Kelly Criterion for Prediction Markets: My Actual Bankroll Math

The Kelly Criterion for prediction markets is the single most misapplied piece of math in event trading, mostly because traders plug in the wrong probability and get a wildly wrong stake. If you're active on Kalshi or Polymarket, position sizing is the difference between an edge that compounds and an edge that gets erased by three unlucky weeks. This isn't a theoretical exercise. Below is the actual bankroll math — full Kelly, fractional Kelly, and the adjustments that matter when your "probability" input is itself an estimate with error bars.

What the Kelly Criterion Actually Solves For Kalshi Traders

The formula is simple to write and easy to misuse: f* = (bp - q) / b, where p is your true win probability, q is 1-p, and b is the net odds received (payout relative to stake). On Kalshi, a contract trading at 62 cents that you believe resolves YES 70% of the time gives you b = (1-0.62)/0.62 = 0.613. Plug in p = 0.70, q = 0.30, and f* works out to roughly 21% of bankroll on that single position.

Almost nobody should ever stake 21% on one market. The formula assumes your probability estimate is exact, that outcomes are independent across trades, and that you're maximizing long-run geometric growth rather than avoiding short-term ruin. All three assumptions break down in prediction markets, which is why the raw output of the formula is a ceiling, not a target.

Bankroll Management Kalshi Traders Actually Use: Fractional Kelly

In practice, position sizing on event contracts should run at a fraction of full Kelly — typically quarter to half Kelly depending on how confident you are in your probability estimate. Half Kelly cuts variance dramatically (roughly 75% reduction in the volatility of your growth rate) while only giving up about 25% of the theoretical long-run growth rate. That trade-off is almost always worth taking.

Here's the actual math on the example above: full Kelly says 21%. Half Kelly says 10.5%. Quarter Kelly says 5.25%. If your bankroll is $10,000, that's the difference between a $2,100 position and a $525 position on a single market. Given that most retail traders' probability estimates carry meaningful error, quarter Kelly is the more defensible starting point until you have a verified track record of calibration on a specific market category.

The reason this matters more on Kalshi and Polymarket than in traditional markets is that event contracts often have wide bid-ask spreads and thin order books at the tails. A full-Kelly stake sized for a 62-cent fill can end up costing you 65-68 cents once you've walked the book, which silently erodes the edge the formula assumed you had.

Position Sizing Event Trading: Adjusting for Probability Uncertainty

The biggest practical flaw in applying Kelly to prediction markets is treating your probability estimate as ground truth. A model, a forecast, or your own read on a political race is an estimate with a confidence interval, not a fact. There's a well-established adjustment for this: shrink your edge toward the market-implied probability in proportion to your uncertainty.

If you believe a market should be priced at 70% but the current price implies 62%, don't plug 70% into the formula. Instead, weight your estimate against the market's own signal. A reasonable approach: if you're moderately confident, blend 70/30 in favor of your number; if you're only somewhat confident, blend 50/50. That moves your effective p closer to something like 66-68%, which meaningfully reduces the calculated stake and protects you against overconfidence — the single largest driver of blown bankrolls in event trading.

This is also why serious traders track their own calibration over time. If you say "70% confident" across 50 markets and those markets only resolve favorably 55% of the time, you have a systematic overconfidence problem that no position-sizing formula can fix — you need to correct the input, not the math.

Correlated Positions: Where Kelly Math Falls Apart on Kalshi and Polymarket

Standard Kelly assumes each bet is independent. In practice, a trader running multiple positions across Fed rate decisions, multiple congressional races in the same election cycle, or several NFL prop markets in the same game is holding correlated risk, even if each position looks small on its own. Three "5% of bankroll" positions that all resolve based on the same underlying event are effectively one 15% position with correlated variance — and the Kelly formula, applied position-by-position, will systematically oversize your book.

The fix is to size at the portfolio level, not the position level. Group markets by their actual driver — a single election outcome, a single economic data release, a single game — and apply Kelly logic to the group's aggregate exposure rather than to each contract individually. This is one of the areas where structured analysis across a full market book matters more than any single-trade calculation, and it's a common gap in manual spreadsheet tracking.

How PillarLab AI Fits Into This

Manually recalculating fractional Kelly stakes across a moving Kalshi and Polymarket book — while also tracking calibration, correlation, and live price movement — is exactly the kind of structured, repeatable process that benefits from automation. PillarLab AI runs a 9-pillar structured analysis on any market you drop in, pulling real-time data directly from the Kalshi and Polymarket APIs rather than relying on stale screenshots or manual entry.

The framework evaluates each market across dimensions that feed directly into position sizing: liquidity depth and spread cost, historical resolution patterns for similar contract types, current order book pressure, cross-platform price divergence, and a probability assessment that's explicitly separated from the raw market-implied price. That separation matters — it's the same input distinction the Kelly formula requires, and most manual research blends the two without realizing it.

Instead of a single "buy" or "pass" signal, the output is a structured breakdown you can actually plug into a bankroll model: an independent probability estimate, a confidence read on that estimate, and the market's current implied price side by side. That's the exact pair of numbers the Kelly formula needs — and having them generated consistently, market after market, is what makes fractional Kelly sizing practical to run as a repeatable process rather than a one-off spreadsheet exercise. For traders comparing platforms as part of that process, this Kalshi vs Polymarket comparison is worth reading alongside the tool itself.

Building a Practical Bankroll Framework

Putting this together into something usable on a daily basis means setting a few hard rules before you place a single contract:

• Cap single-position sizing at quarter Kelly unless you have a verified, multi-month calibration record in that specific market category.
• Shrink your probability estimate toward the market price in proportion to your confidence — never plug in your raw gut number.
• Group correlated markets and size at the portfolio level, not the individual contract level.
• Recompute stakes when prices move, since b changes continuously and a position sized correctly at entry can become oversized after a favorable move.
• Track resolution outcomes against your stated confidence so you can correct systematic overconfidence rather than just adjusting the formula's fraction.

None of this requires exotic math. It requires discipline in the inputs and consistency in applying the same framework across every market you touch, which is where most manual processes break down under volume. Traders who've compared structured tools for this kind of repeatable analysis often land on the same conclusion — see the breakdown in this betting AI tools comparison for how PillarLab stacks up against alternatives.

Common Sizing Mistakes That Erase the Edge

A few patterns show up repeatedly in traders new to structured position sizing on event contracts. First, treating a market's current price as irrelevant once you've formed your own view — the price is data, and ignoring it in your probability blend is how overconfidence compounds. Second, resizing positions emotionally after a win streak, which is functionally the same error as chasing losses, just in the opposite direction; Kelly sizing should be re-derived from the same disciplined inputs regardless of recent results. Third, ignoring platform-level differences in fee structure and settlement mechanics, which change the effective b in the formula even when the quoted price looks identical across Kalshi and Polymarket — a gap covered in more depth in this prediction apps comparison.

The common thread across all three mistakes is the same: bankroll math only works if the inputs feeding it are disciplined and repeatable. A formula applied inconsistently is worse than a rough rule of thumb applied consistently.

Frequently Asked Questions

What is the Kelly Criterion formula for prediction markets?

f* = (bp - q) / b, where p is your estimated true probability, q is 1-p, and b is net odds based on the current contract price. It outputs the theoretically optimal fraction of bankroll to stake.

Should you use full Kelly or fractional Kelly on Kalshi?

Fractional Kelly, typically quarter to half Kelly. Full Kelly assumes perfect probability estimates, which rarely holds, and produces stakes with unacceptable short-term volatility for most traders.

How do you handle correlated positions in Kelly sizing?

Group markets by their shared underlying driver (same election, same data release, same game) and size at the portfolio level rather than treating each contract as independent.

Does the Kelly Criterion work the same way on Polymarket as Kalshi?

The math is identical, but effective odds (b) differ due to fee structures and settlement mechanics, so recalculate b per platform rather than assuming price parity means identical sizing.

What's the biggest mistake traders make applying Kelly to event contracts?

Plugging in an unadjusted personal probability estimate instead of shrinking it toward the market-implied price based on confidence, which systematically oversizes positions.

Structured, repeatable analysis is what makes any of this sizing math trustworthy over time, and running it manually across dozens of markets a week doesn't scale. Start free with 10 credits and run a full 9-pillar analysis on your next Kalshi or Polymarket position before you calculate the stake — you'll have both the probability estimate and the market data side by side in one structured output.