On XBRL again!

I have a hypothesis why XBRL has not been as successful as it should have been. I have said that might be a possibility many times before, on AECM and at many conferences where many XBRL advocates, myself one of them, spoke.

When XBRL was designed, the initial taxonomies were based on the audit manuals of the big accounting firms. Accountants have a fetish for numbers, and therefore the taxonomy was far too numbers oriented, and had a lot less semantic data than what the users were demanding. The designers were looking through a rear-view mirror and not what was happening in the real world.

I had suggested over a decade ago that the taxonomies should be empirically derived, ie., based on a statistical analysis of the text in the standards, news reports, and footnotes. Unfortunately, I was not able to convince any one. I applied for funding many times (NCAIR, NSF, PWC, ...), but was denied. Ultimately, I wanted to do it on my own, but around 2005 I got stuck with administrative responsibilities as department chair and I had to put the project on a back burner. Then I moved on to Computing, discouraged that the possibility of research funding in accounting (other than navel-gazing)  was non-existent. I also was unable to convince my doctoral students that it is a great topic.

I remember watching the roundtable of SEC a few years ago when the CEOs, especially Nooyi  (Pepsi), flatly declared that they would not be interested in implementing the footnotes part of the taxonomies. One only has to look at the taxonomy to realise why not.

The analysts have a way of statistically analysing the numbers in financial statements to squeeze out of them the last ounce of information; you can not squeeze blood out of a turnip. On the other hand, analysts would love a way to extract information out of the footnotes. Studies have suggested that the accuracy of  fraud classification using the textual data in the footnotes and other non-numeric information may be superior.

It is tragic that XBRL chose to bark the wrong tree. Even now it is not too late, but if we choose to redo the taxonomies, it will be a major disruption. However, the world has moved on. Semantic web was just an idea those days, but is now close to reality. The same analysis I am suggesting can enable us to build empirically determined ontologies for accunting. And that will be an exciting development since it also facilitates communicating systems to operate with less human intervention.