Physical crude has no consolidated tape. The price of a grade that did not trade today is inferred - from a related instrument that did, from a quote nobody had to honour, or from a relationship that held the last time anyone checked. That is a workable way to produce a number. The problem is that the line does not say which one it was.
Summary
Open a position report and each line carries a price. What the line does not say is where that price came from — whether somebody traded at it, quoted it, published it, or calculated it from something else. Physical crude has no consolidated tape: deals are negotiated bilaterally and stay private unless a party chooses otherwise, grades and delivery months routinely go through a day without a single transaction, and the numbers filling those gaps are inferred — from something related that did trade, from a quote somebody was willing to show, or from a relationship that held the last time anyone checked. A price arrived at this way is not wrong. It is evidence of a particular strength, and the strength is not recorded next to it.
This article sets out how a price for physical crude actually comes to exist: why the visible part of the market is a self-selected sample rather than a representative one, why discovery happens in moments rather than continuously, what it means that a derived price keeps printing after the relationship behind it has stopped holding, and what has to sit alongside a number before anyone can say how much weight it will carry. Whether a mark is a price you could exit at is a separate question and a separate cost; this is about where the number came from in the first place. It is written for the people who have to defend the numbers — traders forming a view in a market with few visible prints, risk and middle office who mark and report the book, and anyone who has been asked where a figure came from and found that the answer was not stored anywhere.
If you want the mechanism, start with There is no tape, so the price is an inference. If you already work with thin markets and want the part that fails silently, start with A derived price does not stop printing when it stops being true.
There is no tape, so the price is an inference
In physical crude there is no consolidated record of transactions. What is called the price is an inference drawn from the part of the market that was visible, and visibility is not the same thing as representativeness.
An equity market has a tape because trades happen on a venue that records them. Physical crude does not work that way. A cargo is negotiated between two parties, often through a broker, and the terms are private unless somebody chooses to make them public. There is no venue that saw it, no rule that it be reported, and no register that would let an outsider count what happened that day.
What exists instead are windows and reports. Some participants trade in a public process where bids, offers and deals are visible to everyone watching. Some tell a price reporting agency what they did. Some tell a broker, who tells a market. Exchange contracts, which are recorded and cleared, sit alongside all of this and are visible in full — but they are a different instrument from the cargo, and being visible is not the same as being the thing you hold.
So the price of a physical grade on any given day is built from a fraction of the transactions that occurred, plus quotes that never became transactions at all. That is not a defect in the process; it is the process. The people who assemble these numbers are explicit about it, which is why a methodology screens, weights and disregards rather than simply averaging what it was told. The defect only appears downstream, at the point where the output of that process is copied into a system and treated as though it were a reading from an instrument.
The visible market is the part that had a reason to be seen
A participant chooses whether to be visible, and that choice is itself a trading decision. The observable sample is therefore selected rather than random, and it is selected by people with positions.
Consider what showing a price costs and buys. A seller who puts a cargo into a public window gains a genuine chance of a better bid and gives up the knowledge that they are a seller. A buyer who reports a deal contributes to a benchmark that may price their other contracts. Someone who is quietly accumulating has an obvious reason to stay out of the light, and someone whose book benefits from a particular level has an obvious reason to be in it.
None of that has to be improper to matter. Ordinary commercial motives are enough to make the visible sample unrepresentative in ways that shift from week to week. The sample that produces the price is not a random sample of the market; it is the part of the market that had a reason to be seen that day — and the reasons change.
This has a practical consequence that gets missed because it sounds theoretical. When a market thins, the number of participants whose motives are shaping the visible sample gets smaller, and the influence of each one grows. The published number keeps arriving on schedule and keeps looking like the numbers that came before it. Nothing about its appearance tells you that the population behind it changed.
Discovery happens in moments, not continuously
A price line updates all day. Actual discovery is concentrated into short periods, and outside them there are prices but very little new information.
In a crude market there is typically a moment when participants who intend to transact do so, and where quotes are firm because someone is prepared to be hit. Around that moment, the market genuinely learns something. Away from it — early, late, on a holiday one side observes and the other does not, in a delivery month too far out for anyone to have a reason to trade — the screen still shows a number, and that number is largely yesterday's information wearing today's timestamp.
Two things follow, and both of them are ordinary rather than exotic.
The time a price was formed matters more than the time it was recorded, and systems record the second one. A value collected at the end of a session and stored with a collection timestamp looks equally fresh regardless of whether it came out of an active window or out of a quiet afternoon in which nothing happened. The distinction that matters to anyone relying on it has been discarded at the point of capture, and it cannot be recovered later.
A quiet market and a stable market produce the same picture. A number that does not move because the market has settled and a number that does not move because nobody looked at it are indistinguishable on a chart. They imply completely different things about the risk of the position they are marking.
A position report is a list of prices, and only some of them were traded
Each grade, location and delivery month is a separate instrument. The market does not print a price for each of them every day, so the ones that did not trade are placed relative to the ones that did.
This is arithmetic rather than a criticism. A book can hold positions across several grades, in several locations, across a run of delivery months, and each combination is its own instrument. No market prints a fresh price for that many things daily, and none tries. What happens instead is that a small set of liquid reference points are observed, and the rest are placed relative to them — by a spread, a differential, a curve shape, a ratio, or a rule of thumb somebody encoded years ago and nobody has revisited.
Every one of those relationships is an assumption, and assumptions are load-bearing in exactly the way that is easy to forget: nothing on the line says it is there. The line shows a grade, a month, a quantity and a price. It does not show that the price was produced by taking a liquid marker and applying a relationship that was last examined in a different market regime.
The proportion of a book that is observed rather than derived is not a mystery. It is a number a firm's own systems could produce, if anyone asked them to record the distinction in the first place. Asking for that proportion is the single most informative question in this article, because the answer tells you how much of your reported risk is a measurement and how much of it is a model whose author may have left.
A derived price does not stop printing when it stops being true
When the relationship behind a derived price breaks, the price does not disappear or flag an error. It carries on arriving, plausible and on time, and it is simply no longer connected to anything.
This is the failure mode that makes the previous section worth caring about. A missing price announces itself: a cell is blank, a report fails, somebody investigates within the hour. A wrong-but-derived price announces nothing. It sits in the same column as every other price, formatted identically, and the only evidence that anything has changed is that a position which used to behave predictably has started producing results nobody can explain.
The breaks themselves are ordinary. A grade that used to move with a marker stops doing so because refining economics in its region changed. A spread that was stable becomes seasonal, or stops being seasonal. A relationship calibrated in a period of ample supply is applied through a period when the constraint is somewhere else entirely. A curve that could be interpolated across neighbouring months is asked to reach into a month where nobody is trading for reasons that have nothing to do with price.
The general shape is worth stating in one line, because it applies well beyond this subject: a number that is calculated rather than observed inherits the health of its inputs and none of their alarms. Whether the relationship still holds is a question that has to be asked deliberately, on a schedule, by someone whose job it is — because the number itself will never raise its hand.
In a thin market, asking is participating
Discovering the price of something that rarely trades means showing a price. What you show is information you have given away, and it moves the thing you were trying to measure.
In a liquid instrument this hardly arises: the market is there, and looking at it costs nothing. In a grade that trades occasionally, between a handful of counterparties who all know each other, price discovery is an active process. You call brokers, you show an indication, you find out who has an axe. Every one of those actions tells somebody something about you — that you are a buyer, that you have a position to move, that you have a deadline.
Two consequences deserve to be in the open rather than in individual heads.
The prices you gather this way are not independent of each other. Three brokers consulted in the same hour may be reflecting the same underlying interest back at you, and the agreement between their answers reads as confirmation when it may only be an echo. That agreement feels like evidence, and it is the kind of evidence that is strongest exactly when it is least reliable.
And an indication given while discovering a price can outlive the enquiry. It gets remembered, repeated and occasionally reported. What was intended as a question can end up as a data point in someone else's picture of the market — including, eventually, in the picture that comes back to you.
Where this leaves the middle and the back office
None of this is unusual and none of it is anyone's fault. It is what price formation looks like in a market where transactions are private by default and a great many instruments trade only occasionally. What makes it expensive is that the strength of each number is known at the moment it is captured, and does not travel with it.
The trader knows whether today's level came out of a real window or out of a quiet afternoon. The person who built the spread knows which relationship it rests on and when they last examined it. The analyst who gathered three broker quotes knows they may all trace back to one enquiry. All of that is in the room while the number is being formed, and none of it is in the field where the number is written down. What arrives at the risk report, the valuation and eventually the auditor is a bare figure with a timestamp.
The consequence is a specific and familiar kind of afternoon: somebody asks where a number came from, and the honest answer requires finding the person who produced it and hoping they remember. A number without its provenance is not a weaker version of a number with it — it is a different object, because you cannot tell how much weight it will carry. The cost is not paid on the day the price is captured. It is paid on the day the price is challenged.
Six ways a number gets onto a price line
| What it is evidence of | What it is silent about | How it goes stale | |
|---|---|---|---|
| A completed transaction | That two parties agreed this level for a specific parcel | Whether the parcel resembles yours, and whether either side was under pressure | Immediately, if the market is moving; slowly, if it is not |
| A firm quote somebody would have honoured | That a level was available to be hit | What size it was available in, and for how long | Within the session, usually within minutes |
| An indicative quote from a broker | That somebody's impression of the level is this | Whether it reflects an actual buyer or an echo of your own enquiry | Ambiguously, because it was never firm to begin with |
| A published assessment | That an administrator applied a methodology to what it could see | Which transactions were screened out, and how thin the underlying evidence was | On the schedule of its publication, not the schedule of the market |
| A number derived from a relationship | That a rule was applied to a liquid reference | That a rule was applied at all, and when it was last examined | Silently, and without changing appearance |
| A value carried forward from an earlier session | That nothing has replaced it | Whether the market moved in the meantime | Continuously, while looking unchanged |
The bottom two rows are the ones to sit with. They are the two that never look stale, and they are the two that an ordinary report does not distinguish from the first row.
What has to sit next to a price before anyone can rely on it
These are fields on a stored record rather than terms in a contract. Each one is known at the moment of capture and irrecoverable afterwards.
| Field | Why it is load-bearing |
|---|---|
| Exactly which instrument the number belongs to — grade, location, delivery period, basis | A price attached to an ambiguous instrument cannot be checked against anything |
| Which of the six sources above it came from | This is the whole of the difference between a measurement and an inference |
| The moment it was formed, not the moment it was collected | These diverge most in exactly the quiet periods where the divergence matters |
| Whether it is observed or derived, as a stored flag rather than a convention | A convention lives in the head of whoever set it up; a flag can be counted |
| If derived: from what reference, and by what relationship | Without this, the number cannot be re-derived, questioned or repaired |
| If derived: when that relationship was last tested against something real | An untested relationship is an assumption with a date on it, and the date is the useful part |
| For gathered quotes: how many independent sources, and whether they were independent | Three echoes of one enquiry are one data point wearing three hats |
| Who is told if the source stops publishing or the window stops being used | Otherwise the answer is nobody, and the discovery happens at the next audit |
A price carrying these eight fields can be defended by someone who was not there when it was captured. A price carrying none of them can be defended only by the person who produced it — which works, until the question is asked about a date nine months ago.
Questions people ask about this
Where does the price come from for a grade that did not trade today?
From something that did trade, plus a relationship. Typically a liquid marker is observed and your grade is placed relative to it using a spread or differential that somebody established at some point. That is a reasonable way to produce a number and often the only available one. The thing worth knowing is that it is what happened, because the line on your report looks identical whether the price was observed directly or constructed this way.
Why do two people here quote different prices for the same cargo?
Usually because they are drawing on different sources rather than disagreeing about the market. One may be using the last completed transaction, another a published assessment, another a broker's indication from this morning, another a curve-derived value. All four can be defensible and all four can differ. The disagreement is not really about the level; it is about which kind of evidence counts, and that argument is much easier to settle in advance than in the middle of a valuation.
If we ask three brokers for a price, have we discovered the price?
You have gathered three impressions, and you should find out whether they are independent before treating them as three. In a market with few active participants, brokers consulted within the same hour may be reflecting the same underlying interest — sometimes your own enquiry, having travelled in a circle. Agreement between them feels like confirmation. It is worth asking what would have to be true for them to disagree, and if the answer is "nothing much", the agreement is not carrying the weight you thought.
Is a published assessment a real price?
It is a real output of a stated process, which is not the same thing as a record of what traded. An assessment reflects what an administrator could see, screened and weighted according to a published methodology. That makes it well-defined and often the best number available. It does not make it a measurement of the whole market, and on days when the evidence behind it was thin, nothing about its appearance says so.
How much of our book is actually marked on observed prices?
It is a question that tends to have no ready answer, and it is answerable. The obstacle is not analysis but capture: if the distinction between observed and derived is a convention rather than a stored field, no report can count it afterwards. Start by recording the distinction going forward, then produce the proportion. Whatever the answer turns out to be, it tells you how much of your reported risk is evidence and how much is a model — which is the single thing most worth knowing about a risk report.
What actually goes wrong when a market gets thin?
The number keeps arriving, which is the problem. Fewer participants means each one's motives weigh more heavily on the visible sample, spreads between the sources widen, and derived relationships get applied further from the conditions in which they were built. None of that changes how the price looks. The practical signal is not in the price itself but in what stands behind it — how many independent observations there were, and how far the derivation had to reach.
Our system shows a price for everything. Is that a good sign?
Not by itself. A system that always has a number is either observing a very liquid market or filling gaps, and filling gaps is far more common. The useful question is whether the system can tell you which of the two happened for a given line on a given day. If it can, completeness is genuinely reassuring. If it cannot, completeness is just an interface that never admits to uncertainty, and that is a different thing entirely.
What should we start recording next to a price?
If you do only one thing, record the source class — observed transaction, firm quote, indication, published assessment, derived, carried forward — as a stored field on the price rather than as a habit. Almost every other question in this article becomes answerable once that field exists, and none of them are answerable without it. It costs one column and a decision about who fills it, and it is the difference between a book you can explain and a book you can only assert.
Where this lands in a trading system
Nothing above is an argument for buying software. It is an argument about what is known at the moment a price is captured and what survives that moment. Some of it maps onto system capabilities, and it is worth being exact about which parts, and about where the mapping stops.
The first conclusion is that a mark and the position it values have to be held together before anyone can ask how much of the book rests on evidence. That is the territory of a CTRM/ETRM system providing complete trading lifecycle management for commodity traders, integrating physical trade, financial hedging, risk control, and settlement in one platform; Time Dynamics' Fusion is one, and the part of it that bears on this article is comprehensive risk assessment and monitoring capabilities applied to a book whose marks carry their own provenance rather than arriving bare.
The second conclusion is that price sources live outside the trading system — in a publisher's feed, a broker's message, a spreadsheet somebody maintains — and that the fields which make a price defensible are lost at the point of capture rather than later. That is what X-Ray is for: a non-invasive data processing and analysis platform designed specifically for enterprise clients, whose collection toolkit XDK does non-invasive automated data collection from databases, Excel files, and web interfaces, into a store offering unified storage with unlimited scalability for structured and unstructured data. The property that matters here is that capture and annotation happen in the same step, because a provenance field added afterwards is a guess. It will also collect data without disrupting existing systems or workflows, which is the relevant constraint when the sources in question are somebody's working mailbox and somebody else's morning spreadsheet.
The third conclusion is that a derived price rests on a relationship, and a relationship that is never tested against what actually happened is an assumption with a date on it. X-BT is a Python-based model development and backtesting environment with VSCode integration, and Feynman Language is a custom programming language for time series data visualization and processing. Whether either is pointed at a marking relationship rather than at a trading model is a configuration question rather than something a platform decides for you — but testing a rule against the history of what it was supposed to predict is an ordinary shape, not an exotic one.
No system on this page discovers a price. It will not tell you where the market is, find you a counterparty, or make a grade that trades twice a month behave like one that trades continuously. It does not know that a relationship has stopped holding — it can test one only if a person decides which relationship, against what, and how often, and that decision is judgement rather than configuration. It has no view on whether three broker quotes were independent, and it cannot supply a provenance field for a price that was captured last year without one. What a system can settle is narrower and duller than the problem in this article: whether the source of each number is recorded at the moment it is taken, whether derived and observed can be told apart afterwards, and whether the person defending a figure can find its history without finding the person. That is a real part of the gap. It is not the whole of it, and the difference is worth knowing before you buy anything.
A note on the evidence in this article
There are no statistics here. That is deliberate, and in an article about the strength of evidence behind numbers it would be an awkward place to make an exception. The observations underneath it — how price sources are captured in practice, what is known at the moment of capture and lost immediately after — come from material we are not able to publish as a citable figure. A proportion nobody could check would make this read as more authoritative while making it worth less.
So the argument runs on mechanism, and it is built to be tested against your own operation rather than believed. Three checks are proposed above and they are the intended use: what share of your book is marked on observed prices rather than derived ones, whether the source class of a price is a stored field or a habit, and when the relationship behind your most-used derived mark was last examined against something that actually traded. If your own book answers those differently from what is written here, your book is the better evidence.