Finance is a Latent LLM

How market simulators connect to LLMs

As a practitioner in finance, who is also an active researcher in large language models (LLMs), I’ve found that finance and LLMs are closely intertwined. In a sense, the finance system can be seen as a latent large language models in the following aspects:


Environment: Market Simulators vs. Token Simulators

In finance, a market simulator is a synthetic finance world model that generates prices, order flows, volatility, and liquidity and indices to test your strategies and models.

LLMs simulate discrete token sequences to model language and reasoning trajectories.


Reinforcement Learning

RL functions as risk control in finance versus behavior alignment in LLMs.


Optimization & Allocation

Both are constrained allocation problems under uncertainty.


Scale & Infrastructure

At scale, systems engineering dominates algorithmic details.


Alpha–Beta vs. Scaling Laws

Finance uses alpha for excess returns and beta to model market exposure.

LLMs rely on scaling laws to predict final loss and determine when to stop training.

Both guide capital and compute allocation.


State & Control

Finance infers the state; LLMs are the state.


Prompt Engineering vs. Technical Analysis

Control without retraining.


Safety

In both systems, tail risk matters more than average performance.

Fundamental analysis v.s. pretraining

model training

pretraining + fine-tuning.