Power Dynamics in Organizations

Li et al. 2017

Introduction
Definition of Power in Organizations
Formal Authority (enforced by 3rd parts)
Informal Authority as relational contract (self-enforced, delegation)
Recall Baker et al. (1999) “Informal Authority in Organizations”
Motivating Case
Powerful divisions in firms still enjoy their power even after the end of their heyday.
A Paradox of Organizational Inefficiency
• Good power structure has an impact on performance.
• Power structures can be imitated.
• The agent enjoys private info, but some trend/opportunity are publicly available.
Then why inefficient power structure still exists?
1
Introduction
A Theory of Path-dependency
• Monetary transfer as incentive is constrained by the environment (law for instance)
• Power as payment, instead, has a long history in Sociology, Organization Studies.
• By promising to let the agent abuse their power tomorrow, the principal creates
incentive of the agent to make good use of his power today.
• Even if there will be public opportunities tomorrow, the principal promises to
abandon her power to leverage them in the future to co-opt the agent today.
Explain Organizational Differences
• The current power structure is the result of earlier events.
• Delegation (Centralization) occurs when agent brings (not) enough profits in the past.
• The previous performance of the agent was also influenced by opportunities (luck).
1) The power structure of any firm today is to some extent random.
2) Established firms have a hard time adjust themselves, because they are already
bounded by power promises in the past. 2
Model
Model Attributes
• Stochastic game
There are two state of the world:
1) Availability of the principal’s preferred project (baseline + extension);
2) Availability of a publicly observable good project for the principal (extension only).
• A repeated game with imperfect public monitoring
The agent has private information (moral hazard, on/off schedule deviations) that only
he knows the state of the world.
• Players can recall the history
The players uses what happened in the past to adjust the strategies.
• Monetary transfer is not available
The principal manipulates the agent’s continuation payoff by award/deprive power in the
future. Mechanism design using optimal relational contract of power arrangement,
rather than can commit to a formal contract of salary.
3
Model
Methodology: The Perfect Public Equilibrium (PPE) Approach
Developed by APS 1990, FLM 1994.
• Basic idea
“Decomposition” in dynamic programing.
Breaking a complex optimization problem (derive the optimal relational contract) into
more tractable problems (characterize the payoff set, derive its frontier, then see what
combination of “action pairs” maximize the payoff).
• The advantage of PPE
Public strategy: The players only act according to public signal/actions in the history.
For repeated games with imperfect monitoring, PPE is the most intuitive and tractable.
Why?
1) If other players follow public strategies, you have a public strategy as best response.
2) Every sequential equilibrium in pure strategy is payoff-equivalent to some PPE.
3) Recursive property: the PPE payoff set available at time 0 is the same at any period.
4
Model
Set up
• Two players: the principal (she) and the agent (he).
• Stochastic time periods:
• Stage game:
There are 4 projects, .
Give the agent and the principal if successful, if fail.
Projects/Payoffs Successful (B>b, Fail
net of cost c)
Default
Agent preferred
Principal preferred w/p
w/p
Disastrous

Information Stage Both players know n=0 is the default project, only the agent know
which is which of the remaining projects, and whether the principal’s preferred project is
5
available.
Model
Set up
Projects/Payoffs Successful (B>b) Fail
Default
Agent preferred
Principal preferred w/p
w/p
Disastrous

Advice Stage The agent send to the principal. (cheap talk)

Choice Stage The principal chooses to carry out a project.
Implementation Stage Players respectively choose Pay a cost if exert effort. The project
is successful iff both exert effort. Then payoffs are realized.
Public Randomization A public randomization device is realized. (uid)
The public randomization device is to help the two players coordinates, which is a
common and easy way to convexify the combination of payoff vectors.
6
Model
Preliminaries

At the beginning of time t, the per period average payoffs:

7
Model
Preliminaries
Denote the PPE payoff set as , any equilibrium payoff pair is supported by a pure action
and a particular continuation payoff.
Question: How does this payoff set look like? Spoiler:

8
Model
We choose 4 types of pure actions of the stage game (power structures)
1) Centralization ()
The principal always choose the default project. The agent also chooses the default one.
2) Restricted Empowerment ()
The principal rubberstamps what the agent recommends. The agent chooses principal’s
preferred project when available, the default otherwise.
3) Cooperative Empowerment ()
The principal rubberstamps what the agent recommends. The agent chooses principal’s
preferred project when available, his preferred project otherwise.
4) Restricted Empowerment ()
The principal rubberstamps what the agent recommends. The agent always chooses his
preferred project.
- Notice that in the latter 3 pure actions, both players exert efforts.
- Do we need to consider other pure actions? No, because later we’ll see they are enough
9
to support an optimal relational contract. (pick and verify???)
Model
Preliminaries: Constraints
There are constrains to be satisfied if a payoff pair is supported by a pure action + a
continuation payoff.
Take Centralization () and Cooperative Empowerment () for examples.
• Centralization
WLOG, we discuss the situation at time 0 (recursiveness of PPE).
The payoff of the principal under centralization is:

The payoff of the agent under centralization is:

1) Clearly, the continuation payoff must also be PPE payoff (self-enforcement constraints),
thus satisfy :

10
Model
Preliminaries: Constraints
• Centralization
2) No Deviation (incentive compatible, )
Two types of deviation: On/Off schedule
Because the principal does not listen to him anyway, WLOG, the agent chooses m=0.
And he does not have an incentive to deviate.
Due to the existence of the disastrous project, the principal also does not deviate.

3) Promise Keeping: If your action at the current period is optimal, then the payoff
should satisfy

11
Model
Preliminaries: Constraints
• Cooperative Empowerment
1) Self-enforcement:
2) No Deviation
The agent can deviate by choosing his own preferred project when the principal’s
preferred project is actually available. To compensate, the continuation payoff after he
reports faithfully must be higher than after such deviation:

3) Promise Keeping:

12
Model
Preliminaries: Constraints
• Randomization between different pure actions
The intuition is randomization is that, suppose the principal’s strategy is as follows:
“Well, since you perform well, from the next period on:
1) With probability , you can always choose your favorite project (entering );
2) With probability , you must still choose my project when available (still ).”
And the agent coordinates with her by looking at the randomization result last period.

13
Model
Preliminaries: Assumptions

• A1 ensures the sustainability of and , preventing deviations. Only when A1 holds,

under , the principal can manipulates and such that the gap is large enough to
prevent the agent from deviation (incentive compatibility).
• A2 and A3 makes it optimal that the game starts under . This is more interesting
because once we enter , , , the rest of the game is stationary.

14
Model
Maximization Problem
We define the (enforceable) payoff frontier as:

Which is the largest feasible payoff of the principal under a certain payoff of the agent.

• Part 1
As a convex combination of finite pure actions, the PPE payoff set is compact.
• Part 2
The public randomization device ensures so.
• Part 3
The agent’s worst payoff is under forever, and the best is under forever.
Which gives him respectively .

15
Model
Maximization Problem

Intuition:
By moving below the frontier, players actually means that “some of us deviates, but we
don’t know who did, so let’s punish everyone.”
However, here the principal does not have any on-schedule deviations.

16
Model
Maximization Problem

17
Model
Maximization Problem
Now that we have the functions of the agent’s continuation payoff, we accordingly have
the same for the principal:

18
Model
Maximization Problem
Now we stating the maximization problem:

Question: Why it ensures that the convex combination of these 4 pure actions can reach
the maximum of the principal’s payoff? The 4 actions are manually picked out.
19
Model
The Optimal Relational Contract

20
Model
The Optimal Relational Contract

21
Model
The Optimal Relational Contract

22
Model
The Optimal Relational Contract

23
Model
Extension: Failure to Exploit Public Opportunities
Now suppose that starting from a publicly observable period , a “super good” new
project becomes available w/p . And once it becomes available at some period, it stays
available (post-opportunity phase).
• In the post-opportunity phase, two new power arrangement arises:
1) The principal always chooses the new project (smarter Centralization)
2) The principal chooses the old preferred project when available, and the new project
otherwise (Restricted Empowerment, but improved since the default project is replaced
by the new project)
• Why the principal does not choose the new project w/p one?
Because the logic stays the same: by sacrificing future payoffs, the principal locks in the
performance of the agent today.
If the agent performs really well, under some parameters settings, the principal rewards
him by promising not to choose the new project (instead still letting him abuse the
24
power).
Takeout points
Recall previous paper we read:
Power can be divided into 4 steps.
1) Initiation (by the agent)
2) Ratification (by the principal)
3) Implementation (both)
4) Monitoring (by the principal)
The paper only discusses the power dynamics of 1) and 2).
In the literature, the focus on 1) and 2) is common in relational contract papers, while
the discussion of 3) and 4) is common in “classical agency models” about moral-hazard
and adverse selection.
My intuition is that, there are theoretical gaps to fill by studying different combinations.

25
Takeout points
A possible application/extension of the model’s idea
The principal has no profitable deviation in the model, since the agent can always punish
her by shirking in the implementation .
However, the implicit setting here is that the agent has significant power in
implementation (being able to sabotage).
What if the agent’s power to punish the principal is also dynamic?
• 鸟尽弓藏 vs 功高震主
• An alternative way to explain the guardianship dilemma?
Not only an exogenously powerful agent can kill you, but any successful project he
carries out inevitably increase his ability to sabotage in the future (endogenously
growing power following good performance).

26
Takeout points
Re-examine dynamic models using MPE?
The PPE can handle imperfect monitoring, but requires recalling capacity.
Are there any previous papers using MPE that are actually substantively more suitable
under PPE?

Is PPE a technique worthy of attention?

In light of recent aesthetic trends in formal models, is PPE even a good idea?
Does it violates our pursuit of parsimony?
Any alternative set up that are both rich and parsimonious for the question?

The denial of historical determinism

In this model, crucial institutional outcomes are decided by random events in an
organization’s early days.
Does the same logic apply to political transition?
27
What kind of model can we come up with?