Bankroll risk: simple models to estimate your chance of going broke

Bankroll risk (probability of ruin) is the chance your bankroll hits a failure threshold (often zero) before your edge has time to show up. Simple models-Bernoulli/binomial, gambler's ruin, and Monte Carlo-let you approximate that risk and set safer stakes. They also reveal limits: wrong inputs, changing games, and tilt can invalidate neat "risk of ruin calculator" outputs.

Core Concepts: Probability of Ruin at a Glance

• Ruin is defined relative to a threshold (zero, a stop-out level, or "can't play your usual stakes").
• Any estimate depends on your time horizon (sessions/hands/trades) and your bet sizing rule.
• Even with a positive edge, variance + oversized bets can make ruin likely.
• A bankroll risk calculator is only as good as your win-rate/variance inputs and assumptions (stationary conditions, independent trials).
• Use models to choose conservative stakes, then enforce them with operational bankroll management rules.

Defining Bankroll Risk and Time Horizon

Bankroll risk (chance of going broke) is the probability that your bankroll process hits a lower boundary before a chosen endpoint. The boundary can be literal zero, or a practical stop-out like "below 20 buy-ins I must move down." The endpoint is your time horizon: one day, one trip, the next 50k hands, or a full year.

Two players with identical skill can face very different ruin risk if they choose different bet sizes or horizons. "I can't go broke in one session" and "I won't go broke this year" are different questions. That is why poker bankroll management advice often specifies both buy-ins and volume.

A compact way to state the question is:

P(ruin) = P(bankroll hits B_min at any time before T)

Example: You start with 100 units, stop-out at 40 units (move down), and you care about the next 200 trials. That is not "risk of hitting zero"; it is risk of hitting 40 before 200 trials end.

• Define ruin as a specific threshold (0 or stop-out level), not a feeling.
• Set a time horizon (hands/sessions/month) before modeling.
• Separate ruin (hit boundary) from drawdown (temporary dip).

A Simple Bernoulli/Binomial Model to Estimate Ruin

This model treats each trial as a win/loss with fixed probabilities and fixed payoffs. It is crude, but it forces discipline about inputs and exposes how bet size drives risk.

Assume:

• Each trial wins with probability p and loses with probability q = 1 - p.
• Win adds +a units; loss subtracts -b units.
• You start at bankroll B, and ruin is at 0 (or convert to "steps" to a stop-out).

A quick safety check is your expected change per trial:

E[Δ] = p·a − q·b

If E[Δ] ≤ 0, long-run ruin is not a "tail event"; it is a default outcome unless you stop early.

Numerical example (simple): let p = 0.52, a = 1, b = 1. Then E[Δ] = 0.04 units/trial. You have an edge, but if you risk 10 units per trial instead of 1 (same p), variance scales up and the "distance to ruin" shrinks in trials.

Where the binomial view helps in practice:

1. Estimate p conservatively (from a large sample, or discount your observed p).
2. Convert your bankroll into number of loss-steps: N ≈ B / b.
3. Decide a horizon T and ask: "How many net losses would break me within T?"
4. Use the binomial distribution to bound probability of extreme loss counts, or treat it as a screening tool before deeper modeling.
5. If you rely on a "risk of ruin calculator," verify it matches your payoff asymmetry (a ≠ b) and stop-out rules.

• Don't model until you've fixed p, a, b, B, T.
• Reject any setup where E[Δ] ≤ 0 unless your goal is short-term entertainment with a stop.
• Run the model with a discounted edge (e.g., p minus a buffer) to reflect uncertainty.

Gambler's Ruin: Random Walk Intuition and Closed Forms

Gambler's ruin frames bankroll as a random walk between boundaries: you move up or down by steps, and you ask the probability of hitting the lower boundary first. This is the backbone intuition behind many poker bankroll management rules and several closed-form approximations.

Typical scenarios where this model applies well:

1. Cash games with approximately fixed buy-in units: each "step" is a fraction of a buy-in (or a fixed bb amount).
2. Sports betting with consistent staking: each bet is a fixed unit, outcomes roughly independent.
3. Promotional grinding / bonus clearing: many small, similar trials with a modest edge.
4. Shot-taking with a hard stop: upper boundary = "move up", lower boundary = "move down."
5. Multi-day trips: lower boundary = "trip bankroll depleted," horizon is the trip length.

Compact closed-form intuition (even-money steps): with p > 0.5, ruin probability decreases roughly exponentially with the number of steps of bankroll you have. If p ≤ 0.5, the walk is not drift-positive and the lower boundary is eventually hit with high probability unless you stop early.

Numerical example: If each loss costs 1 unit and you have 50 loss-units in bankroll, then "how safe you are" is primarily about (1) how far 50 units is relative to typical downswings, and (2) whether you ever increase step size when winning.

• Use gambler's ruin when you can express the game as step up/step down with stable p.
• Always model two boundaries if you have move-up/move-down rules.
• If p is unstable (table quality, fatigue), treat the model as optimistic and add buffers.

Kelly Criterion vs Fixed-Fraction Strategies: Ruin Implications

Kelly sizing maximizes long-run log growth under strict assumptions (known edge, repeatable bet, independence). It is not a guarantee against large drawdowns, and using "full Kelly" with noisy estimates can be dangerous. Fixed-fraction strategies are simpler and often safer under uncertainty.

Kelly-style sizing: strengths and sharp edges

• Strength: If your edge and variance are correctly known and stable, fractional Kelly can balance growth and drawdown.
• Risk: Misestimating your edge turns "optimal growth" into overbetting, which sharply increases ruin probability.
• Practical note: Many players adopt half-Kelly or smaller to reduce sensitivity to estimation error.

Compact formula (even-money bet): Kelly fraction f* ≈ 2p − 1. Example: p = 0.52 → f* ≈ 0.04 (4% of bankroll per bet). If you instead stake 10%, you are betting ~2.5× Kelly, which increases drawdown/ruin risk substantially if p is overstated.

Fixed-fraction / fixed-unit sizing: robustness over theoretical optimality

• Strength: Easy to implement, easier to audit, compatible with hard stop-loss and move-down rules.
• Risk: If fraction is too high, you can still hit ruin quickly; if too low, growth is slow and opportunity cost rises.
• Best use: When your edge is uncertain, conditions change, or you want predictable risk limits.

• Prefer fractional Kelly only after validating edge/variance with enough data.
• For uncertain edges, default to small fixed fractions plus strict move-down triggers.
• Any sizing method fails if you allow ad-hoc increases after losses (tilt/martingale behavior).

Monte Carlo and Stress Testing for Tail Events

Monte Carlo simulates many possible bankroll paths using your assumed win-rate, variance, and staking rules, then estimates how often you hit the ruin boundary. It is the practical way to approximate complex rules (stop-loss, move down, shot-taking) that closed forms don't capture cleanly.

Common mistakes and myths:

• Myth: "My simulator says 1% ruin, so I'm safe." If inputs are optimistic, the output is fiction.
• Ignoring non-stationarity: your edge changes with game selection, fatigue, and field strength.
• Under-modeling tail risk: real outcomes can be streakier than an IID model (tilt, table changes, correlated decisions).
• Forgetting operational constraints: cashout delays, rake/fees, limits on moving down, and minimum buy-ins.
• Overfitting to recent results: using last month's win-rate as if it were a law of nature.

Compact example: simulate 10,000 paths of 5,000 trials each with your staking rule; estimate P(ruin) as (ruined paths / 10,000). Then re-run with p reduced and variance increased to see sensitivity. This is what a "bankroll risk calculator" should effectively be doing under the hood.

• Run at least base and stress scenarios (lower edge, higher variance).
• Model your real rules (move down, stop-loss, max tables), not idealized behavior.
• Use simulation to choose a stake where stress-case ruin is still acceptable for you.

Operational Rules: Position Sizing, Stop-losses and Risk Limits

Models reduce surprise; rules prevent self-sabotage. The safest bankroll management systems translate "ruin probability" into constraints you can execute every session, especially when emotions or time pressure rise.

Mini-case: you play a game where your edge is modest and variable. You decide ruin is "dropping below 30 buy-ins for my current stake," and you want low risk over the next 3 months of volume. You implement fixed-fraction sizing with strict move-down points.

Simple pseudocode you can actually follow:

if bankroll < 30 buyins_at_current_stake:
    move_down_one_stake()
max_buyins_lost_per_day = 3
if daily_loss_buyins >= max_buyins_lost_per_day:
    stop_playing_for_24h()
stake_buyin = min(current_stake_buyin, bankroll / 50)

How to operationalize with tooling: use a spreadsheet or bankroll management software to log sessions, enforce thresholds, and avoid "silent" stake creep. The tool matters less than the rule clarity and adherence.

• Define move-down levels in buy-ins and treat them as non-negotiable.
• Add a daily stop to prevent tilt-driven overbetting.
• Cap stake as a function of bankroll (e.g., bankroll/50) and update it automatically.
• Track results with simple tooling; don't rely on memory when enforcing limits.

Self-check before trusting your ruin estimate

• Did I state ruin as a specific threshold and a specific horizon?
• Did I re-run the estimate with a worse edge and higher variance than I expect?
• Is my bet sizing rule mechanical, or do I increase size when emotional?
• Do I have an enforced move-down + stop-loss plan that matches the model?
• Will my tracking (spreadsheet or bankroll management software) catch stake creep immediately?

Concise Clarifications on Common Bankroll Pitfalls

Is "risk of ruin" the same as having a bad downswing?

No. A downswing is a drawdown that can recover; ruin means hitting a boundary that forces you to stop or drop stakes.

Can a positive edge still lead to going broke?

Yes. If your sizing is too aggressive or your edge is smaller than you think, variance can push you to the ruin threshold before your advantage manifests.

How should I use a risk of ruin calculator responsibly?

Treat it as a sensitivity tool: run conservative inputs, include your real stop-out rules, and test worse-than-expected scenarios rather than trusting one "best guess."

Does poker bankroll management require different modeling than sports betting?

Often yes. Poker has changing game conditions and skill-dependent variance; sports bets may be closer to independent trials if you keep bet types consistent.

Is Kelly Criterion "safe" for bankroll management?

Only when your edge and variance are reliably estimated and stable. With estimation error, full Kelly can be an overbetting trap; fractional Kelly is more forgiving.

What's the biggest operational reason models fail in real play?

Breaking the sizing rule under stress-chasing losses, shot-taking without a stop-out, or quietly increasing stakes after wins-invalidates the assumptions faster than any math error.