Who are we?
I am a math master's student obsessed with football (let's go ravens flock) and finding edges using data. Starting from the NFL draft all the way until the playoffs, I dream about drafting and managing the perfect fantasy team. Naturally, I've always seen posts about using Vegas lines to project points, but the methods I have seen used are quite basic, and I'm not sure if their projections are even that useful because of it. So, I thought of some ways that I could apply my statistics background to try and improve on the existing framework. Shoutout to my software developer friend who helped me actually build something that works and other people can use.
Why use betting lines for projections?
Sportsbooks react to news, injuries, matchups, and weather faster than any one model. By translating those market prices into
probabilities we can build full distributions for each stat instead of working off of a single number.
- Market-informed: Sportsbooks are incentivized to move their as soon as news breaks.
- Calibrated: We remove the book’s margin (“devig”) to get fair probabilities.
- Uncertainty-aware: Alt lines tell us how likely different outcomes are, not just one median.
How we turn lines into an expectation
-
Start with prices, not just the o/u. For a player you’ll often see several “alt” numbers (e.g., 45, 55, 65 yards) each with an Over/Under price.
After devigging, those prices become probabilities like “40% chance to go over 65.”
-
Fit the curve those probabilities describe. We choose a sensible curve for the stat
(yards ≈ bell-shaped but skewed; counts are discrete). We tune that curve so its
“chance of beating each alt line” matches the market.
-
Average across all outcomes. Once we have the curve, the expected value
is just the weighted average of every possible outcome:
mean = Σ(value × probability of that value).
That’s the number we rank on.
Beating the Median with the Mean
Previous methods just found the median of each o/u line, which ignores how the data is shaped around the center. On skewed stats, which football stats are, the mean properly credits upside, which the median does not take into account. This should project more accurately compared to previous Vegas-based methods.
Since this is a hypothesis, we are still tracking data for the other methods to verify which method is more accurate. Keep in mind that some lines may be missing for games that are further out but will be added as the week goes on.
Models under the hood (for nerds)
- Yardage props: a skew-friendly curve for yards (log normal) fitted from o/u + alt lines, then take the mean yards.
- Count props (receptions, pass TDs): a discrete, over-dispersed model (negbin) tuned to the alt lines; we use its mean.
- Touchdowns: the Anytime TD price implies a scoring rate; that rate is the expected TDs (poisson model).
Fantasy scoring
We convert each stat’s expected value into points using your scoring settings and
rank by expected fantasy points. This should help make decisions about who to start, especially if there are.
More to Come
If this tool is popular, we will add
custom scoring options, ability to
import teams from platforms, and
separate players into tiers for even easier decision making. We also have plenty of other analytics-powered ideas in the chamber. Feel free to tip to show that there is interest.