H7: Principal-Agent theory
Introduction: information economics
=> In many business applications, agents must take decisions while at an
information disadvantage (hiring specialists, investing in early business…n
- Last chapter: agents facing opponents with private information as given
- However, players can interact and learn or deliver information, reducing this
asymmetry
=> Information economics: studies the strategic value of information
- How valuable is it to send a credible message
- How much is it worth paying for information
- Is it better to monitor or to incentivize
- What market segments are unprofitable under asymmetric information
=> Focus on category of games “Principal-agent” models
Principal-agent models
Algemeen
- Principal has a goal, but needs the help of an agent to achieve it
- Agents have their own goal and either know something more (i)
(hidden information) or can’t be monitored while acting (ii) (hidden action)
- Principal’s goal: induce the agent to take the desired action or to reveal its
information
Principal-agent taxonomy: manager decisions
1
, Sequential games
=> Dynamic game with asymmetric information
Categorization
- Bayesian games: some players are better informed than others, and
uninformed players need to form belief about the other players’ type
- Extensive games: players can observe the other player’s moves and react
- Sequential (extensive + Bayesian): Players “learn” by observing other’s
actions or can “bluff” => Mislearning/deception can be induced and is a
possible strategy now
=> In general, sequential games can be very hard to solve, and have many
(uninteresting) equilibria
- We will see:
- One-sides asymmetric information
- Two stages
- Two player
=> Sufficient for most Principal-agent problems
Abstract form sequential games (simplified)
- A set of players (e.g. P = {BurgerKing(BK), McDonald(MD)})
- One-sided asymmetric information: we assume that only one player has
more than 1 type
=> e.g. 𝛀BK = {Efficient(BKE), Inefficient(BKI)}
=> Uninformed player (MD) have beliefs over types: MD believes P(BKI)=0.7
- For each combination of types, we have an extensive game (with histories
and player function)
=> Histories: Hi = {ø, Enter, NotEnter, EnterEnter, EnterNotEnter,
NotEnterEnter, NotEnterNotEnter}
=> Players functions: BK chooses first, MD chooses after seeing “Enter” or
“NotEnter”
=> Players’ payoffs for each terminal history
- A strategy for a player specifies an action for each non-terminal-histort and
each type of that player
=> BK strategy: “Enter if efficient, NotEnter if inefficient” is valid strategy
=> MD strategy: “Enter if BK entered and NotEnter if BK NotEnter” valid strat.
Sequential games: extensive form
2
Introduction: information economics
=> In many business applications, agents must take decisions while at an
information disadvantage (hiring specialists, investing in early business…n
- Last chapter: agents facing opponents with private information as given
- However, players can interact and learn or deliver information, reducing this
asymmetry
=> Information economics: studies the strategic value of information
- How valuable is it to send a credible message
- How much is it worth paying for information
- Is it better to monitor or to incentivize
- What market segments are unprofitable under asymmetric information
=> Focus on category of games “Principal-agent” models
Principal-agent models
Algemeen
- Principal has a goal, but needs the help of an agent to achieve it
- Agents have their own goal and either know something more (i)
(hidden information) or can’t be monitored while acting (ii) (hidden action)
- Principal’s goal: induce the agent to take the desired action or to reveal its
information
Principal-agent taxonomy: manager decisions
1
, Sequential games
=> Dynamic game with asymmetric information
Categorization
- Bayesian games: some players are better informed than others, and
uninformed players need to form belief about the other players’ type
- Extensive games: players can observe the other player’s moves and react
- Sequential (extensive + Bayesian): Players “learn” by observing other’s
actions or can “bluff” => Mislearning/deception can be induced and is a
possible strategy now
=> In general, sequential games can be very hard to solve, and have many
(uninteresting) equilibria
- We will see:
- One-sides asymmetric information
- Two stages
- Two player
=> Sufficient for most Principal-agent problems
Abstract form sequential games (simplified)
- A set of players (e.g. P = {BurgerKing(BK), McDonald(MD)})
- One-sided asymmetric information: we assume that only one player has
more than 1 type
=> e.g. 𝛀BK = {Efficient(BKE), Inefficient(BKI)}
=> Uninformed player (MD) have beliefs over types: MD believes P(BKI)=0.7
- For each combination of types, we have an extensive game (with histories
and player function)
=> Histories: Hi = {ø, Enter, NotEnter, EnterEnter, EnterNotEnter,
NotEnterEnter, NotEnterNotEnter}
=> Players functions: BK chooses first, MD chooses after seeing “Enter” or
“NotEnter”
=> Players’ payoffs for each terminal history
- A strategy for a player specifies an action for each non-terminal-histort and
each type of that player
=> BK strategy: “Enter if efficient, NotEnter if inefficient” is valid strategy
=> MD strategy: “Enter if BK entered and NotEnter if BK NotEnter” valid strat.
Sequential games: extensive form
2
