How a bid is formed

What a bid is made of

A quote from this desk is not a price arrived at and then defended. It is a reference notional — the size of the position multiplied by the price of the asset that position is denominated in — with five things subtracted from it. The counterparty supplies the reference. The desk supplies the subtractions, and published them before it was asked.

The five terms, in the order applied: venue, which asks whether a market for the asset exists at all; out of hours, which asks whether the market the reference came from is open at this moment; size; curve, which asks how much depth is left underneath the position; and the desk’s own charge. Four price risk. One is revenue, and it is the smallest line in the table.

The schedule

A dash is not a zero. A zero means the term applies and costs nothing; a dash means the schedule declines to price that case at all, and what comes back is a reason rather than a bid.

Term one — venue

Of the 240,350 live curves at block 55,223,440, 125,256 hold a pair against ether on the canonical factory. For those the schedule returns nothing at all: there is a market, the automated one is a better price than a warehouse can be, and the desk points at it. That is the Ether-quoted row, and its blank is deliberate.

The gap between the two remaining rows — 145 basis points against 420 — is the business written as a difference. A position the desk can offset against a listed instrument is warehoused until the next open, and the next open has a date on it. A position with no pair, no borrow and no offset is warehoused until an exit exists, and nothing in this repository can say when that is. The extra 275 basis points is the price of not having a date.

Term two — out of hours

The curve trades every hour of the week. The equity it prices against trades 32.5 of 168, so for 80.65% of the week the reference is a price nobody can transact at. The desk charges for the crossing rather than pretending the reference is live.

The term is z × σ, and both are stated: σ = 2.4% is the desk’s assumption for a single-name overnight move on a retail-heavy US equity, and z = 0.84 places the quote at roughly the eightieth percentile of that distribution. Multiplied out, 0.024 × 0.84 = 0.02016, which rounds to 202 basis points — the only rounding anywhere in the waterfall.

Both numbers are arguable, which is the point of printing them. The term is linear in each, so a reader who believes overnight σ for these names is twice the desk’s assumption believes the term should be twice as wide — a disagreement one multiplication long, rather than a conversation about whether the desk is fair.

Term three — size

Bands rather than a smooth function, and the reason is not modelling: a continuous curve cannot be quoted over the phone and cannot be verified on paper. Below $2,500 of reference notional the term is zero — not generosity, but the observation that a ticket that size does not move the desk’s book. Above $250,000 the schedule ends, with no extrapolated fifth band, because the desk would not honour a number it had extrapolated.

Term four — curve

Every Pons curve opens with a virtual quote reserve of 1.68 and graduates at 4.2 raised, by which point 71.43% of the 1,000,000,000 supply has been sold. Remaining depth is what the desk would be selling back into if an exit ever appeared, and it thins as the raise approaches that threshold. Past it the curve stops accepting sells and the position becomes a V3 liquidity problem — a different desk, and this schedule says so by refusing rather than by quoting wide.

Term five — the desk

25 basis points, flat, on every ticket the schedule will quote. It does not scale with size and it does not scale with tier. A desk whose own charge is the largest number on its schedule is not warehousing risk, it is selling access.

Why the terms add

The net bid is one line of arithmetic:

gross = reference × size
net   = gross × (1 − Σ bp × 0.0001)

The terms add. They are not compounded down the waterfall, each applied to what the one above it left behind. Compounding is marginally the more defensible construction, and at the headline case it produces 669.8 basis points against the additive 682 — so the additive schedule is the wider of the two by 12.2 basis points, in the desk’s favour, and the desk keeps it anyway.

The reason is not that the difference is small. It is that a schedule which cannot be checked by hand is not published, it is announced. Five whole numbers that sum to a sixth can be verified by a counterparty with a phone and no trust in the desk at all; a product of five factors cannot, and the moment verification requires the desk’s own calculator, the publishing was decorative.

A worked example

The waterfall below was not typed. It is the return value of quote() called at build time on a 1,625-unit position in the stranded tier, out of hours, on a curve at 50% of its graduation raise. The size is the reserve the deepest curve on the chain reports — 1625 AMC.

The ether line converts at the rate recorded with the survey at block 55,223,440, stated so the conversion can be redone rather than trusted. It will be stale by the time you read it, which is why the schedule is denominated in basis points and not in ether.

What the schedule refuses

Four inputs make the same engine return a reason instead of a number. These are its own words, not a description of them:

• Ether-quoted positions have a venue. Use it — the desk is a worse price by construction.
• Below the 250 USD minimum ticket: settlement gas would be a material share of the trade.
• Above the 250,000 USD schedule ceiling — quoted by a human, not by this page.
• The curve has graduated. The position is a V3 LP problem, not a curve problem.

A refusal is more useful than a wide quote. A wide quote is a desk hoping you will take it; a refusal is a desk telling you where its book ends.