Poker Bankroll Management: Fixed Percentage, Kelly and Stop-Loss

Bankroll management is a handful of numbers you fix before you sit down: how many buy-ins you hold, how much of the bankroll a session can cost, and the point at which you drop a stake. Most players set those numbers by feel, and feel does not survive a 40 buy-in downswing.

This article runs the three methods serious players actually use, fixed percentage, the Kelly criterion and the stop-loss, through the same player's numbers, so you can see where they agree, where they disagree and what risk each one leaves you with.

Key takeaways

• The bankroll you need is an output of your win rate and your variance, not a round number from a forum.
• Fixed percentage sizes your stake, Kelly tells you the most aggressive stake that still makes sense, and a stop-loss caps a session. They answer different questions.
• Risk of ruin turns "I feel safe at this stake" into a number you can check.
• The common 20 to 30 buy-in advice is a floor for proven winners, not a plan: for a 3 bb/100 winner it leaves a 10 to 23% chance of going broke at fixed stakes.
• Your win rate is an estimate with wide error bars until you have a large sample, so plan with a cautious one.

The three numbers every method needs

Every method below needs the same three inputs, in the same units. For cash games the convention is big blinds per 100 hands, written bb/100.

• Win rate. Your average profit per 100 hands. Measure it in bb/100, not in money: money changes when you change stakes, bb/100 does not.
• Standard deviation. How far your results swing around that average, per 100 hands. In no-limit hold'em most players sit between 80 and 100 bb/100; this article uses 90.
• Bankroll. Money set aside for poker only, counted in buy-ins of 100 big blinds at the stake you play.

The examples use one player throughout: a 3 bb/100 winner with a standard deviation of 90 bb/100 and a bankroll of $5,000. Tournament players can follow the same logic with ROI per tournament and a standard deviation in buy-ins; there is a section on that further down.

Method 1: fixed percentage

You never risk more than a fixed share of the bankroll at once. A common version is 2% per session, with a session's buy-in as the amount at risk.

On $5,000, 2% is $100, which is one 100 big blind buy-in at NL100. The rule therefore says: play NL100 while you have at least 50 buy-ins for it. At this player's numbers, holding 50 buy-ins at fixed stakes carries a risk of ruin of about 2.5%.

The useful part is what happens after a loss. If you recompute the 2% from your current bankroll, a downswing shrinks the stake you are allowed to play. Stakes come in steps, so in practice it reads: when the bankroll drops below 50 buy-ins of NL100, play NL50 until it recovers. A player who follows that never goes broke; the downswing costs stakes, not the whole roll.

The weakness is that 2% is the same number for everyone. It does not know whether you win 1 bb/100 or 6, or whether you play a game that swings twice as hard. That is what the other two methods are for.

Method 2: the Kelly criterion

The Kelly criterion answers a different question: which stake makes the bankroll grow fastest in the long run? For a game with a win rate μ and a standard deviation σ per 100 hands, the answer is a big blind worth μ / σ² of the bankroll. Put the other way round, the Kelly bankroll is σ² / μ big blinds.

Kelly bankroll = σ² / μ = 90² / 3 = 2,700 big blinds = 27 buy-ins

So at full Kelly this player can play a big blind of up to $5,000 / 2,700 = $1.85: NL200 is just over the line, NL100 comfortably inside it. That is far more aggressive than the 2% rule, and it comes with a cost: if you hold 27 buy-ins and never move down, the risk of ruin is about 13.5%.

Full, half and quarter Kelly for a 3 bb/100 winner

Sizing	Buy-ins of 100 bb	Biggest big blind on $5,000	Risk of ruin at fixed stakes
Full Kelly	27	$1.85	13.5%
Half Kelly	54	$0.93	1.8%
Quarter Kelly	108	$0.46	0.03%

Three things make full Kelly a ceiling rather than a target:

1. Your win rate is an estimate. Kelly assumes you know it. If your true edge is half of what you think, your "full Kelly" stake is really double Kelly, and at double Kelly the bankroll stops growing at all while the swings keep getting bigger.
2. Kelly assumes you resize constantly. The no-ruin property only holds if every loss immediately moves you down. Most players move down late, if at all.
3. The swings are brutal. Even played perfectly, a full-Kelly bankroll has a one-in-two chance of halving at some point on the way up.

That is why players who use Kelly at all use a fraction of it. Half Kelly means half the stake, or twice the bankroll: 54 buy-ins and a 1.8% risk of ruin at fixed stakes. Quarter Kelly means 108 buy-ins. Note how close half Kelly lands to the 2% rule: two different arguments end up near 50 buy-ins for this player.

Method 3: the stop-loss

A stop-loss is a rule for leaving a session: stop after losing three buy-ins, for example. It is popular, and it is often credited with things it does not do.

At fixed stakes a stop-loss does not lower your risk of ruin. Every hand has the same expectation whether you are up two buy-ins or down three, so cutting a session short changes how your results are split into sessions, not where the total ends up.

What a stop-loss does change is how many hands you play while you are playing badly. If you know your decisions get looser or more aggressive after a big loss, a stop-loss removes exactly those hands, and that raises your real win rate. Treat it as a tool for decision quality: set it before the session, at a level that only triggers on genuinely bad days, and never move it during the session.

The same logic applies to a "stop-win". Quitting while ahead to lock in a result feels prudent, but if the game is good and you are playing well, leaving early only costs you hands at a positive expectation.

Where the three methods disagree

Run the same player through all three:

Fixed percentage and half Kelly agree; full Kelly is the aggressive edge of what is defensible; the stop-loss is not a sizing method at all. A sensible plan uses the first two to choose a stake and the third to protect the plan from a bad night.

Risk of ruin: the number behind every method

All of the risks above come from one approximation, the same one the risk of ruin calculator uses:

risk of ruin ≈ e^(−2 × win rate × bankroll / standard deviation²)

With the win rate and standard deviation in bb/100 and the bankroll in big blinds, you can solve it for the bankroll that holds the risk at any level you choose. For this player:

Bankroll for a 3 bb/100 winner at each risk of ruin

Risk of ruin	Bankroll (big blinds)	Buy-ins of 100 bb
5%	4,045	41
1%	6,217	63
0.1%	9,326	94

Now compare that with the advice most players get. Published guides put the minimum for no-limit cash games at 20 to 30 buy-ins: The Poker Bank says 20, and PokerNews' introduction to bankroll management gives 25 to 30 as the minimum and 40 to 50 as a middle setting. Here is what those numbers leave you with, depending on how well you actually win:

Risk of ruin by win rate and buy-ins held

Win rate	20 buy-ins	30 buy-ins	50 buy-ins	100 buy-ins
1.5 bb/100	47.7%	32.9%	15.7%	2.5%
3 bb/100	22.7%	10.8%	2.5%	0.06%
5 bb/100	8.5%	2.5%	0.21%	below 0.01%

For a strong winner at 5 bb/100, 30 buy-ins is a 2.5% risk, which is fine. For a 1.5 bb/100 winner, the same 30 buy-ins means roughly a one-in-three chance of going broke if they never move down. The rule is the same; the risk is not, because the rule ignores the one input that matters most.

Three assumptions sit inside every number in this section: the stake stays fixed, you never withdraw, and the win rate is your true one. None of them holds forever. Moving down after a downswing lowers your real risk; withdrawing money raises it; and the win rate is the weakest link of all.

Your win rate is an estimate with error bars

A win rate measured over a sample is a guess at your true one, and poker samples are noisy. The rough 95% range around an observed win rate is ±1.96 × σ / √(hands / 100). At 90 bb/100 that is ±17.6 bb/100 after 10,000 hands, ±7.9 after 50,000 and ±5.6 after 100,000.

In other words, a player who is +3 bb/100 over 50,000 hands could be anywhere from a losing player to a strong winner. That is the strongest argument for a conservative bankroll: whatever method you use, feed it a cautious win rate, not your best estimate. The hourly rate calculator shows the range for your own sample, and downswing or leak covers how to read your results while the range is still wide.

Tournaments need far bigger bankrolls

The same formula works for tournaments with ROI per tournament as the win rate and a standard deviation measured in buy-ins. The difference is the size of that standard deviation: a few large prizes and many small losses make tournament results swing far more than cash.

For a 20% ROI player, holding a 5% risk of ruin takes about 120 buy-ins in fields of 100 to 200 players and about 367 in fields of a thousand or more. The poker bankroll calculator fills in a typical standard deviation for each format so you can check your own case.

A bankroll plan in six steps

1. Measure your inputs. Win rate and standard deviation for the format and stake you play, from your own results.
2. Choose your risk. 5% if you can refill the bankroll from income, 1% or lower if you cannot.
3. Size the bankroll. Use the risk of ruin with a cautious win rate, and sanity-check it against half Kelly.
4. Set a move-down point. Use a second, higher risk level for it. This player plays NL100 while the bankroll holds the 5% figure, 41 buy-ins, and moves down if the bankroll falls below the 10% one, 32 buy-ins. Write both lines down.
5. Set a stop-loss for sessions, if losses affect your play, at a level that only triggers on bad days.
6. Review every few thousand hands. Recompute the inputs and adjust the stake to the bankroll, not to your mood.

Moving up has its own rules, because a short heater says very little about a new stake: when to move up in stakes covers how to take a shot with a fixed budget and a fixed exit.

Every method depends on your own numbers

All of the above runs on three inputs: a real win rate, a real standard deviation and a real bankroll history. Guess any of them and the output is decoration.

That is what tracking is for. Log every session, with buy-ins, cash-outs and hours, and the win rate, hourly rate and swings come from your results instead of from memory. PokerRoll does that for live and online play, for free, and the calculators linked here run on the same engine as its PokeAI assistant. For a second opinion on the same principles, PokerFuse's bankroll basics covers buy-ins by format and risk of ruin, and The Poker Bank works out how long moving up takes at a given win rate.

Track first, then decide, in that order.