Auto-bidding, Auctions, and Ad Allocation

on YouTube ↗42m

" summary

Autobidding agents model targets (e.g., Target CPA) via an integer program relaxed to an LP whose dual yields a simple uniform bid scaling — bid = multiplier × conversion-rate — implementable in truthful auctions. Using LP duality and equilibrium analysis, second-price/VCG equilibria suffer price of anarchy 2 (can halve target-weighted conversions), but a randomized truthful auction (RAND) cuts that to ~1.89 for two bidders and a randomized nontruthful design reaches ~1.8; as bidders grow large the bound reverts to 2.

AI-written summary · about

" chapters (16)

1. 0:00Ad auctions and traditional bidding
2. 3:19Bidder constraints and budgeted allocation
3. 5:53Autobidding and its role
4. 7:21Target CPA autobidding
5. 9:58Optimizing autobidding agents
6. 17:10Uniform bidding formula and optimality
7. 18:13Matching with autobidders
8. 20:47Welfare in autobidding equilibria
9. 24:25Price of anarchy in auctions
10. 28:45Improving auction welfare
11. 33:25Randomized auctions and limits
12. 35:15First-price auctions inefficiency
13. 36:17Randomization and nontruthfulness
14. 37:05Future directions in mechanism design
15. 39:12Open questions and conclusion
16. 40:26Q&A on autobidding strategies