AI And The Municipal Bond Market

AI is rapidly reengineering the $4 trillion Municipal Bond Market. No part of the market is going to be untouched. Trading, pricing, underwriting, credit analysis, compliance, disclosure, regulations—every bit of it is being transformed. In five years, the municipal bond market as it exists today will look vastly different. Wringing out existing inefficiencies and opaqueness, AI will ultimately create billions in value for investors and save issuers billions in interest expenses.

Okay, okay. That sounds a bit hyperbolic, but I’m going to stand by it. Yes, AI has been hyped to atmospheric levels, with grandiose pronouncements that it will surpass human intelligence in just two years. On the other hand, one wag quipped when I mentioned I was writing about Artificial Intelligence in the muni market, “I thought most intelligence in the muni market was artificial.”

Snark all you want or dismiss this as starry eyed overenthusiasm for the latest shiny new toy, but AI is driving the world forward in incalculable ways. Be it robotics or breast cancer detection or fashion choices, AI often does it better than its human counterparts.

The municipal bond market is not immune. AI is moving the market into the 21st century, whether some like it or not.

Vise-Grip

My confidence stems from two powerful forces gripping the municipal bond market in an ever tightening vise: economics and technology.

At the close of 2023, Morningstar Morningstar Direct and the Federal Reserve reports show close to 72% of $764.3 billion in municipal bonds assets under management (AUM) by open-end mutual funds were held in the portfolios of the top 10 municipal bond fund managers. Morningstar Direct also reports roughly $496.7 billion of municipal bonds are held in hundreds of thousands of separately managed accounts. Other sources have the number at nearly triple that.

But wait, there’s more. The Federal Reserve report also notes ETF assets at $122.3 billion AUM. It’s a stunning 700% increase in over the last decade.

That’s $1.3 trillion of assets under professional management, which doesn’t include the billions of bonds held by individuals in brokerage and trust accounts.

The operational management alone keeping track of vast sums of money requires technology for accounting, recordkeeping, compliance, valuation, trading, and the myriad of other day-to-day processes.

(For the intrepid, data on municipal bond holdings of the entire market is in the Municipal Securities section of the Board of Governors of the Federal Reserve System, Federal Reserve Statistical Release Z.1 Table L.212 Financial Accounts of the United States, Flow of Funds, Balance Sheets, and Integrated Macroeconomic Accounts).

Because of this combined asset growth and concentration, money management has become a saturated, hypercompetitive ecosystem, forcing managers to limbo dance ever lower on fees as they have to vault ever higher to find investment returns.

Put yourself in the seat of a senior executive at an asset management company faced with this situation. You quickly realize you need solutions and processes offering automation, speed, execution, and efficiency at a cost lower than whatever you are doing currently.

The AI Arms Race

Technology is the solution that checks all those boxes and it’s AI doing the checking. With massive amounts of data growing at a pace never before seen and with hitherto unimagined compute speeds to analyze it, AI brings increasingly sophisticated quantitative analytics to create solutions.

Hence a technological arms race to apply AI is on. Unfolding at breathtaking speed amongst broker dealers and money managers, each competitor realizes that without a technological advantage, they’re going to lose.

Amid all this sits the $4 trillion municipal market—fragmented, opaque, inefficient, and stale—redolent with time-honored (and increasingly anachronistic) traditions, embedded biases, dated processes.

A Big Fat Ripe Target

In short, it is a big, fat target ripe for an overhaul by financial technology companies armed to hilt with AI and not afraid to use it.

But the question is how? After all, the municipal bond market didn’t get and stay the way it is for no reason. Entrenched incumbents, be they bond lawyers or rating agencies or bond issuers, think they see no incentives to alter their behavior or business models. They have built thick barriers to entry. Or so they confidently assume.

Also, there have already been some pioneering fintech firms who, seeing this messy market, thought they could “disrupt” it by imposing technological organization the same way a software engineer codes an app. It didn’t work. They were right in intent, just off on the timing.

Why is today different? To answer that, this series of articles over the next few weeks presents the five top-level components of the municipal bond market where fintech and AI are driving advances in the most essential market functions and processes. These components are, in order, Data, Pricing, Credit, Platforms, and Algorithmic Trading.

Visualize them as a three dimensional Venn diagram. Inextricably linked, they are a neural network, each independent and interdependent, acting and reacting, an infinite feedback loop. This is why they are driving the market.

But first, a fundamental question. What is AI?

What is AI?

When most of us think of AI, maybe we think of Chat GPT. Enter a question and the algorithm spits out an answer, accurate or otherwise. We see it when sentences get autocompleted writing emails. Or maybe it’s that picture of the Pope wearing a puffy coat. Or a trout.

The fact artificial intelligence is now a buzzword already simply referred to as “AI” (you didn’t have to Google Google what AI meant when you saw it in the title of this article, right?) is in and of itself a comment as to how fully integrated a part of our culture it has become.

But none of that actually explains what AI is. AI is Large Language Models, preceded by Machine Learning which isn’t Deep Learning but Deep Learning is a type of Machine Learning which has Neural Networks with Neurons, Multilayer Perceptrons (with Input and Weight Vectors), which, applying Backpropagation and/or Gradient Descent to train the network—which are sometimes but not always integral in training Recurrent Neural Networks, Convolutional Neural Networks, and Generative Adversarial Networks—on either structured or unstructured data for predictive analytic outcomes.

Did you get all that?

What AI Is. And Isn’t.

Actually, it’s really pretty simple. At its core, AI is just a set of various methods to analyze data and generate outcomes.

The data component is essential. No data, no AI. So what is data? Anything a number can be attached to—and there is nothing a number can’t be attached to. Human behaviors to Shakespeare’s plays to photos of cats to bond prices. Everything.

AI analytic methodologies and functions applied to that data generate outcomes. Outcomes may be to answer questions, test hypotheses, find trends, identify groupings, determine correlations or, well, anything you can do with data analysis. That includes putting a puffy coat on a Pontiff to predicting the weather.

Perhaps not so much on that last one, at least not with perfect accuracy, which makes a key point. Artificial intelligence, for all of its many strengths and capabilities, is not perfect. Depending on how it is used, it can be 100% right, just not all the time. Like anything, it has its pitfalls and limitations. However even if it is mostly right most of the time, it is far better than anything we have today without it.

For those intrepid souls seeking a deeper understanding of AI, IBM IBM , MIT, and Stanford offer numerous free YouTube lectures by world renowned faculty. They explain the specifics far better than can be related here, so you are encouraged to pursue those. Which I highly recommend you to do.

But for our purposes, it’s more important to grasp the implications of how AI is being applied in the municipal bond market than to get lost in understanding how it does it. Put another way, you can still use your iPhone or Droid without a thorough knowledge of IEEE 802.11 for WLANs, MAC, and PHY functions. (That’s the technical term for WiFi—see what I mean?)

With that, let’s go.

An AI View of the Municipal Bond Market:

Data, Pricing, Credit, Platforms, and Algorithmic Trading

Since without data there is no AI, the first article in the series starts with data. And when it comes to data, the municipal bond market has oceans of it. Trading levels, financial performance, deal structures, yield curves, professional publications. Just about name a type of data and it exists. And, from an AI standpoint, the data is very, very good.

The second article will focus on pricing. Ask nearly any municipal bond market investment professional their opinion about pricing in the market and a phrase that rhymes with “it pucks” is likely to be in the first words out of their mouth.

The frustration of market participants with bond pricing and valuation services is legion. It stems from what is genteelly referred to as the market’s “structural problem”: barely 2% of the market’s outstanding bonds actively trade, and those that do vary in block sizes from the millions to only a few thousand, with no up-to-the minute posting for prices or yield curves. All of that make it nearly the perfect problem for eager AI engineers to tackle. Which they are. With vigor.

The third article examines credit analysis. The municipal bond market considers itself credit driven, with fundamental credit analysis at its core. Unfortunately, that analysis is of financial information at times over a year old and reported in PDF form. Extruding the pixelated information requires a process akin to unscrambling an egg. Moreover, lacking a consistent reporting taxonomy, credit analysis is hobbled in comparative testing to determine the probabilistic weights of its metrics. In an AI world, analog credit assessment methodologies increasing look like something resembling cephalomancy. More to the point, if an investment grade rated municipal bond’s default rate is 0.08%—not even 1/10^th of 1.00%, exactly what credit is there to analyze?

Alternative Trading Systems (ATS) platforms have grown dramatically over the last decade, the implications of which are explored in the fourth article. As MSRB reports show, at the close of the first quarter of 2024, almost 15% of all customer trades were executed with dealers associated with an ATS. In 2015, it was a mere 2.9%. Readily scalable and offering automation, connectivity, best execution, trading efficiency, lower cost, speed, and transparency, it is rapidly becoming the trading platform of choice. Traditional broker dealers, who had handled much of the market’s trading volume with human traders, have taken the “if you can’t beat ‘em, join ‘em” approach, replacing traders with computers and hiring quants holding Ph.D.’s. As ATS platforms continue to grow, it is not hard to imagine these platforms coming to resemble a municipal bond market exchange in the not-so-distant future.

The final article closes with algorithmic trading. Using data, pricing, and ATS, algorithmic trading in many ways closes the circle. Perceptions of algo trading (as the market slang calls it) are usually conjure up images of a mysterious, secretive cabal of Ph.D.’s generating endless lines of thickly quant-driven code, their computers roaming the market to find and profit by trading misvalued bonds.

Other than the mysterious and cabal part (well, maybe still a bit mysterious), the perception is generally accurate. Fragmented and opaque, but with copious amounts of data, the municipal bond market is somewhat of an easy mark for algorithmic trading. That’s why algo-trading strategies drawn from AI methodologies are sweeping into the market. Be it hedge funds or proprietary trading desks (“hedgies” and “prop desk” respectively—the market loves catchy abbreviations) to broker/dealers executing trades on behalf of clients, it’s self-propogating: the more it gets used the more others will have to use it. With ATS platforms simplifying access to bonds combined with ever increasing asset growth in SMAs and fund managers, algo trading is going to continue to take an ever larger share of customer trades.

The Good News

All of this is good news for investors and issuers alike. AI unlocks trapped value to open greater liquidity with better pricing, tighter bid-ask spreads, and lower execution costs. Improving the valuation on the $4 trillion of outstanding municipal bonds by just one basis point—1/100^th of 1%—fattens wallets by billions.

Equally, issuers will find better price discovery and more precise deal structuring when they come to market. In 2023, $325 billion of tax exempt municipal bonds came to market. If issuers shave off just one basis point of interest expense, the savings would top $300 million. Two basis points…well, you get the idea.

A lot got covered here and there is a lot more to go. Buckle up. It’s going to be quite a ride.

Next in the series, Data. The municipal bond market has oceans of it. But who has it, who can access it, and what are collectors and users doing with it?