A Polymarket backtest runs a trading rule against past data to see how it would have performed. It is harder than backtesting a stock, because each market resolves once, has a fixed end date, and its price depends on liquidity. A backtest is useful only if you know where the data comes from and what it leaves out.
How to run a Polymarket backtest: what you need
To test a rule properly, you need four things:
- Price history for each market, ideally with timestamps.
- The resolution outcome of each market, so you know which trades would have paid out.
- Trading costs, including fees and the spread between the best bid and the best ask.
- Order book depth, so you can check whether a trade of your size could have been filled at the price you assumed.
Most simple backtests use only the first two. That is the first source of optimism in the results.
Where the history comes from
Polymarket's public API can return price data, but the coverage is not always complete. One developer testing a bot on the 15-minute crypto UP/DOWN markets found that the historical endpoints returned no data for those markets, so the backtest produced no trades. Their workaround was to record live best-ask prices while the bot ran, which produced about 6 GB over four days. This is one developer's account, so check the current Polymarket API documentation before you rely on any endpoint.
For the rules that decide how each market resolves, which a backtest must reflect, read Polymarket's documentation.
Public datasets exist too. One post describes a free dataset of more than a billion Polymarket trades published on GitHub. Check its licence and coverage before you use it.
Tools for tracking your own results
Tracking your own trades helps you compare real results with the rule you tested. Before you run a polymarket backtest, make sure the data includes the trades you need. Polymarket analytics tools can also show how a market moved before it resolved. A Polymarket account tracker or analytics tool can record each trade, so you can check whether your live history matches the backtest.
Common backtest mistakes
Assuming a fixed fee. Polymarket's fees depend on the category, and the schedule changed in 2026. A backtest with one flat fee can be far off.
Ignoring the spread. A strategy that buys at the midpoint looks better than one that pays the ask. Use the actual bid and ask at each timestamp, not the last traded price.
Ignoring depth. A large order in a thin market moves the price against you. A backtest with no depth model assumes you can always trade your full size at the quoted price.
Overfitting. If you tune parameters until the backtest looks perfect, you have fitted noise, not a rule. Keep a holdout period that you do not use for tuning, and test the rule on it once.
Missing markets. Cancelled or annulled markets can be missing from a dataset, which makes the results look cleaner than reality.
A checklist before you trust a backtest
- Confirm that the data covers the markets and periods you want to test.
- Use bid and ask prices, not midpoints.
- Model fees by category and add a slippage assumption.
- Separate a tuning period from a test period.
- Write down what would make you stop the strategy before you see the results.
Backtesting and paper trading are different tests
A backtest shows how a rule would have behaved in the past, under the assumptions you put in. Paper trading tests your process in real time, without money. A good plan uses both. For practice without risk, see how to practise Polymarket with no real money.
Read next
To understand what drives the prices you will test, read how Polymarket resolves markets and how the order book works.
Key takeaways
- A backtest is only as good as its data and its cost assumptions.
- Use real bid and ask prices, and model depth.
- Protect a holdout period from tuning.
- A good backtest is a reason to test further, not a reason to trade.
A polymarket backtest is a test of a rule, not of the market itself, so keep both in view when you read the results.
Data sources, tools and fee schedules change. Check them before you rely on any result.


