Causal ML

Read these first

1. 01Confounding in Financial Data

   A confounder is a hidden variable that drives both a strategy's signal and its outcome, making a correlation look like a tradeable edge when it is really just two effects of the same underlying cause.

2. 02Policy Learning and Optimal Treatment Rules

   Estimating a treatment effect answers "did this help on average"; policy learning goes one step further and asks "who specifically should get it", a genuinely different, harder question.

3. 03Uplift Modelling

   Uplift modelling predicts how much a specific action changes an individual's behaviour, not just how likely they are to behave that way, the difference between targeting people who will respond and people who will respond because of you.

4. 04Double Machine Learning

   Double machine learning estimates the effect of one variable on an outcome while letting flexible ML models absorb every confounder, using a two-stage residual trick that keeps the final effect estimate from inheriting the ML models' own overfitting bias.

5. 05Estimating Conditional Average Treatment Effects With ML

   An average effect can hide the fact that a treatment helps some people and hurts others, CATE estimation uses machine learning to predict how the effect itself varies across individuals, not just its overall average.

Then the rest

1. 06Causal Features vs Merely Predictive Features
2. 07Causal Trees and Honest Splitting
3. 08ML Propensity Scores and Weight Clipping
4. 09Positivity and Overlap Violations
5. 10Qini Curves and Uplift Evaluation
6. 11Selection Bias in Trading and Execution Data
7. 12The DR-Learner
8. 13The S-Learner and T-Learner
9. 14Cross-Fitting and Sample Splitting
10. 15Deep IV and ML Instrumental Variables
11. 16Generalized Random Forests
12. 17Invariant Risk Minimization
13. 18Neyman Orthogonality
14. 19Off-Policy Evaluation of Decision Policies
15. 20Regularization Bias in Causal Estimation
16. 21Targeted Maximum Likelihood Estimation
17. 22The R-Learner and Robinson's Decomposition