Blog
Blog
Price momentum worked in crypto until the moment it didn't, and the desks that found out late are still explaining the drawdowns to their LPs.
The edge eroded faster than anyone modelled. Strategies that printed for eighteen months went flat inside a quarter, then started bleeding. The survivors are not the ones who ran better price models. They are the ones who switched data sources.
What they switched to is the same thing equities quants spent a decade building infrastructure for. Crypto is compressing that timeline into roughly two years.
Most institutional desks that entered crypto between 2020 and 2023 came in through the same door: momentum. Price breaks a level, volume confirms, position goes on. It worked because the market was thin, directional, and populated mostly by retail participants who responded predictably to price levels.
The assumption was that crypto price data was sufficient because crypto was a price-driven market.
That assumption was wrong in equities by the mid-2000s. It is wrong in crypto now.
Price tells you what happened. Order flow tells you who did it and how much conviction they had. A $400 million Ethereum position being built over four hours looks almost invisible in the price chart if it is being worked carefully across exchanges. It is not invisible in the order flow. The aggressor side of every fill leaves a trace: order flow imbalance accumulates, bid-ask depth shifts asymmetrically, the cost of aggressive buying starts rising before price does.
Quants who built around price data kept fitting better and better models to a signal that was already deteriorating. The signal was not the price. The signal was the order flow underneath it.
In traditional finance, the transition from price-based to microstructure-based strategies happened in stages. The theoretical groundwork, market impact models, information asymmetry in spreads, and the relationship between order flow and short-term price discovery, was in academic literature by the late 1990s. Implementation at scale came later, as exchange data became structured and accessible enough to build on.
By 2014, researchers like Cont, Kukanov and Stoikov (Journal of Financial Econometrics, 12(1), 47-88) had formalised the measurement of order flow imbalance as a predictor of short-term price changes, demonstrating that the imbalance between buyer-initiated and seller-initiated volume at the top of the book carries predictable information about near-term direction. Not as a trading rule. As a measurable structural feature of how prices actually move.
This work did not surprise equities practitioners. They had already built it into their systems empirically. The paper confirmed what execution desks already knew.
Crypto quants are arriving at the same destination, under pressure, in a fraction of the time.
Equities had natural friction that slowed the convergence of strategies toward microstructure. Fragmentation across venues was there, but it was regulated and relatively stable. Tick sizes were standardised. Market maker obligations existed.
Crypto has none of those guardrails. A single large asset trades simultaneously across dozens of venues with different fee structures, liquidity profiles, and matching engine behaviours. Arbitrage is aggressive and automated. The moment a price-level strategy becomes crowded, the market structure punishes it immediately, not over quarters, over days.
Consider an illustrative scenario, recognisable to anyone who has run a live book in crypto. A systematic desk enters a long on a clean breakout above a resistance level that has held for three weeks. The signal is textbook. Within 45 seconds the position is $800 down as the price tests back through the level, triggers stop clusters from other breakout players, and then immediately reverses and continues higher. The breakout was real. The timing was wrong. Price gave no indication that a stop-sweep was coming. Order flow, specifically the absence of genuine aggressive buying in the book before the break, gave every indication.
This is not bad luck. It is what happens when strategy density at a given signal type exceeds market capacity to absorb the aggregate position. Price signals converge. Microstructure signals are harder to replicate at scale because they depend on execution context, not observation alone. That crowding dynamic is the same one described in the alpha decay problem: the signal that everyone can cheaply observe usually compresses first.
The desks that are moving away from price feeds are not abandoning quantitative approaches. They are upgrading the input layer.
The core shift is from observing prices to observing the mechanics that produce prices. Order flow imbalance (OFI) measures the net aggression differential between buyers and sellers over a given window, specifically whether the buying side is hitting the ask harder than the selling side is hitting the bid, and by how much. VPIN (Volume-Synchronized Probability of Informed Trading) estimates, on a rolling basis, what fraction of recent volume is likely to be informed, meaning it is moving against uninformed liquidity rather than providing it. These are not price derivatives. They are structural measurements of what participants are actually doing. That shift becomes easier to understand once you put it beside what market depth actually measures and VPIN explained, because both show why the chart alone leaves out the part desks now care about most.
Whale markers, large single-order prints that cross the spread or consume multiple levels of depth, are separately meaningful because they indicate direction and commitment both. A market order that sweeps three levels of the ask book is a different signal than twenty small aggressive orders that add up to the same volume. The structural signature differs. The information content differs.
Combined with cumulative volume delta (CVD), which tracks the net difference between buyer-initiated and seller-initiated volume across a session, these inputs give a picture of market participation that price alone cannot provide.
The objection that comes up here: microstructure signals are noisy and short-horizon. True. They are not meant to replace fundamental or medium-term positioning frameworks. They are meant to improve entry timing, detect regime changes before price confirms them, and avoid stop-sweep environments that look identical to genuine breakouts in the price chart.
Equities quants had access to structured, normalised order book data through exchange feeds and consolidators for years before they built systematic strategies on top of it. The data was there. The infrastructure evolved to use it.
Crypto had the opposite problem for a long time. The raw data existed, exchanges publish L2 books and trade tapes, but normalisation across venues was a significant engineering problem. Every exchange has a different API structure, different update cadence, different handling of partial fills and cancelled orders. Aggregating a coherent microstructure picture across Binance, Bybit, OKX and other major venues simultaneously, in real time, with the latency tolerance that live trading requires, was not trivial.
That infrastructure problem is largely solved now. Platforms that specialise in real-time microstructure data, normalising order book depth, OFI, VPIN, CVD, and whale markers across multiple exchanges into a single structured feed, have removed the engineering barrier that kept many desks on price data longer than they should have stayed.
The shift is now less about whether to use microstructure data and more about how quickly you can integrate it.
Institutional desks rebuilding around microstructure data changes the environment for everyone trading the same assets.
Price-level strategies face a more hostile environment as the aggregate participation at those signals increases and microstructure-aware participants exploit the predictable stop clusters they create. This is not a new dynamic in equities. It is arriving in crypto on an accelerated schedule.
The desks that transition early gain two things. First, they operate on a less crowded signal set during the transition period, because microstructure signals are still underused in crypto relative to equities. Second, they develop the internal capability to read market structure diagnostics that price-only frameworks cannot generate, which compounds in value as the market matures.
The desks that stay on price data face a choice that does not improve with time: keep fitting better models to a deteriorating signal, or rebuild the input layer under time pressure.
In equities, that choice was spread across a decade of gradual pressure. Crypto is giving participants roughly two years before the same dynamics that made microstructure data standard in equities make it standard here.
Because price-only signals crowd quickly and miss the execution context that often determines whether a setup survives live trading.
They show aggression, liquidity quality, informed-flow context, and how much stress the orderbook can absorb before price fully reacts.
No. It upgrades the input layer and timing context around strategies that were previously relying too heavily on visible price alone.
Because crypto is more fragmented, cheaper to access programmatically, and less protected by structural friction that once slowed crowding elsewhere.
If the data layer stays shallow, the team ends up competing on the same crowded chart signals that larger desks are already moving beyond.