OUR EXPERIMENT Betting Math Published · 11 min read · 2,506 words

Even a Winning Bettor Goes Broke by Betting Too Big: 5,000 Simulated Careers

We gave every simulated bettor a genuine edge. The only thing we changed was the stake. Some compounded, some were cut in half, some finished below where they started.

Betting 11.1% of your bankroll on each 55%-win, 1.91-decimal wager leaves you no better off after 1,000 bets - the median bankroll is exactly the starting amount.

Our simulation of 5,000 careers shows that the fraction of bankroll risked per ticket decides whether a modest 5.0% edge compounds or collapses.

WHAT WE FOUND
  • Betting 1.4% of bankroll (quarter Kelly) yields a median bankroll 1.8 times the start after 1,000 bets.
  • Betting 2.8% of bankroll (half Kelly) yields a median bankroll 2.9 times the start after 1,000 bets.
  • Betting 5.5% of bankroll (full Kelly) yields a median bankroll 4.1 times the start after 1,000 bets.
  • Betting 11.1% of bankroll (double Kelly) leaves the median bankroll unchanged at 1.0 times the start after 1,000 bets.
  • Betting a flat 10% of bankroll produces a median bankroll 1.6 times the start after 1,000 bets.

A bettor with a real edge

Imagine a player who wins 55% of his wagers at decimal odds 1.91. That win-rate translates into an edge of 5.0% per bet, meaning each stake is expected to grow by five percent on average.

The Kelly formula tells a bettor with a 5.0% edge to wager 5.5% of the current bankroll on every bet. That fraction maximises the exponential growth of the bankroll while keeping the risk of ruin as low as the model allows. In the simulated world, the “Full Kelly” rule therefore stakes 5.5% of the bankroll per bet.

When the stake is reduced to a quarter of the Kelly recommendation - 1.4% of the bankroll per bet - the median bankroll after 1,000 bets climbs to 1.8 times the start. Only 1.9% of simulated players ever see a drawdown of 50% or more, and 6.9% finish below their initial bankroll. Halving the Kelly fraction to 2.8% per bet improves the median outcome to 2.9 times the start, but the risk rises sharply: 42.4% of players suffer a 50% drawdown and 11.2% end below where they began.

At the full Kelly level - 5.5% of the bankroll per bet - the median bankroll after 1,000 bets reaches 4.1 times the start. The upside comes with a near-certain downside: 98.9% of simulated bettors experience a drawdown of at least 50%, and 20.0% finish below their original bankroll. Doubling the Kelly stake to 11.1% per bet collapses the advantage entirely; the median bankroll stays at 1.0 times the start, every player (100.0%) endures a 50% drawdown, and 51.0% finish below the start. A flat-10% stake behaves similarly: the median bankroll grows to 1.6 times the start, yet 100.0% of players see a 50% drawdown and 43.1% end below where they began.

Quarter, half, full: growth against pain

Choosing how much of your bankroll to risk per wager is a pure trade-off between compounding the edge and exposing yourself to volatility. The Kelly formula tells you the fraction that maximises exponential growth, but every step away from that fraction reshapes the distribution of outcomes. When the stake is a quarter of Kelly, the bet size is 1.4% of the bankroll; at half Kelly it doubles to 2.8%; at full Kelly it climbs to 5.5%. The edge that fuels the growth is the 5.0% per bet you earn from winning 55% of 1.91-decimal odds.

The low stake keeps each swing small, so the bankroll drifts upward while the probability of a crippling loss stays negligible.

At half Kelly the median climbs to 2.9 times the initial bankroll, but the price of that extra growth is a dramatic rise in painful drawdowns. The larger bet size amplifies both wins and losses, so the upside improves while the downside becomes a frequent reality.

The edge is still present, but the bankroll spends most of its time in the danger zone, making recovery a matter of luck rather than skill.

The practical upshot is that “more growth” only matters if you can survive the swings. If you can tolerate deep dips, half Kelly offers a higher median but at the cost of nearly half of your bankroll being wiped out at some stage. Full Kelly should be reserved for bankrolls that can absorb almost total erosion without ending the session.

Start every session by fixing your stake at 1.4% of the current bankroll - exactly the quarter-Kelly fraction. Recalculate the stake after each win or loss, and never let a single bet exceed that proportion.

Double Kelly: the edge disappears

Betting 11.1% of the bankroll on every flip of a 55%-win-rate, 1.91-decimal line pushes the stake far beyond the Kelly optimum. The Kelly fraction for that edge is 5.5% of the bankroll; doubling it means each win adds a large chunk, but each loss shaves off an even larger slice. The simulation shows the cost instantly: after 1,000 bets the median bankroll is exactly 1.0 times the start, so the edge that should have compounded to a higher balance is gone.

Those numbers mean that the negative variance introduced by the oversized stake overwhelms the 5.0% edge per bet. In log-space, the expected growth rate turns negative when the stake exceeds the Kelly solution, so the bankroll behaves like a fair game that is being eroded by volatility rather than being amplified by the edge.

The math is simple. Kelly maximises the expected logarithm of wealth, which translates into the fastest sustainable growth. When you bet twice that fraction, the term that represents the edge is halved by the square of the stake proportion, while the variance term is doubled. The result is a net expected log-return of zero or less, which explains why the median stays flat and why half the players end up poorer. The mechanism does not depend on any particular sequence of wins; any reasonably long run will expose the overbetting to the same destructive swing.

The safe path is to respect the Kelly fraction or, better yet, to use a fraction of it.

Calculate the Kelly stake (edge divided by odds-1) and then apply a multiplier no larger than one. Never let a single bet exceed the Kelly-derived percentage, and you will keep the edge alive.

A flat ten percent

Putting a flat 10% of the bankroll on every selection looks tidy. The simulation shows why that neatness is an illusion.

When simulated players staked 10% per bet, the median bankroll after 1,000 wagers was 1.6 times the start. Every simulated player (100%) endured a drawdown of at least 50% at some point, and 43.1% finished the run below their original bankroll. The median growth looks respectable, but the universal half-bankrupt drawdown tells a different story: a flat 10% stake does not protect the bankroll from the inevitable swing of a 55% win-rate.

Contrast that with the quarter-Kelly rule, where the stake was 1.4% per bet. The flat 10% rule sits between those two points: it offers more upside than quarter Kelly but sacrifices the safety margin that the Kelly-derived fraction provides. Because the stake is fixed as a round percentage, it does not shrink when the bankroll contracts, so a losing streak can erase the buffer that a Kelly-scaled stake would preserve.

The mechanism is simple. A flat percentage treats each bet as an independent gamble, ignoring the fact that the edge is expressed as a fraction of the bankroll. When the bankroll falls, the absolute amount risked stays at 10% of a smaller base, meaning the same dollar loss represents a larger proportion of the remaining funds. The result is a feedback loop: a loss reduces the bankroll, the next 10% bet is now a bigger hit relative to the new total, and the cycle accelerates. The round number therefore becomes a plan that amplifies volatility rather than a disciplined staking rule.

If you want the edge to compound instead of being wiped out by swing, replace the flat 10% with a Kelly-derived fraction. Then apply a conservative multiplier, such as a quarter Kelly, which translates to 1.4% of the current bankroll per wager. Recalculate the stake after every result so the bet size always reflects the actual bankroll size. This simple adjustment turns the “nice round number” into a mathematically grounded safety net.

Losing runs arrive even at 55%

Even a modest edge does not stop a long string of losses from showing up.

On a fair coin flip the chance of at least one run of ten losses within 500 flips sits at 21.5%. Extending the horizon to 1,000 flips raises the probability of a twelve-loss run to 11.4%, and stretching to 5,000 flips gives a 15-loss run a 7.3% chance of occurring at least once. Those odds mean that in a career of a few thousand bets, a bettor who thinks “ten losses in a row will never happen” is wrong more than one time in ten.

Roulette’s red/black outcome, with a 48.6% win probability, produces even higher streak odds. Within 500 spins the chance of a ten-loss run climbs to 26.4%, and a twelve-loss run appears in 1,000 spins with a 15.0% probability. Over 5,000 spins a 15-loss stretch is expected with a 10.5% chance. Crash games that pay out at a 48.5% win rate behave similarly: a ten-loss run in 500 rounds has a 27.0% chance, a twelve-loss run in 1,000 rounds a 15.4% chance, and a 15-loss run in 5,000 rounds a 10.9% chance. The numbers show that a losing streak of a dozen or more is not a rare glitch; it is a regular feature of any long sequence.

Low-paying slots make streaks inevitable. When only 26% of spins produce any win, an eight-loss run in 100 spins occurs with 95.2% probability. A ten-loss run in 500 spins, a twelve-loss run in 1,000 spins and a 15-loss run in 5,000 spins all reach a 100.0% probability of appearing at least once. Moreover, a run of 15 dead spins appears within 100 spins with a 25.3% probability, and a run of 30 dead spins appears within 1,000 spins with a 3.7% probability.

A streak that feels impossible in a single evening becomes near certain over a long playing history because every extra round is another chance for it to start.

Most bettors do not have the edge at all

A no-skill bettor faces a negative edge from the moment a single wager is placed. At decimal odds 1.91 on a true 50/50 outcome the bookmaker hold is 4.5%, so the expected loss is $4.50 for every $100 staked.

When the same 4.5% margin is embedded in a multi-leg parlay the loss accelerates. A five-leg parlay carries a hold of 20.6% and loses $20.56 per $100 wagered; a ten-leg parlay’s hold jumps to 36.9% and loses $36.90 per $100. The fair payout for a ten-leg parlay of 50/50 legs is 1,024x, yet the offered payout is only 646x, a shortfall that directly translates into the 36.9% hold.

Our simulation of 100 one-unit bets by no-skill players shows how the negative edge manifests in outcomes. On single bets 30.2% of simulated bettors finished ahead, but the average result was -4.5 units. On five-leg parlays 38.1% ended ahead, yet the average result deepened to -20.4 units. On ten-leg parlays only 9.5% were ahead and the average result plunged to -35.4 units. The higher proportion ahead in five-leg parlays does not offset the larger expected loss per $100; the edge remains negative and the bankroll erodes faster as the number of legs grows.

Stake size only changes the speed of that erosion. Betting 10% of a bankroll on a single at the 4.5% hold loses $4.50 for every $100 risked. Betting the same 10% on a ten-leg parlay loses $36.90 for every $100 risked. The direction of the outcome - loss - does not flip; the magnitude expands proportionally with the hold embedded in the bet type.

If you are determined to place parlays, calculate the expected loss per $100 first and cap your stake accordingly. For a ten-leg parlay, an expected loss of $36.90 per $100 means a $10 stake on a $1,000 bankroll will, on average, shave $3.69 off the bankroll each bet. Adjust the stake to keep the per-bet loss within a tolerable budget, or avoid parlays altogether.

A staking rule for real life

Even a modest advantage evaporates if you let a single losing streak dictate the size of the next wager. The math shows that the only thing separating a thriving career from a busted bankroll is how much of the edge you risk on each ticket. The rule that survives the simulation stress-test is simple: measure your edge conservatively, bet only a fraction of the Kelly stake, and never increase that fraction to chase a drawdown.

Start with the edge itself. Use your own historical data or a transparent model, then shave a safety margin off the raw estimate. This buffer protects against estimation error, which in a long series of bets can otherwise turn a positive expectation into a negative one. The simulation that used a full Kelly fraction (5.5% of bankroll per bet) produced a median bankroll 4.1 times the start but 98.9% of simulated players suffered a 50% drawdown. By contrast, the quarter-Kelly run (1.4% per bet) still grew the median to 1.8 times the start while only 1.9% hit a 50% drop. The difference is the conservative edge estimate combined with a smaller stake.

The Kelly formula translates edge into a proportion of bankroll: f = (bp - q)/b, where b is the decimal odds minus 1, p is win probability and q = 1 - p. Plug in the conservative p and you obtain the “full” Kelly fraction. A quarter-Kelly or half-Kelly stake keeps the growth curve smooth enough for a 1,000-bet horizon while limiting ruin risk.

Never let a losing streak dictate a larger bet. Raising the stake to “recover” a loss merely adds the same volatility that produced the loss in the first place, and it does so at a magnified scale because the bankroll is already smaller.

THE VERDICT

Stick to a fraction of Kelly - or less - and never increase stakes to chase losses.

The data behind this piece

Every figure in the text comes from the runs below. They are computer simulations and exact calculations, not real-money play, and they can be reproduced: the model parameters are stated and the random seed is fixed.

Kelly staking: growth against drawdown for a bettor with a real edge

How we ran it: 5,000 simulated bettors with a genuine 55% win rate at decimal odds 1.91 each place 1,000 bets, staking a fixed fraction of the current bankroll. Random seed: 2027

Same edge, different stake sizes
Staking ruleStake per betMedian bankroll after 1,000 betsSuffered a 50% drawdownFinished below the start
Quarter Kelly1.4%1.8x1.9%6.9%
Half Kelly2.8%2.9x42.4%11.2%
Full Kelly5.5%4.1x98.9%20.0%
Double Kelly11.1%1.0x100.0%51.0%
Flat 10% of bankroll10.0%1.6x100.0%43.1%

How normal long losing streaks are in a fair game

How we ran it: Exact probability of at least one run of consecutive losses in a series of independent rounds, computed by recurrence for four win rates. No simulation is involved.

Chance of seeing the streak at least once
Game8 losses in a row within 100 rounds10 in a row within 500 rounds12 in a row within 1,000 rounds15 in a row within 5,000 rounds
Coin flip (50% win)17.0%21.5%11.4%7.3%
Roulette red/black (48.6% win)20.2%26.4%15.0%10.5%
Crash cash-out at 2x (48.5% win)20.6%27.0%15.4%10.9%
Slot spin that pays anything (26% win)95.2%100.0%100.0%100.0%

How a 4.5% margin compounds inside a parlay

How we ran it: Exact bookmaker hold for parlays built from fair 50/50 legs priced at decimal odds 1.91 (the standard -110 line, rounded), plus a simulation of 20,000 bettors each placing 100 one-unit bets as singles, 5-leg and 10-leg parlays with no skill. Random seed: 2027

Bookmaker hold by number of legs
LegsPayout offeredFair payoutBookmaker holdExpected loss per $100 staked
11.91x2x4.5%4.50
23.65x4x8.8%8.80
36.97x8x12.9%12.90
413.31x16x16.8%16.82
525.42x32x20.6%20.56
648.55x64x24.1%24.14
8177.12x256x30.8%30.81
10646.15x1,024x36.9%36.90
122,357.22x4,096x42.5%42.45
1516,424.82x32,768x49.9%49.88

Questions players ask

What stake should I use if I have a 5% edge on 1.91 odds?

The Kelly formula recommends wagering 5.5% of your current bankroll on each bet when you have a 5.0% edge at decimal 1.91. Our simulation shows that this full-Kelly stake grows the median bankroll to 4.1 times the start after 1,000 bets, but it also produces a 98.9% chance of a 50% drawdown along the way.

How often will I lose half my bankroll using full Kelly?

Running 5,000 simulated careers with a 55% win-rate at 1.91 odds shows that 98.9% of bettors who stake the full Kelly fraction (5.5% of bankroll) experience a drawdown of 50% or more at some point during 1,000 bets.

What happens if I bet double Kelly?

After 1,000 bets the median bankroll is exactly the starting amount, every simulated player suffers a 50% drawdown, and 51.0% end below where they began.

Are parlays profitable for a bettor without an edge?

A 5-leg parlay at true 50/50 odds carries a 20.6% bookmaker hold, and a 10-leg parlay a 36.9% hold, meaning you lose $20.56 or $36.90 per $100 staked respectively. In 5,000 simulated no-skill runs, 38.1% of 5-leg parlays were ahead after 100 bets, but the average result was a loss of 20.4 units, confirming parlays are a losing proposition without an edge.