Chapters
Every chapter, grouped by Part. Use the card metadata to calibrate how much trust to place in a chapter's specific claims.
Foundations
-
Potential Outcomes and the Frisch-Waugh-Lovell Theorem
Causal estimands, the fundamental problem of causal inference, identification under unconfoundedness + overlap, and the Frisch-Waugh-Lovell theorem as the bridge to Double Machine Learning.
-
Neyman Orthogonality and Double Machine Learning
Regularization bias in ML nuisance estimation, Neyman orthogonality as the fix, cross-fitting for valid inference, the full DML algorithm, EconML implementation, CATE, sensitivity analysis, and the influence-function view.
-
Comprehensive Validation Framework
A seven-method validation battery on synthetic DGPs with known ground truth: confounding-strength and sample-size sensitivity, cross-fitting and nuisance-model selection, CATE recovery, power analysis, runtime benchmarks, and a practitioner decision tree.
Integration
-
Cross-Sectional Application: Price Elasticity with Sensitivity Analysis
DML on the Dominick's orange-juice data: price-elasticity estimation, OLS baselines, overlap diagnostics for continuous treatment, Rosenbaum bounds / E-values / partial identification, a contrastive fragile gasoline example, brand- and income-level heterogeneous sensitivity, and a reporting checklist.
-
Temporal PLR DML for Time Series
Extending DML to time series: why random K-fold leaks future data, time-series cross-validation (expanding/sliding/purged windows with gaps), HAC/Newey-West standard errors with kernel and bandwidth choice, the TemporalPLRDML scalar estimator (lagged controls + temporal CV + HAC), and a Monte Carlo coverage check. Not recursive dynamic g-estimation.
-
Panel DML and Rolling Window Methods
DML for panel data: the within (fixed-effects) transformation, Panel DML combining fixed effects with ML nuisance models, cluster-robust standard errors, Rolling Window DML for time-varying effects and structural-break detection, with insurance and macro applications.
-
FRED Integration: Macroeconomic Controls
Integrating FRED macroeconomic controls into time-series DML: the macro-confounding framework and omitted-macro bias, the FREDLoader (series sets, caching, frequency conversion, transforms, missing data), augmenting controls with Ridge vs. Lasso and HAC under persistence, plus insurance-rate and monetary-policy applications. Uses synthetic FRED data for reproducible examples.
Synthesis
-
Competitor Pricing: An Insurance Application
Synthesizing the toolkit on insurance/annuity competitor pricing: a configurable synthetic DGP (simple/moderate/full), Panel DML with product fixed effects and cluster-robust SEs, FRED macro controls, rolling-window regime detection, and an end-to-end pipeline plus a real-data handoff checklist.
-
Heterogeneity Analysis
Conditional Average Treatment Effects: CATE identification and the R-learner, CausalForestDML and the Best Linear Predictor (as external-package / optional methodology), heterogeneity visualization, and subgroup/policy trees — applied to insurance pricing. EconML code is external illustration, not a verified repo API.
-
Research Pipeline Utilities for Causal Inference
Research-pipeline patterns for deployed causal inference (book-companion, not a production framework): why effect accuracy is unmonitorable, DML model serialization/versioning, CATE inference-API sketches, causal-specific monitoring (overlap, treatment shift, nuisance degradation, effect stability), causal retraining triggers, an end-to-end InsuranceDMLPipeline, and a case study.