Future Research Topics

Related Topics

• Optimal Execution

  • Optimal liquidation or acqurization within 15-30 minutes with agent trained on the simulated environment.

• Optimal Scheduling with Predicted Trading Volume

  • In vwap strategies, split the task into smaller sizes according to the predicted trading volume.

• Order-flow Generating

  • Mathematical Perspective: Order flow as a general spatial point process

  • Time-series Forecasting for Order Flow

  • Order Flow Generating by Large Language Models

• Representation Learning

  • Representation Learning for the States/Observations

  • In the Optimal Execution, e.g. the all snapshots of limit order book in the past 30 seconds.

  • VAE, GAN and other encoder models.

• Price Impact Research through Market Clearing

  • Mathematical Perspective: Market clearing as a deterministic operator acting on the distributions of buy and sell orders.

  • Calculate the price impact without the assumption of impact function

• Indirect Market Impact

  • Agent’s Impact on Triggering the Modification of other Agents’ Actions

  • Different from the price impact, which is the direct maret impact.

• Agent Based Modelling/Simulation

  • Generative adversarial network approach simulation

  • Market Simulation

• Recover Trader’s Reward Function

  • Recover Trader’s Reward Function by Inverse RL

• Unsupervised Environment Design

  • Adversarial Learning by the differentiable environment

Related Papers

ChatGPT:

• Generative Agents: Interactive Simulacra of Human Behavior

• Large Language Models as Simulated Economic Agents: References

All related ICAIF and OMI Research Newsletter papers are included.

• Related Sections in ICAIF

  • Simulated Markets

  • Learning Trading Strategies

  • Forecasting Financial Data

High Related ICAIF Papers:

• ICAIF2022

  • Dynamic Calibration of Order Flow Models with Generative Adversarial Networks

  • Learning to simulate realistic limit order book markets from data as a World Agent

  • Cost-Efficient Reinforcement Learning for Optimal Trade Execution on Dynamic Market Environment

• ICAIF2021

  • Towards a fully RL-based Market Simulator

  • Towards Realistic Market Simulations: a Generative Adversarial Networks Approach

  • Learning to simulate realistic limit order book markets from data as a World Agent.

  • Bit by Bit: How to realistically simulate a Crypto-Exchange

• ICAIF2020

  • Get Real: Realism Metrics for Robust Limit Order

  • Multi-Agent Reinforcement Learning in a Realistic Limit Order Book Market Simulation

  • Deep Reinforcement Learning for Automated Stock Trading: An Ensemble Strategy

• Mid Related ICAIF Papers

• OMI Research Newsletter

  • Market Microstructure

    • Many learning agents interacting with an agent-based market model, April 2023

    • Asynchronous Deep Double Duelling Q-Learning for Trading-Signal Execution in Limit Order Book Markets, February 2023

    • Model-based gym environments for limit order book trading, Octorber 2022

    • A simple learning agent interacting with an agent-based market model, Septermber 2022

  • Time Series Forecasting

• Other related papers

  • StockGAN: Robust Stock Price Prediction Using GAN Algorithm.

  • Generating Realistic Stock Market Order Streams

  • Deep Reinforcement Learning in Agent Based Financial Market Simulation

  • Optimal execution with stochastic delay

Related Techniques

• Transformers

  • Time Series Forecasting with Transformers:

  • Transformers in Time Series: A Survey

  • Transformer in Low Signal-noise Ratio System:

  • Generating Long Sequences with Sparse Transformers

• Long Sequence Modelling

  • Efficiently Modeling Long Sequences with Structured State Spaces(S4)

  • Simplified state space layers for sequence modeling(S5)

• Unsupervised Environment Design

  • Emergent Complexity and Zero-shot Transfer via Unsupervised Environment Design

• Behavior Cloning

  • BC-IRL: Learning Generalizable Reward Functions from Demonstrations

Related Issues

Hard to generalize. There might be several reasons jointly contribute to this situation:

1. The signal-to-noise ratio of financial market data is much lower than that of other artificial intelligence fields.

2. The financial market is not a closed system and will evolve on its own.

3. The financial market is a derivative of the economy and therefore can be impacted by external factors.