Excel Homework Due 11/29 at 11 PM

Please watch this video first.  It gives the algebra of the principal-agent model and will make you comfortable with the notation and the basic ideas.

Then do the Excel homework, which does the same analysis but this time with a graphical approach.

------

This is the simplest possible model of the principal-agent, with low or high effort and low or high output.  I am going to describe possible extensions of the model for you to consider (and which we might discuss in class) but which we won't model.

The first is to make output a continuous variable.  This, it turns out isn't conceptually that much harder, as long as higher output makes it more likely that the agent chose high effort.  In that case, the reward scheme should also rise with output.  In many real world reward schemes, the scheme is linear.  For example, it entails a base pay and then some percentage share of the output.  A lot of sales people work under schemes of this sort.  Linearity is hard to explain from the model, but it is easy to understand in that it can be readily communicated.  More complex schemes would be harder to articulate and to understand.

The second is to make effort a continuous variable, with higher effort making higher output more likely.  This too is not that hard to conceptualize, but then you lose the notion of shirking in the model, unless you establish a cutoff effort level below which the agent is said to shirk.

Further extensions make both output and effort multi-dimensional.  This is definitely harder to consider theoretically, but it is far more realistic, especially for knowledge work.  As I have said in class before, professional knowledge workers typically don't shirk (play solitaire on their computers all day or do Facebook, though they might do it as a diversion from time to time to refresh themselves).  Such employees have a strong sense of autonomy.  The issue for the manager is to align what these workers are doing with organizational objectives.  Monitoring happens to achieve alignment, not to prevent shirking.

The last extension I will mention is considering what happens when output has a large random component to it so it is not a very informative signal.  A little bit of randomness doesn't change the story much.  A lot of randomness does.  (The agent is risk averse and if you make a risk averse individual absorb a lot of risk, you need to pay a substantial risk premium to compensate for that.)  So instead, one looks for other solutions and then not rely on rewards for high output.