Robust Inference for Games via Theoretical Guarantees
summary
Defines an empirical price of anarchy that takes the worst-case welfare ratio over all value distributions consistent with the observed bid distribution, uses partial identification plus a revenue covering bound computed from inverse price‑per‑click thresholds, and produces a cheap diagnostic that avoids solving high‑dimensional equilibria. Applied to Bing data (11 keywords) the bound is near 1 for most keywords but flags a few monopolized queries as problematic, so the method works on real data and gives practical guarantees.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:02Structural inference
- 3:41Practical questions
- 6:45Inference via identification
- 14:20Inference for First-price Auction Inference Equation
- 18:04Set inference and partial identification
- 38:03Revenue covering approach
related talks
27mEconometrics for Learning Agents
29mIndividual Fairness in Advertising Auctions through Inverse Proportionality
19mMarket Algorithms for Autobidding
01h 07mMulti-Dimensional and Non-Linear Mechanism Design and Approximation I
15mFundamental Limits Of Learning In Data-Driven Problems
09mMechanism Design Via Machine Learning: Overfitting, Incentives, and Privacy
43mLearning about Auctions
01h 14mCompetitive Analysis Meets Stochastic Input: Secretary Problems And Prophet Inequalities