The Fed will cut in December. Not a guess — the data window is CLOSED. Full analysis → open.substack.com/pub/tommijohns…

The Fed just cut rates but the labor market is far weaker than the headline jobs report suggests. Underlying job growth may be closer to 39k/month. Layoff signals are rising. Claims data is lagging reality. This makes the 2026 rate path WAY murkier than markets think 👇

The market reacts to headlines before you can read them. That’s not intuition — it’s measurable. We’re building a research-grade sentiment indicator using FinBERT + market data (no magic, no hype). 🧵👇 Read: open.substack.com/pub/tommijohns…

Wall Street has always been part math, part mood. What changed? Machines are now better than humans at reading the mood. Gen-AI is quietly turning headlines, earnings calls, and “vibes” into tradeable signals — sometimes before prices move. We break down the evidence 👇

📉 Why most “sentiment signals” in finance are totally misleading and how synthetic data exposes the gap between tone and real investable insight 👉 open.substack.com/pub/tommijohns…

When Powell speaks: Humans get a headline. AI gets a probability shift. Markets listen to the second one. 🔗 open.substack.com/pub/tommijohns…

Most sentiment models used in investing are solving the wrong problem. Tone is not attribution—and it matters more than people think. Experiment + results ↓ open.substack.com/pub/tommijohns…

AI isn’t just a “nice to have” in finance anymore, it’s becoming the default way markets are analyzed and traded. If you work with stocks, you’re already feeling it. 👇 Read more open.substack.com/pub/tommijohns…

Markets aren’t getting safer. They’re getting more synchronized. AI is quietly creating “attractor markets” that look stable… until they snap. Read this before the next “random” unwind: open.substack.com/pub/tommijohns…

Sentiment isn’t a property of text. It’s a property of who the sentiment is about. That single mistake explains why FinBERT-style models confidently mislead investors. We show why — and how to fix it — here: open.substack.com/pub/tommijohns…

The market doesn’t just run on earnings. It runs on vibes you can actually measure. There are 40+ academic ways to quantify “investor mood” — from GIFs and Reddit posts to IPO activity and options skew. This is the map. 👇 open.substack.com/pub/tommijohns…

FinBERT is wrong 83% of the time when it calls a headline “positive.” We validated 991 NVDA/AMD headlines over 120 days. It’s great at detecting tone. It’s terrible at detecting price catalysts. A 4B local LLM fixed it. +24.8% accuracy improvement 94% precision on positives $0

FinBERT’s “positive” calls on semis → –0.37% next day returns. Signal isn’t weak. It’s inverted. Hybrid (FinBERT filter + Claude Sonnet) delivers ±1.26% spread at 1/5 cost of pure LLM. 12 tickers · 2,287 articles · 30 days. Architecture > benchmark accuracy. Full study:

Investor sentiment DOES predict stock returns. But actually profiting from it? Brutal. • 93% of alpha disappears after costs • Most gains come from shorting • Crowding can trigger crashes • LLMs are accelerating decay The research is far more sobering than FinTwit

We tested AI sentiment on 13,207 financial news articles. What we discovered: The companies with the best signals had nothing in common. Not sector. Not size. Not industry. They shared one structural trait. And it predicts whether sentiment models work.

Markets don’t move because investors feel bullish or bearish. They move when investors feel something they didn’t expect to feel. That gap — the sentiment surprise — may be one of the most underappreciated drivers of returns. New research 👇 open.substack.com/pub/tommijohns…

Two completely different models looked at tech stocks… They only agreed 14 times out of 87. That’s the signal 👇 open.substack.com/pub/tommijohns…

Goldman Sachs just dropped a brutal Iran war scenario note. The real bombshell? Markets may be pricing the inflation shock... but not the growth shock. Strait of Hormuz is the whole game. Read this before the next oil move: open.substack.com/pub/tommijohns…

A single late-night post can move billions in market value. But here’s the real takeaway: Markets are not rational processors of fundamentals, they are real-time sentiment engines. We tested two independent AI systems across 321 companies and found something surprising: They

We built 2 AI systems to read the market. They disagreed 76% of the time. Both were right. One predicts stock returns… for exactly 1 day. This changes how you think about “efficient markets.” open.substack.com/pub/tommijohns…