Retail market making economics on modern perp DEXs

Modern perp DEXs — Paradex, Aster, Hyperliquid, GRVT — are priced and flow-engineered to tax API market makers. This is a structural observation, not a complaint. The design is visible in how each venue's fee schedule and flow-routing rules slot retail takers, designated market makers, and API makers into distinct economic roles. For a retail-scale operator, this shapes which strategies are viable before any code is written.

The three-way design

On any perp DEX where a significant share of volume comes from retail flow, the venue has three economic actors:

• Retail takers. Price-insensitive, unsophisticated, and the venue's customer in the commercial sense. Fees kept at or near zero to retain flow.
• Designated / professional market makers. Capital-committed liquidity with explicit program membership. Receive deep rebates and often preferential flow routing; the venue's structural partner.
• API market makers in between. Retail-scale capital, running maker strategies programmatically, without program membership. On venues indifferent to this middle class, standard maker rebates apply and economics are what your alpha and markout make them. On venues designed with retail-vs-DMM routing as a core feature, the middle class pays for both.

Modern perp DEXs are the second kind.

Venue by venue

Paradex — explicit retail/DMM routing via RPI

Paradex's Retail Price Improvement mechanism routes retail orders to Professional Market Makers through a hidden-order layer. API makers without PMM designation are structurally cordoned off from the benign retail flow and quote only against the remaining order book — dominated by non-retail, informed flow. Retail takers pay effectively zero; the PMM program has its own preferential rebate. For an API maker, RPI is the mechanism by which the venue's best flow is unavailable by design.

Aster — MM rebate gated behind a $100M/month program

Standard retail maker fees on Aster are positive (~0.5 bps per fill after token-based discounts). The MM rebate tier is gated behind volume thresholds that are inaccessible to retail capital by several orders of magnitude. The effect is similar to Paradex's RPI, achieved through the fee schedule rather than routing: the small maker pays every fill; the large maker earns on every fill.

Hyperliquid — vault staking as the price of admission

HL's most favourable MM tier requires validator- or vault-level economics equivalent to thousands of USD per month in opportunity cost. This is a capital-commitment bar, not a monthly fee. Retail API makers run at default tier, where per-fill economics do not close against observed flow toxicity on most tokens.

GRVT — tier 1 too small, tier 2 out of reach

GRVT's public retail tier pays a small maker rebate (~0.01 bps). Tier 2 (~0.04 bps) requires $100k/month volume, which at retail capital is ~200 trading days of full capacity. The tier-1 rebate is two orders of magnitude smaller than the adverse-selection markout we have measured at the venue level. The rebate as-of-today does not close the gap.

Why spread + rebate doesn't close

In any passive maker strategy, per-fill net decomposes as:

per_fill_net = gross_spread_capture + rebate − markout − slippage

On modern perp DEXs every term in this equation works against the retail API maker:

• Spread capture is microstructure-bounded. On tick-floored markets (BTC, ETH on most major DEXs) there is no spread to capture. The minimum tick dominates quoted width. Strategy cannot manufacture capture from zero.
• Rebate tier is gated. Retail tier is near-zero on most venues; the meaningful rebate tier is off the table without capital that retail operators don't have.
• Markout is venue-level, not symbol-level. We ran a small cross-symbol study on a mid-sized perp DEX — three symbols of different microstructure profiles (fast retail, slower retail, thin TradFi-adjacent), a few hundred fills each — and per-fill markout landed in a tight band around −1 to −2 bps at the venue level. Changing symbols did not change markouts meaningfully. Adverse selection appears to be a property of the venue's flow composition, not of the instrument.
• Slippage shows up in fill rate and post-only rejections. On tight books, minimum-reprice thresholds and post-only cross protection suppress fills that would otherwise be profitable. A second-order tax on an already-negative primary term.

Take the numbers together and the shortfall is structural, not tactical. At a venue where the rebate is 0.01 bps and the observed markout is −2 bps, reaching break-even requires roughly a 200× rebate increase. That is not a tuning problem.

What this means operationally

For an operator with retail-scale capital, the space of viable strategies on modern perp DEXs is narrow. The following things don't work:

• Passive retail market-making on any major-pair perp DEX without a designated program.
• Symbol-shopping within a tick-floored venue.
• Tuning alphas to overcome venue-structural markout on tight books.

The following things can work:

• Cross-venue latency arbitrage, where capital requirement per trade is small, the leg on the faster venue compensates for retail fee tiers on the slower venue, and the economic driver is location-dependent signal lead rather than spread capture.
• Strategies on venues where the tick floor is far enough from the typical spread that retail-fee economics close before adverse selection. These venues are rare but they exist — Bluefin, as of our April 2026 measurement, is one.
• LP strategies where the venue's design deliberately pays the small participant (vault products, LP tokens, incentive farms). A different role in the three-way design — LP, not API maker.

None of these are “just run a maker bot harder.” The correction isn't in the strategy code; it's in the role the capital is playing in the venue's design.

Closing

Treat the venue's economic design as a first-class input to strategy selection. Fee schedules and flow-routing rules are not boilerplate — on modern perp DEXs they are the game. Reading them first, before writing any quoting logic, saves the expense of learning the same lesson in live P&L.