Most of the information that moves markets is qualitative. We use LLMs to unlock this data for systematic strategies for the first time.
Locked in text
Systematic strategies run on numbers. But what moves a stock often never becomes one: a shift in strategy, a new risk, a promise quietly dropped. Sentiment scores and keyword counts capture how that language sounds, not what it says. Understanding it still takes an analyst, reading one company at a time.
Write a feature the way you'd brief an analyst.
Agents read whatever it takes, from one earnings call to every filing, transcript, article and report the question touches.
Define the signals you care about, and Stated measures them across your entire universe, turning the written record into data your models can use.
| date | subject | demand_corroborated | debt_due_2y | kpi_dropped |
|---|
Built for backtests
LLMs know how history turned out. Ask one to predict a bank's prospects in March 2023 and it may simply remember what happened next. That lookahead bias looks like alpha in a backtest and disappears in live trading.
So our agents never predict anything. They describe a company as it stood on a date: guidance withdrawn, prices raised, a disclosure dropped. The forecasting is left to your models. And because knowing the ending doesn't help describe the starting point, every description is point-in-time by construction.
Read the paper PDFWe are reaffirming our full-year guidance for loan growth and net interest income.
MaintainedDeposit outflows since March 10 will bring net interest income below our prior outlook.
LoweredGiven the uncertainty, we are withdrawing our outlook for the remainder of the year.
WithdrawnWe now expect net interest income to decline 8 to 10% for the full year.
ReinstatedWe are raising the low end of our range for net interest income.
RaisedOpen source, and it runs on the AI agents you already use.