1 pointby shatzakis6 hours ago1 comment
  • PaulHoule6 hours ago
    You familiar with the idea of

    https://en.wikipedia.org/wiki/Kelly_criterion

    ?

    This is the standard way of turning a probability-calibrated predictive model into a trading or betting strategy. Even the best AI model you're going to make is going to be making guesses about the direction of the financial market so you have to (1) make it smart about "knowing what it knows" and (2) build it's predictions into a systematic risk-managed trading strategy.

    • shatzakisan hour ago
      Hi Paul, thanks for chiming in. While risk management is included in the model framework, and we are using RAG (since before RAG was even a term, to eliminate hallucinations), the position sizing is relative to the users balance and their objectives (we dont give investment advice). We are testing different tools for risk management and position sizing, in terms of parameters that the user can configure, but the key value prop from this paper is using MCP for separating the probabilistic plane (parsing user intent with llms) from the deterministic plane (securely executing orders and managing positions etc..).