Maker-Taker Fee Optimization
Maker-taker optimization asks whether to post liquidity or remove it after including the full economic outcome. The visible spread is only a starting point. A one-mill rebate can make passive execution attractive, while a taker fee can erase a small signal; yet a rebate is a poor bargain if the fill systematically precedes an unfavorable price move.
Fee schedules are venue-, symbol-, participant-, and monthly-volume-specific. Store a versioned schedule keyed by venue and effective date. Do not hard-code a single bps number: tier qualification, inverted venues, odd-lot rules, routing fees, and regulatory charges create discontinuities precisely when trading volume changes.
Write the economic objective
For a buy order, compare expected implementation shortfall from arrival price:
\[ C{\rm post}=Pf-A-r+E[\Delta P\mid fill]+C{\rm nonfill} \] \[ C{\rm take}=Pa-A+f+C{\rm impact} \]
Here \(r\) is a maker rebate, \(f\) a taker fee, and \(\Delta P\) is the post-fill adverse move. Prices are signed so a cost is positive. The passive decision requires a queue-aware fill estimate, not the assumption that a displayed quote is available. See queue position modeling for that layer.
| Term | Passive order | Aggressive order |
|---|---|---|
| Explicit fee | Rebate or maker fee | Taker fee |
| Spread | May capture | Pays half/full spread |
| Certainty | Stochastic fill | Usually immediate |
| Information risk | Often high when filled | Lower conditional exposure |
| Market impact | Low displayed impact | Can walk the book |
Routing is an optimization, not a fee sort
The best rebate venue may be slow, shallow, or low in queue priority. A router should score expected all-in cost per child order:
def venue_score(fill_p, edge_if_fill, rebate, adverse_cost, nonfill_cost):
expected_value = fill_p * (edge_if_fill + rebate - adverse_cost)
return expected_value - (1.0 - fill_p) * nonfill_cost
Features include displayed depth, own queue estimate, recent trade intensity, venue latency, fill-to-cancel ratio, and market-wide imbalance. A smart order router also must avoid sending identical marketable orders to multiple venues without protection against overfill.
Fragmentation makes NBBO data insufficient for research. You need venue timestamps, protected quotes, and a causal view of which liquidity was reachable. These details sit inside market microstructure, not in a generic commission field.
Inverted venues and rebate capture
On an inverted market, displayed liquidity may pay the remover and charge the poster. This reverses simple “make versus take” intuition. It can make aggressive routing economically efficient, but only after accounting for the displayed quote’s stability and the chance that a fast competitor removes it first.
Avoid optimizing a backtest against fee leakage. Common errors are:
- Applying today’s tiers to historical orders.
- Crediting rebates for orders that a realistic queue model would not fill.
- Ignoring fees on partial fills and cancels.
- Treating an internalization or midpoint fill as a lit-exchange maker fill.
- Netting a monthly tier without verifying the strategy can actually reach it.
Evaluate net alpha by venue
Decompose results by route and regime:
| Metric | Why it matters |
|---|---|
| Gross edge after fill | Signal quality independent of schedule |
| Fee/rebate per share | Direct schedule contribution |
| Post-fill markout | Adverse-selection measurement |
| Fill and cancel rate | Queue and routing realism |
| Effective spread | Combined price improvement and fees |
If a strategy’s net edge is mainly rebates, stress test fee changes, lost tier status, and lower fill rates. Such strategies are operationally fragile and may create crowded behavior, a concern shared by crowding risk.
Operational safeguards
Reconcile broker confirms to the router’s fee estimate daily. Broken venue codes, incorrect tier assumptions, and partial-fill allocations are common sources of systematic P&L attribution errors.
Key takeaways
- Optimize expected all-in cost, never fee or rebate alone.
- Version schedules by venue and date; tiers make fee economics nonlinear.
- Route selection needs fill probability, latency, markout, and non-fill cost.
- Inverted venues reverse labels but not the need for adverse-selection controls.
- Attribute live performance into gross edge, fee impact, fills, and post-fill markout.