I am attempted to forecast the fantasy points scored by players in the upcoming NFL season, using Python. I have historical fantasy points data for the players from 2009 to 2017, although obviously this data is not available for all of the players as not all of them have played every season in the time frame. I also have other data which I would like to be used, if it were to improve the accuracy of the forecast, such as age, position, height and weight. So for each player I have up to 9 data points (2009-2017), and across the whole dataset I have this for over 1000 players. Is there a way to make predictions for each player by using the patterns found in the other samples? For example age is likely to have an impact, as presumably players’ points peak at a certain point before dropping as they pass their peak. Additionally, I am attempting to do this in Python, but any solution is much appreciated.
If your data meets the required assumptions you can perform linear regression and use the model to predict points - see here for an example Testing Hypothesis with Time series and Location Data
If the relation between dependent and independent variables is constant in time, you could also pool the data points and build a model which is no longer a time series.
Another approach is to use date dummy variables and allow the intercept to change...
- More complex models such as dynamic regression models, exponential smoothing etc.. may provide better forecasts, but may be more difficult to interpret from an inferential perspective
A few useful resources
- chapter 13 Wooldridge Introductory_Econometrics_A_Modern_Approach
- Hyndman's new book: https://otexts.org/fpp2/least-squares.html