When someone says a prediction market is "pricing in a 63% probability," they're doing a simple calculation: the YES contract is trading at 63¢, and YES pays $1 if the event happens. 63 divided by 100 is 0.63, or 63%. That's the implied probability.
But implied probability is a market-price-derived number, not a scientific measurement. It has real limitations. Here's how to use it properly.
What implied probability IS
It's the collective current opinion of the traders participating in that market, expressed as a probability. If the market is deep and liquid, and if traders are rational profit-seekers, the price tends to track actual likelihood closely. That's the theory. Most of the time, it holds.
What implied probability IGNORES
Implied probability doesn't account for:
- The bid-ask spread: the "true" probability is somewhere between the YES bid and YES ask. In thin markets, that spread can be 10-20 points wide. A market "trading at 50%" might really be 45% to 55% depending on which side you're trading.
- Fees: Kalshi takes 1-7% per trade. Polymarket takes gas and maker/taker fees. After fees, the "true" break-even probability is somewhat worse for the retail participant than the raw price suggests.
- Liquidity risk: if you want to exit before settlement, you may have to cross the spread or sit in the order book. That's a cost.
- Settlement risk: is the resolution rule unambiguous? Some prediction markets have settled controversially (Polymarket's Ethereum ETF market, various Kalshi ones where the data source was ambiguous).
What implied probability ISN'T
Not a forecast. The market isn't predicting anything, it's reflecting the current collective bet. If fresh information arrives, the price moves. What the price was 10 minutes ago is irrelevant to what it is now.
Not a sports line with vig. Sports betting lines include the bookmaker's house edge (usually 4-10% built in). Prediction market contracts settle at $1 or $0 cleanly; the equivalent of "vig" is the exchange fee, which is separate from the price.
Not a lock on any individual event. A 70% implied probability means 30% of the time you lose. Over many bets on 70% markets, you'd expect to win ~70% of them. On any single one, you won't know until resolution.
When implied probability is most trustworthy
- Deep markets: high volume, narrow bid-ask spread, significant open interest
- Events with clear resolution: "Did the Fed cut rates at the May meeting?" = yes or no, no gray area
- Time close to resolution: short-dated markets have less time for new information to upend the price
- Multiple independent venues agreeing: if Kalshi and Polymarket both price an event at ~65%, that's a stronger signal than one market alone
When to be suspicious
- Wide bid-ask spreads (>5 points), the price isn't stable
- Low volume (<$10k total), a single trader can move it
- Volatile prices, big swings hour-to-hour suggest people are speculating on thin info
- Ambiguous resolution rules, you may be betting on something slightly different from what you think
- Long time to resolution, more time for new information to change the picture
How to cross-check a market's implied probability
If you want to know whether a market is pricing an event correctly, compare it to:
- A qualified external model (NOAA for weather, see our mispricings page)
- Historical base rates (how often has this type of event happened before?)
- Another prediction market for the same event (see our arbitrage pairs)
- Expert analyst surveys (Bloomberg consensus for economic data, for instance)
When those agree within a few percentage points, the implied probability is probably right. When they disagree a lot, something interesting is happening, either the market knows something the model doesn't, or the market is wrong.