The FSA and TSE have been assiduous in encouraging more engagement between investors and Japanese companies, and in highlighting the problems raised by the ever-increasing share of funds invested on a passive basis in the Japanese market – which is leading to a sort of “hollowing out” of meaningful feedback from institutional investors. I would encourage anyone who reads Japanese to read the most recent Action Plan for corporate governance, especially including the reports by the Secretariat in the FSA’s May 16th meeting. This is very commendable.
On the other hand, there is a stark contradiction between this stance and a big defect in the machine-readability of the Corporate Governance Reports (CG Reports) submitted by Japanese companies to the JPX/TSE, which is regulated by the FSA . The defect renders a major portion of these reports almost entirely useless for rigorous analysis by computers… even though I pointed it out some six years ago. In a word, the 11 (or more) different “disclosure items” required to included in CG Reports, which account for close to half of the meaningful information in each report, are all mashed together into one XBRL “barrel” that does not even have a standardized format.
This means they are not separable from one another, and therefore cannot be added as separate items in a database. In terms of machine readability, important policies for 11 different topics such as “nominations”, “compensation”, “cross-shareholdings,”, “related party transactions”, or “director training” are all just jumbled together in a a single 11,000-character data field with a single XBRL tag… a barrel which is thereby rendered machine-un-readable. Because it its nearly impossible (except by hand) to separate out and compare each of these policies between companies, very few investors do that in a rigorous manner. Knowing that rigorous analysis is not being done, many companies’ disclosure is low-quality and vague, because they know that few investors are comparing it in depth using modern text-analysis techniques or even simple ranking methods.
It is a vicious circle. Data quality remains low because few investors are comparing it as much as they would if the process was easier and quality was higher. From an investor’s point of view, why invest huge resources in analyzing data that is low-quality and extremely hard to collect and compare? From a company’s point of view, why take the time to give meaningful details if no one ever asks about it?
Obviously, engagement can only be effective if investors can first analyze companies and compare them to many others, and then have dialogue with each company. Otherwise, most engagement time will just be spent in confirming (verbally only) what the company “says” its practices are, making it very inefficient. In today’s AI and computer-driven world, this analysis is not all done by the human eye, in the first stage. Computers do the initial work, that of collecting and normalizing the data — and if possible, comparing it. Therefore, analyzing data first requires that it be easily (a) collectible and (b) separable into different governance practices and topics.
When I proposed the CGC in 2013-2014, enhancing the quality and quantity of data that reveals “what is going on inside the black box of each company’s board”, and making comparable, was one of my major goals. In fact, this concept was clearly set forth in the very first line of my presentation to the LDP Subcommittee in Feb, 2014. To achieve that goal in a “big data” world (and in a country with 3,700 listed companies), requires that the aspect of (b), “separable” be adhered to for each different data item. Otherwise, analysis is impossible.
But eight years, later, my vision has still not been realized, even though adding 11 additional XBRL tags would be a very simple, low-cost thing for the JPX/TSE to do. I have mentioned this problem to the TSE, parliamentarians, and the FSA…. even Prime Minister Kishida. But the issue is never addressed.
To his great credit, Administrative Reform Minister Taro Kono (in his own words) has “declared war on hanko and fax machines”. And on a relative basis, the FSA has generally been a global leader in the use of XBRL based standard taxonomy so as to enable much better machine-readability. As Japan surges forward with its policy to “digitalize” , I am hoping that Mr. Kono will meet with the FSA and also declare war on “data that is disclosed but in a format that cannot be digitally analyzed”.
Unless it can be digitally analyzed – which would be easy to make possible (!) — such disclosure feels like a waste of time for market participants, and reduces trust in policies and their effectiveness.
Nicholas Benes
PS and update – Please note that this is NOT a problem that can be solved using generative AI, because within the single-XBRL-tag “disclosure items” barrel, each company uses its own format, and many companies also insert links to documents on their web sites, and those pages are totally different from one another in format, ordering, and topics. Generative AI would generate output which, – even if it were able to separate out 70% of items, albeit with unrelated text attached – would be worded differently from the original disclosure and therefore, in many cases, even worse “disclosure” than what we have now. It would also be subject to the risk of hallucinations. For any specific company, it would be unreliable. These issues are even greater because the base data is not very large, in the context of AI. Moreover, investors cannot possibly go back to companies complaining about disclosure that is worded differently from what the company disclosed in the first place, with different (generalized) meaning and nuances.