Collected molecules will appear here. Add from search or explore.
Implementation of a Model-based Reinforcement Learning framework (VAE + MDN-RNN) for simulating and predicting Limit Order Book (LOB) dynamics, based on the J.P. Morgan AI Research 'World Models' paper.
Defensibility
stars
11
forks
5
This project is a personal implementation of a 2019 research paper from J.P. Morgan AI Research. With only 11 stars and zero activity for over 1000 days, it serves primarily as an academic reference rather than a production-ready tool. The defensibility is low because the core 'World Models' architecture (VAE/RNN/Controller) has since been superseded in the finance domain by more advanced Transformer-based architectures and more robust LOB simulators like Gym-LOB or DeepLOB. Frontier labs (OpenAI, Google) are unlikely to compete here as LOB modeling is a highly specialized financial niche. The primary 'moat' in this space is access to high-fidelity, tick-level historical data, which this repository does not provide. From a competitive standpoint, any quantitative trading firm would likely build a more optimized version of this internally rather than utilizing an unmaintained 3-year-old codebase. It represents a 'reimplementation' of existing research without novel additions or significant community traction.
TECH STACK
INTEGRATION
reference_implementation
READINESS