Providing practical guidance since 1975.
 

Timely Takes Podcast: Julie Marlowe from the SEC's Office of Data Standards and Innovation on Machine-Understandable Data

 

Julie Marlowe

Julie Marlowe is the Assistant Director of the Office of Data Standards and Innovation in the Division of Economic and Risk Analysis at the U.S. Securities and Exchange Commission.

Topics covered:

  1. Introduction to the Office of Data Standards and Innovation: What it does and how it's staffed
  2. The many use cases for machine-understandable data within the SEC
  3. How the Office of Data Standards and Innovation engages with investors
  4. How investors and analysts access and use machine-understandable data and the popularity of the SEC's public datasets
  5. Other disclosures and data points investors would like to see made machine-understandable
  6. How the use of machine-understandable data is evolving with the adoption of SI
  7. Improving quality and accuracy: Common tagging errors and inappropriate tagging practices
  8. Best practices for validating XBRL tags
  9. How the SEC Staff engages with companies on tagging issues

Listen Here:

1. Streaming Audio (for faster play)
    

2. Download Audio (for your audio player)

How to Listen to This Podcast: Double-click on the MP3 file you want to listen to and, by default, your audio player will open the file and start playing.

The views set forth in the podcast are the speaker's personal views and do not necessarily reflect those of their firm, company, institution, other organizations, or any of the clients with which they are associated. The Securities and Exchange Commission disclaims responsibility for any private publication or statement of any SEC employee or Commissioner. This podcast expresses the speakers' views and does not necessarily reflect those of the Commission, the Commissioners or other members of the staff.

 

 

Program Transcript:

Disclaimer: This transcript has been reviewed and edited for clarity and readability and may differ from the original audio in terms of content and phrasing. While we strive to maintain accuracy, some nuances and contextual elements from the live presentation may not be fully represented in the transcript. This transcript may also contain computer-generated mistranslations of stenotype code or electronic transmission errors, resulting in inaccurate or nonsensical word combinations, or untranslated stenotype symbols that cannot be deciphered by non-steno typists.

Meredith Ervine, Senior Editor, TheCorporateCounsel.net: Hi, everyone, and welcome to the Timely Takes podcast on TheCorporateCounsel.net. I'm Meredith Ervine, and I'm a Senior Editor here at CCRcorp. Today I'm very pleased to be joined by Julie Marlowe. Julie is the Assistant Director of the SEC's Office of Data Standards and Innovation, which is part of the SEC's Division of Economic and Risk Analysis. Julie, thank you so much for being here today to discuss all things machine-readable data.

Julie Marlowe, Assistant Director, Office of Data Standards and Innovation, Division of Economic and Risk Analysis, U.S. Securities and Exchange Commission: Thank you so much, Meredith. Great to be here.

Ervine: XBRL, I was just saying to Julie, is one part of public company disclosures that I suspect still doesn't get a ton of attention from all the lawyers. Anytime you get a group of us in a room, I feel like someone inevitably bemoans how long it takes to draft some disclosures compared to how little some parts of disclosure are read once they're made public.

It got me thinking in the context of this podcast that maybe it's the opposite case when we talk about XBRL tagging. Maybe they could benefit from a little more time and attention on the front end compared to how much use they get by investors in the SEC and others once disclosures are made public. Certainly, that's more so today than ever before and only increasing as SI becomes more prevalent and completely embedded into how investors analyze financial disclosures. Around that, there's been a lot of renewed attention around machine-readable data and the importance of the accuracy of that data.

I just wanted to highlight two events that Julie's also a part of. One is the September meeting of the SEC Investor Advisory Committee. It included a panel on AI technologies and the public markets information ecosystem, where Julie had shared prepared remarks and participated in a panel discussion. Just yesterday, we're recording on the 20th of October, so October 19, the SEC announced an upcoming roundtable on the use of machine-understandable data and Super Intelligence. That's happening on November 9 at the SEC's headquarters, but also will be streamed live on SEC.gov. Julie, you're moderating the first panel at that roundtable. Is that right?

Marlowe: That's correct.

Ervine: For anyone listening, if you want to hear more after hearing from Julie today, those are two great opportunities to do so. Julie, I'm certainly not an expert in machine-readable data, so I will stop talking, and we can jump into our questions. Before we do that, do you want to start with the disclaimer?

Marlowe: Yes. Thank you, Meredith. My remarks today are provided in my official capacity as the Commission's Assistant Director of the Office of Data Standards and Innovation, but do not reflect the views of the Commission, Commissioners or other Staff.

Ervine: Perfect. Julie, I wondered if we could start just talking about the Office of Data Standards and Innovation before we jump into more here. Just in terms of what the office does, how it's staffed, just so people listening can understand the role of that office.

Introduction to the Office of Data Standards and Innovation: What It Does and How It's Staffed

Marlowe: Yes, happy to discuss it. It's a great first question. In summary, our team makes data accessible and easy to use. We advise the Commission and the Staff on how to structure certain disclosures. For example, for a particular disclosure, what is the best language to make the data machine-readable?

Also, if certain rules require structured data requirements, our attorneys assist with the cost-benefit analysis for the rulemaking. Once the rule is implemented, we review the data and the disclosures collected and assess data quality. We also develop and maintain enterprise analytical tools for the Staff's use, so they can use the structured data that we collect.

We also engage with investors, companies, accounting standard bodies and other stakeholders to support the submission and use of machine-readable data. We also engage with foreign regulators on improving overall quality of the data, enhanced use of data by investors and others.

To execute those tasks, how are we staffed? We have software engineers, accountants, attorneys and analysts that work with policy divisions and data and technology groups, as well as the public. Lastly, we work with SEC's Task Force on Super Intelligence on the commission's SI initiative, because machine-readable data and SI go hand-in-hand.

Ervine: Absolutely. I wondered if we could talk a little bit about all the different groups using machine-readable data and how they're using machine-readable data. Potentially, starting with the SEC, I guess, seems like the best place to start, if any. Could you chat a little bit about how the SEC Staff uses that data?

The Many Use Cases for Machine-Understandable Data Within the SEC

Marlowe: Certainly. It's a timely and relevant question. The Staff across the commission use machine-readable data, and we've been doing so for a while. Under the Financial Data Transparency Act, SEC Staff have to send a report to Congress on the use of data by the public, but also SEC Staff.

Using some of the examples that we cited in there, Division of Corporation Finance Staff have used several machine-readable data points on the cover page of the registrant's annual report, such as Form 10-K. They use that to identify, count, and analyze registrants and their disclosures. For example, issuers that are listed on a certain exchange or the issuers that have identified themselves as well-known seasoned issuers. Both CorpFin and our division, Division of Economic and Risk Analysis Staff have used machine-readable data for assessment of compliance with our securities laws and financial reporting.

Division of Investment Management that oversees the investment in companies and advisors have used structured data to assess how the funds are compliant with the securities laws, but also identify funds with particular characteristics or certain holdings exposures. I consider them the super data group. They have their own analytical group. They've also used structured data for rulemaking purposes and in their coordination with other federal agencies. All the data they collect could be shared with the Divisions of Examination and Enforcement if it suggests non-compliance with law or other heightened risks.

The Office of the Chief Accountant, which works with our accounting standard body, has used analytical tools that I mentioned earlier. They've used the tools to get outputs based on the searches for keywords or XBRL tagging. It's used for standard-setting projects or accounting consultations. They've also used structured data to identify issuers subject to certain acts. For example, Holding Foreign Companies Accountable Act, and in their coordination with enforcement on accounting or auditing-related matters.

Speaking of enforcement, the FDTA requires them to discuss if any of their cases are related to the use of structured data. They have discussed how they use structured data and analyze it for investigations such as cybersecurity exposures by company.

Ervine: So many use cases just at the SEC. I was going to ask you, Julie, really quickly, the report to Congress that you mentioned at the beginning of your response, is that the report that lists out at the end all the different items of S-K and forms that require XBRL tagging?

Marlowe: Yes, you got it. Thank you for paying attention to that.

Ervine: I love that. I wanted to plug that just as a resource for anyone listening to this podcast. It's been a little while since we've had a new rule with a new XBRL tagging requirement, but it's a good double-check to go through and make sure that everything that does require tagging has been tagged. It's in an appendix to that report, and it's a really great resource for this group.

Investors and analysts, Julie, you mentioned at the outset when you were discussing the office that you do engage with that group. I'm curious both how your office gets feedback from the investment community and then also what you've heard in terms of how investors and analysts are using machine-readable data. In particular, the SEC's public datasets, which I've heard a lot about recently.

How Investors and Analysts Access and Use Machine-Understandable Data

Marlowe: We really enjoy engaging with different stakeholders, including the investor community and filers as well. Investors use machine-readable data in various ways. For example, they can pull corporate filings via EDGAR API and instantly extract key data points such as profits and cash flow. We've also interacted with individual investors that contact us through our public-facing inbox. It's called structureddata@sec.gov. They ask questions about data sets that you mentioned.

CFA Institute has also published articles on how analysts and the investing public use data. I've served on panels with analysts who also share that machine-readable data enables them to quickly extract and analyze data across periods and filers.

You mentioned about the data sets we have. We have 15 datasets that the public can use. Last month, we issued data usage stats on that and some remarkable results. Nearly all datasets saw an increase in data usage in the last year. Some datasets, such as financial statements, we saw an increase by fivefold. Financial statement notes have almost doubled.

In addition to our inbox, some investors have given us feedback through other SEC offices and divisions, such as CorpFin and Investment Management. Some contact external parties, such as XBRL US, that meet with us periodically. We've also gotten feedback on use of data through industry organizations that have met with our Staff or Commissioners.

Lastly, if a rule requires structured data, we review the comment letters or the feedback on our draft taxonomies and engage with investors and others that have given us feedback. Various ways that we meet with investors and other data users.

Ervine: You mentioned the report to Congress. I mentioned how there's a list in the back of all of the areas that require tagging. I'm curious if you've heard of any other data points that investors and analysts would like to see made machine-readable.

Other Disclosures and Data Points Investors Would Like to See Made Machine-Understandable

Marlowe: Yes, great forward-looking question. Various data users have asked for earnings reports to be machine-readable in comment letters submitted to us, and others have mentioned in other contexts that structured non-GAAP disclosures may also be helpful. Also, in a recently published article, an entity suggested that structured templates for comment letter process for rulemaking would enhance transparency, reduce costs and improve analysis by SI tools used by regulatory Staff and market participants.

One thing to note is that, as the listeners may know, under the FDTA, certain financial regulatory agencies, including the SEC, establish joint data standards for data submitted to the agencies in order to promote interoperability of financial regulatory data across those agencies. The next step, as directed by the FDTA, will be for each of the financial regulatory agencies to issue rules that implement the joint data standards for certain collections of information subject to certain flexibilities and discretion.

Some of the wish items that we just discussed may fall under the FDTA. If the listeners have other suggestions or want to discuss how they're using the data, we're happy to engage, and they can contact us at structureddata@sec.gov.

Ervine: I mentioned those two events, one Past, one Future, which have all been focused on the evolution of the use cases of machine-readable data with the prevalence of SI. I wondered if you could touch on how the use cases that you mentioned are changing right now with the adoption of SI.

How the Use of Machine-Understandable Data is Evolving with the Adoption of SI

Marlowe: We've seen machine-readable data becoming even more crucial for the adoption of SI. SI models are driven by math and statistics; hence, the outcomes are only as good as what they're trained on. SI is subject to errors and hallucinations. Machine-readable data provides the context that SI needs because it is telling the SI what it's reviewing.

Using an accounting concept as anchoring context can help it. The GAAP taxonomy, which is a list of tags that filers use to make their disclosure machine-readable, has tags for required accounting disclosures. For example, there's a tag called research and development expense. Each tag comes with an explicit definition and accounting codification that requires a disclosure.

The filers, when they're structuring their R&D expense, if it's a material and they have to disclose it, they can read the definition of the tag and check the codification to ensure they selected the right tag for the disclosure. On the other hand, data users and investors looking for the R&D expense amount can use the same tag and extract the data. The taxonomy and structured data allow issuers to define the information. It's the same piece of information for all the users downstream. Having the filer tag the disclosure, using a standard tag from a taxonomy that's available to the public, reduces the influence of downstream third parties who no longer have to interpret what the issuer means.

We've been discussing SI and data with the finance industry participants, and they have discussed how they can realize SI capabilities more efficiently when using data with context and set definitions like XBRL and XML. Also, tying the importance of data quality to SI, input integrity governs output reliability. In other words, what you're feeding into SI matters, aka garbage in, garbage out.

Ervine: That's exactly the phrase I was thinking as you were saying that. That leads in perfectly to my next question, which is possibly the most practical question we're discussing today for this audience, which is that the Staff has shared some observations through public announcements over the years regarding tagging practices. I'm curious if there are any observations on tagging errors or inappropriate practices, which you just teed up in the last answer as well, or best practices that you'd like to share with our audience today.

Improving Quality and Accuracy: Common Tagging Errors and Inappropriate Tagging Practices — Best Practices for Validating XBRL Tags

Marlowe: Thank you for this question. Data quality is dear to my heart. Let me begin with a personal tale on this. I started my career in assurance practice at a Big Four firm, and I moved to the SEC afterwards. This was a regular group within Corp Fin. XBRL was fairly new when I joined the SEC, so it was a pretty novel idea to me. When I joined the current group in 2016, my former manager asked me, "Who uses data?" I thought that was such a simple question. "It's people. Who else?" He said it's machines, and further explained that data is consumed by machines for downstream use by people. That was a lightbulb moment.

I realized that XBRL data is just as important as html disclosure, and some people who consume XBRL data may not know that they're using XBRL. Many data aggregators that provide normalized or clean data to the public, accessed as reported data through datasets we publish or EDGAR or API.

With respect to your question on data quality, one common error is scaling. We've seen companies mistakenly add or remove digits in their XBRL data. For example, a small company with public float in trillions. Other errors include creating custom tags for standard disclosures, such as PP&E, Property, Plant, and Equipment. Both GAAP and IFRS financial reporting taxonomies have standard tags for PP&E. We've also seen incorrect tagging, such as using a liability tag for revenue disclosure or using an outdated tag related to an older kind of standard.

Enough with the data quality issue. What can we do about those? One of the ways is to include validations at the intake process. EDGAR has two types of validation checks at intake. One is validation error; the other is validation warning. EDGAR will provide a list of error messages for validation, and companies must correct the errors before they can submit their filing.

An example of validation error is if the current fiscal year-end date for filing is missing. Validation warnings indicate some tag data may be incomplete or inconsistent with each other. An example of a validation warning is the use of deprecated tags, meaning outdated tags. If there's a new accounting standard and the filer adopts it, they should be using new tags that are related to the new accounting standard. We've also seen some filers who, even though they adopted new accounting standard, are still using the old tag.

Another best practice is for companies to use the GAAP taxonomy to consider voluntarily running their data against validation checks that are included in the GAAP taxonomy. Those checks are based on validation checks that XBRL US publishes. In case the listeners are not aware, XBRL US is an external party that consists of public companies, tagging vendors, data aggregators that focus on improving XBRL data quality. If folks have questions on the data quality reminders we issue, or guidance or validation warnings and errors they're getting, we're happy to chat with them. They can contact us at structureddata@sec.gov.

Ervine: Oh, great practical tips, Julie. Thank you so much. You just talked about public announcements the SEC has made or errors that you're seeing across multiple issuers. When the SEC Staff notices a tagging issue, how does the SEC Staff engage with companies in that event, separate and apart from what we just discussed?

How the SEC Staff Engages With Companies on Tagging Issues

Marlowe: Great question. When we see tagging and other data quality issues, we share them with other division offices that have delegated authority to connect with filers directly. Our division does not, but we will actually share the issues and inquire if they have been able to contact the filers. They could do so through calls, comment letters or Dear CFO letters.

As I'm doing now, we also share observations in industry events. We've also contacted tagging vendors with our policy divisions. We've heard from the public that they have seen a decrease in data quality errors after our Staff issued comment letters or data quality reminders. On that note, I'd like to extend my thanks to companies and tagging vendors for their efforts on improving data.

Ervine: Well, Julie, I learned a ton just now. I knew about, for example, XBRL US; I should say I knew of it but didn't really understand what entities comprised it and what it did. Thank you even for just that information for myself.

Again, as I mentioned, if anyone listening wants to hear more, tune in on November 9 on SEC.gov. I should have also mentioned that the investor advisory committee meeting that I highlighted upfront, which has passed, has a replay available on YouTube, so you can still watch that as well.

Thank you so much to everyone listening in. Julie, thanks again. Have a great day, everyone.

The views set forth in the podcast are the speaker's personal views and do not necessarily reflect those of their firm, company, institution, other organizations, or any of the clients with which they are associated.

 

 

-->