If you've read anything about profitable betting, you know expected value (EV) is the foundation. Every profitable bettor calculates EV. Every sharp betting tool shows EV. It's the single most important concept — until it isn't.
There's a deeper concept that EV doesn't capture. It's the reason the Kelly Criterion exists, the reason professional bettors don't bet their entire bankroll on the highest-EV opportunity, and the reason parlays can be mathematically worse than straight bets even when they have higher EV.
That concept is expected growth — and understanding it will change how you think about every bet you place.
🎲 The Bet-Everything Paradox
Let's start with a thought experiment. You have a $100 bankroll and a coin that lands heads 55% of the time. It pays even money. This is a +EV bet — you should take it every time.
Question: How much should you bet?
If you're maximizing EV, the answer is simple: bet everything. A $100 bet at 55% gives you +$10 EV. A $50 bet only gives you +$5 EV. More is always better... right?
The Bet-Everything Paradox
Strategy A: Max EV
Bet 100% of bankroll every time
Strategy B: Kelly
Bet 5.5% of bankroll each time
Strategy A has 18x higher EV per bet — but goes broke almost every time. Maximizing EV without considering growth leads to ruin. This is the paradox that Kelly solves.
This is the bet-everything paradox. Maximizing expected value tells you to bet your entire bankroll every time. But anyone with common sense knows that's insane — one loss and you're done.
The problem isn't with your edge. The problem is that EV doesn't account for what happens to your bankroll over multiple bets. It treats each bet in isolation. It doesn't care that you need to survive to bet again.
📐 The Arithmetic vs Geometric Mean
Here's where the math gets interesting. Imagine two different bet outcomes:
- Bet A: Win $50 half the time, lose $40 half the time
- Bet B: Win $10 every single time
Both have the same EV: +$5 per bet. But they behave completely differently over time.
With Bet B, your bankroll grows linearly: $100 → $110 → $120 → $130. Steady and predictable.
With Bet A, something counterintuitive happens. Say you win, then lose:
$100 → $150 → $110 (net: +$10)But what if you lose, then win?
$100 → $60 → $90 (net: -$10)Same bets, different order, different result. Over many cycles, your bankroll's path depends on the geometric mean, not the arithmetic mean. And the geometric mean is always less than or equal to the arithmetic mean — the gap being driven by variance.
This is the core insight: EV measures the arithmetic average. Your bankroll compounds at the geometric average. They are not the same thing.
🚨 When EV Lies to You
The gap between EV and actual bankroll growth gets dramatic in extreme cases:
When EV and Growth Disagree
$100 bankroll, betting 100% each time
EV says breakeven. Growth says you shrink.
EV overstates the real compounding rate.
Infinite EV. Certain ruin. The ultimate trap.
The most famous example is the St. Petersburg Paradox. A game offers you $2 if a coin lands heads on the first flip, $4 if the first tails and second heads, $8 for two tails then heads, and so on — doubling each round. The expected value of this game is literally infinite.
But no rational person would pay $1,000 to play it. Why? Because the median outcome is terrible. You win $2 half the time. You win $4 a quarter of the time. The astronomical payouts that drive EV to infinity almost never happen.
In betting, the same principle applies on a smaller scale. A 10-leg parlay at +50000 might have higher EV than 10 straight bets. But the expected growth of your bankroll is probably negative — because you're losing 99.9% of the time, and no single massive payout can undo the compounding damage of those losses.
📉 The Drawdown Problem
Here's the brutal asymmetry that EV ignores:
The Drawdown Asymmetry
% gain needed to recover from a loss
Losses and gains are not symmetric. This is why variance kills compounding.
A 50% loss requires a 100% gain to recover. A 75% loss requires a 300% gain. The deeper the hole, the harder it is to climb out — and this asymmetry gets worse exponentially.
This is why variance is not neutral. EV treats a +50% gain and a -50% loss as washing out to zero. But your bankroll disagrees:
$1,000 × 1.50 × 0.50 = $750 (not $1,000)You gained 50% and lost 50%, but your bankroll is down 25%. This "variance drag" is the difference between the arithmetic average (0%) and the geometric average (-13.4%). The higher the variance, the bigger this drag.
This is the fundamental reason high-variance strategies (like big parlays or overbetting) can destroy bankrolls even when every individual bet is +EV.
🧠 Enter Expected Growth
Expected growth fixes this by using the logarithm of wealth instead of wealth itself.
Where EV calculates: E[W] = p × (win amount) + q × (loss amount)
Expected growth calculates: E[log(W)] = p × log(win amount) + q × log(loss amount)
Why logarithms? Because the log function naturally captures two critical realities:
- Diminishing returns: Winning $1,000 when you have $100 matters more than winning $1,000 when you have $1,000,000. Log scales proportionally.
- Asymmetric losses: Log penalizes losses more heavily than it rewards equivalent gains. A 50% loss hurts more than a 50% gain helps — exactly matching how compounding actually works.
When you maximize E[log(W)], you're maximizing the long-run geometric growth rate of your bankroll. In the language of economics, you're maximizing expected logarithmic utility — and that turns out to be equivalent to maximizing the rate at which your bankroll compounds.
This is also why there's truth to the famous saying: "The first million is the hardest." It's not just motivational fluff — it's a mathematical fact about geometric growth. When your bankroll is small, the compounding engine is small. Kelly tells you to bet a fixed percentage of your bankroll, so a $1,000 bankroll at 5.5% Kelly is risking $55 per bet. But a $100,000 bankroll is risking $5,500 per bet — same edge, same percentage, 100x the dollar growth. Each dollar you add accelerates the compounding of every dollar after it. The hard part isn't finding the edge. It's surviving long enough for the geometric growth to take off.
The terminology: In academic economics, this is called "expected utility" under logarithmic utility preferences. In the sports betting world, it's usually called "expected growth" because that's more intuitive — it describes what's actually happening to your bankroll. Same math, different framing. We'll use "expected growth" because you're a bettor, not an economist.
🔑 Kelly Criterion: The Bridge Between EV and Growth
The Kelly Criterion is what you get when you solve for the bet size that maximizes expected growth. It's not a separate concept — it's the direct consequence of optimizing E[log(W)] instead of E[W].
Kelly Criterion — The Bridge
Kelly maximizes
E[log(W)] = p · log(1 + bf) + q · log(1 − f)
not E[W] = p · (1 + bf) + q · (1 − f)
Optimal fraction
f* = (bp − q) / b
At 55% on -110
f* ≈ 5.5%
Kelly tells you to bet a fraction of your bankroll proportional to your edge. Crucially:
- Bigger edge = bigger bet. A 10% edge warrants a larger fraction than a 2% edge.
- Worse odds = smaller bet. The same edge at +200 gets a different Kelly fraction than at -110.
- It automatically protects against ruin. You never bet 100% because log(0) = -infinity. Kelly bets approach zero as your edge approaches zero.
This is what Ganchrow explained in his classic SBR forum posts in the early 2000s: Kelly doesn't maximize how much you expect to make. It maximizes how fast your bankroll actually grows. Those are different things when variance exists — and variance always exists.
💡 The Practical Difference
EV vs Expected Growth — Side by Side
| Expected Value | Expected Growth | |
|---|---|---|
| Measures | Average profit per bet | Bankroll compounding rate |
| Formula | E[W] | E[log(W)] |
| Optimizes | Arithmetic mean | Geometric mean |
| Bet sizing | Bet everything | Kelly fraction |
| Weakness | Ignores variance & ruin | Lower per-bet “profit” |
Here's what this means in practice:
Two bets, same EV, different growth:
Two simultaneous straight bets at 55% on -110 — one on each game, Kelly-sized at 5.5% each. Combined EV: +$10.00 per $200 wagered. Combined growth: +0.00275.
One 2-leg parlay of the same two 55% legs at +264 — Kelly-sized at 3.9%. EV: +$10.25 per $100 wagered. Growth: +0.0019.
The parlay has slightly higher EV but 45% less growth. The straight bettor has a "split" outcome (one wins, one loses) that barely dents the bankroll. The parlay bettor counts that same scenario as a full loss. That cushion lets the straight bettor size more aggressively and compound faster.
This is the entire parlays debate resolved in one comparison. (We covered this in depth in our parlays article.)
⚖️ Fractional Kelly: What the Pros Actually Do
Full Kelly maximizes expected growth. But it comes with brutal swings — 50% drawdowns are not just possible, they're expected over a long enough timeline. Most professional bettors use a fraction of Kelly:
Fractional Kelly — The Practical Compromise
Growth rate relative to full Kelly
½ Kelly gives you 75% of the growth at a fraction of the variance. Going above full Kelly actually decreases your growth rate — at 2x Kelly, your expected growth is the same as not betting at all.
The key insight: ½ Kelly gives you 75% of the growth at dramatically lower variance. The growth-variance tradeoff is extremely favorable at smaller fractions. You give up a little compounding speed for a much smoother ride.
At 2x Kelly, you actually decrease your expected growth rate back to zero — you're overbetting so badly that the variance drag completely cancels your edge. Go above 2x Kelly and your expected growth turns negative, even on +EV bets. This is the mathematical proof that more EV does not always mean more growth.
📋 The Five Rules of Expected Growth
Put this all together and five rules emerge:
1. EV tells you if a bet is worth taking. Expected growth tells you how much to bet.
Never use EV alone to determine position size. A +EV bet sized incorrectly can be -EG (negative expected growth).
2. Variance is not free.
EV treats variance as noise. Expected growth treats it as a cost. High-variance strategies need proportionally higher edges to justify themselves.
3. Survival comes first.
You can't compound your edge if you go broke. Kelly automatically ensures survival. Flat-betting based on EV alone does not.
4. Fractional Kelly is almost always better in practice.
Theoretical Kelly assumes perfect edge estimation. Real edges are uncertain. Betting half-Kelly protects you from overestimating your edge — which everyone does.
5. The highest-EV bet is not always the best bet.
A moderate-EV bet that you can size correctly will often outperform a high-EV bet with massive variance. This is why straight bets beat parlays below ~57% per-leg win rates, and why seasoned pros prefer volume over home runs.
🎯 Why This Matters for You
If you're placing +EV bets but not thinking about expected growth, you're leaving money on the table — or worse, you're risking ruin.
- Sizing too large on high-EV spots? You might be above Kelly, which means negative expected growth despite positive EV.
- Stacking 5-leg parlays because the combined EV is massive? The variance drag might mean your bankroll never actually compounds.
- Ignoring correlation in your bets? Correlated exposure is hidden concentration — your effective bet size is bigger than you think.
Expected growth isn't a niche academic concept. It's the operating system that every professional bettor and quantitative fund runs on. Kelly isn't popular because it's elegant — it's popular because it works. And it works because it optimizes the right thing: how fast your bankroll actually grows, not how much you hypothetically win on average.
🏁 Key Takeaways
- 📊 EV measures average profit. Expected growth measures compounding speed. They can disagree — and when they do, growth is right.
- 💥 The bet-everything paradox proves EV alone is insufficient. Max-EV bet sizing leads to certain ruin.
- 🌊 Variance drag is real. High variance reduces your geometric growth rate below your arithmetic EV. This is not theoretical — it affects every bet you place.
- 🔑 Kelly Criterion maximizes expected growth, not expected value. That's why it works for long-term bankroll building.
- ⚖️ ½ Kelly is the sweet spot. 75% of the growth, fraction of the risk. Virtually all professional bettors use fractional Kelly.
- 🛡️ Overbetting is worse than underbetting. Going above Kelly decreases growth. Going above 2x Kelly makes growth negative. Always err on the side of smaller bets.
Have questions? Join our Discord and ask the community. 💬
The difference between a bettor who understands EV and a bettor who understands expected growth is the difference between knowing which bets to take and knowing how to actually build a bankroll. Now you know both. 🚀