Expected Value (EV): The Core of Value Betting

Published Tue Sep 02 2025

Expected Value (EV) is the average amount you expect to win or lose per bet if you could repeat the same wager infinitely. In value betting, only bets with positive EV (+EV) are taken intentionally.

Basic Formula (Decimal Odds)

EV = (p * (o - 1)) - (1 - p)

Where:

  • p = your estimated true probability
  • o = decimal odds offered

Expressed as a percentage of stake:

EV% = ( (p * o) - 1 ) * 100

Example: Probability 0.55, odds 2.00:

EV% = (0.55 * 2.00 - 1) * 100 = +10%

Another: Probability 0.38, odds 3.10:

EV% = (0.38 * 3.10 - 1) * 100 = (1.178 - 1)*100 = +17.8%

Higher EV% does not always mean better decision if volatility or correlation increases bankroll risk—Kelly sizing will adjust.

Edge vs EV

Sometimes “edge” is used:

Edge = (o * p) - 1  (same as EV%)

Estimating True Probability

This is the hard part.

  • Market-derived (reverse engineer sharp books & adjust vig)
  • Modeling (statistical / ML)
  • Hybrid (market anchor + subjective adjustments)

Market-Derived Outline

  1. Collect odds from sharp book(s).
  2. Remove vig: convert odds to implied probabilities, renormalize to 100%.
  3. Apply situational adjustments (injuries, weather, motivation) if you believe market underreacts.
  4. Compare to soft book odds → identify +EV discrepancies.

Modeling Approach

  • Feature engineering (team strength ratings, rest days, travel, player availability).
  • Fit model (e.g., logistic regression, gradient boosting, Bayesian hierarchical).
  • Backtest out-of-sample; record calibration (Brier score, reliability plots).
  • Blend with market probability if model uncertainty high.

Probability Calibration

You need forecasts to be well calibrated. Example: All events you label 55% should win ~55% over large samples. Tools:

  • Reliability diagram / calibration curve.
  • Brier Score (lower is better): mean((forecast - outcome)^2).
  • Log Loss for sharper penalty on overconfidence.

If your 60% bucket only wins 52%, you are overconfident → adjust downward or incorporate shrinkage toward the market.

Common Mistakes

  • Using implied odds of a single soft book as “true” probability.
  • Overfitting models with small samples.
  • Ignoring correlation between markets (double counting edge).
  • Not logging results—no feedback loop.
  • Treating narrow confidence intervals as fact (underestimating uncertainty).
  • Chasing edges smaller than operational costs (time, limits, friction).

Distribution of Outcomes vs EV

Expected value is a mean. Realized results form a distribution around that mean. For a fair coin with +2% EV (somehow), after 50 flips you can easily be negative. The variance of returns shrinks with sample size: Standard Error ≈ σ/√n. Hence patience is integral to “unlock” EV.

Portfolio / Correlation Considerations

Two 5% EV bets that are perfectly correlated behave like one larger bet; your aggregate risk grows faster than your EV estimate if you ignore correlation. Use diversification across sports, bet types, and time windows.

EV vs Edge vs Closing Line

Triangle of validation:

  1. Model/Market produces probability → EV.
  2. Stake sized via risk framework (e.g., fractional Kelly).
  3. Track CLV as independent process metric; persistent +EV with negative CLV flags probability mispricing or execution lag.

Worked Mini Case Study

You log 400 bets, each staked 1 unit, average EV% = +4%. Expected profit ≈ 16 units. You are up only 5 units. Is the edge fake? Maybe not. If per-bet standard deviation of returns is ~1 (typical near even odds), standard deviation of total ≈ √400 * 1 = 20 units. Being +5 instead of +16 is only (16-5)/20 = 0.55 standard deviations below expectation—statistically ordinary.

FAQs

My EV projections look great but results lag—quit? Check CLV, calibration, sample size. If those are healthy, stick with process.

Should I always bet every +0.5% edge? Micro edges may be erased by line movement latency and limits. Set a minimum actionable threshold (e.g., 1%+ after vig).

Can I sum EV across correlated bets? Not naively. Use scenario / simulation or reduce correlated stakes.

Should I include bonuses/free bets in EV? Yes—treat them as positive expected return instruments; amortize signup bonuses over initial volume to understand adjusted baseline edge.

Variance & EV

Positive EV does not guarantee near-term profit. Variance can mask edge for long stretches. Bankroll discipline keeps you solvent until the math plays out.

Takeaway

EV is the compass. Without a credible probability estimate, you are gambling, not exploiting mispricing.