Whoa! Right off the bat: liquidity is everything. Seriously? Yes. If your execution costs, tail risk, and inventory bleeding don’t line up with your edge, you’ve got no edge. My instinct said that most platforms pitch shiny APYs, but underneath those numbers the plumbing is what kills P&L. Initially I thought picking the deepest book was enough, but then I realized that depth alone is only part of the story—latency, funding mechanics, fee rebates, and settlement cadence matter just as much.

Here’s the thing. Perpetual futures changed the game by letting traders lever directional and relative-value positions without rolling futures. That convenience comes with hidden levers: funding rates, oracle cadence, and liquidations that interact with liquidity provision in weird ways. On one hand you can capture fees from providing two-sided quotes. On the other hand—though actually—your unrealized inventory exposure to funding swings can swamp nominal fee income.

Okay, so check this out—I’ll be honest: I’ve run market-making stacks and I’ve been burned by funding-rate arbitrage gone wrong. Hmm… somethin’ about overconfidence in models bugs me. You can design a strategy that looks profitable on paper but collapses under concentrated liquidations or sudden volatility. That part bugs me. And not just me; I’ve seen desks on both coasts misread risk.

Short primer: liquidity provision means you’re committing capital to facilitate trades. Market making is the active, often automated, process of posting bids and asks to capture spread, rebates, and order flow. Perpetuals layer in funding payments that transfer between long and short holders to peg the contract to spot. These pieces interact. They rattle each other. They sometimes align beautifully. And sometimes they don’t.

Trade-offs matter. Low fees attract volume, but low fees also attract noise traders who flip positions in milliseconds—excellent if you have tick capture, not so great if you run a high-latency hedge. Depth interests hedge funds. Tight spreads attract arbitrage desks. Volume attracts slippage—very very important to model this correctly.

Practical anatomy: what pro traders should measure

Really? You want metrics? Fine. Monitor five core things: realised spread capture, adverse selection rate, inventory decay rate, effective funding carry, and tail-loss exposure. Track them daily. Weekly is too slow. Monthly is too complacent. These metrics tell you whether your strategy is actually harvesting microstructure profits or just subsidizing better tech.

Measurement needs nuance. Adverse selection is not just “you get picked off.” It’s the rate and size of picks during regime shifts. Inventory decay is how fast your positions mean-revert when market moves. Effective funding carry combines funding rate history with your typical directional bias. If you are long during persistent positive funding, your carry can be a tax. Hmm…

Initially I assumed funding was symmetric. Actually, wait—let me rephrase that: funding shows structural flow. Perp funding tends to reflect leverage demand on one side. During rallies, longs often pay; during dumps, shorts pay. So if your MM strategy accumulates inventory in one direction, your funding exposure compounds P&L in non-obvious ways.

Strategy patterns that hold up

Short burst: Whoa—diversify across venues. Medium: Spread capture plus hedged directional exposure works when your hedge latency is low. Longer thought: A robust approach is to run two complementary strategies—one passive, low-touch quoting to earn spreads with strict inventory caps, and one agile, event-driven hedger that adjusts size and delta as funding and volatility signals change, because that splits the frictional risk from the tail-event risk and reduces correlated drawdowns.

Rules of thumb that I trust: keep inventory skew limits tight in thin regimes, widen spreads during elevated implied vol, and always calibrate quote size to local depth (not to your overall capital). On many DEXs, posting too large a quote pushes you into the “maker of last resort” role—bad. Also, simulate liquidation cascades and funding shocks, not just normal spreads.

Tools matter. Use order-book replay, Monte Carlo with clustered volatility, and stress tests tied to funding-rate spikes. If your stack can’t simulate a 20% one-day move and the resulting funding flip, you lack a necessary safety net. Most teams skip that. I’m biased, but that’s a killer oversight.

Perpetual specifics: funding, oracles, and settlement nuances

Funding rates are a tax or subsidy. They’re also a signal. Medium: watch 1H and 8H funding patterns and volume behind them. Longer: if oracle updates are sparse or susceptible to manipulation, funding can swing violently when a patch of liquidity dries up or when a MEV burdened block hits the chain. On-chain perps add latency variables that CEXs don’t have—so your hedges need to be adjusted for that.

One common trap: assuming funding will revert fast. Historically, funding can stay one-sided for days if leverage builds. Another trap: assuming oracle lag is trivial. Not so. Oracles that publish every N seconds or upon new block headers can introduce jumps that coincide with large trades—timing matters.

Also, fee models differ wildly. Some venues pay maker rebates; others charge taker fees only. On some DEX perpetuals, liquidity providers can collect protocol-level incentives that mask underlying slippage—dangerous if those incentives dry up. I’ve seen APYs collapse when incentives ended. That was unpleasant… and a good lesson.

Execution mechanics: latency, sizing, and TCA

Quick sentence: latency kills market makers. Medium: quantify round-trip latency to top liquidity, and measure update-to-execute time for your hedges. Longer: if your quote update loop is 200ms and the venue matches at 50ms, then your quote is stale more often than you think—leading to systematic adverse selection that eats spreads and inventory buffers.

Order sizing should respect local depth. A naive rule: don’t exceed 20–30% of immediate visible depth for aggressive two-sided quoting. But this varies with tick size and typical order-sweep behavior. Use adaptive sizing that scales down when time-weighted volume rises suddenly.

And don’t forget TCA (transaction cost analysis). For perpetuals, include funding and funding volatility in your TCA, not just spread and slippage. TCA that ignores funding is incomplete. Hmm… I’m not 100% sure about the exact thresholds for every market, but the principle stands.

Capital efficiency and leverage

Perps offer leverage; liquidity provision consumes capital. These interact. If you pledge capital as collateral to provide liquidity, your usable margin shrinks. Medium: optimize collateral allocation across spot, perp, and isolated margin to maximize return per unit of capital at acceptable risk. Longer thought: some protocols offer concentrated liquidity features or tokenized LP positions that free up capital; consider them if they reduce opportunity cost without introducing counterparty exposures you can’t quantify.

Leverage amplifies both funding benefit and funding pain. If you tailor sizing to leverage-adjusted drawdown tolerances you avoid surprise blowouts. Many desks only stress test at nominal leverage. Don’t be that desk.

DEX vs CEX: different beasts

Short: DEXs have composability and atomicity advantages. Medium: on-chain execution gives you transparency and settlement guarantees, but brings MEV, gas dynamics, and oracle quirks. Longer: CEXs provide speed and consolidated liquidity but centralize counterparty risk and custody—so your operational playbook needs to reflect trade-offs. Use both if you can, but route intelligently.

For DEX-focused market-making, check out platforms that prioritize deep order books and low fees. If you want a place to evaluate, I’ve spent time with a few emerging venues—one being hyperliquid official site—and they highlight how architecture choices change the calculus for makers.

Final thought: market making and perpetuals reward people who think in systems, not single trades. The edges are small, but they compound. Stay humble. Test with real-world, adversarial scenarios. Keep the plumbing tight. And remember—some models look good in backtest because they ignore the human and technical failures that show up when markets panic… somethin’ to chew on.