General

  • similar to Backpropagation in Machine Learning
  • solve from the “deepest” subgame to the “shallowest” subgame
  • as long as there is no indifference i.e. a player can chose either way and have equal payoffs, there is always exactly 1 subgame perfect strategy
    • if there is just 1 Nash Equilibrium, it must be subgame perfect
    • if there is just 1 subgame i.e. the game itself, the Nash Equilibrium is subgame perfect
    • always applicable
  • eliminates Non-Credible Threats
    • only in 1-shot game
    • reputation can bring the empty threats back into perspective
  • in simultaneous games all Nash Equilibrium are subgame perfect

Process Repeated Games

  • start at last decision
  • treat as one-shot game and solve for Nash Equilibrium
  • take a look at next last decision
  • taking the nash of the previous game into account, what is the nash now?
  • repeat until you found a loop
    • mostly will be right away with second to last decision