MIT14_126S10_lec13 - Learning—Adjustment with persistent...

Info iconThis preview shows pages 1–7. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Learning—Adjustment with persistent noise 14.126 Game Theory Mihai Manea Muhamet Yildiz Main idea ¡ There will always be small but positive probability of mutation. ¡ Then, some of the strict Nash equilibria will not be “stochastically stable.” 1 General Procedure Stochastic Adjustment 1. Consider a game. A B 2. Specify a state 1. space Θ , e.g., the A 2,2 0,0 # of players playing a strategy . B 0,0 1,1 2. Θ = {AA,AB,BA,BB} 2 Stochastic Adjustment, continued 3. 3. Specify an adjustment AA AB BA BB dynamics, e.g., best- response dynamics, 1 0 0 0 0 0 1 0 0 1 0 0 0 0 0 1 AA with a transition matrix AB P, where P = P θ,ξ = Pr( θ at t+1| ξ at t) BA BB φ = a probability distribution, a column vector. Stochastic Adjustment, continued 4. Introduce a small 4. AA AB BA BB noise: Consider P ε , continuous in ε and P ε → P as ε → 0. Make sure that there P ε = exist a unique φ ε * s.t. φ ε * = P ε φ ε *. (1−ε) 2 (1−ε)ε (1−ε)ε ε 2 (1−ε)ε ε 2 (1−ε) 2 (1−ε)ε (1−ε)ε (1−ε) 2 ε 2 (1−ε)ε ε 2 (1−ε)ε (1−ε)ε (1−ε) 2 φ ε * = (1/4,1/4,1/4,1/4) T . 3 Stochastic Adjustment, continued 5. Verify that lim ε→0 φ ε * = φ * exists; compute φ *. (By continuity φ * = P φ *.) 6. Check that φ * is a point mass, i.e., φ *( θ *) = 1 for some θ *. The strategy profile at θ * is called stochastically stable equilibrium . Kandoori, Mailath & Rob 4 Coordination game A B q α ∗ = 1/3 A B (2,2) (0,0) (0,0) (1,1) α ∗ = 1/3 p Adjustment Process ¡ N = population size. ¡ θ t = # of players who play A at t . ¡ u A ( θ t ) = θ t /N u(A,A) + ( N- θ t )/N u( A , B ) ¡ θ t +1 = P( θ t ), where P( θ t ) > θ t Ù u A ( θ t ) > u B ( θ t ) & P( θ t ) = θ t Ù u A ( θ t ) = u B ( θ t ). ¡ Example: ⎧ ⎪ N if u A ( θ t ) > u B ( θ t ) P ( θ t ) = BR ( θ t ) = ⎨ θ t if u A ( θ t ) = u B ( θ t ) ⎪ ⎩ 0 if u A ( θ t ) < u B ( θ t ) 5 Noise ¡ Independently, each agent with probability 2ε mutates, and plays either of the strategies with equal probabilities....
View Full Document

This note was uploaded on 11/28/2011 for the course ECONOMICS - taught by Professor Muhammadyildiz during the Spring '05 term at University of Massachusetts Boston.

Page1 / 18

MIT14_126S10_lec13 - Learning—Adjustment with persistent...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online