Jane Street 式研究流程里用 Jev 筛假设

hammertime @hammertime_one原推发布于 2026.10.10

在 X 上打开原推

Jane Street generated $39,600,000,000 in revenue in 2025. That's more trading revenue than JPMorgan, Goldman Sachs or Morgan Stanley individually. And most retail traders couldn't explain how Jane Street actually makes money. The firm started in 2000 with a small group of traders and grew into one of the most powerful market makers on Wall Street. Its business isn't built around predicting whether Bitcoin or Nvidia will be higher next month. A major part of it involves pricing thousands of instruments, managing inventory, hedging exposures and exploiting small differences between related markets. The margins on individual trades can be tiny, but the operation runs at enormous scale. What's interesting is the amount of research required before those trading decisions ever reach production. A pricing model needs historical data, realistic transaction costs, sensitivity testing and rules for when it should stop trading. Every improvement competes against the existing system, and an apparently profitable change can be rejected because it increases risk or behaves poorly in certain market conditions. Claude makes it possible to automate pieces of that research. JEV engineering could organize the workflow so simple checks run cheaply, failed hypotheses get discarded early and more expensive reasoning is reserved for strategies that deserve investigation. Instead of repeatedly asking a model for better trading ideas, you give it a process for determining which ideas are worth pursuing. The part I'm interested in is how much of this becomes accessible to someone without an institutional research department. Horizon lets you start with a strategy written in plain English and examine its historical performance, rather than spending the weekend building a backtesting environment. Jane Street spent 25 years building its trading infrastructure. The interesting development is how much cheaper the first stage of quantitative research is becoming.

这条是引用着另一条写的：x.com/hammertime_one/status/2108916760673964533