Chapter 13 of 33
The Short Version
Of roughly 1.5 million wallets that have traded on Polymarket, about 7.6% finish profitable and 84.1% lose money. Ask the profitable minority what separates them from the losers, and you will hear the same answer over and over: position sizing. Not picks. Not timing. Not secret alpha. The boring question of "how much do I put on this trade?" is the single biggest driver of whether you end the year up or broke.
This guide gives you a system. First, the math: the Kelly Criterion, the formula that tells you the mathematically optimal fraction of your bankroll to bet given a real edge. Second, the reality: why nobody actually uses full Kelly, and why quarter Kelly is the professional standard. Third, the cheat sheet: simple tables you can use today even if you never want to touch a calculator again. By the end you will have a default sizing rule for every trade you place.
What you'll learn in this guide
- The Kelly Criterion formula adapted for Polymarket's price format
- Three fully worked examples (clear edge, modest edge, small edge)
- Why full Kelly destroys real traders and why quarter Kelly wins
- Bankroll sizing tables from $500 to $250,000
- How to handle multiple positions, correlation, and category concentration
- When to size up, when to size down, and the one hard rule you never break
- The $2M case study that proves sizing matters more than being right
- Eight of the most common position-sizing questions, answered
Quarter Kelly sits in the sweet spot: most of the growth, fraction of the volatility.
Part 1: Why Sizing Matters More Than Picking Winners
A now-famous Polymarket trader lost more than $2 million in 2024 despite winning 51% of his individual trades. Think about that for a second. He was better than a coin flip. He had a real edge. And he was crushed anyway.
The reason is brutal but simple: he bet small when he was right and huge when he was wrong. A handful of oversized losses on high-conviction trades overwhelmed hundreds of correctly-sized wins. The math does not care how often you are right. It cares how much you bet when you are wrong.
The arithmetic of uneven sizing
Imagine 100 trades, 51 wins at $500 profit each and 49 losses at an average of $1,200 each (because you sized up on the ones you felt most strongly about).
- Wins: 51 x $500 = +$25,500
- Losses: 49 x $1,200 = -$58,800
- Net: -$33,300
You won more than half your trades and still lost a third of your bankroll. That is the entire story of why sizing matters.
Position sizing answers the only question that matters when you click Buy: "How much should I put on this?" Get it right and a modest edge compounds into real money. Get it wrong and even world-class analysis leads to ruin.
Why “being right” isn’t enough — the sizing column is what determines PnL.
Part 2: The Kelly Criterion — The Mathematical Answer
The Kelly Criterion was derived by Bell Labs scientist John Kelly Jr. in 1956 while working on signal-to-noise problems for long-distance phone lines. Ed Thorp later adapted it for blackjack and the stock market. Today it is the foundation of professional bankroll management in every edge-based game: poker, sports betting, options trading, and prediction markets.
The classic formula
f* = (b x p - q) / b
- f* = optimal fraction of bankroll to bet
- b = net odds received on the wager (profit per $1 risked)
- p = your estimated probability of winning
- q = probability of losing (1 - p)
The Polymarket simplification
On Polymarket, every share is priced between $0.01 and $0.99 and pays $1.00 if the outcome resolves Yes. That makes Kelly much cleaner:
f* = (p - c) / (1 - c)
- c = current share price (implied probability)
- p = your estimated probability
If p > c the formula gives you the fraction of your bankroll to deploy. If p < c the result is negative -- which is the formula politely telling you not to take the trade.
The Polymarket Kelly formula. Negative result? Skip the trade.
Part 3: Three Fully Worked Examples
Example 1: Clear Edge (20 points)
Market: "Will Fed cut rates at March meeting?"
- Current price: $0.40 (40% implied)
- Your estimate: 60% (based on recent CPI, Fed speeches, futures curve)
- Edge: 20 percentage points
f* = (0.60 - 0.40) / (1 - 0.40) = 0.20 / 0.60 = 0.333
Full Kelly says bet 33.3% of bankroll. On $10,000 that is $3,333.
Example 2: Modest Edge (10 points)
Market: "Will Lakers make the playoffs?"
- Current price: $0.65 (65% implied)
- Your estimate: 75%
- Edge: 10 percentage points
f* = (0.75 - 0.65) / (1 - 0.65) = 0.10 / 0.35 = 0.286
Full Kelly: 28.6% = $2,860 on a $10K bankroll.
Example 3: Deep Favorite (8 points)
Market: "Will BTC close above $50K on Dec 31?"
- Current price: $0.80 (80% implied)
- Your estimate: 88%
- Edge: 8 percentage points
f* = (0.88 - 0.80) / (1 - 0.80) = 0.08 / 0.20 = 0.400
Full Kelly: 40% of bankroll. The formula gets aggressive on deep favorites because the per-share payout multiplier (1 / c) is smaller, so Kelly compensates by increasing size. This is exactly where real traders get wrecked -- more on that in a moment.
Part 4: Why Full Kelly Destroys Real Traders
Full Kelly maximizes long-term geometric growth rate. In theory it is optimal. In practice it is a suicide pact. Three reasons:
| Problem | What it means |
|---|---|
| Volatility is brutal | Full Kelly has a ~33% chance of halving your bankroll before doubling it. Drawdowns of 50-80% are routine, not tail events. |
| Estimation errors cascade | Kelly assumes you know the true probability. You don't. If you overestimate your edge by just 5 points, full Kelly becomes ruinous. |
| Emotional impossibility | Nobody, including you, can watch a correctly-sized full-Kelly portfolio draw down 60% without deviating. Deviation in fear is worse than under-sizing. |
Real talk
No professional poker player, no serious sports bettor, no hedge fund I have ever heard of uses full Kelly. It is a theoretical upper bound, not a strategy. Treat the Kelly formula's output as the maximum size you would ever consider, not the number you actually bet.
Full Kelly vs quarter Kelly: same expected growth path, wildly different drawdowns.
Part 5: Quarter Kelly — What Professionals Actually Use
The community consensus across poker, sports betting, and prediction markets: use quarter Kelly (f*/4). Some more aggressive traders use half Kelly (f*/2) on their highest-conviction trades. Almost nobody goes above that.
Why quarter Kelly wins in the real world
- Drawdown cut by ~75% — expected worst drawdowns go from 60%+ to 15-20%
- Growth rate cut by only ~50% — you give up half your theoretical return but keep most of the compounding
- Survives estimation error — if your 60% probability estimate was really 55%, quarter Kelly is still profitable. Full Kelly might already be bleeding.
- Emotionally sustainable — you can actually stick to your system when the drawdowns are small enough to sleep through
Quarter Kelly applied to our examples
| Example | Full Kelly | Quarter Kelly | Dollars on $10K |
|---|---|---|---|
| Fed cut (40c, 60% est.) | 33.3% | 8.3% | $833 |
| Lakers (65c, 75% est.) | 28.6% | 7.1% | $714 |
| BTC $50K (80c, 88% est.) | 40.0% | 10.0% | $1,000 |
These numbers feel right. $700-1,000 per trade on a $10K bankroll is aggressive enough to matter when you win and survivable when you lose.
Print-and-stick reference — the cheat sheet used by most active traders.
Part 6: The "I Don't Want to Do Math" Cheat Sheet
If staring at the Kelly formula makes your eyes glaze over, use these community-tested tables. They approximate quarter Kelly for typical edges (8-15 percentage points) and force portfolio diversification.
Standard sizing table
| Bankroll | Conservative (2-5%) | Moderate (7-10%) | Maximum (15%) |
|---|---|---|---|
| $500 | $10-25 | $35-50 | $75 |
| $1,000 | $20-50 | $70-100 | $150 |
| $5,000 | $100-250 | $350-500 | $750 |
| $10,000 | $200-500 | $700-1,000 | $1,500 |
| $25,000 | $500-1,250 | $1,750-2,500 | $3,750 |
| $50,000 | $1,000-2,500 | $3,500-5,000 | $7,500 |
| $100,000 | $2,000-5,000 | $7,000-10,000 | $15,000 |
| $250,000 | $5,000-12,500 | $17,500-25,000 | $37,500 |
The one hard rule you never break
Never risk more than 15-20% of your bankroll on a single trade, no matter how certain you feel. This rule has saved more traders than every signal service combined. The moments you feel most certain are the moments you are most likely to be catastrophically wrong.
Balanced portfolio: active size, patient limits, permanent cash reserve.
Part 7: Managing Multiple Positions
Kelly tells you how to size one trade in isolation. Real portfolios have 5-20 positions at a time. Three concepts turn single-trade Kelly into a complete system.
Rule 1: Keep cash in reserve
Never deploy more than 75-80% of your bankroll at once. You want dry powder when a fat opportunity appears -- and they always appear when you are fully loaded.
Rule 2: Treat correlated positions as one
Two "Trump wins" markets across different platforms are one bet, not two. Two "BTC above $50K by year end" markets with different expiries are highly correlated. If you are already 10% deployed on Trump, don't add another 8% on the same state's Senate race for the same party. Aggregate exposure matters, not position count.
Rule 3: Cap category concentration
Don't put more than 35-40% of deployed capital in a single category (politics, crypto, sports, etc.). Categories have shared risk factors: a CPI surprise hits every rate market, a protocol exploit hits every crypto market. Diversification across categories is real diversification.
Sample portfolio on $10K bankroll
| Slot | Allocation | Purpose |
|---|---|---|
| 5-8 active positions | ~$700-1,000 each, $5,000 total | Core trades, quarter Kelly sized |
| Pending limit orders | $2,500 | Patient entries waiting to fill |
| Cash reserve | $2,500 | Opportunity fund + safety buffer |
Part 8: When to Size Up, When to Size Down
Size DOWN when any of these are true
- High estimation uncertainty — you think there might be an edge but can't articulate it clearly
- Illiquid market — daily volume under $50K or spreads wider than 3 cents
- Long time to resolution — capital locked for 3+ months increases opportunity cost
- Recent losing streak — could be variance, could be broken model. Shrink until you know which.
- Emotional state off — tired, angry, high, or revenge-trading. Cut size by half.
- New category for you — first 10 trades in a new domain, size at 50% of your normal until you verify calibration
Size UP (toward half Kelly) only when ALL of these are true
- Extremely high conviction with documented evidence — rules-reading edges (like the Trump Says China case) where the wording forces resolution
- Imminent resolution — minutes to hours, not weeks
- Deep liquidity — $500K+ daily volume, tight spreads
- Verified calibration — you have 100+ closed trades and your 70% confidence picks actually win ~70% of the time
Part 9: Bankroll Growth Rules
As you win, your bankroll grows. As it grows, your position sizes grow proportionally -- but not faster. Discipline here is what separates the 7.6% from everyone else.
The recalculation rhythm
- Weekly: total bankroll value including open positions at mark
- Before each trade: use current bankroll, not yesterday's peak
- After losing streaks: size on the depressed bankroll. Do not try to "win back" losses by upsizing.
- After winning streaks: same rule. New peak becomes the base, but percentage sizing stays constant.
How compounding actually looks
Starting bankroll: $5,000. You average 10% monthly growth with quarter-Kelly sizing.
| Month | Bankroll | Typical position (8%) |
|---|---|---|
| 0 | $5,000 | $400 |
| 6 | $8,860 | $710 |
| 12 | $15,690 | $1,255 |
| 24 | $49,200 | $3,940 |
After 2 years of 10% monthly compounding with consistent quarter-Kelly sizing, a $5K bankroll becomes ~$49K -- and your position sizes have grown 10x without you ever changing your percentage rule.
The survival curve: what a 10-trade losing streak does to your bankroll.
Part 10: The Most Important Rule
Position sizing exists to protect you from the only scenario that truly ends your career: a string of losses that wipes you out.
Check the math on survival at different sizes. Assume a brutal but possible 10-trade losing streak:
| Size per trade | Bankroll after 10 losses | Can you recover? |
|---|---|---|
| 2% | 82% | Easy |
| 5% | 60% | Hard but doable |
| 10% | 35% | Brutal — need 186% gain to recover |
| 15% | 20% | Near-impossible — need 400% gain |
| 25% | 6% | Game over |
Losing streaks of 5-10 trades happen. They will happen to you. They are not a signal that you are broken -- they are a feature of any edge-based game with positive variance. Your job is to size small enough that variance cannot kill you.
Survival is the prerequisite for profitability. Size to survive first, profit second. Everything else is details.
Part 11: The Liquidity-Adjusted Size Cap
The Kelly number is theoretical. The book is real. Before you size up, check the actual bid/ask stack: a perfect Kelly size of $3,000 is wrong if the top-of-book shows $800 of size. The two caps that matter on Polymarket:
- Never consume more than 25% of visible depth on your side of the book in a single order. Doing so moves the price against you and signals your intent to market-making bots.
- Skip the trade if executing your Kelly size would require more than 0.5% of daily volume. That level of impact implies a book too thin to deserve your best edge.
A practical workflow: compute Kelly, compute quarter-Kelly, then cap at 25% of top-of-book depth. Break the remainder into staged limit orders over the next 15–60 minutes. You forfeit a small amount of expected value to execution patience; in exchange you get a much tighter realized PnL distribution. Polymarket’s own documentation confirms there are no platform-imposed trading limits — slippage is the only constraint, and you are the one who has to enforce it.
A simple heuristic that works: if a single market-buy would move the price more than 1 cent on a market trading between 20¢ and 80¢, break the order into two or three pieces. At prices below 10¢ or above 90¢, the same 1-cent move represents 10%+ of the payout spread, so tighten further — split into five or more slices and favor limits inside the spread over aggressive takers.
Part 12: Pro Habits From the 7.6%
Validated sizing habits you can adopt today
- Set a hard per-trade dollar cap. Even on huge edges, most professionals cap a single trade at 2–5% of bankroll regardless of what Kelly says. “I would lose X and not change my behavior” is the right size.
- Log the Kelly input before every trade. Write down the market price, your probability estimate, and your sized position in a journal. After 100 trades, you’ll discover whether your probability estimates are systematically too optimistic.
- Never size up on a loser to recover. Revenge-trading doubles your size after a loss. The correct response to a loss is to check whether your thesis still holds, then enter at the same size or smaller.
- Rebase weekly, not trade-by-trade. If you recalc bankroll after every win, you get size creep; if you recalc after every loss, you cut too aggressively. Weekly mark-to-market on a fixed day smooths the ride.
- Pair Kelly with a stop-loss discipline. Kelly assumes you can lose the full stake. Combine with the −40% stop-loss rule (see Selling Positions) and your real downside is half of full Kelly’s.
- Account for fees in the edge calculation. A 2pp edge on a Crypto market (1.80% taker) is not the same as a 2pp edge on Geopolitics (0%). Subtract the round-trip fee before computing Kelly.
What's Next?
Sizing is the skeleton. The meat is the edge itself and the discipline to execute it consistently. Keep reading:
