Buy pressure

A flow, with units

Write buy pressure as dollars per period and it becomes usable. Sum every recurring source of demand over a month: tokens bought because the protocol requires them, tokens bought and removed by a revenue-funded buyback, tokens taken out of circulation by staking or lockups, and net purchases by new holders. Then sum every recurring source of supply over the same month: allocations unlocking, emissions being sold by whoever received them, fee recipients converting to cover costs, and any inventory lent to a market maker that is being quoted with.

The difference between the two is the number worth arguing about. It is not a forecast of anything. It is a statement about how many dollars of each side are expected to reach the market in a given month, and it is only as good as the assumptions underneath each line.

The reason to hold the units firmly is that buy pressure is otherwise a rhetorical device. Almost any mechanism can be described as creating buy pressure. Very few can be described as creating a specific number of dollars of it per month, and that is the version that changes a design decision.

How a flow becomes a price

On a pooled venue the mechanism is not mysterious. Price is the ratio of the reserves, and a net imbalance of buying over selling shifts that ratio in proportion to the fraction of the reserves it consumes.¹ That is why the same monthly demand produces a very different chart in a shallow pool and a deep one, and why depth belongs in any buy-pressure model rather than being treated as a separate topic.

It also means the flow and the venue interact. A protocol that generates $200,000 a month of genuine token purchases into a pool with $200,000 of quote-side reserve is generating a materially different price effect than the same flow into a pool ten times that size. The mechanism designer controls both halves.

What the research supports, and what it does not

There is a real empirical literature on order flow imbalance in crypto markets, and it is narrower than the way the term gets used. Work on explainable patterns in cryptocurrency microstructure finds that order flow imbalance, bid-ask spread and depth account for a meaningful share of very short-horizon return variation, with imbalance having a largely monotone positive effect on returns and concavity at the extremes, meaning each additional unit of imbalance moves price less than the last.² The same work documents short-lived pressure and reversion toward prior levels.

A separate 2026 study builds alpha signals from spread, Kyle's lambda, the Amihud ratio, order flow toxicity and order flow imbalance, and reports that imbalance and toxicity measures survive formal stability selection over a six-month window at minute frequency.³ Both findings are worth having. Neither is a claim about tokenomics.

Here is the boundary, stated plainly. These results establish that short-horizon buying imbalance is measurable and partially predictive of returns over minutes, and that much of it reverts. They do not establish that a buyback programme, a burn, or a staking lockup produces a durable price effect over months. We looked for a verified, dated, named example of a token mechanism whose price effect was documented against primary sources, and did not find one we would put our name to. Anyone presenting that link as settled is going further than the evidence goes.

The measurement problem nobody solves in a slide

Reported volume is not net flow. It counts both sides of every trade and it includes activity created for the purpose of being counted, a practice charged as wash trading for hire in an October 2024 federal action against four firms operating as market makers.⁴ Any buy-pressure figure derived from a volume feed inherits whatever is in that feed.

Onchain you can do better, because net token flow into and out of a specific pool is computable per period. What you cannot do from that data alone is attribute the flow to a cause. Knowing that $400,000 net entered the pool last month is a fact. Saying it entered because of the staking programme is an interpretation, and it needs a separate argument.

Modelling it without pretending

Build the supply column first, because it is the one with dates on it. Every unlock cliff, emission tranche and market-maker loan has a schedule in a document somebody signed. Convert each to dollars at the current price, note that the conversion moves with the price, and total by month.

Then build the demand column with explicit assumptions and a low case. Protocol-usage demand should be derived from usage you already observe rather than from a target. Speculative demand belongs in the model, because it is real, and belongs in a separate line, because it is unreliable and does not scale with the business.

Where the supply column exceeds the demand column month after month, the design has a structural gap and only two honest responses. Grow the business generating the demand, or change the schedule generating the supply. There is no third option that consists of communication, and in our view the projects that discover this late are the ones that treated the token as the product rather than as infrastructure attached to one.

What we will not claim

We will not tell you a mechanism will move a price, or by how much. Nothing in the accessible literature supports that at the horizons a founder cares about, and the firm does not publish forward statements about token value.

What the model is actually for is narrower and more useful. It tells you the size of the demand a design is relying on, in dollars, so you can ask whether the business could plausibly produce it. That question is answerable. It is also the question that gets skipped, because the answer is often no.