What Institutional-Grade Crypto Trading Actually Means

HFT Market Making in Crypto Markets

Market making in crypto, much like in traditional finance, is basically described as “post bids and offers, collect the spread, manage inventory.” At a strategy level, that is true. There isn’t a huge difference between the theoretical mechanics of a traditional or crypto market making strategy. But this high level thinking hides the two things that actually decide whether the strategy works in production:

1. Microstructures (what information you’re reacting to, and how fast)

2. Economics (what you earn or pay per fill once the fee schedule hits you)

In this blog, we’ll be looking at how Blockhouse deals with these two concerns.

The Core Strategic Theory

CrossX is Blockhouse’s cross-exchange market making system built around a simple observation: order books on different venues do not update simultaneously. Those small timing differences create short-lived dislocations, which are tiny windows where one venue is “behind” another in terms of pricing. If you can observe that state reliably and update quotes fast enough, you can turn those moments into a repeatable source of edge.

The core thesis is an information-asymmetry one:

Crypto exchanges are fundamentally fragmented due to the sheer number of them and the difficulty of transferring information reliably from exchange to exchange

As such, different exchanges incorporate information at different times.

That lag shows up as measurable imbalance in the book and in trade flow.

We quote on one venue informed by what’s happening on another continuously to harvest this information latency

CrossX runs an always-on microstructure loop. We ingest a live stream per venue, i.e. ticks, trades, book updates, and compute signals that are useful at market-making horizons such as:

• Rolling realized volatility

• Trade arrival rate

• Cumulative volume delta (CVD)

• Order book imbalance

These are computed per exchange and consolidated into a unified cross-venue state. That state feeds directly into quoting decisions: spread and skew are updated every cycle, not as a periodic “model refresh,” but as part of the tick loop. These signals function as Blockhouse’s view of the underlying microstructures and quantifies some of the latent information embedded into what is actually observable. With this information, our market making system is able to autonomously adjust and place profitable trades.

If trade intensity spikes on Venue A, or imbalance leans one-way, our quotes on Venue B widen, skew, or pull, depending on the regime and current inventory. The goal isn’t to predict direction, it’s to avoid being the stale quote when the market starts moving.

Market Making Mechanics

So how does this all work in practice? What are the main vectors of concern for the practical implementation?

Everything in CrossX reduces to three main dials:

1) Provide Liquidity and Capture Spread

This is the core profit vector.

We quote two-sided markets. If we’re doing our job correctly, we get filled at prices that are slightly favorable relative to the evolving mid, and we net the spread over many small trades.

The better we do this, the more money we make.

2) Control Adverse Selection

Additionally, spreads aren’t collected in a vacuum. The moment we start quoting, we’re part of the market: our orders change the visible book, interact with other participants’ models, and can nudge short-term price dynamics. A practical way to think about it is: if we show meaningful size on the bid, does that information get interpreted elsewhere as a signal, pulling in additional buyers and lifting the market away from us? And if we do get filled, did we just buy right before the market repriced, i.e., would the next few ticks have looked better if we hadn’t been the one providing liquidity?

You only capture spread reliably when fills are not systematically followed by unfavorable moves. Blockhouse manages that by conditioning our quotes on fast microstructure features: order book imbalance, trade intensity, open interest etc. and by keeping a tight cancel/replace loop so our orders stay aligned with the state we’re observing.

3) Manage Inventory

Inventory adds the third vector to that market making function. Even if you optimize for spread capture and control post-fill markouts on average, you still need to ensure fills don’t accumulate too exposure in one direction. Put simply, we can’t just keep buying because the spreads and adverse selection look better on the buy side. That defeats the purpose of being directionless as we would inadvertently enter a long position.

In CrossX, we track inventory per venue and in aggregate, with explicit targets and limits at both levels. Quoting incorporates inventory through an inventory-dependent skew (and, when needed, size throttles): if inventory is long relative to target, we shift the bid away and the ask closer to bias fills toward selling; if inventory is short, we do the opposite. The exact skew mapping is regime-dependent, primarily a function of short-horizon volatility, spread regime, and expected impact, but the structure is consistent: inventory error feeds directly into quote offsets and sizing so exposure reverts toward target without introducing unnecessary directional bias.

But so far, these 3 vectors are things that all profitable market making systems must consider. So what’s special about the crypto space? And that’s where the microstructures and economics really come into play.

Competitive Advantage #1: Rebates

A lot of backtests implicitly assume something like “spread capture minus fees.” In crypto, that’s often the wrong simplification.

For market making, fee tiers and maker rebates can dominate the unit economics. When your edge is thin, moving from “small rebate” to “larger rebate” isn’t incremental, it can be the difference between:

• Positive expectancy per fill vs. negative

• Being able to quote tight vs. needing to sit wider

• Scaling volume safely vs. strict capacity limitations

CrossX’s execution layer is built to optimize for maker economics across venues:

• We treat fee schedules / tiers / rebates as inputs to quoting and routing decisions

• We engineer for maker intent, because accidental taker behavior is the fastest way to erase edge.

• We evaluate expected value per fill as:

  spread + rebate – expected adverse selection – expected inventory cost

If you only optimize the spread term, that’s ignoring a significant second vector of positive Pnl on the table. And this is one of the biggest differences between traditional and crypto market making, market makers are significantly more rewarded for participating in the many exchanges of the digital space.