Polymarket Expected Value: The Trap in Positive EV

Polymarket expected value, or EV, is the average monetary result of a purchase under a stated probability estimate. For an ordinary binary share paying $1 or $0 and held to resolution, the calculation per share is EV = probability of winning − execution price − costs. The harder task is justifying the probability. Entering an optimistic number does not establish a profitable opportunity.

This guide works through a hypothetical purchase, tests its sensitivity to error and builds a short pre-trade worksheet. Every price, probability and cost below is illustrative. These are not signals, current quotes or real-wallet results.

How to calculate Polymarket EV

Let q be the estimated probability that the purchased outcome wins, p the execution price per share and c additional costs per share. Use decimal probabilities: 60% is 0.60. For a NO purchase, q refers to the probability of NO, not YES.

A winning share earns 1 − p before costs; a losing share loses p. Weight those results by their probabilities:

EV per share = q × (1 − p) + (1 − q) × (−p) − c = q − p − c.

Here c is fixed for the chosen scenario and paid regardless of the outcome. If costs depend on the outcome, probability-weight them too. For N identical shares, position EV = N × (q − p − c).

This model assumes ordinary final $1/$0 resolution and holding through that resolution. Split payouts or a different exit plan require the actual possible cash flows. See Polymarket’s positions documentation for the mechanism. The guide to Polymarket edge explains the broader concept of an advantage over price; this article focuses on a reproducible cash calculation.

Worked example: 100 shares bought at $0.55

Assume a 60% probability estimate, a $0.55 execution price and a hypothetical $0.01 additional cost per share. Buying 100 shares of one outcome costs $55 plus $1 in costs, or $56 altogether.

EV = 0.60 × $44 + 0.40 × (−$56) = +$4. The short formula gives the same answer: 100 × (0.60 − 0.55 − 0.01) = $4.

Expected profit divided by purchase cost including the assumed costs is $4 ÷ $56 ≈ 7.14%. This is a modeled return on this position’s cost, not an annual return or a promised gain. Under ordinary resolution, the whole position produces +$44 or −$56, not exactly +$4. Owning 100 shares of one market does not create 100 independent trials: all shares share the same outcome.

EV describes an average across possible outcomes if the model is correct. A positive expected value still permits a loss. It also does not specify how much loss a particular person can afford.

Use executable prices and account for costs once

Use the price available for the intended quantity. The displayed market quote may differ from the buying price, and a larger quantity may fill across several order-book levels. The distinction is explained in the official prices and order-book guide.

When portions fill at different prices, weight each price by share count. For example, 40 shares at $0.54 and 60 at $0.56 give (40 × 0.54 + 60 × 0.56) ÷ 100 = $0.552. With q = 0.60 and hypothetical c = 0.01, EV on 100 shares falls to $3.80.

If p uses actual execution prices, slippage is already reflected in p. Do not subtract it twice. Account separately for applicable fees and other additional costs. The one-cent assumption in our examples is not a Polymarket fee quote. Check the current official fee documentation; the platform-cost guide provides further context.

A limit price constrains execution but does not guarantee a fill. An EV estimate should not assume that the entire order will fill at a favorable price when that is only an unverified assumption.

Where does the probability estimate come from?

q must apply to the specific outcome, decision time and information available then. A market quote is a reference for comparison, not independent confirmation of your estimate. Setting q equal to the execution price gives zero EV before additional costs and negative EV after positive costs.

An overall historical UP frequency is not the probability of UP at any arbitrary moment in the current window. The move already observed, remaining time and exact resolution rules may matter. A pattern discovered after reviewing the results needs evaluation on a subsequent period with rules fixed in advance.

A wallet’s historical win rate is not automatically q either: prices, markets, entry times and win-counting methods differ. The win-rate and breakeven guide explains those limitations. Without a defensible probability estimate, an EV table is a conditional “what if q were…” exercise, not an identified profitable trade.

Check whether estimation error changes the sign of EV

Keep the $0.55 price and hypothetical $0.01 cost, changing only the probability:

If a defensible range of assumptions is 54% to 62%, the result ranges from −$0.02 to +$0.06 per share. The sign changes. An attractive central estimate conceals substantial uncertainty.

A range you simply choose is not a statistical confidence interval. Even positive EV at the lower endpoint offers no guarantee: that endpoint could be wrong. Record why those assumptions were chosen, then separately test worse execution and higher costs.

Derive an entry-price threshold from the model

With fixed q and c, zero EV occurs at p = q − c. In our example, that is $0.59. A lower price produces positive EV under the stated assumptions; a higher price produces negative EV.

To require a modeled margin m per share, the condition EV ≥ m becomes p ≤ q − c − m. With q = 0.60, c = 0.01 and m = 0.02, the condition is p ≤ $0.57. This is an algebraic threshold, not an instruction to place an order. It cannot repair an incorrect probability estimate.

When costs change with price or quantity, c cannot remain fixed across every alternative. Recalculate the full cost at each price and size. For the relationship between quoted price and implied probability, see how to read Polymarket odds.

What the simple EV calculation leaves out

• Selling before resolution. An early-exit plan needs a model of sale prices and exit costs, rather than only the final $1/$0 payout.
• Related positions. Expected monetary results can be added, but this does not make risks independent. Several linked positions can lose together.
• Execution changing with size. EV on a small quantity does not scale linearly if a larger order gets a worse average price.
• Holding time. The same expected monetary gain with different capital lock-up periods does not have the same economic meaning.
• Model error. The formula does not assess q’s quality, data coverage or whether an old market regime applies to the future.

Size and potential loss need a separate assessment. The position-sizing guide explains why EV alone cannot set the purchase amount. To review completed actions, use wallet P&L and history, not a probability selected after the outcome.

A pre-trade EV worksheet

1. Record the specific outcome, resolution rules and exit plan: hold or sell early.
2. State the source of q and the range of assumptions. Do not label a scenario input a measured probability.
3. Choose the quantity and estimate the average executable price for that quantity.
4. Add costs without double-counting execution effects. Identify unknown inputs.
5. Calculate EV and both monetary outcomes. Test sensitivity to probability, execution price and costs.
6. Save the time and assumptions before the market result. Later compare predictions with observations without rewriting the original inputs.

A useful calculation makes its assumptions visible. If a small change in an unproven probability removes positive EV, that limitation should be apparent before a decision rather than after a loss.

Sources and calculations

Mechanics checked on 30 October 2026: Positions & Tokens, Prices & Orderbook and Fees. The tables use hypothetical inputs calculated for this article. This guide explains a model, does not recommend specific trades and makes no claim of verified strategy returns.