Guidelines on Team-Selection Projects

* Read the book "Moneyball" by Michael M. Lewis; start at
http://en.wikipedia.org/wiki/Moneyball:_The_Art_of_Winning_an_Unfair_Game

* Team-Selection projects where your choices don't depend on other
teams' choices are usually fairly straightforward, but not exciting.
It is not too hard to get a good grade, but getting a great grade 
is a little harder than with some other projects.
In particular, the default Topic Ambition score is a 2.

* Team-Selection projects with a draft (where your choices depend
on which players get drafted by other teams) are much more difficult
and exciting.  Start with the paper
Michael J. Fry, Andrew W. Lundberg, and Jeffrey W. Ohlmann.
"A Player Selection Heuristic for a Sports League Draft" 
Journal of Quantitative Analysis in Sports 3.2 (2009).
http://works.bepress.com/mike_fry/1 
http://www.bepress.com/jqas/
(ask Prof. Ross for a copy, and go looking for more recent
papers that cite that one.)

* A really strong project combines two parts:
 i) Forecasting each player's statistics for the upcoming year
    based on past data, and
 ii) Using those forecasts to select your team.
 
 You should start with part (ii), but using the most recent year's
 data as if it was the upcoming year.  Get that model working, then
 go back and do your forecasting to see how things change.

* E-mail your professor as soon as you can with a full set of the
rules of your fantasy league.  These rules should also be included
in your proposal and your project write-up.

* You are not allowed to make up a projected-points system off the top
of your head.  For example, you can't just say that each player's
rating is (taking baseball as an example)
  3*RBIs + 8*batting avg + 2*slugging - 1.5*strikeouts
where you just invent the coefficients out of thin air.  Instead,
you should try to identify optimal coefficients using something like
regression on past years' data.

* You are not allowed to base your project on randomly generated data.
You should use real data.  In some cases, you might want to compare
results from real data to results from random data, but you must always
have real data in the project.

* Your report should have a formal definition of your mathematical model,
including decision variables, objective function, and constraints.
You may not simply rely on your spreadsheet formulation to get the
point across (though of course you should explain your spreadsheets too).

* You should do some sort of sensitivity analysis.  The most common one
to do is varying the salary cap, re-optimizing each time, and plotting
how the predicted # of points varies with the salary cap.

* You will probably need to use one of the free alternatives to Excel,
like Gnumeric or OpenOffice, since you will probably have more
than 200 variables.