November 16, 2026 · execution

Why Your Strategy Optimizes for the Wrong Fee Tier

Why Your Strategy Optimizes for the Wrong Fee Tier

A strategy can pass optimization because the backtest charged it fees from a trading tier the account has never reached. The mistake is easy to miss: fee schedules are public, the rate is real, and the equity curve is mathematically correct. The account assumption is what’s wrong.

Fee tiers often depend on trailing volume, token holdings, or both. A strategy that trades $100,000 a day may qualify for a better rate after building enough volume, but it cannot claim that rate on day one. If the edge only survives at the discounted tier, the backtest has quietly assumed a trading history it hasn’t earned.

Mistake 1: charging the best tier from the first trade

Suppose a venue charges 0.10% per side at the base taker tier and 0.06% at a higher-volume tier. Your round trip costs 0.20% at the base rate, before spread and impact. At the discounted rate it costs 0.12%. A strategy with 0.16% gross edge per round trip loses money at the base tier and appears profitable at the discounted one.

If the backtest applies 0.06% from its first fill, it has given the strategy a subsidy. The account may eventually qualify for the discount, but only after generating the required rolling volume. That volume has a cost: it comes from trades the strategy must actually make, and the rate may rise again if activity falls.

0.08%round-trip cost gap between these tiers
0.16%gross edge in the example

The wreckage usually shows up in a sensitivity run. The equity curve bends sharply when you replace the assumed tier with the account’s current one. A useful model tracks eligibility through time using the venue’s actual rolling-volume rules, or reports separate results for clearly stated tiers. If qualification depends on activity outside the strategy, treat that activity as an explicit assumption.

Mistake 2: treating maker fees as a discount you can select

Maker fees are tempting in a backtest. Put a limit order near the market, label it maker, and replace a taker fee with a smaller charge or even a rebate. On paper, a few basis points can rescue a fast strategy.

But a resting order earns maker treatment only if it adds liquidity when it executes. A limit order that crosses the spread may be charged as a taker. Even a genuinely resting order can sit unfilled while the market moves away, fill only on the wrong side of a price move, or miss the trade entirely. The filled subset is not a random sample of all placed orders.

I once reviewed a replay where the limit filled whenever the next bar touched its price. The strategy looked wonderfully patient; the market had been made wonderfully cooperative. A fill model needs to distinguish fee classification from fill probability and adverse selection. Model queue assumptions and missed fills where the data supports them, then show a taker-only result as a useful stress case. A maker rebate is revenue only on fills that could plausibly have happened.

Mistake 3: blending fees from different products and account modes

One venue can publish different rates for spot, perpetual futures, options, and special account programs. A backtest that loads one generic “exchange fee” for all instruments can be wrong even when every symbol points to the same brand.

What changesWhy it mattersCheck in the model
ProductSpot and derivatives can have separate schedulesFee lookup keyed by instrument and product
Liquidity roleMaker and taker rates differClassify each execution, not each order
Account statusVolume or holdings can unlock discountsEligibility over the relevant time window
Fee currencyA discount token can fluctuate against the account currencyRecord the fee amount and conversion rate

The wreckage is a cost report that looks precise but can’t explain which schedule produced its numbers. Keep fee rules versioned by venue, product, date, account mode, and execution role. When a schedule changes, the historical backtest should still use the rules that applied then, while a forward paper run should use the current ones.

What should a fee-aware backtest report?

Show the fee assumption beside the result, including the tier, maker share, and how tier eligibility was determined. Then rerun the strategy under plausible alternatives: current account tier, taker-only execution, and a conservative fee schedule. If the strategy survives only one optimistic combination, that is a useful research finding. It tells you exactly what needs to be true before the paper result can be compared with the backtest.

Fees are not a constant attached to a ticker. They’re a contract between a venue and an account, with conditions that can change over time. Model the contract the strategy can actually meet, and the backtest’s cost line becomes something you can try to reproduce.

trading feesbacktestingstrategy optimizationexecution costs
← All posts