Building an investment tool from scratch
Day 2 of developing Tracking Alpha
First things first: knowing which merger arbitrage deals are live in the US at any given time.
Finding them is easy. They all go through the SEC and get published on EDGAR.
The hard part comes after that. There are thousands of documents. Some deals have already closed, others are still open, and some don’t interest us: for now, we only want the ones paid in cash.
So I’m building a semi-automated system that gathers that information, filters it, and sends it to me to review.
That last step is key. Every data point comes with the exact quote from the document it came from, and the system checks that the quote appears exactly as is in the original. Two different AI models read the same thing, and if they don’t agree, I decide. No deal gets published without me hitting the button.
That’s how I’m refining the system with real cases, so that when it runs on its own, it makes as few mistakes as possible.
On the one hand, this is pretty easy thanks to AI.
On the other, it’s still complicated, even with AI.
If we did what almost everyone does, it’d be easy: grab 5 deal details (price, buyer, estimated date...) and move on.
We want the whole deal, down to the last detail: every condition, every pending approval, every date, and the fine print. We specialize in merger arbitrage and want to save you the hours it takes to get the full picture.
The information exists, but it’s scattered across dozens of documents. Many have nothing to do with the deal, though sometimes those bastards hide an important clue.
That’s why it’s such a pain in the ass:
📄 You have to read lots of very different documents: the announcement, the merger agreement, the proxy, the tender offer...
🔤 Names don’t match from one document to another.
🏢 The buyer signs through a company set up just for the deal, even though there’s an established company behind it.
🔎 Every condition has fine print, and you have to check it.
Doing it by hand is a nightmare: hours for every deal. And the real challenge is doing it better than anyone else without costs going through the roof.
The good part? Imagine the end result.
All live deals in one list. Each one with its conditions, what still needs approval, the expected closing date, and the spread. And every data point with its source.
And what I think is even more interesting: over time, a database of past deals, catalogued in meticulous detail, with how each one ended and the sources to prove it.
To build the best tool, we need the best data: pulled from the original source and reviewed one by one (for now).
If you’re interested, I’ll leave the link in the next tweet. Joining the waitlist is free, and the first 20 people to become members will get launch pricing: $20/month instead of $29, and they keep that price.
I’ll keep you posted on how it’s going.
