pitchfork CLAW-back
Moontower #306
In this issue:
half-baked AI thoughts
making your own teacher
a satisfying derivation of the 80% straddle rule
Friends,
You can’t swing a cat these days without hitting a prediction of what AI means for humanity. Insofar as it’s possible for someone writing for a wide audience, I’ll share what amounts to some half-baked thoughts that I keep coming back to.
“YOU GET A UI, YOU GET A UI”
I say this as someone who feels some dissonance building an analytics product with charts and tables. The future is just APIs talking to each other. In the future, Zillow’s UI is a mere suggestion. You want the data in a different format? You want inference beyond what Zillow decided goes into a Zestimate? All of that is getting cheaper and accessible to non-technical people.
You’ll go to a site. Maybe. Instead it wil be Siri or Jarvis or Alexa or whatever “I want to see XYZ” and your client-side listener will construct it but it’s going to need access to API that has the data. The data is increasingly all the value while rigid presentations become pointless.
Again, the future is APIs talking to each other. Data becomes increasingly locked down.
On Data Lockdown…
Scarce or exclusive data’s value increases as its complement, inference, gets cheaper. The big futures and stock exchanges are some of the original 2-sided platform businesses. Can see them flexing their quasi-monopolistic might on the data side.
…which might get dark
The fight over data will intensify as well. If you place a bid, the exchange claims that’s their data. But is it yours? Well, it’s of little value to you but whose gonna galvanize the white-collar movement around “hey AI trained on all this data that had little value in isolation, but so much value in aggregate then used it to disrupt US”. It started with the Hollywood writers guilds, but is it crazy to imagine rolling protests as automation eviscerates industry by industry? To see picketers with dystopian slogans like “my data, my choice”.
It’s a different argument than p(doom) objections to AI. It’s not “you shouldn’t have built this” but you had no right to cut me out. It seemed like a bargain when we got “free” email or “free” social media. The tone deaf tech mogul will undoubtedly claim it was fair at the time and maybe that can fly intellectually. But it’s not the clawback by court that will decide. That one happens by pitchfork.
AI Immune
As SaaS gets wrecked, we wonder retains value in the singularity. My working model is:
trust and accountability
Maybe AI can sell my house but realtors have survived fee compression and technology far longer than anyone expects. I think this hints at a still-valid truth. People want to have someone to yell at, appeal to, or simply talk to when it comes to lumpy, rarely repeated transactions.
It’s the “shit umbrellas” theory. The human’s value is not in doing the work but in retaining liability. AI can read the tax code, but my accountant will stand by his work in court.
If you travel extensively for work your relationships are hard-earned. The proof of work is miles traveled or other exclusionary behavior where there was no substitute. Relationships are repos of accumulated, unfakeable work.
The more you can position yourself as accountable the more value you can retain. Being trustworthy and reliable don’t go out of fashion. They will get even more valuable as so much else can be faked.
A strange corollary to this:
Things that were always fake but valuable will stay that way. Like astrology.
A final thought in this thread:
Anyone burning their reputation to the ground thinks either the world is ending or the tides won’t ever turn. Hmm, how would you act if you couldn’t afford for the tides to ever turn?
Art
Art is a big question. There’s will always be a positional scarcity component of it and there will always be genius. The question of whether there will be a surplus of robot genius around as well. We may finally get an answer to would a million monkeys write Shakespeare.
Live performance and sports will stay important. At least until everyone over the age of 10 today is dead. Then all bets are off. I never thought watching people play video games would be popular. I’m now open to the possibility that future people may perfectly prefer robots playing video games or anything for that matter over people playing.
I leave you with this interview with the creator of Open Claw. Listen specifically from 8:30 to 10:17
107 seconds and I quote, “Holy fuck”
If you are hanging your hat on cleverness for its own sake or rigid definitions of intelligence, your time is over. I’m not saying intelligence is losing value, but that its truest definition will become obvious — the ability to get what you want out of life. This is the only definition that will matter and the bright side of that is that its more inclusive than whatever school thinks it is.
I was at a local book fair last week where you dump as many books as you can fit into brown shopping bag for $8. It was an epic haul and a great incentive to just snag boogs that seem even remotely interesting.
I picked up this 1957 classic by C. Northcote Parkinson. I was familiar with his eponymous law, which states "work expands so as to fill the time available for its completion".
This entire book is an extraordinary, laugh-out-loud, pull-no-punches satire. Parkinson would have been an absolute master at Twitter. The law itself is satire wrapped around a specific observation. The way he formalizes the argument is pure art, even ending with an insane equation (he constructs hilarious equations throughout making the book feel like a tongue-in-cheek treatise on social physics).
Satire, notwithstanding, it sure feels like the very thing Parkinson’s Law pokes fun also holds the key to our salvation if AI just does all the work. So that’s where I am now. Placing humanity’s hope on a joke by a British naval man whose skill with the pen is such that I want to vibecode a Parkinson writing voice app.
Money Angle
As an example of “everyone gets their own UI” I whipped up a Dashboard page where I can add modular tools.
The first tool I added is something I call the “Financial Analyzer”. The seed of the idea came from an education POV. I’m teaching this Investment Beginnings Class, but I myself am a novice at reading balance sheets and income statements.
Pre-AI I would want a teacher sitting next to me explaining what sticks out to them. In minutes I was able to make a site where I give it a ticker and range of years and it pulls the filings from the SEC’s Edgar database. But the best part is Claude can act as the teacher and write its own summary of what it sees from year to year. I’m not saying this is going to be human expert level analysis but that is not the bar. I’d just like to pull up a stock and get a quick orientation through the years.
I could even feed the analysis back into an LLM to have it explain it to me even simpler.
I happened to pick up Neal Stephenson’s Diamond Age from that book fair I went to. You can build your own “Young Lady's Illustrated Primer” now. It’s a bull market for learners out here.
The video below shows how the tool works. I’m not sharing the tool publicly because the analysis layer uses Anthropic tokens but this description is from Claude and is probably a good enough prompt to make it yourself:
A browser-based tool that pulls structured financial data for any US public company directly from the SEC’s EDGAR XBRL API — the same data companies are required to submit with their 10-K and 10-Q filings. No third-party data providers, no API keys.
Enter a ticker, pick a year range, and get collapsible income statement, balance sheet, and cash flow tables going back to 2005.
Toggle between annual and quarterly views.
Every line item shows the exact XBRL concept tag it maps to, so you can trace any number back to the original filing. A “Raw XBRL Data” tab exposes the complete set of concepts the company filed — not just the ~40 our template covers — with search and filtering. Remainder rows flag where our template’s sub-items don’t sum to the reported total, so nothing silently disappears.
An AI-generated analysis summarizes trends across the full time series.
Stack
Single HTML file (no framework, no build step)
Vercel serverless functions for the SEC proxy and Claude API calls, hosted on Vercel’s free tier.
Data endpoints
www.sec.gov/files/company_tickers.jsonfor ticker lookupdata.sec.gov/api/xbrl/companyfacts/for filing data.
Money Angle for Masochists
My new podcat series with Outlier Trading is up. The first is a short one just to set the stage for what to expect from weekly episodes.
We released a second one on Friday because, well, oil is interesting.
Visual Derivations
This week I re-published a foundational option post on X articles.
It’s a beast of a post that I orginally published in the format of a Socratic homework but in the X format you can basically read straight through it.
It covers:
☑️The relationship between MAD (mean absolute deviation) and standard deviation
☑️how to approximate a straddle value without a model
☑️a visual derivation of the approximation
☑️see how the straddle is the MAD
☑️gain an intuition for how skew and fat-tailed distributions distort the relationship between straddle prices and volatility
☑️see practical situations where ATM straddles and therefore volatility misrepresents risk
Sometimes these learning posts go over a lot of material so I think it’s helpful to point out what parts of them are most interesting to me personally. In this post it’s the section Lessons from a Skewed Coin and how standard deviation and in turn straddles are deeply misleading representations of risk when a distribution is highly skewed such that the mean is balancing many frequent events in one direction verse rare but large events in the other. I talk about how this shows up in familiar investing scenarios.
The other concept in here I like to emphasize, in no small part, because it’s legit fun is the visual derivation of the straddle approximation that states the straddle is 80% of the vol.
The derivation:
The mean of the distribution
We want to estimate the straddle. The mean of the underlying stock distribution is centered around the forward price not the at-the-money price.
We will estimate the at-the-forward (ATF) straddle.
This means we are estimating the straddle struck at the ATF strike.
The ATF strike occurs at the ATF price:
Approximating the ATF call option
This is the meat of the work.
[It requires no more than pre-algebra.]
Let’s go.
While we want the straddle, let’s start with the ATF call option.
We established 3 identities that occur at-the-forward:
Now we just plug these back into the B-S formula for the call.
Hmm, this looks fairly docile. Stare at it hard. The next section will feel good if you like geometry!
The underlying distributions for B-S is that stock prices are lognormal. The prices are lognromal but logreturns are normally distributed.
This is handy because normal distributions are familiar to work with.
d1 and d2 are like Z-scores on a Gaussian (bell) curve of logreturns!
The probability density function (PDF) for a bell curve:
The center of our distribution is an expected logreturn of 0 corresponding to the forward Seʳᵗ
The peak of a bell-curve at that forward price corresponding to a logreturn of 0. For the standard normal curve we can assume σ = 1
Plug 0 into x of the PDF:
Let’s bring this all together into a picture:
Understanding the picture
The value of the ATF call is the integral of the PDF between d1 and d2 but we can estimate it!
height x base x forward price
Note: This will slightly overestimate the value of the call (see the overestimated region in the picture
To go from call price to the straddle, we remember that at-the-forward strike the call and put are equal because of put-call parity!
The rest is easy:
The straddle is the MAD!
The volatility, which is computed just like a standard deviation, gives large moves extra weight. But the straddle is a better reflection of what move size we typically see.
It will cost you .80 of the standard deviation to buy a fairly priced straddle. Let’s plug that into a normal curve’s cumulative distribution function:
💡Theoretically, if the straddle is fairly priced:
it will expire in-the-money ~ 42% of the time
despite the low “hit rate”, it’s fairly priced because the payoff on larger moves balances the expectancy
Stay groovy
☮️
Moontower Weekly Recap
Posts:
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 data is increasingly all the value". I agree. But I think it goes deeper. The logic used by orgs also has value (investment analysts process, insurance underwriting, mortgage underwriting, etc.). So knowing how to get a better inference also has value. You are getting 80% of the financial analyst value know with a dashboard prompt. What about the 20% you are missing? What's the edge in that? And the second thing that goes deeper is all the niches out on the fringes with non public and hard to collect data. As a corollary to your note, as broad data sets lose value as they commoditize, these specialty data sets gain value. And for individuals, those hard earned pattern matching and analysis skills in thinly covered but valuable areas are likely provide value, not unlike that achieved by the best pro basketball players.
Will the podcast be available on Apple eventually?