ML for Trading & Execution

Read these first

1. 01Learning Execution Cost Models From Fills

   A trading desk's own fill history is a goldmine for predicting execution cost, but it's also a biased sample, because the orders that got filled aggressively are not a random draw from all the orders the desk might ever want to place.

2. 02Learning Market Impact From Your Own Fills

   Estimating how much your own trading moves the price sounds like a simple regression problem, but your fills are entangled with the very price moves you're trying to measure, so the naive estimate is biased in a way that a bigger dataset does not fix.

3. 03Learning Quoting Policies for Market Making

   A market maker's quote is a decision made thousands of times a day under a feedback loop no ordinary supervised model captures, where you quote changes who trades with you, which changes your inventory, which changes where you should quote next.

4. 04Short-Horizon Price Prediction From the Order Book

   Predicting whether the price ticks up or down in the next second from order book imbalance is one of the cleanest supervised-learning problems in finance, and one of the easiest to accidentally solve by cheating, because the book itself reacts to trades almost instantly.

Then the rest

1. 05Classifying Adverse Selection at Fill Time
2. 06Counterfactual Evaluation of Execution Policies
3. 07Features for Execution Cost Prediction
4. 08Fill Prediction: Features, Labels and Evaluation
5. 09Fill Simulator Fidelity for Training Execution Models
6. 10Intraday Volatility Forecasts for Execution Scheduling
7. 11Intraday Volume Prediction With Gradient Boosting
8. 12Label Design for Microstructure Prediction
9. 13Learned Flow Toxicity Scores
10. 14Learned Participation-Rate Policies
11. 15Learning Cancel-and-Replace Policies
12. 16Learning Inventory Skew From Fill Data
13. 17Nonparametric Impact Curve Estimation
14. 18Normalising Order-Book Snapshots for Neural Models
15. 19Predicting Markouts With Machine Learning
16. 20Trade Sign and Aggressor Flow Features
17. 21Training Cost Models on Parent Order Data
18. 22Venue Selection With Contextual Bandits