Not that it would drastically change the conclusions, but do the numbers for "fund i" include the forms that say "fund ii" etc (by virtue of the fact that "fund i" is a substring of "fund ii" etc)?
You were actually right. I went and checked to see that some (not all) values were double counted. I've updated the graph to reflect this, and added a note. The trend remains identical, despite this change. Thanks for inspiring me to double check.