Ask a bettor how their season is going and they will quote you a win rate or an ROI. Ask a portfolio manager how their fund is doing and they will quote you a return and the risk they took to earn it. That second habit is the one sports bettors most need to steal, because a return figure with no measure of the risk behind it is only half the story — and it is usually the half that lets a dangerous strategy masquerade as a good one.
This post is about the number that fills in the missing half: a Sharpe-style ratio, borrowed straight from finance, that scores a betting strategy by its return per unit of risk. It is the difference between "this made money" and "this made money in a way my bankroll can actually survive."
What ROI leaves out
Return on investment — profit divided by total amount risked — is a fine headline. A 3 percent ROI means that for every dollar you put through the market, you kept three cents. Over thousands of bets, that is a genuine, hard-won edge. The problem is not that ROI is wrong; it is that ROI is silent about the journey.
Two strategies can earn the exact same ROI while delivering completely different experiences. One grinds out small, consistent wins betting Chiefs, Bills, and Ravens point spreads, with your bankroll ticking gently up and down. The other swings for the fences on longshot parlays, with your bankroll rocketing up on a hit and cratering through long dry spells. On an ROI leaderboard they are tied. In real life they are not remotely the same bet, because the volatile one can bust you before the average ever arrives.
This is the same lesson that runs through bankroll management and drawdown planning: survival is a precondition for compounding. A strategy that earns a great average but occasionally vaporizes your roll has an ROI you will never actually collect, because you got wiped out during one of the swings.
The Sharpe-style ratio, in plain English
The fix is to divide your return by its own volatility. In finance this is the Sharpe ratio; adapted for betting it is simply:
risk-adjusted score = (average return per bet) / (standard deviation of returns per bet)
The numerator is your edge. The denominator is how bumpy the ride was. A higher score means you earned more return for each unit of gut-churning variance you endured. The chart below makes the point with three strategies that all post an identical 3 percent ROI but carry wildly different volatility.
Same ROI, radically different quality. The steady flat-spread strategy earns its 3 percent with low volatility and scores highest. The longshot-parlay strategy earns the same 3 percent but with such enormous swings that its risk-adjusted score is a fraction of the flat approach's. If you could only run one of these strategies with your actual bankroll, the choice is obvious once you look past the ROI headline — and completely invisible if you do not.
Variance is measured in drawdowns you actually feel
The abstract "standard deviation" number becomes concrete when you translate it into drawdowns — the peak-to-trough dips your bankroll takes along the way. A low-variance flat-spread strategy might see its worst losing streak dent the bankroll by 10 or 15 percent, a dip you barely notice and easily ride out. A high-variance longshot-parlay strategy earning the identical average can routinely draw down 40, 50, even 60 percent before a big hit bails it out. Same expected return, wildly different survival odds — and a bettor who is not psychologically prepared for the deep drawdown will bail at the bottom, locking in the loss and never collecting the average. This is the whole reason two strategies with the same ROI are not interchangeable: the one whose drawdowns you can actually stomach is the one you will still be running when the edge finally pays off. Plan the depth of the dips in advance, and keep an honest log of every result — the kind of record described in how to track your bets — so you measure your real drawdowns instead of guessing at them.
How to compute it for your own bets
You do not need finance software. Log the return of every settled bet as a fraction of what you risked:
- A win at -110 returns about +0.91 units (you risked 1 to win 0.91).
- A loss returns -1.00 units.
- A push returns 0.
- A +150 winner returns +1.50; a +150 loser returns -1.00.
Now you have a column of per-bet returns. The average of that column is your mean return; its standard deviation is your volatility; divide the first by the second and you have your risk-adjusted score. Any spreadsheet or the tracking tools in /tinker will compute both in one line once you have logged the outcomes. The number becomes trustworthy at a few hundred bets — the same sample-size caveat that applies to every metric in betting.
Why the ratio changes how you bet
The risk-adjusted score is not just a nicer way to keep score. It directly informs how much you should stake, because lower-variance edges can be pressed harder for the same risk of ruin.
This is exactly what fractional Kelly staking captures. Kelly sizes a bet in proportion to your edge and inversely to the variance of the outcome. A steady 3 percent edge on point spreads can safely carry a larger bankroll fraction than a boom-bust 3 percent edge on parlays, precisely because its lower volatility means smaller drawdowns. So a higher Sharpe score does double duty: it tells you the strategy is better and that you can size it more aggressively — which compounds your bankroll faster even when the ROI headline is identical.
It also reframes what "improving your model" means. Cutting variance can be as valuable as raising ROI. If you can earn the same return with steadier results — by trimming the highest-variance bets, diversifying across markets, or avoiding correlated parlays — you have made a materially better strategy without moving the ROI number at all. The Sharpe ratio is the metric that gives you credit for that work.
Where the ratio fits among your metrics
Risk-adjusted return is one piece of an honest evaluation dashboard, not the whole thing. Pair it with:
- Closing line value — the forward-looking check on whether your picks beat the number the market closes at. A strategy can have great CLV and still be volatile; the two metrics answer different questions, and both matter. See the CLV explainer.
- Calibration — whether your stated probabilities match reality. A model that says 60 percent should win about 60 percent of the time. Miscalibration hides in a good ROI just like variance does. Compare your probabilities against the market in model vs consensus edges, and read glass-box vs black-box models for how to inspect what your model is actually doing.
- Sample size — every one of these metrics is noise until you have hundreds of bets behind it. A great Sharpe ratio over 30 bets is a coin flip with a tailwind.
Together these give you a picture ROI alone never can: not just whether you made money, but whether you made it with a real, well-calibrated edge and a risk profile your bankroll can live with.
The bottom line
ROI answers "did this work?" The Sharpe-style ratio answers "did this work in a way I can survive and repeat?" — and for anyone betting a real bankroll, the second question is the one that decides whether you are still in the game next season. Two strategies with the same return are not equal; the steadier one is better, sizes larger, and compounds more reliably. Start logging your per-bet returns, compute the ratio in /tinker, and rank your strategies by return per unit of risk rather than by the ROI headline alone. Then compare your risk-adjusted numbers against the public models on the leaderboards to see where your edge really stands.
Bet responsibly — set limits, never chase losses.
Named modeling examples
A model page is more useful when the feature examples are concrete. Josh Allen rushing attempts, Ja'Marr Chase target share, Nikola Jokic assist rate, Tarik Skubal strikeout projection, Igor Shesterkin starter confirmation, and Islam Makhachev control time are all different prediction problems. A single “player form” feature cannot explain them all, so the model needs sport-specific inputs and review notes.
- NFL: separate route participation, pressure rate, and red-zone role from box-score volume.
- NBA: separate usage, minute projection, pace, and back-to-back fatigue.
- MLB: separate starter skill, handedness, park, weather, and lineup confirmation.
- NHL and UFC: late confirmations and fight-week news can matter more than a season average.
Model inputs worth naming
Use names as evidence, not decoration. The useful SEO win is that Josh Allen, Ja'Marr Chase, Bijan Robinson and Puka Nacua and Chiefs, Bills, Ravens, Eagles and Lions appear inside decisions, thresholds, and internal links instead of being dumped into a keyword list.
- NFL model: route participation for Ja'Marr Chase, rushing attempts for Josh Allen, pressure rate allowed by the Bengals, and red-zone carry share for Jonathan Taylor should be separate features.
- NBA model: usage, projected minutes, rest, and pace should move Nikola Jokic or Shai Gilgeous-Alexander props differently than a one-number power rating.
- MLB model: Tarik Skubal strikeout projection, Coors Field park factor, lineup confirmation, and bullpen rest need their own columns.
- Review loop: grade entry price, closing price, bet result, and model error separately so lucky results do not hide bad forecasts.
Build or audit the workflow in Tinker and review it with CLV.
Research note board
Use this model-audit board to keep features, validation, and bet sizing from collapsing into one confidence score.
| Model layer | What to inspect | Example input | Downgrade when |
|---|---|---|---|
| Feature | Whether the variable maps to the sport and market | Josh Allen role data or PPR price movement | The feature is a proxy for something you can measure directly |
| Validation | Out-of-sample error, CLV, calibration, missing data | Chiefs market movement after injury news | Wins come without beating the close or improving calibration |
| Sizing | Bankroll, confidence interval, correlation, market limit | closing line value exposure compared with related tickets | Multiple bets repeat the same thesis at full stake |
Model calibration: predicted vs observed
Predicted win probability bucket vs the empirical win rate inside that bucket on the test set. Points on the y=x reference line are perfectly calibrated; points below mean the model is overconfident in that bucket.
EV per $100 across win rate × odds grid
Expected value of a $100 stake at each combination of true win rate and market odds. Anywhere the cell is positive you have a long-run profitable bet; the magnitude shows how aggressive Kelly will size it.
