf generator libraries. Free and simple stupid solutions are generally better, as a "richtext" document format I also like using a subset of html or rtf that is unfortunately not an option this time but I would ask the client about the html because that would have only advantage: easy editing/modification, there is a viewer basically on every machine/smartphone...
pdf generator libs:
http://www.jagpdf.org/downloads.htm[
^]
http://sourceforge.net/projects/libharu/[
^]
With google you can find plenty of them.
Another option when I want to exploit the comfort provided by office products: The Office 2003 suite introduced some xml formats that are very easy to parse by any program!
http://en.wikipedia.org/wiki/Microsoft_Office_XML_formats[
^]
What we used is the Excel 2003 XML format. We saved out a few tables, analized the resulting xml and then we generated such xmls from our program and we allwed the user to edit the xmls and the program was able to read it back. Note that the xml is full of formatting and totally unneeded elements that don't necessarily have to be there when the xml is generated by your program. You can try and use the same trick with the Word 2003 XML format if you are interested.