Fees and queue position in Polymarket backtests

• The strategy object includes taker_fee and trade_lag_seconds
• Omitting the fee inflates high-turnover strategies the most
• trade_lag_seconds models the delay from signal to order reaching the book
• Book capture is at 250ms, so a one-second lag spans up to four book updates

Every taker order pays a fee, and the backtester applies it when taker_fee is set on the strategy. A backtest that leaves it at zero is only defensible for a strategy that never takes liquidity, which on short-horizon binaries is rare.

The fee hurts hardest where turnover is highest. A strategy that trades many times per window pays the fee many times, and a small per-trade edge can vanish entirely. Set the fee, then look at how the strategy's edge changes.

Make the fee a first-class field rather than a mental adjustment. Reasoning about a fee that lives only in your head is how it gets forgotten in the version you eventually publish.

The lag is the delay between your signal and your order reaching the book. Set it to at least the honest end-to-end latency of the system you would run - data fetch, decision, submission. A lag of zero silently assumes you see a price and trade at it simultaneously.

On a market that reprices in under a second, the lag decides whether the opportunity still exists. Bothered to measure it? Compare the same strategy at increasing lags: a strategy whose edge survives a two-second delay is far more robust than one that only works at zero.

For resting orders, the archive cannot tell you your place in the queue - that depends on arrival order relative to everyone else. A maker backtest that fills the instant the price touches your level assumes you were first in line, which is the best case and rarely true.

The conservative assumption is that you are behind the resting size and only fill if the market trades through your level. For strategies that take liquidity, queue position is irrelevant, which is one reason taker strategies are easier to backtest honestly.

The realistic plan sets all three: a nonzero taker fee, a latency that reflects your stack, and a conservative queue assumption for any resting orders. Each one alone is a modest haircut; together they routinely turn a marginal strategy from profitable to not.

That is the point of a backtest - to find the not-profitable strategies on paper rather than in the market. If a strategy survives the fee, the lag, the queue and a realistic fill model, it is worth a closer look.

import requests

BASE = "https://api.polyorderbooks.com/v1"
HEADERS = {"X-API-Key": "your-key", "Content-Type": "application/json"}

# Does the edge survive fee and latency?
for fee, lag in ((0.0, 0), (0.002, 1), (0.002, 3)):
    body = {"strategy": {
        "market": {"coin": "xrp", "market_type": "5m"},
        "side": "yes",
        "entry": [{"type": "reference", "change": 0.001}],
        "exit": {"type": "hold_to_expiry"},
        "taker_fee": fee, "trade_lag_seconds": lag},
        "sampleSize": 100, "days": 7}
    r = requests.post(f"{BASE}/ai/backtest", headers=HEADERS, json=body)
    r.raise_for_status()
    print(fee, lag, r.json()["report"]["total_pnl"])

The Starter plan is free, includes order books, prices, and metrics at 250ms resolution, with 3 days of history, 60 requests/min, 1,000/day, 1 free AI backtest, and 3 strategy backtests. No credit card required.

Paid data windows from $19/mo extend history to 30–120 days at higher throughput. Backtest AI add-on is +$19/mo or standalone at $29/mo.

• Fill model
• Slippage model
• Backtest AI