Forecasting Receiving Yards Using Facebook's Prophet
I am planning on releasing a YouTube video on this in the upcoming weeks but Facebook has created a library that allows Python and R users to create simple yet accurate forecasts. I thought I would try it out on NFL receiving yards and compare the outputs against what actually happened. To do this, I took the historical receiving yards of players going back to 2015 and sampled some of the highest targeted receivers in the 2020 season. Week 10 predictions and actual yards are below:
For using only one data point (receiving yards) the model a fairly accurate. Without taking into account important variables such as weather, quarterback quality, and defense, the model is able to get within 10-12 yards for 4 of 11 players. From a betting perspective, most players outperformed their lower bound by wide margins so the lower bound can serve as a baseline when assessing risk.
So with the Vikings squaring off against the Bears, here is what the model is predicting:
And the current props:
I like Thielen and Jefferson 59+ and 57+, respectively. In those cases even if the model is off by 20 yards they are still hitting. The risk there of course is if Dalvin Cook goes off again they may never need to throw the ball.
Data provided by www.pro-football-reference.com
Lines provided by www.bovada.lv