Portfolio construction strategies

[Henry Harrison]

Maybe we can start off but just describing the strategies we use. That will help us figure out where are opportunities for graphical aids.

My typical strategy, for NFL GPPs, is to choose a QB-WR/TE stack for each lineup. Then I like to get an WR from the opposing team if possible, as passing-game scores of opposing teams are correlated. I try to include RB-DST-K stacks, or at least two of them. I do this many times and try to differentiate lineups from each other. Besides figuring out the stacks, there's not a lot of analysis to it. The best stacks are easy enough to determine with some projections and a brute-force search. I use points per dollar as a metric.

One common strategy I see is figuring out roughly what percentage of your lineups should include each player, then "reverse-engineering" lineups. This is something we could pursue. The task then becomes two-fold: first, optimizing the choices of these percentages from our estimated probability distributions, and second, creating aids to tell us what percentage we're currently at so we know which players are under- and over-represented.

A related strategy I've thought about is starting with a value for each player in a basic points-per-dollar sense. Then, whenever you include a player in a lineup, bump their value down a bit to reflect the fact that you already have some ownership of him. Eventually the less valuable players will rise to the top and you'll end up with similar situations as above--a larger representation of the more valuable players but still some representation of less valuable players. The task for pursuing this strategy would be coming up with an algorithm for adjusting value based on your previous lineups. It should probably go further than individual players and consider what groups of players you have not included, or you have included too often.