Bankroll Management for Event Trading (Kelly Criterion)

Put $2,000 on a single Kalshi contract because you're "sure" the Fed cuts in September, watch the data surprise the other way, and you've just erased 20% of a $10,000 bankroll on one trade. Bankroll management trading rules exist precisely to stop that — not by predicting outcomes better, but by sizing positions so no single wrong call, and no cluster of correlated wrong calls, takes you out of the game. The tool traders actually use to do this systematically is the Kelly criterion, adapted for binary YES/NO contracts on Polymarket and Kalshi.

What bankroll management means for event contracts

Event contracts pay exactly $1 or exactly $0 at resolution. That binary structure makes position sizing cleaner than in most markets — there's no need to model a distribution of outcomes, just a probability of one of two states. But it also means a mis-sized position is unforgiving: there's no partial recovery, no "it'll come back." If you're wrong, the stake on that contract goes to zero. Bankroll management is the set of rules — position size, category caps, drawdown limits — that keeps one wrong probability estimate from compounding into account-ending losses.

The Kelly criterion for binary contracts

The Kelly criterion sizes a bet to maximize the long-run geometric growth rate of a bankroll, given a real edge. For a YES contract priced at P (where P is between $0.01 and $0.99, representing the market's implied probability) and your own independent probability estimate p, the full-Kelly fraction of bankroll to stake is:

f* = (p − P) / (1 − P)

The numerator, p − P, is your edge — how much more likely you think the event is than the market does. The denominator scales that edge by the payout odds: the cheaper the contract, the bigger the multiple you're wagering on, so Kelly sizes it down accordingly. If p equals P exactly — you agree with the market — f* is zero. There's no edge, so Kelly says don't bet, full stop.

Worked example: sizing a $10,000 bankroll

Say a Kalshi market on a specific Fed rate decision is trading YES at $0.40. Your own model, built from CME futures pricing and recent Fed commentary, puts the true probability at 0.52. Your edge is 0.52 − 0.40 = 0.12.

Full Kelly: f* = (0.52 − 0.40) / (1 − 0.40) = 0.12 / 0.60 = 0.20, or 20% of bankroll.

Bankroll	Kelly fraction	Stake	Contracts @ $0.40
$10,000	Full Kelly (20%)	$2,000	~5,000
$10,000	Half Kelly (10%)	$1,000	~2,500
$10,000	Quarter Kelly (5%)	$500	~1,250

A 20% full-Kelly stake on a single event contract is aggressive by almost any standard — and that's before accounting for the fact that your 0.52 estimate is a guess, not a certainty. This is exactly where fractional Kelly earns its keep.

Why fractional Kelly, not full Kelly

Full Kelly assumes p is correct. It never is exactly — it's a model output, a forecast aggregate, or a judgment call, and it carries estimation error. Overestimate your edge by even a little and full Kelly oversizes the bet, and because Kelly-sized losses compound geometrically, a string of overconfident full-Kelly bets can carve a bankroll down fast even when your average edge is real. Running at half Kelly (10% in the example above) sacrifices some theoretical growth rate but cuts the variance of that growth rate roughly in half. Quarter Kelly (5%, or $500 in the example) is the more common choice among traders who've been burned by an overconfident model once already — it trades some upside for a bankroll curve that doesn't wipe out on a bad month.

Position and category caps

Kelly math handles one bet in isolation. Real event-trading books hold dozens of open positions, and many aren't independent. The 500+ active 2026 midterm markets on Kalshi and Polymarket, for instance, are heavily correlated — a single generic-ballot polling shock moves House, Senate, and individual race markets together. Treating ten of those as ten separate full-Kelly bets understates your combined exposure to one underlying event. Two caps fix this regardless of what the Kelly formula outputs for any individual trade:

1. Per-position cap: a hard ceiling (commonly 3-5% of bankroll) on any single contract, overriding Kelly if the formula suggests more.
2. Per-category cap: a ceiling on total exposure to a correlated theme — e.g., no more than 20% of bankroll across all midterm election markets combined, no matter how many individually attractive edges you find within that theme.

Ruin risk and the drawdown circuit breaker

Risk of ruin is the probability that a losing streak takes your bankroll below a level you can trade back from. It's driven by three things: position size relative to edge, the correlation between your open positions, and how long you keep sizing up after a string of losses without questioning your model. The practical defense is a drawdown circuit breaker — a rule that says, for example, "after a 15% peak-to-trough drawdown, stop opening new positions until I've reviewed what changed in my calibration." This is a mechanical rule, applied the same way every time, precisely because "just one more trade to get it back" is the instinct that turns a normal Kelly drawdown into ruin.

Running this by hand across dozens of open positions and two or three venues is where most manual bankroll discipline breaks down — you lose track of category exposure or forget the circuit breaker was supposed to trigger. PolyMarketMaker's automated market-making quoter includes a drawdown auto-disarm: set a bankroll drawdown threshold once and the quoter stops pulling new positions the moment it's hit, without needing you to be watching. PolyMarketMaker runs Simulation at $149/mo to test sizing rules before funding a live book, or Live Trading at $299/mo with the safety rails active.

Building the discipline into a routine

None of this works as a one-time calculation. Calibration — how close your p estimates track realized outcomes — is a skill you build by logging every trade's predicted probability against what actually happened, then adjusting. Traders who do this consistently tend to find their early estimates were overconfident in the extremes (too close to 0 or 1) and well-calibrated near 50%, which is itself useful information for scaling future position sizes. Combine that ongoing calibration check with expected value thinking on every trade candidate, and Kelly sizing stops being a formula you look up and becomes the default lens for every position you open.

Where to go next

Everything above applies whether you're taking directional positions or running an automated quoter — market making explained covers how the same sizing and drawdown discipline applies when you're providing liquidity instead of taking it. Before committing real capital to a Kelly-based sizing rule, run it against historical data first with backtesting prediction markets, and if you're new to event trading generally, common Polymarket beginner mistakes covers the sizing errors that show up most often in a first live account.

FAQ

What is the Kelly criterion in event trading?

A formula for sizing a bet to maximize long-run bankroll growth given a real edge. For a binary contract bought at price P with true probability p, full Kelly is f* = (p − P) / (1 − P).

Why use fractional Kelly instead of full Kelly?

Because p is an estimate, not a certainty. Overestimating your edge causes full Kelly to oversize a bet; running at half or quarter Kelly trades some theoretical growth for materially less drawdown risk.

How much of my bankroll should I risk per trade?

It depends on your edge and your confidence in p, but most disciplined traders cap any single position at 3-5% of bankroll regardless of what Kelly's raw output suggests.

What is risk of ruin in prediction market trading?

The probability a losing streak drops your bankroll below a recoverable level. It rises with oversized positions and with correlation between positions you're treating as independent.

Does the Kelly criterion work for correlated event contracts?

Not without a category cap. Kelly assumes independence between bets; many event contracts (all 2026 midterm markets, for example) move together on shared news, so total exposure per theme needs its own ceiling.

This article is for educational purposes only and is not financial advice. Trading involves risk of loss.