Chapter 4 · 9 min read
Chapter 4 — Value bets and bankroll: winning the single bet vs surviving
What separates long-term winners from people who disappear is almost always bankroll management, not pick quality. You can be 60% accurate and fail by risking too much per bet. You can be 53% and prosper by sizing your bets correctly. This chapter is the math that separates hobby from investment.
1. What a value bet is (and what it ISN'T)
A value bet is NOT a bet that will win. It's a bet where your estimated probability is HIGHER than the probability implied by the odds. Formula: edge = (your_probability × odds) - 1. Example: model says Milan wins 60%; odds 1.95 = implied 51%. Edge = (0.60 × 1.95) - 1 = +0.17 = 17% value. Across 100 bets like this you win 60 and lose 40, but the odds pay more than the market implied. The single bet can lose — the value-bet portfolio wins over time. PronoStats flags value bets when edge > 10%.
2. How to calculate edge — practical examples
Three examples. Case A: model 55%, odds 2.10. Implied 47.6%. Edge = 0.55 × 2.10 - 1 = +15.5%. Strong value. Case B: model 70%, odds 1.40. Implied 71.4%. Edge = -0.02. No value, market is right. Case C: model 40%, odds 3.50. Implied 28.6%. Edge = +0.40 = +40%. CAREFUL: edges this large are often model errors, not real value. PronoStats caps flagged edges at 10-25% to avoid 'too good to be true' bets.
3. Kelly criterion — how much to actually bet
Odds tell you what they pay, edge tells you if it's worth it, Kelly tells you HOW MUCH of your bankroll to risk. Formula: f = (b × p - q) / b, where b = odds - 1, p = your probability, q = 1 - p. Example: odds 2.10, prob 55%. b = 1.10, p = 0.55, q = 0.45. f = (1.10 × 0.55 - 0.45) / 1.10 = 14% of bankroll. Kelly is mathematically optimal but psychologically brutal: a losing streak destroys you mentally. Almost every pro uses ½ Kelly (7%) or ¼ Kelly (3.5%): grows slower but drawdown is far more manageable.
4. Bankroll management ground rules
Five non-negotiable rules. (1) Separate the bankroll: betting money isn't grocery money. (2) Never more than 5% of bankroll on a single bet, even with high edge (safety cap against model errors). (3) No emotional chasing: 3 losses in a row, do NOT double. (4) Stake decision BEFORE seeing the outcome, not after. (5) Weekly review: track ROI, win rate, average odds. If any of those worsens for 4 weeks straight, something is broken — take a break.
5. The 4 PronoStats AI bankroll strategies
Premium-only: (1) Conservative — Kelly ¼, max 3 bets/day, edge > 12% only, targets steady +5-8% monthly ROI. (2) Balanced — Kelly ½, max 6 bets, edge > 8%. (3) Aggressive — full Kelly, max 10 bets, edge > 5%, high drawdown accepted. (4) Combo Hunter — 2-3 leg accumulators with positive expected value, max 1-2 combos/day. Each has a live equity curve in the Bankroll panel: see how it grows, where drawdowns happen, how many bets are needed for statistical convergence.
6. The psychology that ruins you (and how to defend)
Three enemies. (1) Recency bias: yesterday's match weighs more than 3-month trends. Defense: write your thesis BEFORE seeing odds. (2) Loss aversion: a -€100 hurts more than +€100 feels good. Result: closing winners too early, holding losers too long. Defense: fixed staking rule, no live improvisation. (3) Tilt: after a losing run you raise stakes to recover. Defense: daily stop-loss (e.g. -3 units = stop until tomorrow). PronoStats AI bankrolls all apply a daily cap to prevent tilt.
Chapter quiz
Check if the concepts stuck — nothing tracked, just for you.
1.Model 55%, odds 2.10. Is there value?
2.What fraction of bankroll does full Kelly suggest with edge 15.5% at odds 2.10?
3.After 3 losses in a row, you...
Frequently asked questions
Does a value bet always win in the long run?
It will WIN if your probabilities are actually better than the market's. If your probabilities are miscalibrated, even apparent 'value' loses you money. That's why calibration (Chapter 3) is critical.
How much initial bankroll do I need?
For ¼ Kelly with 1-3% stakes, you need a bankroll giving you at least 100 units (e.g. €200 with €2 stakes). Below 50 units you're in a real ruin-risk zone.
Can I use Kelly with my own probabilities, not the model's?
Yes, but know that humans systematically over-estimate their own accuracy by 15-20%. Use conservative estimates or stay on ¼ Kelly to cushion.