•
Cheridito & Weiss (2025) use actor-critic RL to mix market and multi-level limit orders in impactful LOB simulations. Plain-language intro. Not investment advice.
Takeaway. Modern execution is not only "how fast to market-order"—agents can learn when to post multi-level limits vs cross with market orders inside a simulated, impactful LOB.
Reinforcement Learning for Trade Execution with Market and Limit Orders (Patrick Cheridito, Moritz Weiss, 2025) studies that mix with actor-critic methods and carefully chosen action distributions.
Fit for learners: Actor-critic RL allocating market and multi-level limit orders via logistic-normal policies in impactful LOB sims.
This Hub note is a plain-language intro. Not investment advice. Classics first: Almgren–Chriss, then Double DQN optimal execution if you want earlier RL context.
Pure market-order schedules ignore resting liquidity you could have provided. Pure limit schedules risk non-fill. Hand-tuned hybrids are brittle. RL promises adaptive policies—but only inside a faithful simulator, never as a magic profit button.
Read the PDF for architecture and experiments. Educational curiosity only.
RL + LOB is a popular project theme. This paper shows a research-grade problem framing: action space design matters as much as the network.
Tiny evening exercise: without RL, write a rule-based hybrid: if spread ≤ 1 tick post a limit at touch else send a tiny market child; run it on a fake mid + random fills loop and log fill rate vs slippage proxy.
Landing: arXiv:2507.06345
Not investment advice. This Hub note is educational only. The paper is research—not a licence to trade, quote, or execute with real capital.
Single course CTA: SoloQuant Academy's Introduction to Python Programming—get comfortable with functions, files, and plots first; then these microstructure PDFs become much less scary.