Games and Economic

Behavior

Unit 3: Games with Sequential

Moves
 Game Trees
n Games with Sequential Moves are represented
by game trees.
¨ In game theory terminology: the game tree
represents the game in extensive form (all moves
are known).
n All game trees consist of:
1. Initial node (root) – the starting point of the game.
2. Action nodes – any node where a player chooses
from two or more decisions.
3. Branch – the representation of one action, moving
away from a node
4. Terminal Node – the representation of the end of the
game, where payoffs are realized.
 Game Trees
n Example: Who Picks up the Check?
game.
1. You and your friend have finished lunch.
2. The check is delivered when your friend is in
the bathroom.
3. You can pay now or defer to your friend.
4. He can pay when he gets back, or if he defers
you split the bill.
 Game Trees
You initial node or root (1)
branches (4)
action or decision nodes (2)
pay wait
Friend
-1,0
pay split
terminal nodes (3)
0,-1 -1/2,-1/2

payoffs (3)
n Which are the
1. initial nodes?
2. action nodes?
3. branches?
4. terminal nodes?
5. payoffs?
 Game Trees
n An outside player called
“Nature” can be used as
an external,
randomizing force in a
game.
n In all games, Nature has
no payoffs or interests.
n In this game, Chris
could think about
expected payoffs when
choosing between
“safe” and “risky.”
n In this game, payoffs
are written as (Ann,
Bob, Chris, Deb). It is
important to keep track
of these conventions.
 Moves vs. Strategies
n A single action
taken at each node
is a move.
n Having a move for
every node (and
thus every
eventuality) is a
strategy.
n For Bob, Chris, and
Deb, since they
move only once,
moves and
strategies are
equivalent.
 Moves vs. Strategies
n Ann will only move
twice if she chooses
stop.
¨ But it still is
important to list her
moves at the other
node in her strategy
(either GU or GD).
¨ Also applies to Bob
and Deb’s moves.
n We need to know
Ann’s motivations
for equilibrium.

The Bottom Line: List ALL Moves with Strategies, List ALL Strategies with Equilibrium
 Rules for Trees
A. Trees consist of nodes and branches
1. Branches connect to nodes
2. Each game has one start: a decision node called the initial
node.
3. Each game has at least one end called the terminal node that
has payoffs for all the players associated with it.
B. Branches represent all possible actions in the game.
1. Each decision node must have at least one node flowing from
it. All non-initial nodes must have exactly one branch leading
to it.
2. Terminal nodes have exactly one branch flowing to them.
C. Game trees are typically drawn left to right (or top-
down), but other conventions are fine.
Example S1
 Solving Games by Using Trees
n Carmen is a teenager who is deciding whether
to smoke.
n She has two decisions
1. Whether to smoke
2. Whether to continue smoking
 Solving Games by Using Trees
n Carmen likes to try new things, so being able to try smoking gives
her a positive payoff.
n But smoking causes long term health risks so continuing long term
has negative payoffs.
n But this analysis ignores the problem of addiction, once Carmen
has tried smoking for a while she will have different tastes. This is
no longer a decision, it is a strategic game.
 Solving Games by Using Trees
n If Carmen chooses to smoke, her future self will
experience withdrawal when quitting, changing her
payoffs.
n Carmen should see that here future self will choose to
continue to smoke.
n Thus, we don’t need to think about a specific branch and
node. We cross it out.
Future
Carmen
prefers
continuing
 Solving Games by Using Trees
n Today’s Carmen must choose between smoking
forever and not smoking.
n She prefers not trying smoking.
n We may cross out two more branches.
n Outcome: TC chooses N, FC chooses C, (0,0)

Today’s Carmen
prefers not smoking
 Solving Games by Using Trees
n Checkmarks are an alternative, and my
preferred method.
¨ Much clearer
n Always start at the end and work back.
 Rollback Equilibrium
n Rollback – The method of looking ahead and
reasoning back to determine behavior in
sequential move games. AKA backward
induction.
n Rollback equilibrium – the collection of
strategies outcome that arises when all players
optimally use rollback analysis. AKA subgame
perfect equilibrium (Selten).
n Theorem: All finite, sequential move games
have at least one rollback equilibrium (most
have exactly one).
n Rollback Equilibrium: TC chooses N, FC
chooses C.
1. In each of the
following games,
how many pure
strategies
(complete plans of
actions) are
available to each
player? List all of
the strategies for
each player.

2. Identify the
rollback
equilibrium
outcome and
complete strategy
for each player.
 Example of Game with 3 Players
n Emily, Nina, and Talia all live on the same
street.
n Each is asked to contribute to a flower garden.
n Each prefers bigger gardens, but also prefers
not to give a flower because of the cost (effort,
money).
n If all contribute the garden will be beautiful, if
none contribute, hideous. To be “pleasant” >1
must contribute.
n Emily moves first, Nina second, Talia third.
 Ordinal Rankings
n For each player there are four outcomes:
1. She does not contribute, but the others do (pleasant
garden, no contribution).
2. She contributes, and one or both of the others do
(pleasant garden, cost to contribute).
3. She does not contribute, only one or neither of the
others do (unpleasant garden, no contribution).
4. She contributes, but neither of the others do.
(unpleasant garden, cost to contribute).
n Clearly, 1>2 and 3>4, we’ll assume 2>3, so
1>2>3>4. Assign ordinal payoffs 4,3,2,1.
Game Tree
Rollback Step 1-Talia
Rollback Step 2-Nina
Rollback Step 1-Emily
 Rollback Outcome
n What happens: Emily
plays don’t, Nina
plays contribute, and
Talia plays contribute.
n Payoffs: (4,3,3)
n Emily has a clear
advantage due to
moving first!
 Important terms for this game
n Path of play – A route through
the game tree linking nodes,
that is consistent with the rules
of the game.
n Equilibrium path of play – path
of play when players choose
their rollback equilibrium
strategies.
n Our equilibrium path is affected
by a player’s knowledge of
what players do on non-
equilibrium paths.
¨ Need complete knowledge of
strategies for this result.
 Rollback Equilibrium
n To have an
equilibrium we need a
listing of strategies
which are complete.
¨ Emily – D (a)
¨ Nina – DC (b,c)
¨ Talia – DCCD (d,e,f,g)
n There are other
possible non-
equilibrium strategies.
 Other Strategies
n Possible Strategies
¨ Emily: C,D (2)
¨ Nina: CC, CD, DC, DD (4)
¨ Talia: CCCC, CCCD, CCDC, CCDD, CDCC,
CDCD, CDDC, CDDD, DCCC, DCCD, DCDC,
DCDD, DDCC, DDCD, DDDC, DDDD (16)
n To find their optimal strategies, Nina
figures out Talia’s from 16 possibilities.
Emily finds Talia’s and then Nina’s.
 Conclusions
n In solving a sequential game from a tree, you
will find three distinct concepts:
1. The listing of the strategies available to each player.
2. The optimal strategy from that listing.
3. The actual path of play.
n Emily had a first-mover advantage in that
game, a strategic advantage from making the
first move.
¨ There may be a second-mover advantage as well
(think rock-paper-scissors).
S9
Adding Multiple Moves for Each Player
n More complicated sequential games
allow players to alternate moves back
and forth.
¨ Our examples will be tic-tac-toe, chess and
checkers.
n Tic-tac-toe example will be a simpler
version of the game on a 2x2 grid.
 2x2 Tic-Tac-Toe
n Rollback (or common sense) shows us
that the strategies chosen do not matter,
the first player always wins.
 3x3 Tic-Tac-Toe
n In the 2x2 game there are only 4x3x2x1
moves (24)
n In the 3x3 game there are
9x8x7x6x5x4x3x2x1 moves (362,880)
¨ 504 moves after just first two (9x8x7)
n The 3x3 game can be simplified so that
moves are corner, middle or side.
¨ 2ndplayer can always force a tie.
¨ Game has been solved. (pg. 65)
 Checkers
n Checkers has 5x1020
possible combinations to the
game.
¨ The game was solved by the
Chinook program in 2007.
¨ Program used both rollback
and intermediate valuation
function.
n The game is “weakly solved”
1. Program proved a tie can be
generated from the start.
2. Once there are 10 pieces
left, all positions can be
solved by rollback. (39x1012)
 Chess
n Chess is a finite game.
¨ Rules state that a draw
occurs if the board
position repeats 3 times.
n By our theorem, there
must exist a Subgame
Perfect Equilibrium
(Rollback Equilibrium)
n Moves
1. Not all possible moves
have been counted.
2. Estimated at 10120.
3. 400 possibilities after two
moves.
4. A supercomputer would
take 10100 years to solve.
 Chess
n How a computer plays
without a solution:
1. Look ahead at all
possibilities 5-10 moves
ahead.
2. Assign values to each
position using an
intermediate-value
function, a rule that
assigns such payoffs.
3. Rollback and solve for
those 5-10 moves.
 Chess
n People use similar
techniques:
1. Openings can be
memorized, so first
dozen moves are known.
2. Points can be assigned
to pieces.
3. Engames can be solved.
4. Instinct from repetition
matters.
n These techniques can
be modeled into
computers.
 Chess
n Artificial intelligence,
machine learning and
deep learning have
greatly altered
computer performance.
n Computers are fed data
and make up own
algorithms to
synthetically create
knowledge.
n A computer that only
played itself (400
times!) was able to beat
the best chess
program.
 Go
n Go is more complex than
chess; it features 10172 board
positions and 10360 nods.
n Considered too complex for a
human-programmed
computer to beat an human
expert.
¨ Brute force cannot work in this
level of complexity, need
intuition.
n However, recent AI advances
have led to self-trained
computers to beat expert
humans.
Simplifying Games
 Ultimatum Game
n Player 1 has $1 to divide, between him and
player two.
¨ Player 1 chooses allocation (50/50, 90/10, 99/1)
¨ Player 2 can reject and both players receive nothing.
n Player 1 often offers above 99/1.
n Player 2 often rejects.
n 50/50 most common.
n Failure of Rollback?
 Centipede Game

n Often game goes beyond first node.

n Failure of Rollback?