Bayesian Learning in Social Networks

on YouTube ↗57m

" summary

Revisits the 2011 Review of Economic Studies model (with Acemoglu, Dahleh, Ozdaglar) where agents observe past actions through a stochastic network; proves a necessary and sufficient condition for asymptotic learning when signals are unbounded private beliefs and identifies network properties that guarantee learning regardless of signal structure. Uses a martingale aggregation argument for global results and an improvement function / Lyapunov construction (a computable Z map and threshold rules) to show expanding observations (largest past neighbor → ∞) drives learning, while bounded signals require aggregators or growing neighborhoods.

AI-written summary · about

" chapters (15)

1. 0:00Intro
2. 1:42The Starting point of the Social Learning Literature
3. 2:41The BHW 1992 Model: An Example
4. 10:15The Martingale Approach
5. 12:47Learning over Social Networks
6. 17:48Our Model - The Social Network
7. 19:19An Example of a Social Network
8. 20:35Solution Concept
9. 21:47Bayesian Learning over Networks is Hard to Analyze
10. 23:26Our Approach: An Improvement Function
11. 25:35Observing a Single Decision
12. 28:58The Lyapunov Function
13. 31:45Generalizing to Deterministic Networks
14. 48:55Two Follow-Up Papers (with Evan Sadler)
15. 50:54Concluding Thoughts