ICI Compliance, Risk, and Legal Conference

Thank you Hope and good afternoon. It is a pleasure to join the Investment Company Institute today and to spend some time with people who care deeply about the strength and integrity of our capital markets.

Before I begin, let me give our standard disclaimer. My remarks are provided in my official capacity as the Chief Economist and Director of the Division of Economic and Risk Analysis but do not necessarily reflect the views of the Commission, the Commissioners, or other members of the staff.

High-quality information

The U.S. capital markets are the largest and most dynamic in the world. They allow trillions of dollars in assets to be traded, innovative companies to raise capital, and millions of Americans to save for retirement.

A critical input in all of that is high-quality information. Investors rely on high-quality information to make decisions. Markets rely on high-quality information to allocate capital. And regulators rely on high-quality information to understand markets and evaluate policy.

Much of that information begins with issuers and market participants like many of you in this room.

Form N-PORT. Form N-CEN. Form N-MFP. Form PF. You are all familiar with these forms and the data they contain.

You may occasionally wonder what happens to all that information after you hit submit. In some cases, the SEC uses it to conduct market monitoring. In other cases, members of the public use it to make investment decisions. Sometimes the information may support rulemakings. But regardless of who ultimately uses the information, machines are playing an increasingly important role in how these data are accessed, processed, and understood.

Economic analysis, and thus data, sits at the heart of what my division-the Division of Economic and Risk Analysis (more commonly known as DERA) does. DERA works at the intersection of economics, statistics, data, and policy. We support the Commission through economic analysis, risk assessment, litigation economics, and data science. And we manage extensive public and private datasets.

This spring, we renamed several offices within DERA. Two of the new names, in particular, say something about where our work at the Commission is going.

Our Office of Data Science became the Office of Advanced Analytics and Artificial Intelligence. And our Office of Structured Disclosure became the Office of Data Standards and Innovation.

One of these offices focuses on strengthening our analytical capabilities using advanced technologies and techniques. The other focuses on the quality, structure, accessibility, and usefulness of the information those capabilities depend on. The work of these two offices is highly complementary. One strengthens our analytical capabilities, while the other strengthens the data those capabilities rely on.

And that is really the idea I want to explore today. Artificial intelligence is changing the economics of information. It is changing how quickly information can be processed, how broadly it can be compared, and how much can be extracted from it.

One indication of how quickly information consumption is changing comes from the SEC's own public datasets. The SEC publishes numerous datasets to make information reported to the Commission easier for the public to analyze. These include data drawn from investment company filings, as well as data from business development company submissions, financial statements, and many other filings.

These datasets have long been popular. But recently, downloads have increased dramatically.

Through July of this year, the SEC's Form N-PORT registered fund dataset was downloaded more than 14 times more often than during the same period last year.

The Form N-MFP money market fund dataset was downloaded nearly seven times more often, while the Form N-CEN registered investment company dataset was downloaded nearly four times more often.

And downloads of the SEC's business development company dataset increased the most among these. It was downloaded nearly 75 times more often than during the same period last year.

These are extraordinary increases over the prior year.

Of course, that does not mean there are suddenly thousands more analysts downloading these files by hand. A lot of that growth is probably automated downloads by machines. But that is not really a caveat to the download growth. In fact, it is part of the point.

Financial information increasingly has two audiences: people and machines. For most of the history of securities regulation, disclosure was designed primarily around the first audience. A person opens a document, reads the text, studies the tables, and uses that information to make a decision.

People still do that, of course. But financial market information is also being pulled into databases, analytical systems, algorithms, and AI models. Those systems can process information at a scale that would have been difficult to imagine even a decade ago. But to do that well, they need information that is consistent and structured.
As our analytical tools become more powerful, the quality of the underlying information matters more, not less. This may be counterintuitive at first as some might think that with AI we do not need to worry as much about data quality. This is not true.

A person can look at two slightly different labels and recognize that they mean basically the same thing. For an AI model, that connection is not always obvious. AI systems today can be probabilistic, not deterministic. These models work by determining what meaning is most likely, not by calculating a single predetermined answer. So, the less ambiguity we give these machines, the better.

Structured data provides context for AI. Standardized data makes comparison easier. And consistent definitions make it possible to analyze thousands of observations systematically rather than one document at a time.

If you have ever watched a show with automatically generated captions, you know that most of the time, they are right. But every so often, a name or a technical term comes out as a completely different word. And the captions do not tell you they are unsure. They display the wrong word in exactly the same clean type as all the right ones. This is what can happen when a model has to infer meaning that was never made explicit. And it is a primary reason that structured data matter. When it comes to financial information that can affect investment decisions, we would rather tell the machine what a number means than make the machine guess.

This is not abstract for us or for investors. This information is used to make important, real-life financial decisions.

Information that begins as a regulatory filing can become part of a dataset that machines process and analyze, which helps investors allocate capital, allows economists to identify developments across markets, and market participants better understand the information available to them.

Increasingly, investors, market intermediaries, academics, and regulators are accessing and analyzing information through automated systems. That makes it more important that the information be presented in a form machines can understand reliably. Otherwise, we increase the risk that it will be misread, misclassified, or misunderstood.

Artificial Intelligence

And that leads me to my second point. AI is a wonderful tool, but it is not magic.

AI can lower the cost of processing information. Large volumes of information can now be searched, summarized, combined, and compared at very low marginal cost. And AI may also lower some of the costs involved in preparing disclosures.

It can therefore be tempting to conclude that if disclosures are becoming easier for machines to process, and perhaps somewhat less costly to prepare, the economic case simply points toward producing and disclosing more information.

But embracing new tools does not mean losing sight of the principles that anchor our disclosure regime.

At its heart, materiality balances investors' need for information with two enduring considerations: the cost of producing information and the risk that excessive detail overwhelms rather than assists investors. Some may argue that because AI makes disclosures easier and cheaper to process, and lets machines, not just people, read filings, the traditional logic for limiting disclosures to material information becomes weaker.

But that argument overlooks the realities that still govern disclosure.

AI can make disclosures much cheaper to process. But it does not make the underlying information free to produce, verify, or stand behind.

Information still has to be collected. It has to be verified and reconciled. Systems have to be built and maintained. Definitions have to be applied consistently. Errors have to be identified and corrected. Sensitive information has to be protected.

And many of you in this room are responsible for standing behind the accuracy of such information.

Registrants also face economic and legal stakes associated with accuracy, confidentiality, and the potential disclosure of competitively sensitive information. In other words, AI may lower some of the costs of preparing and processing disclosures. But it does not eliminate many of the other costs associated with producing reliable information and making it public.

The answer to the power of AI is not an indiscriminate flood of information. AI does not necessarily strengthen the case for more information. It strengthens the case for better information.

The goal should remain a disclosure system grounded in materiality and credibility, reinforced by strong verification practices that ensure investors receive high-quality information they can trust.

Economic Analysis in Commission Rulemaking

The importance of high-quality information does not stop with investors or markets. It also matters for the Commission's own decisions.

When the Commission promulgates rules, input from market participants can help us sharpen the economic analysis, test assumptions, and better understand effects that may not be fully visible in public data.

Congress has directed the Commission to evaluate how our rules affect efficiency, competition, and capital formation.

Doing that well takes more than good intentions. Under this Administration, the SEC is placing renewed emphasis on detailed, transparent, and reliable economic analysis that carefully examines how rules affect investors, issuers, registrants, and markets.

That analysis draws on the high-quality data I have been discussing. But it can also be strengthened by information from market participants, like this audience, who see operational realities and market effects that may not be apparent from available data alone.

The Commission has access to a tremendous amount of information.

But we cannot observe everything.

Many of you know things about the operation of these markets that we cannot readily learn from a regulatory filing.

You know what it actually takes to modify a reporting system. You know which costs involve a one-time build and which recur year after year. You know whether a new requirement can be incorporated into a process you already have or requires an entirely new one.

You know whether the same obligation falls differently on a large fund complex or a small fund. And you are well positioned to anticipate how registrants or investors will change their behavior in response to a regulatory requirement.

That kind of institutional knowledge can be enormously valuable to economic analysis.

ICI and many of the firms represented here routinely provide the Commission with detailed data through the comment process. I want to tell you, from the other side of the table, which parts of that information can be particularly valuable to economists.

In my experience, when commenters provide economic information, it can be especially useful when it answers one of four questions.

First, how large is the effect, and what drives it?

It is useful to know that something will impose a cost or create a benefit. It is even more useful to understand what determines its magnitude.

Consider compliance costs. If a cost is largely fixed, it may fall proportionally more heavily on smaller firms. If instead it scales with assets or transaction volume, its competitive effects may look very different.

The structure of the cost matters, not simply its dollar amount.

Second, who ultimately bears the effect?

Averages can sometimes hide important details.

The party that initially incurs a cost is not always the party that ultimately bears it. Costs incurred by an adviser, fund, or intermediary may be passed through in whole or in part to investors. Understanding who ultimately bears those costs can be important to evaluating a policy's economic effects.

Third, how will behavior change?

The economic effect of a regulation is not simply the check someone writes to comply with it. People respond to incentives. Firms may change products, processes, prices, business models, or investments. Investors may change their behavior. Entry may increase or decline. Competition may change.

In some circumstances, those behavioral responses can matter as much as the direct compliance expense.

And fourth, what evidence supports the conclusion?

That evidence can take many forms.

It may be internal data. It may be a survey that commenters conduct. It may be historical experience with an earlier regulatory change. It may come from another market or another jurisdiction.

And some effects cannot be quantified reliably. In those cases, DERA conducts careful economic reasoning that clearly explains the mechanisms at play.

No commenter has every answer. Nor should they. The responsibility for conducting the Commission's economic analysis belongs to the Commission.

But market participants sometimes possess information we simply do not have. When that information is provided with enough context for us to understand how it was generated, the assumptions behind it, and its limitations, it can improve the record on which our analysis is based.

That matters because the comment process is not simply a procedural step in rulemaking. It is one of the ways the Commission learns.

Ultimately, better information improves the quality of the choices available to the Commission. And that brings me back to where I began.

Conclusion

We are entering a period in which analytical technology will allow all of us to do things with financial information that were recently difficult, expensive, or impossible.

That is genuinely exciting. But greater analytical power does not reduce the importance of reliable data and information, it raises it.

AI makes structured, high-quality data more valuable, while also changing the costs of processing and using that information.

You see those challenges from one side through the work of producing, validating, protecting, and standing behind information.

We see them from another through the work of analyzing that information to understand markets and evaluate policy.

But the underlying economics are the same.

Better information produces better analysis. Better analysis produces better decisions. And better decisions help keep U.S. capital markets the deepest, most innovative, and most trusted in the world.

Thank you.