Task

Objective

Predict point spread together with an interval (confidence/prediction) in the 2025 NCAA conference games for Duke, NC State and UNC.

Instructions

  1. Download a copy of the google sheet here. Click File, Download, Comma Separated Values (.csv) to download your prediction spreadsheet as a csv to your computer.

  2. Only edit cells that contain NA. All the information in the first three columns should not be altered at all. The games should remain in their original order. Your predictions may not be considered if you alter the information in these first three columns.

Important
  • Enter your predicted point spread (home team minus away team) in the pt_spread column.
  • Enter the lower bound of the interval in the ci_lb column.
  • Enter the upper bound of the interval in the ci_ub column.
  • Point spreads and intervals must be numeric values. They do not need to be integers. For example, you may predict the spread to be somewhere between -8.25 and 3.75 points.
  • Insert your team name in the first cell of the team_name column.
  • Insert the names of all the students on your team in the team_members column. This includes the name of the person who submitted. If you are submitting solo, this column should only have your name in the column.

  1. Rename your file to the name of your team. Upload your file as a csv to the dropbox link here. Submission of this is due by midnight on January 10, 2025. Submissions after midnight will not be considered at all.

  2. Submit a one-page (maximum) write-up detailing how you came up with your predicted point spread and associated intervals, e.g. what type of model did you use? Additionally, discuss why you chose your methodology. Please upload this as a .pdf or .docx document here. Again, be sure to rename your file to the name of your team. Submission of this is due by midnight on January 10, 2025.

Evaluation

Your predictions (point estimates and intervals) will be evaluated separately by two metrics described below. Your goal is to minimize each of the following metrics.

  1. Mean Absolute Error (MAE)

\[ MAE = \frac{1}{n} \sum_{i=1}^n \left| \mu_i - \hat{\mu}_i \right| \]

where \(\mu_i\) is the actual point spread outcome (home - away points) and \(\hat{\mu}_i\) is your predicted point spread for the \(i\)th game. The average is taken over all \(n = 42\) total games for which you are predicting the outcome. In words, the \(MAE\) is the average number of points your predictions are away from the truth. Smaller \(MAE\) indicates accurate point predictions.

  1. Interval Width (CIW)

In order for your confidence intervals to be considered, at least 70% of your intervals must contain the true spread. Given this criteria, we heavily recommend reporting intervals with a minimum of 70% coverage.

\[ CIW = \begin{cases} \frac{1}{n} \sum_{i=1}^n \left| ub_i - lb_i \right|,& \text{if Coverage} \geq 70\% \\ & \\ \text{Disqualified (DQ)}, & \text{if Coverage} < 70\% \end{cases} \]

where \(ub_i\) is the upper bound of your interval and \(lb_i\) is the lower bound of your interval for the spread of the \(i\)th game. The average is taken over all \(n = 42\) total games for which you are producing confidence intervals. In words, the \(CIW\) is the average width of all your confidence intervals. Smaller \(CIW\) indicates more precise confidence intervals.