Polymarket strategies: a probability-first trading framework

The most defensible Polymarket strategies begin with one question: is your probability estimate meaningfully different from the price you can actually trade? A YES share bought for 0.43 has a maximum resolution value of 1 if YES wins and 0 if it loses. If your evidence supports a 0.52 probability, the simplified expected value is 0.52 - 0.43 = 0.09 per share before fees, spread, slippage, funding or transfer costs, taxes, and estimation error. If you cannot explain why 0.52 is better than the market's view, there is no demonstrated edge—only a preference.

A complete decision therefore needs five parts: a precise reading of the resolution rules; an independent probability range; the order book's executable price for your intended size; a position limit that accounts for related bets; and a written condition for updating or exiting. The displayed probability is a market price translated into probability-like terms, not an oracle. Polymarket's own explanation says prices are formed by trading, while its technical documentation explains that orders rest in an order book and match with other participants.

This framework applies across the specific Polymarket trading strategies below. None removes uncertainty, and a good process can still produce a losing trade. The goal is positive expected value across many independent decisions—not certainty on the next contract.

Start by defining a reference class. For an election, that may combine base rates, polling quality, economic indicators, institutional rules, and time remaining. For a product launch, it may combine the company's prior delivery record, the wording of its announcement, supply-chain evidence, and the exact deadline. Record each input before looking at the market when possible; otherwise the current price can anchor your estimate.

Use a range rather than one heroic number. Suppose your base case for YES is 58%, your cautious case is 53%, and your optimistic case is 63%. At a 0.49 executable entry, the gross expected value under the base case is 0.58 × 1 + 0.42 × 0 - 0.49 = 0.09 per share. The cautious edge is only 0.04. That 0.04—not the 0.09 headline—is the amount available to absorb costs and model error. NIST's probability primer is useful background on distributions; in practice, your estimate is a model with uncertain inputs, not a measured physical constant.

For a NO position, run the same calculation from the NO price and your NO probability; do not assume the displayed YES quote tells you the price at which your whole NO order will fill. Compare expected value at resolution with the value of exiting earlier. Mark-to-market gains can disappear, and money committed until resolution has an opportunity cost. A forecast can be directionally right yet be a bad trade if the entry price already reflects it.

1. Specialist forecasting. Restrict yourself to a domain where you can evaluate source quality faster or better than a general audience. Build a base-rate model, define what new evidence would move it, and timestamp forecasts. The edge must come from analysis—not merely seeing public news after the price has already reacted. Failure mode: expertise becomes overconfidence, especially on one-off events.
2. Relative-value consistency. Compare contracts that share logical constraints. If one outcome necessarily implies another, their prices should respect that relationship after accounting for different wording, dates, resolution sources, costs, and the ability to execute every leg. A apparent mismatch is not an arbitrage if the contracts can resolve differently. Failure mode: treating similar headlines as identical claims or filling only the unattractive leg.
3. Event-driven updating. Define likely catalysts and estimate conditional probabilities before the event. For example: “If source A confirms X, move the range from 42–48% to 56–62%.” This reduces improvisation when a fast market moves. Failure mode: crossing a wide spread on an unverified headline, then discovering the source or contract language says something else.
4. Patient liquidity provision. Place limit orders at prices with a sufficient cushion and accept that they may never fill. A resting order can earn a better entry than immediately taking the best offer, but it also exposes you to adverse selection: informed traders are most eager to trade with your stale quote. Cancel or revise orders when information changes. Failure mode: confusing many fills with a profitable edge.

These strategy families can overlap, but the thesis should identify which one is doing the work. “The market is wrong” is incomplete. “My preregistered model gives 61–66%, the conservative all-in break-even is 55%, and the discrepancy comes from a documented base-rate error” is testable. Keep a record of forecasts and compare predicted probability buckets with outcomes over time. A strategy that cannot be audited is difficult to improve.

An order book has bids from buyers and offers from sellers. The difference between the best bid and best offer is the spread. If YES is bid at 0.47 and offered at 0.51, a trader buying immediately pays 0.51 while a trader selling immediately receives 0.47. The 0.49 midpoint can summarize the book, but it is not guaranteed to be executable. Polymarket documents this order-book price formation directly in its official prices and order-book guide.

Depth matters as much as the top quote. Imagine 100 shares offered at 0.51, 300 at 0.53, and 600 at 0.57. An immediate order for 1,000 shares would have a volume-weighted average price of (100 × 0.51 + 300 × 0.53 + 600 × 0.57) / 1,000 = 0.552. The 4.2-cent difference from the best offer is slippage. A 56% forecast looks attractive against 0.51, but not against a 0.552 average once costs and uncertainty are included.

Before submitting, inspect cumulative depth to your worst acceptable price, calculate average fill, and decide whether a partial fill is useful. Use a limit price derived from your conservative probability—not a desire to “get in.” Recheck the live fee page because charges and eligible markets can change. Liquidity can evaporate around news or near resolution, so your planned exit may be unavailable at a reasonable price.

Binary contracts make maximum contract loss easy to see, but that does not make a position appropriately sized. Begin with a small fixed risk budget per independent thesis and reduce it when the probability range is wide, liquidity is thin, resolution is ambiguous, or several positions share the same driver. Full Kelly sizing is highly sensitive to estimation error; a small fraction of any model-implied size, or a simpler fixed cap, is generally more robust for uncertain forecasts.

Correlation often hides behind different market titles. YES on a candidate winning a primary, YES on that candidate becoming the nominee, and YES on an associated policy announcement may all lose after one adverse event. Adding three contracts does not necessarily diversify anything. Investor.gov describes diversification as spreading exposure to reduce risk, while also warning that it cannot guarantee against loss. Map positions to common scenarios before adding them.

• Write the maximum loss for each position at resolution.
• Group positions by shared catalyst, person, institution, data source, and resolution mechanism.
• Stress a plausible scenario in which every position in a group moves against you simultaneously.
• Include open limit orders; if they all fill during breaking news, they become exposure.
• Keep an uncommitted reserve rather than assuming you can always exit a thin book.

A practical cap might say: no single thesis can lose more than a predetermined fraction of the practice bankroll, and no correlated event cluster can lose more than a larger portfolio cap. The exact percentage is less important than choosing it before the trade and enforcing it consistently. Never increase size solely because a losing price now “looks cheaper”; update the probability and resolution analysis first.

Read the complete market description before forecasting. Identify the exact proposition, time zone and deadline, named resolution source, definitions, treatment of delays or cancellations, and any clarification history. Polymarket's help center explains its resolution process and separate guidance covers market clarifications and disputes. A real-world outcome can match your intuition while the contract resolves the other way because its wording or source used a narrower definition.

Common ways an otherwise reasonable-looking trade fails include:

• Forecast failure: your probability model omitted a base rate or overweighted a vivid source.
• Price failure: the idea was correct, but the executable entry already priced it in.
• Execution failure: spread and slippage consumed the estimated edge or only one leg filled.
• Liquidity failure: you planned to exit but no acceptable bid existed when needed.
• Correlation failure: several “different” trades lost on the same underlying event.
• Resolution failure: you forecast the headline, not the contract's formal criterion.
• Process failure: you changed the thesis after entry, chased losses, or ignored an invalidation.

Use a premortem: assume the position loses, then write the three most plausible reasons. If one is ambiguous resolution, demand a larger margin of safety or skip it. If one is a thin exit book, size as though you must hold to resolution. If one is model uncertainty, use the low end of the forecast range. “No trade” is a valid outcome and often the most important strategy.

Practice each stage with simulated positions before treating results as evidence of skill. Refract Funding provides simulated balances, positions, orders, and fills; it does not route your order to Polymarket. Start with the Polymarket paper-trading guide, then use the same ticket for every idea: contract wording, forecast range, evidence, executable price, expected costs, maximum loss, correlated exposure, invalidation, and planned review time.

Review forecasts in probability buckets. If events you called 60–69% happen roughly that often over a large, varied sample, the model may be calibrated; a short winning streak proves little. Separately measure execution against the book at decision time: average fill versus intended limit, spread paid, slippage, partial fills, and whether patience helped. Finally, label each loss by cause. This is more useful than rewriting the thesis as “bad luck.”

For the broader economics behind profits and losses, read how making money on Polymarket works. For Refract's simulator rules and eligibility details, consult the current program guide. This material is educational, not investment advice. No strategy or hypothetical example guarantees a profit. Trading, balances, positions, and fills on Refract Funding are simulated. Refract Funding is independent of Kalshi and Polymarket. It uses their published market data as a simulation reference and does not send user orders to either venue.