Algorithmic Trading
The term algorithmic trading is tossed around a lot these days. But what does it mean?
Let’s break it down into its components.
It’s not that defining an algorithm is all that complex. If anything, it is very simple. An algorithm is a set of instructions enabling a computer program to put together different sources of information [data] and generate a result.
If you’re talking algorithms, you’re also talking AI and data. Both are equally straightforward. At its core AI is simply a way to organize, analyze, and apply data for an outcome. Data is anything that has or can be digitized and tagged with a number. And anything—photographs, text, speech—can be digitized and numerically tagged.
Outcomes can be as basic as determining the mean of a data set or as complex as the predictive analytics needed for a self-driving car. Analyzing outcomes and then refining and adjusting the program and data inputs for a better outcome is what machine learning or deep learning are all about.
As for trading, it’s a safe assumption that if you are reading a Forbes.com investment column, you know that it means buying and selling bonds, stocks, options, and other financial instruments.
Algorithmic Trading in Practice
So how does all this come together as algorithmic trading?
James Pruskowski, Chief Investment Officer & Co-Founder at 16Rock Asset Management offered a general example. Integrating AI and algorithms, the firm’s municipal bond trading models regularly monitor market data—inputs like trades, volume, spreads, bid wanteds, request for quotes. One model is programmed to seek out arbitrage opportunities based on pricing anomalies discerned by analyzing market data. With regularly updated buy and sell parameters, the computer bids/buys and asks/sells.
A fully automated process, all of this can happen in a few seconds. The new trade data gets uploaded, is analyzed, and the model is updated to determine what the best opportunities might be given the new information. And the cycle continues.
Noting AI and algorithmic trading are liquifying the municipal odd lot market (an odd lot is a block of bonds trading at $100,000 par or less), James also observed that separately managed account (SMA) asset growth as well as municipal ETFs were compelling many market participants to be price takers.
Be it a Financial Advisor or ETF fund manager, when faced with a pressing need to invest or sell, operational complexities force both—and their executing traders at large retail wire houses—to be price takers. These prices tend to reflect the higher yielding asking price versus the lower yielding bid price. Correspondingly, levels on market benchmark curves (such as the ICE US Municipal AAA Curve) derived from these trades reflect these price takers’ lower price, higher yielding trades. This is how, in his view, price takers define the opportunity set in today’s market.
Josh Rosenblum, Head of Algorithmic Trading at Brownstone Investment Group (he was formerly head of Municipal Trading Strategies at the firm), offered a different take. Combine the billions of assets under management held by mutual funds and SMA advisors with electronic trading platforms, he sees buy side firms moving from price takers to price makers. Moreover, with those platforms buy side firms could trade with each other directly. The relationship between broker and the client is evolving into more of a partnership to develop mutually beneficial workflows.
He has seen evidence of tighter spreads in the odd lot market due to algorithmic trading, also noting that AI-generated predictive pricing analytics creates a self-fulfilling pricing prophecy. If the model says the next trade on bond has a high probability to be valued at a certain amount, the maximum buy or sell price an investor would be willing take is going to be at that price or within a hair’s-breadth from it. Prediction is driving reality. Moreover, the counterparty “investor” is increasingly likely to be another AI-driven model. While a positive implication of this may be more market liquidity, computer and model dependence has its own pitfalls.
While speaking independently, James and Josh concurred that AI, electronic trading platforms, and algorithmic trading is creating a technological arms race. If you’re not the lead, you’re falling behind, which is accelerating technological adoption because absolutely no one wants to be left behind.
This is particularly true when it comes to execution speeds. If you aren’t trading faster than your worthy competitor, you weren’t going to get bonds. No bonds, no performance. No performance, no investors. Pretty simple.
Broker Dealers Buy and Build
It isn’t just proprietary funds using algorithmic trading.
For broker dealers who want to stay active in the municipal bond market, it’s a classic build-or-buy decision. Perhaps first out of the box was Toronto-Dominion Bank’s (NYSE-TD) acquisition of Headland Tech Holdings, LLC in 2021. Barely a year later, Raymond James (NYSE-RJF) bought SumRidge Partners, LLC in the firm’s ongoing strategy applying technology as the backbone to build out their trading business. Larger broker dealers and asset management firms such as Goldman Sachs and Blackrock, respectively, have generally pursued internal builds to remain competitive, although hiring top software engineers to do so might be interpreted as a form of buying. Then again, at least one bracker firm has thrown in the muni towel: Citigroup exited the municipal bond business in 2023.
Bond Talk
Given its direct market impacts, the market’s focus on algorithmic trading is understandable and justifiable. But there is another algorithm-based technology that might prove equally impactful on the market.
ChatGPT is based on an algorithm, specifically a type of machine learning model called a language model. More precisely, it is built on the GPT (Generative Pre-trained Transformer) architecture, which is a deep learning model designed to understand and generate human-like text. It is not a single algorithm but a complex system involving many components. The overall process can be described as an algorithm that uses machine learning to generate responses based on patterns in the data it has seen.
In fact, that entire prior paragraph was written by ChatGPT.
ChatGPT is a fascinating technology. From planning a Caribbean cruise to generating Python code to the latest procedures in hip replacement surgery, just about anything you can imagine is available at your fingertips. Enter a question and, in a few seconds, ChatGPT replies with an answer.
But interactive technology goes beyond typing in text for a response, as anyone who has talked to Apple’s Siri or Amazon’s Alexa can attest. Flipping that around, Speechify can turn anything into audio—text, pdf, email—and read it to you in high quality digital voices, including Snoop Dogg or Gwyneth Paltrow. In fact, Forbes.com offers an audio of this article for your listening pleasure. (No, it is not my voice.)
Prefer video? Pictory offers text to video, URL to video, PPT to video, and several other to-video abilities. You can copy-paste this entire article into Pictory to see how the technology transforms my jottings into a Hollywood-esque video. (Again, no, I am not featured in it. Sigh. I really need a new agent.)
In short, the abilities of interactivity are boundless. If you can imagine it, it can be done with AI in any communications medium.
That includes talking to municipal bonds.
If you are wondering why you would want to talk to your bonds, ask Robert Kane and Saul Tessler of Munibonds.ai. Founded in 2024, the firm has an AI chat function where you enter a question on a bond and the app generates a very thorough response. This user driven input functionality is far more robust than the limited, predetermined drop-down menu list.
If it has been disclosed, from the Preliminary Official Statement to the most recent quarterly financials, the app finds the data and generates an answer. It’s as close to making the municipal bond market’s PDF-laden disclosure documents into something machine readable as technology will allow. For now.
Munibonds.ai is more than just an interactive search function. Tracking both user queries and other external market data, the firm offers market sentiment analysis on any bond. Or a portfolio of bonds. Or a bid-wanted list. You get the idea.
It also offers a numerical “bond ranking”. While short of a credit rating, with the right use prompts, the report generated by the chat technology—including some fundamental analysis—is pretty close to a standard credit rating report.
Creating the customized need parameters and automating that function, portfolio managers, traders, compliance officers can have credit, structure, use of proceeds, market sentiment, and any other decision-making or reporting information on any bond or portfolio of bonds in nearly real time at any time.
All they have to do is ask.
AI Predictions for the Municipal Bond Market
Imagine a perfect Laplace AI model. Possessing complete knowledge of every piece of data in the universe, past and present, to achieve a state of omniscience, it could predict the future with perfect accuracy. Whether or not the result would be an angel or a demon is up for debate, but read enough about AI and one comes away with the sense this is the ultimate state those developing and applying AI would like to achieve.
It’s impossible, of course. But for the municipal bond market, even if knowledge gleaned from AI leads to asymptotic results, it’s a far better result than what exists now.
So here are the top AI predictions as to what the municipal bond market can expect going forward.
1. The rate of technological advancement and adoption across the market is going to accelerate.
2. Any market processes that can be automated will be.
3. From broker dealers to investment advisors to trading platforms, there will be further market consolidation.
4. The market will become far less opaque and fragmented.
5. A primary market electronic shelf-financing mechanism will develop for municipal bond issuers.
6. The market will adopt standardized, digital official statements, a consistent reporting taxonomy using an XBRL format, and ongoing disclosures will be in a machine readable or similarly algorithmically accessible format.
7. The pricing differential due to the current structural imbalance between odd-lot trade and institutional trade valuation will dissipate.
8. To stay relevant, credit ratings will be attached to numerical rankings that are more probabilistically based.
9. Climate change metrics will have a far greater impact on credit analysis and ratings.
10. These market evolutions will come far faster than many are expecting or are prepared for.
Noting Nobel laureate physicist Niels Bohr’s observation that predictions are difficult, particularly about the future, it is tempting to dismiss any one or all of these or any others that can be thought of with a “that can never happen” hand wave.
Those who would do so are urged to keep in mind that, in technology, the question is not “can it be done” but “how can it be done” and then “how fast can it be done”.
Because if history shows us anything, it is that, in the end, one thing has proven consistently true:
Technology wins.
This is the sixth and final article in the series, AI and the Municipal Bond Market. Prior articles covered AI’s influence on pricing, economic drivers, data (and more data), credit, and Alternative Trading Platforms.
The following Podcasts related to this series are:
Technology is the Solution, The Muni Matrix, Munichain, May 8, 2024
The Dynamics of Municipal Markets with Barnet Sherman, DebtBook/Where Public Finance Works, June 18, 2024
Bond Trading Roundtable—The Future of Bond Trading, Forefront Communications, September 19, 2024
Read the full article here
