Last Updated: 2026-10-25
Risk: These are hypothetical calculations and a research worksheet, not a trading recommendation or evidence of profitable performance. A position can lose its entire stake.
A disagreement with a market price is a hypothesis. To evaluate it, you need the exact contract, an independently justified probability range, a price available for your quantity, and a record made before the result. A wallet leaderboard, a wide spread or a persuasive AI explanation cannot supply those inputs for you.
Define the event before estimating its probability
Save the market ID, outcome token, full rules, named resolution source, deadline and timezone. “A candidate wins the primary” and “the candidate wins the presidency” are different events; a party may nominate someone by another route. A price gap between those contracts is not automatically arbitrage. The platform's resolution documentation explains why the rules, rather than just the headline, determine settlement.
Write the strongest evidence against your view alongside the supporting evidence. Identify when each source became available. If the rules are ambiguous or the evidence is too weak to estimate a range, record insufficient information. Do not convert that gap into a confident 50% forecast.
Separate price, probability and expected return
Assume a fictional YES share costs $0.40, pays $1 if YES occurs and zero otherwise. Let your estimated probability be 50%, and assume $0.01 of total cost per share through settlement.
Scroll sideways to see every column. Keyboard: focus the table region and use the arrow keys.
| Quantity | Calculation | Result |
|---|---|---|
| Expected payout per share | 0.50 × $1 | $0.50 |
| All-in cost per share | $0.40 + $0.01 | $0.41 |
| Expected net gain per share | $0.50 − $0.41 | $0.09 |
| Expected net return on outlay | $0.09 / $0.41 | 21.95% |
| Loss if NO occurs | $0 − $0.41 | 100% of outlay |
The positive expected value depends on the estimated probability being justified. With a probability range of 35–55%, expected net gain ranges from −$0.06 to +$0.14 per share. That uncertainty matters more than a precise-looking central number. Costs in this example are assumptions, not a current platform quote.
For an 8-cent share with a hypothetical 20% chance of paying $1, expected gross payout/cost is 250%; expected net return is 150% before costs. Neither is the chance of winning. A high payout multiple can still accompany a high chance of losing the whole stake.
A price chart is not an executable trade
Record the ask and depth for entry and the bid and depth for an intended exit, both with timestamps. A last-traded price may be stale; total historic volume does not show how much can be filled now. A limit order can cross the book and take liquidity. An accepted order can remain unfilled or fill only partly.
Keep YES and NO units explicit. If NO is bought at $0.60 and later its executable bid is $0.58, selling loses $0.02 per share before costs. A displayed YES quote of $0.42 does not prove that you can sell NO at $0.58: inspect the actual book. Never subtract an entry spread again when it is already included in the paid fill price.
Dependence can overturn an attractive comparison
The probability of A and B is P(A) × P(B given A). Multiplying two marginal probabilities requires independence. If A has probability 60% and B has probability 80% conditional on A, their joint probability is 48%; an unrelated unconditional B estimate does not replace the conditional value.
A basket is locked only if its quantities and rules produce the required payout in every relevant outcome. A national-election position and a state-election position can both lose. Use the separate payoff-table guide to test claimed arbitrage rather than treating correlation as a guarantee.
Choose a hypothesis with a falsifiable failure case
Scroll sideways to see every column. Keyboard: focus the table region and use the arrow keys.
| Hypothesis | Evidence needed | What can defeat it |
|---|---|---|
| A scheduled release has not been incorporated | Original release, publication time, timestamped quotes and matching contract definition | The quote already reflects the news, or the comparator settles differently |
| A news reaction overshot the evidence | A stated fair-value range and a reason the new information does not justify the move | The information is genuinely new; prices need not revert |
| A wallet has information worth investigating | Verified address, fills, cash flows and public research supporting the event | Unseen hedges, unrelated transfers, selection bias or copying after the price moved |
| Passive quoting earns enough to cover inventory risk | Both-side fills, cancelled orders, inventory marks, fees and adverse-selection losses | Informed traders take one side, leaving inventory that falls in value |
These are different experiments. Do not rank them by an invented win rate or assign them a standard annual return. An untested hypothesis can be rejected without ever funding an account.
Expected return is not a position size
For the simplified no-cost binary example with price c and known probability p, the full-Kelly fraction of bankroll is (p − c) / (1 − c), when that expression is positive. At p = 0.50 and c = 0.40, it is 16.67%, not the 25% expected net return before costs. This is a mathematical comparison, not a suggested allocation: unknown probabilities, correlated positions, execution losses and personal constraints invalidate a mechanical use of the result. No fixed percentage here is established as safe or optimal.
Turn the hypothesis into a reviewable record
Scroll sideways to see every column. Keyboard: focus the table region and use the arrow keys.
| Before the outcome | After the outcome |
|---|---|
| Rules, source links, timestamps and probability range | Actual resolution and any rule dispute |
| Executable quotes, quantity, costs and unmatched-leg scenarios | Filled quantities, actual fees and cash flows |
| A reason the evidence may be wrong | Whether that failure occurred, without rewriting the original forecast |
| All forecasts, including rejected or losing ideas | Calibration and net outcomes over the full predefined sample |
For a binary outcome y, a Brier score is (p − y)². Lower average forecast error can be useful, but it does not prove profit after prices and costs. Keep forecasts separate from a paper execution model, and both separate from actual filled trades. Do not report a selected run of wins as a verified strategy.
Access must comply with the platform's current geographic restrictions; a VPN does not establish eligibility. AITR may earn commissions on marked referral links elsewhere. No provider ranking or return forecast is made here.