Poker Tournament Variance: How Big a Sample Before Your ROI Means Anything?

By Rohit Kaul · July 27, 2026 · 11 min read

On July 13, nine players bagged chips in the World Series of Poker Main Event and went home. They come back on August 3 — a three-week gap the WSOP built in so ESPN can air the finale unspoiled (ESPN). As we publish this, no champion has been crowned. What we do know is the arithmetic that got them there: 9,208 entries, 1,382 paid, one bracelet (Las Vegas Review-Journal).

That shape — a huge field, a thin payout tail, one winner — is what makes tournament results so hard to read. It's why a genuinely skilled player can grind for years and still not be able to prove it from their own numbers. So we ran the math ourselves, using the same variance engine that powers our variance calculator, on field sizes drawn from published simulations. The results are worse than most players expect.

What did the 2026 WSOP actually show?

Two things at once, pulling in opposite directions. More people played the WSOP than ever — a record 251,899 entries across its live bracelet events — and yet the total prize money fell by about 2.5% (pokerfuse). The growth all happened in the cheaper events.

The Main Event ran against the same current. Its field was the fourth-largest in 57 years, but it still came in smaller than the year before, and well short of the 2024 record. The WSOP pointed to inflation, tax-law changes and weaker international turnout (Review-Journal). Online bracelets fell harder still, shedding close to a fifth of their prize money in a second consecutive annual decline (pokerfuse).

Why is tournament variance so much bigger than cash variance?

Because of where the money sits. A cash game pays you roughly in proportion to how well you play each hand. A tournament pays almost nothing to most finishers and enormous sums to a few. The 2026 Main Event is a clean illustration. Around one player in seven cashed, and the minimum cash paid $15,000 on a $10,000 entry (WSOP.com). Outlast nearly eight thousand people and you've earned half a buy-in. First place is worth a thousand times the entry fee (Review-Journal). Nearly all of the money sits at the very top, which is another way of saying nearly all of it is out of reach on any given night.

That top-heaviness is the whole story. You lose most of the time, and you're repaid in rare, gigantic scores. Statisticians call the resulting distribution right-skewed; players call it running bad for six months. Jordan "BBZ" Drummond makes the point sharply: a 2,000-tournament sample can hinge on roughly 20 deep runs (BBZ Poker). Twenty. Out of two thousand.

Here's the part that surprised us most while researching this. Tournament standard deviation is not a number you look up. Primedope's tournament variance calculator — the tool nearly every published MTT variance figure traces back to — doesn't accept standard deviation as an input at all. It derives variance by simulating the payout structure (Primedope). Standard deviation is an output, a consequence of the field you chose to play.

So we back-derived it from published simulation results. Converting the totals reported by Anttonen at Upswing and by O'Kearney and Carter into a per-tournament figure gives a swing of under four buy-ins in a hundred-runner field, and more than thirty in a seven-thousand-runner one. Fit a curve through those points and the swing grows with about the square root of the field. That one relationship is why big-field specialists live in a different universe from sit-and-go regulars, and why a bankroll rule that ignores field size can't be right.

How many tournaments before your ROI means anything?

We ran it. Take a player whose true edge is a strong 20% return per tournament, sit them in fields of roughly 2,000 runners, and ask what their results would actually look like along the way. The table below is the range their measured return falls into, 95 times out of 100.

Read the thousand-tournament row again. That's a very good player, and after a thousand events their results are still consistent with anything from a heavy loss to a spectacular win. Close to one in three of them will simply be down money at that point — staring at a losing graph that says almost nothing about how well they played.

How much volume would actually pin the number down? In those fields, narrowing your measured return to within five points of the truth takes north of two hundred thousand tournaments. Even in gentle hundred-runner fields it runs to tens of thousands. Play every day for a decade and you won't come close. For all practical purposes, nobody has a statistically meaningful tournament ROI — not the winners, and not the losers who think a bad year proves something.

One clarification, because it affects how much weight to give this. Working the same scenario by hand gives a standard error of about 38 percentage points, and the same −54% to +94% band. That agreement checks our arithmetic, not our model — both routes run the identical formula. The model itself gets tested against outside numbers further down.

What does a normal tournament downswing look like?

Ugly, and longer than your bankroll plan assumes. We simulated 20,000 careers at each sample size, one step per tournament, tracking the worst peak-to-trough drawdown each career suffered. These are winning players throughout — true ROI 20%.

A median drawdown of 352 buy-ins is not a disaster scenario. It's the middle outcome. Half of all winning players in those fields experience worse. If you're playing $100 tournaments, that's a $35,200 trough on the way to a profitable career — and the 95th percentile is double it.

Granularity matters here in a way it doesn't for the other tables, so it's worth stating what we did. These careers are walked one tournament at a time. An earlier run of ours used a coarser grid of 200 checkpoints, which quietly smoothed over troughs inside each block and reported drawdowns 3 to 5% shallower. Sampling resolution is an easy place for a simulation to flatter itself.

So how many buy-ins do you actually need?

More than the folklore numbers, and the requirement climbs steeply with field size. To keep a winning player's lifetime risk of going broke down to one in twenty, our engine wants a few hundred buy-ins in small fields and roughly a thousand in 2,000-runner ones. Published work arrives from the other direction and lands in the same place: one bankroll that faced only a modest risk of ruin in a 300-runner field carried an 86.5% risk in the 7,000-runner Sunday Million — same buy-in, same player (O'Kearney and Carter). Anttonen recommends 500 buy-ins for large fields.

That needs squaring with the number everyone quotes. The widely cited 100-buy-in MTT minimum is a floor, not a target, and it holds up best in small and mid-sized fields. In 2,000-runner fields it isn't close. Our own guide to how many buy-ins you actually need draws the same line: 100 as the survival minimum, 200 to 500 for conservative or large-field players. The simulations here are what that range looks like once you attach a specific field size to it.

The practical reading isn't "go get a thousand buy-ins." It's that field size, not stakes, is the dominant risk lever in tournaments. Dropping from 2,000-runner fields to 300-runner fields cuts required bankroll by roughly three quarters at the same buy-in. Game selection is bankroll management here.

Why these numbers still understate your risk

Now the honest part. Everything above is a normal approximation of a random walk with drift — the standard model, and the one inside our calculator. Tournament results are heavily right-skewed, so that model is known to understate tail risk. We wanted to know by how much, so we ran the published scenarios through our own engine and compared.

Each comparison below runs at that scenario's own published inputs, not the 20% ROI and 1,000 tournaments used earlier. On the 300-runner field (25% ROI over 520 events) we get a 17.5% chance of losing money against O'Kearney and Carter's published 17.3%. On a 2,000-runner field (30% ROI over 1,000 events) we get 21.5% against Anttonen's 23%. But on the most top-heavy scenario — the 7,000-runner Sunday Million at 25% ROI over 520 events — we get 43.3% where their structure-aware simulation gives 56.9% (O'Kearney and Carter). Our model is optimistic by 13.6 percentage points exactly where the payout tail is thinnest.

This is also why your in-the-money rate deserves less respect than it gets. Two players can cash at exactly the same rate with completely different edges. Cashing one time in ten means something very different in a hundred-player field than in a six-thousand-player one, where the odds of actually winning are some sixty times slimmer (PokerCoaching). ITM tells you how often you outlast the field. It tells you almost nothing about whether you're beating it.

Putting this to work on your own numbers

Stop treating your ROI as a verdict and start treating it as a wide, slow-moving estimate. Three things follow from that.

• Log everything, including the buy-in structure. A tournament ROI is meaningless without the field sizes behind it. Fifty $50 turbos and fifty $500 deep-stacks are not one sample.
• Judge decisions, not results. If 20 deep runs can swing a 2,000-tournament sample, your last month tells you nothing about your play. Review hands on their merits — that's the honest feedback loop, and it's why we built hand review into TableLab rather than leaning on results alone.
• Size your roll to the field, and know your deal equity. If you do reach a final table, chip counts stop mapping to money — our ICM calculator shows what your stack is actually worth in a deal.

You can reproduce every figure in this article. Drop a 20% ROI, a 12 buy-in standard deviation and 1,000 tournaments into our variance calculator and you'll get the same bands, because it runs the same math. Or feed it your own tracked results instead of our assumptions — that's the version that actually tells you something.

The bottom line

• Your swings are set by the size of the field you play, not by your stakes. They grow with roughly the square root of the entries, so the same buy-in can demand three or four times the bankroll depending on the room.
• A genuinely good tournament player can play a thousand events and still show a loss — better than a one-in-four chance of it.
• Your tournament ROI is an estimate, and it will stay one. The volume needed to make it trustworthy runs into the hundreds of thousands of events.
• The typical downswing for a winner in big fields runs to hundreds of buy-ins. That's the middle outcome, not the nightmare.
• Cashing often is not evidence of an edge. Where you finish matters far more than how often you finish in the money.
• The standard variance model is too optimistic about big fields, ours included — we measured it understating the risk of losing by around thirteen points.

When the Main Event final table reconvenes on August 3, one of nine players will win $10 million and eight won't, and the difference between them will be mostly cards. That's not cynicism — it's the same math that makes your own tournament graph so hard to read. The players who last are the ones who stopped asking their results whether they're any good, and started asking their decisions. Start logging yours, or read our honest take on what AI can and can't tell you about them.

Sources

• Vinson, D., "WSOP Main Event is 4th largest in history, features $10M top prize," Las Vegas Review-Journal, 2026-07-07, reviewjournal.com
• ESPN, "2026 World Series of Poker Main Event final table set," 2026-07-14, espn.com
• pokerfuse, "Here's how much the 2026 WSOP generated in prize money," 2026-07-17, pokerfuse.com
• pokerfuse, "2026 WSOP online bracelets: the numbers behind," 2026-07-21, pokerfuse.com
• WSOP.com, "WSOP Updates 2026: The Main Event bubble set to pop today," 2026, wsop.com
• O'Kearney, D. & Carter, B., Endgame Poker Strategy: The ICM Book, excerpted in "Why Playing Smaller Field MTTs is Good For Your Mental Game," Jared Tendler, 2021-12-16, jaredtendler.com
• Anttonen, M., "The 4 Dumbest Career-Ending Mistakes Tournament Players Make," Upswing Poker, 2017-03-17, upswingpoker.com
• Primedope, "Poker Tournament Variance Calculator," retrieved 2026-07-27, primedope.com
• Drummond, J. (BBZ), "The Truth About Variance in MTTs: Lessons from BBZ's Toy Game," BBZ Poker, 2025-10-07, bbzpoker.com
• Jarvis, E., "Beginner's Guide to Tournament Selection, Part 2," PokerCoaching.com, 2021-03-05, pokercoaching.com
• Simulation code: lib/utils/variance_math.dart and tool/variance_report.dart in the TableLab repository — seeded, so every figure above reproduces exactly.

These are simulations built on assumed inputs, not measurements of your game. The standard-deviation figures are derived from published simulation output rather than a real results database, and the underlying normal model understates tail risk in large fields — by our own measurement. Treat every number here as an estimate, and verify big bankroll decisions against your own tracked results.