Automation
Moontower Munchies #132
Friends,
2 munchies to snack on.
Introducing TRADE IDEAS
First, some blatant marketing. The moontower.ai option analytics app is very “opinionated”. It doesn’t have anywhere as much data or coverage as a Bloomberg or TradingView. It’s really born from what I wanted — to look at the options market and quickly cobble together a point of view rather than build a reference where all the data is there, but it’s not organized the way a vol trader “sees”.
[Every pro discretionary option trader that checks out the app says the equivalent of “this is what our in-house tools like”. They may be more custom to what they do but it’s the same gist. It is a lens that discerns based on vol surfaces and asset movements.]
The app is organized like a funnel. Stepping through it feels like a QB going through progressions. What’s the blobs of vols look like? What’s typical, what’s high, what’s low cross-sectionally? Even if everything looks elevated, we can still sort through the blob relatively.
Once we have identified names that stand out, we then see to what extent it’s justified or not. Follow-up questions include:
“how are the names moving?”
“for the names that stick out, do certain parts of their surfaces stick out more or less?”
“how are vols performing relative to other names today?”
Our Primer and Mission Plan documentation spell out the funneling process. It’s Whether your goal is to identify candidate trades or to understand whether your trade idea can be better expressed via options understanding the progressions is important. But once you’ve internalized the signature for various trade setups or what we call “Presets”, you really just want the tickers classified by which “presets” they belong to.
[And this classification should depend on the universe of symbols you care about. All decisions are relative since resources are never infinite. As long as you have constraints, you choose one thing over the other. This idea is baked into our cross-sectional calibrations.]
We’ve now automated these progressions. This short video shows the tool in action:
This post explains it: Our newest feature: TRADE IDEAS
More to come on that stuff, but just wanted to introduce it.
A glimpse behind the scenes of HRT
There was a very smart guy that I clerked for on my first day at SIG. He was young but had a lot of latitude, trading index options including foreign ones with an embedded quanto. After one day, they moved me to SPY. I’ll never know why but I’ve been living with “he knew I was too stupid” for the past 25 years.
Alas, he left SIG as soon as his non-compete was finished and became an early partner at HRT. I hear he retired before 30. It’s not stretch to say he’d be a billionaire if he stayed, which is just a comment on how insanely successful HRT is.
I’ve heard a few stories out of there over the years, and it seems like a deeply intriguing place in the world of finance. Gappy, who has managed risk at Citadel, HRT, and currently BAM has been effusive about HRT standing alone culturally.
PSA: Gappy is now writing on substack…
[Between relatively recent substacks like
, , , there’s an amazing amount of ridiculous trading experience sharing freely.]Ok enough preamble, here’s the second munchie today:
Hudson River Trading’s Head of AI on How Artificial Intelligence Is Changing Markets
I used ChatGPT to re-print my favorite sections but that’s all this is, my favorites. It’s not comprehensive, you should listen yourself.
⚡ Electricity & Turbines
Ian Dunning underscored that power supply—not GPUs—is now the critical bottleneck for scaling AI-driven trading.
“Electricity is quite clearly a very binding consideration when we think about spinning up new GPU-based training data centers… it really feels like: is there electricity?”
“People are spinning up data centers very fast by basically buying as many gas turbines as they can and putting them outside. That’s the only way to get electricity promptly.”
Even for a mid-sized player like Hudson River Trading (HRT), securing tens of megawatts of affordable power is challenging.
💻 Manual Options Trading — “Click Traders” for NVDA
Despite automation, many options markets still rely on human reflexes. Ian had a particular firm in mind but wouldn’t drop names:
“There are options trading firms that have thousands of people that are essentially cyborg trading options… maybe 10 people trading the options for a single big stock like Nvidia.”
“They’re humans staring at feeds and clicking green or red buttons very fast.”
HRT once ran a hackathon with a PlayStation controller to test human reaction speed. Dunning called this “a learnable skill,” but noted that latency-sensitive AI systems are still too slow to replace these traders. In 2025, real-time click-trading remains one of the few frontiers where humans beat machines—barely.
🧠 The Rare Combo of Engineering Talent
HRT competes for a scarce hybrid: researchers who can also build production systems.
“We’re asking for people to know a lot of things—be both good researchers and good engineers. In this AI era the distinction is blurry…Any research idea you have is intimately connected to how you implement it.”
📊 Market Data as the Fertile Ground
Dunning dismissed the mystique of “alternative data.” For high-frequency horizons, plain market data—quotes, trades, cancellations—is the richest signal.
“Every event that happens in markets… that low-level stuff is internet-scale data.”
“By far the most useful thing is just market data—the true expression of everyone’s intents.”
He emphasized that predictive power diminishes with time horizon:
“By definition, there have been more days than months… therefore prediction on a daily basis is richer. That rule of thumb extends all the way down to seconds.”
For intraday models, flows and order behavior dominate fundamentals. Beyond a few days, the environment becomes too sparse—“not a data-rich domain.”
🧩 Lack of Interpretability — and Why It’s Expected
HRT’s models are black boxes, but that’s not a flaw, says Dunning—it’s a function of their superhuman scale and time horizon.
“Our models are not very interpretable—and that’s fine. We’re trading on time frames of minutes or hours.”
“It’s unreasonable to expect interpretability. If I looked at order book data for Tesla, am I really going to be able to tell you better than random what the price will be in a minute’s time?”
Neural networks “learn in a way that’s nothing like how we do.” Trying to impose human reasoning analogies is misleading:
“They’re just big blobs of numbers… anthropomorphizing them—saying they’re reasoning or imagining—is dangerous.”
Interpretability isn’t missing—it’s irrelevant at such short horizons.
⚖️ Non-Competes and the Shift on Openness vs. IP Protection
Trading firms once lost talent to tech companies that offered publishing freedom. That dynamic has inverted.
“PhDs would say: I can go to Google and still publish my research. If I go to HRT, I go behind a veil….Now though, the golden era of being at a big tech company, being paid to publish research, is over. “The papers coming out of the big AI labs are stale or unimportant. The important work is secretive.”
Dunning noted a cultural reversal:
“AI-lab people are now thinking out loud about non-competes—which is amazing, because that was once very antithetical. A lot of money being paid for talent is also paying for IP. Those people know how the soup is made.”
Secrecy used to be a comparative disadvantage in recruiting for quant firms, but it turns out the big tech firms have inched closer to finance’s ways rather than the other way around.
Stay groovy
☮️
Need help analyzing a business, investment or career decision?
Book a call with me.
It's $500 for 60 minutes. Let's work through your problem together. If you're not satisfied, you get a refund.
Let me know what you want to discuss and I’ll give you a straight answer on whether I can be helpful before we chat.
I started doing these in early 2022 by accident via inbound inquiries from readers. So I hung out a shingle through the Substack Meetings beta. You can see how I’ve helped others:
Moontower On The Web
📡All Moontower Meta Blog Posts
👤About Me
Specific Moontower Projects
🧀MoontowerMoney
👽MoontowerQuant
🌟Affirmations and North Stars
🧠Moontower Brain-Plug In
Curations
✒️Moontower’s Favorite Posts By Others
🔖Guides To Reading I Enjoyed
🤖Resources to Get More Out of AI
🛋️Investment Blogs I Read
📚Book Ideas for Kids
Fun
🎙️Moontower Music
🍸Moontower Cocktails
🎲Moontower Boardgaming
The paradox of click traders for NVDA options is fasinating. We automate everything posible in finance but human reflexes still beat AI latency for real time execution. That electricity bottleneck point really underscores how much infrastructure is lagging behind ambition in AI developement. The shift on non competes at big tech labs moving toward finance's secretive model is telling too.
Fantastic. Thank you Kris