Algorithmic Trading and Machine Learning
summary
Reinforcement learning over two state variables—inventory remaining and time remaining—learns executable policies that beat an optimally tuned “submit-and-leave” baseline by 35% improvement on out‑of‑sample historical tests; adding order‑book features yields about a 13% additional gain. Smart‑order‑routing models treat each dark pool’s available liquidity as a distribution, estimate it with the Kaplan‑Meier estimator from censored observations, and use greedy allocation that provably converges (polynomial rate) to the optimal routing; results account only for mechanical impact, not behavioral effects.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 0:58Flash Crash
- 1:25Algorithmic Trading
- 5:09Market Microstructure
- 5:24Canonical Trading Problem
- 8:32Order Book
- 14:31Reinforcement Learning
- 25:01Mechanical Market Impact
- 28:04Features of the Order Book
- 31:56Modern Financial Markets
- 43:09Regulation of Financial Markets
- 45:55Machine Learning Challenges
- 48:31Simulations
related talks
18mKeynote: AI for Financial Services
37mIntelligent Autonomous Trading Systems - Are We There Yet?
16mUsing AI to solve complex business problems: optimizing...
21mLeveraging Multi-Armed Bandit Algorithms for Dynamic Decision Making
41mBetter Learning from the Past: Counterfactual / Batch RL
30mBuying Data Over Time: Approximately Optimal Strategies for Dynamic Data-Driven Decisions
34mReinforcement Learning in Business Applications
23mAlphaDow: Reinforcement Learning for Industrial Production Scheduling