Causal inference and scenario generation within Just Eat Takeaway.com
summary
Just Eat Takeaway.com estimates causal effects from city-level quasi‑experiments using difference-in-difference and synthetic‑control style matching, augments matching with Google's causalimpact (Bayesian structural time series) for credibility intervals, and flags results that disagree across models. For scenarios, a multivariate forecaster—Temporal Fusion Transformer—injects future price or promo paths and produces distributions; synthetic minute‑by‑minute orders come from Gaussian copulas and VAE/GAN options so a router simulation can be stress‑tested; assumptions about parallel trends and sparsity determine confidence.
AI-written summary · about
chapters (8)
- 0:00Start
- 3:55Quasi-experimentation
- 6:55Difference-in-Differences (ctd.)
- 13:11Causal Inference
- 14:29Scenario Generation
- 15:27Temporal Fusion Transformer (TFT)Schoo
- 18:15Temporal Fusion Transformer (ctd.)
- 21:30Synthetic Data Generation (ctd.)
related talks
23mUber's Synthetic Control
11mCausal Inference For Socio-Economic & Engineering Systems
36mWhat-if- Causal reasoning meets Bayesian Inference
41mOn the Path to Causal Inference
32mCausal Forecasting: How to disentangle causal effects, while controlling for ...
35mGeo Experiments and Causal Impact in Incrementality Testing
39mCausal Effect Estimation in Practice: Lessons Learned from E-commerce & Banking
43mBeyond Just Prediction: Causal Thinking in Machine Learning