Digital Financial Reporting

Not sure how I missed this, but I was just made aware of this MIT project.  The blog post describes the MIT prototype as:

Rong is working on a software prototype for translating financial reports from the XBRL format into the OWL language, the idea being to preserve and enhance the implicit semantics in XBRL and enable the logic model of financial reports, according to a soon-to-be-published paper co-authored by Rong.

Personally, I am skeptical but certainly open to being proven wrong.  Why am I skeptical?  Well, read this blog post and be sure to read the PDF pointed to on this post.

The bottom line is this: it is impossible to take something which has not modeled correctly from a semantic perspective, convert the syntax, and magically have the proper semantics.  Now, if the semantics of the SEC XBRL financial filings were correct you could convert all that information into whatever syntax you want including OWL/RDF.

Further, while AI (artificial intelligence) might be helpful in de-tanging the semantic mess of SEC XBRL financial filings, if the semantics are correct, why would you need AI to make the information useful?

Now, I admit that I could be totally missing the point here.  I do agree that someone will be able to use all that SEC XBRL information to compete with Bloomberg, Thomson Reuters, CompuStat, Hoovers and all those other data aggregators who are manually collecting information.  So sure, someone may make money converting the current manual process into some sophisticated process which uses AI and other technologies to over come the poorly modeled SEC XBRL financial filings.  Then, some different company can make money selling the data to others.

But I see two problems with that: (a) is that what the SEC meant when they said investors can more easily use the information?  (b) similar types of processes need to be created for each implementation of XBRL?

Seems to me that the best thing to do is to not fight the symptoms of the problem but rather to fix the problem.  Model the information more appropriately and THEN use all that nifty RDF/OWL and AI stuff to expand what you can do with the well modeled information.

What do you think?

I am currious to see what the folks at MIT come up with. And frankly, I do hope that they prove me wrong.