DFS Prediction Models – Part 1: Quarterbacks

Dave Pratt presents a daily fantasy football prediction model for quarterbacks.

| 1 year ago
AndrewLuckPart1

DFS Prediction Models – Part 1: Quarterbacks


In daily-fantasy leagues, it’s difficult to find an edge. The white noise of the fantasy statistics, rankings, and expert opinions make it hard to pick up on the correct signals. So, is there a “station” that holds the key to finding daily-fantasy value? Well, PFF has found a way to better predict the weekly values of quarterbacks.

Daily-fantasy sites try their best to price QBs in an accurate and fair way. Their methods heavily bias the quarterbacks’ current matchups and their recent fantasy scores. However, a previous week’s fantasy points are not significantly correlated with following-week fantasy points.

Fortunately for savvy fantasy owners, there is value to be found in this shortsighted method of pricing quarterbacks. The following is an analysis of the predictive values of PFF Signature Stats for quarterbacks’ following-week fantasy scores.

To read the entire article, please login or sign up for a PFF Fantasy Subscription

Not a PFF fantasy subscriber? Compare all of our packages and features here.
  • http://www.twitter.com/mikecferrara Mike Ferrara

    Nice article. Nothing better than sports + scatterplots!

  • Richard

    Nice series of articles, I would like to take it a step further and was wondering if there is an easy way to download the dataset that you are using.

  • Joey

    I’m a little confused here – are you plugging in the previous week’s values into the model in order to come up with your projection for this week’s point total? As in, are you taking the yards, completion percentage and accuracy +20 yards from week 4 for QBs and then plugging them into the model to predict their point total for week 5?

  • Travis

    Were the games with 0 passing yards the previous game excluded from the analysis? If so, it would also appear that would knock out many of the 0 results for yards gained over 20 yards, leaving that attribute with a pretty strong positive correlation that should show up in the model. Just my thought looking at the scatterplots.