Kelly Criterion for Prediction Markets: Position Sizing

Most traders on event contract markets have no systematic answer to one question: how much of their available capital should go into any single position? Too small, and winning trades barely move the needle. Too large, and a bad run wipes out months of edge. The Kelly Criterion offers a mathematically principled answer.

Originally developed by J.L. Kelly Jr. at Bell Labs in 1956, the formula links position size directly to your estimated edge. Applied to event contracts on platforms like Gaduin — an offshore exchange where traders take positions on transport delay outcomes and settle in USDT — Kelly provides a framework for sizing that grows capital over time without exposing a bankroll to needless risk.

This article walks through the formula, shows how to adapt it for binary event contracts, explains fractional Kelly as the practical choice for most traders, and ends with a five-step workflow. All numerical examples are illustrative. Nothing here is financial advice.

Before reading further, familiarity with how event contract prices map to implied probability will help.

What Is the Kelly Criterion?

The Classic Formula

The Kelly Criterion answers one question: what fraction of your bankroll should you commit to a position when you believe you have a positive edge?

The formula is:

f* = (bp − q) / b

Where:

f* = the fraction of bankroll to commit
b = the net odds (what you win per unit risked)
p = your estimated probability of winning
q = 1 − p (your estimated probability of losing)

Kelly derived this from information theory. The result is the allocation that maximises the long-run expected logarithmic growth of capital — which turns out to be the fastest sustainable growth rate. No other fixed-fraction strategy grows capital faster over many independent trials.

Why Kelly Maps to Event Contracts

Kelly’s original context was binary: a signal comes in, you either win or you lose, payoffs are known in advance. That structure maps cleanly onto binary event contracts with objective oracle settlement.

On Gaduin, each contract resolves to one of two outcomes — On Time or Delayed — according to a verifiable external data feed. There is no subjective interpretation, no adjudicator discretion. Settlement is automatic when the oracle condition triggers. This objectivity is what makes Kelly directly applicable: the payoff structure is exact, and the only variable is your probability estimate.

The formula was not built for equity markets (continuous prices, variable payoffs, uncertain horizons) or for subjective judgement calls. Event contracts with defined thresholds — a flight delayed more than 15 minutes, a container arriving after a stated port cutoff — match Kelly’s assumptions far more closely.

Applying Kelly to Event Contracts

Price as Implied Probability

On Gaduin, the market price of a contract equals the market’s implied probability of the outcome occurring, expressed as a decimal. A contract priced at 0.30 means the market assigns approximately 30% probability to the event — such as a delay exceeding the contract threshold.

This is the same principle covered in depth at Event Contract Odds & Implied Probability: price × 100 = implied probability in percentage terms.

Your edge exists when your independent probability estimate differs from this market price. If you assess the true probability at 45% and the market is pricing the contract at 30%, you have an estimated edge of 15 percentage points.

Calculating b for a Binary Event Contract

In a standard binary event contract, the payoff structure is straightforward:

You buy at price p_buy
If the contract settles in your favour: you receive 1 USDT per contract, net gain = (1 − p_buy) USDT
If the contract settles against you: you lose p_buy USDT

So the net odds coefficient b is:

b = (1 − p_buy) / p_buy

Illustrative worked example:

Suppose a transport delay event contract is priced at 0.30 (30% implied probability). Your analysis of historical delay data for that route gives you an estimated probability of 0.45.

b = (1 − 0.30) / 0.30 = 0.70 / 0.30 = 2.33
p = 0.45, q = 0.55
f* = (2.33 × 0.45 − 0.55) / 2.33 = (1.0485 − 0.55) / 2.33 = 0.4985 / 2.33 ≈ 0.214

Result: Kelly recommends committing approximately 21.4% of your trading bankroll to this position.

All figures in this example are illustrative. Actual contract prices, your edge, and the resulting Kelly fraction will differ on real markets.

When Kelly Says “Don’t Trade”

If your estimated probability equals the implied probability (price × 100), your edge is zero and Kelly returns f* = 0. You have no mathematical basis for entering a position.

If your estimated probability is below the implied probability — meaning the market is pricing the outcome as more likely than you believe — Kelly returns a negative value. A negative f* does not mean you should short; it means the signal is absent. Walk away.

This is one of the most useful features of the framework: it provides an explicit no-trade signal rather than defaulting to “trade small.”

Fractional Kelly and Bankroll Management

Why Full Kelly Is Too Aggressive for Most Traders

Full Kelly maximises long-run growth but produces extreme volatility along the way. At full Kelly, you should expect drawdowns that feel brutal even if you are consistently correct about your edge. Variance under full Kelly is proportional to the square of the fraction committed — halving the Kelly fraction cuts variance by approximately 75%, while sacrificing only around 25% of expected long-run growth.

There is also an estimation problem. Kelly is only as good as your probability estimate. If you are wrong about your edge — even slightly — full Kelly magnifies that error directly into position size. An estimate of p = 0.45 that is really p = 0.38 does not feel like a small error, but the Kelly fractions it generates will be very different.

Half Kelly and Quarter Kelly

The most common practical adaptation is fractional Kelly: multiply the full Kelly fraction by a constant between 0 and 1.

Multiplier	Position size	Relative growth	Notes
1.0× (Full Kelly)	f*	~100%	Maximum growth, high drawdown
0.5× (Half Kelly)	0.5 × f*	~75%	Recommended starting point
0.25× (Quarter Kelly)	0.25 × f*	~44%	Conservative; suits high uncertainty

For most traders on event contracts, Half Kelly (0.5×) is the recommended default. You give up some expected growth but meaningfully reduce the severity of losing runs. As your confidence in edge estimation improves and is validated by actual outcomes, you can adjust the multiplier upward.

Quarter Kelly suits traders with high uncertainty about their own edge, or those applying conservative risk mandates.

Defining Your Bankroll

The bankroll for Kelly purposes is the capital specifically allocated to trading on Gaduin — not your total wealth, not funds needed for other obligations.

Treating your full liquid net worth as the bankroll would be a fundamental error. Kelly optimises for the designated pool; it assumes the rest of your capital is irrelevant to the calculation. Set a trading bankroll that represents money you can afford to put at risk, segregate it clearly from reserve capital, and apply Kelly only within it.

A practical hard cap: no single position should exceed 20% of your trading bankroll, even if a strong Kelly reading suggests otherwise. Large Kelly fractions often indicate a large but uncertain edge — the hard cap is a safeguard against a systematically overconfident estimate.

Managing Multiple Concurrent Contracts

The classical Kelly formula assumes sequential, independent decisions with a single position open at a time. When you hold multiple event contracts simultaneously, two adjustments matter.

Dividing the bankroll: A practical approach is to divide your available bankroll by the number of active positions and apply Kelly to each share independently. This understates correlation but avoids over-committing the pool.

Correlation risk: Two contracts on flights from the same hub on the same date are not independent. A severe weather event or air traffic control disruption could affect both simultaneously. Positions that share a common risk factor — the same airline, the same infrastructure chokepoint, the same departure window — should be treated as a single correlated exposure for bankroll purposes.

Managing spread and pricing costs is also relevant: Event Contract Liquidity & Bid-Ask Spreads explains how bid-ask spreads affect your effective entry price and therefore your actual edge at the moment of the trade.

Estimating Your Edge — The Hard Part

The Biggest Risk: Overestimating Your Edge

Kelly is a leverage amplifier. It does not generate edge; it scales whatever edge you claim to have. If your probability estimates are accurate, Kelly makes you money faster. If they are inflated, Kelly accelerates your losses.

The sanity check before entering any Kelly-sized position: What informational advantage do I actually have over this market price? Market prices on Gaduin reflect the aggregate of all traders’ estimates, liquidity provider pricing, and historical data already embedded in the pool. Simply having an opinion about whether a flight will be delayed is not an edge. A structured, data-supported reason to disagree with the market price is an edge.

If you cannot articulate exactly why your probability estimate differs from the market’s implied probability, the safer assumption is that you do not have a reliable edge — and the Kelly fraction that assumption produces should be treated as zero.

Sources of Edge in Event Contracts (illustrative — not a guarantee of returns)

Traders on transport event contracts can draw on several categories of verifiable data that may not be fully incorporated into market prices:

Route-specific delay history: airline on-time performance for specific routes and departure windows, available through public aviation databases
Seasonal and meteorological patterns: known congestion periods, weather seasonality at specific airports or ports
Infrastructure constraints: hub airport capacity limitations, air traffic control bottlenecks, known ground handling issues
Vessel and freight data: port congestion statistics, carrier schedule reliability ratings for specific shipping routes

Gaduin’s peer-to-pool market structure — explained in detail at Peer-to-Pool Market Structure — means prices are set by the liquidity pool rather than by a direct counterparty. This can create temporary pricing inefficiencies when incoming information is not yet reflected in the pool’s price.

No category of edge guarantees profitable outcomes. The examples above are illustrative of how traders might approach edge estimation, not a promise of returns.

Risk of Ruin

In continuous Kelly theory, the probability of total ruin approaches zero with an infinite number of trials. In practice, with discrete contracts, limited capital, and estimation error, partial ruin is a real outcome.

A string of losses — even when your edge is real — can draw down a bankroll far enough to impair your ability to continue trading. Three practical rules:

Never commit capital to event contracts that you cannot afford to lose
Track your actual outcomes against your probability estimates; consistent deviation in the wrong direction means your edge model needs revision
Apply the fractional multiplier honestly — 0.5× or 0.25× is not timidity, it is calibrated risk management

A Kelly Workflow for Gaduin Traders

The steps below convert the Kelly framework into a repeatable pre-trade process for individual transport delay event contracts.

Step 1 — Assess the contract

Check the current market price. Apply price × 100 to get the implied probability. Gather your independent data on the underlying outcome — route history, seasonal patterns, any available real-time signal. If you cannot form a defensible estimate that differs meaningfully from the market price, there is no edge and no trade.

Step 2 — Calculate the Kelly fraction

Compute b = (1 − price) / price. Enter your probability estimate as p and q = 1 − p. Solve f* = (bp − q) / b. If f* ≤ 0, skip the trade. If f* > 0.25, cap it at 0.25 — a very high Kelly reading usually reflects an overconfident estimate.

Step 3 — Apply your fractional multiplier

Multiply f* by your chosen multiplier (0.5 for most traders, 0.25 for higher-uncertainty situations). Calculate the USDT amount: multiplied fraction × bankroll. Verify this does not push any single position above the 20% hard cap.

Step 4 — Track portfolio exposure

Before entering, review your active positions. Sum your open USDT exposure. Flag correlated positions (same airline, same hub, same date) and treat them as a single exposure rather than independent entries.

Step 5 — Post-settlement review

After each contract settles, record the outcome against your pre-trade probability estimate. Over time, consistent divergence between estimates and actual outcomes is a calibration signal: your edge model needs revision. Consistent convergence — even across individual losses — confirms the model is working.

Key Takeaways

The Kelly Criterion gives traders on event contract markets a mathematical framework for position sizing consistent with maximising long-run capital growth. Its direct applicability to binary outcomes with known settlement conditions makes it well-suited to transport delay event contracts.

In practice, most traders should use fractional Kelly — Half Kelly (0.5×) as a reasonable starting point, Quarter Kelly for higher-uncertainty situations. Defining your bankroll strictly, capping individual positions at 20%, and tracking outcomes against estimates are the operational disciplines that make the framework work.

Everything here is illustrative. The formula is a tool, not a forecast. Trading event contracts involves real risk of loss, and no position-sizing method eliminates that risk. Nothing in this article constitutes financial advice. Past performance does not guarantee future results.