A forward curve is a list of prices set today for delivery in future periods, not a forecast. Its months are tied together by the cost of carrying goods, its shape is a position, and the points between and beyond its quotes are produced by rules, rebuilt every day.
A forward curve is a list of prices, all set today, for delivery in a series of future periods. It is not a prediction of where prices are going. Its shape says what storage, financing and present scarcity are worth now, and a position valued on it is valued at what could be locked in today. Nor is the list quoted by the market in the form it appears on a screen. Near months trade on their own; further out the market quotes quarters, seasons and years, and beyond those it quotes nothing, so the points between and beyond those quotes are produced by rules, rebuilt every day and loaded into the system that values the book.
This article sets out what a forward curve actually is: why the months on it are tied together by the cost of carrying goods rather than by anyone's expectations, why its shape is a position in its own right, how a block quote becomes monthly prices and how the curve is extended past the last quote, why labels such as front month change meaning at every expiry, and what has to be kept with a curve before a past valuation can be rebuilt. It is written for the people who build the curve and answer for the marks: the traders and analysts who assemble it, the risk and middle office who value the book on it, and the finance teams who are asked to explain a valuation after the fact.
If the idea is new to you, start with A forward curve is what each delivery period costs today, not where prices are going. If you already maintain a curve and want the part that goes wrong quietly, start with The market quotes the curve in blocks, and your book is not held in blocks.
A forward curve is what each delivery period costs today, not where prices are going
Each point on a forward curve is a price at which the market would agree, today, to deliver in a particular future period. The question it answers is what you can fix now, and that is a different question from the one a forecast answers.
A forecast asks what the spot price will be when a month arrives. A forward price asks what can be agreed today for delivery in that month. The two can agree, and on a given day they often do not, because the forward price is set by people who can act on it now: producers selling ahead, consumers buying ahead, and holders of stock deciding whether to sell today or later.
That last group is what ties the months together. Whoever holds goods in storage can sell them for delivery now, or keep them and sell them for delivery later. If a later month is priced above today by more than the cost of storing and financing the goods until then, buying now and selling forward locks in the difference, and that trade itself pulls the two prices back towards each other. So while storage is available, the amount by which later months can stand above nearer ones is held close to the cost of carrying goods between them. In the other direction there is no such anchor. Nobody can borrow goods from next year to sell today, so when the market is short now, the nearer months can stand above the later ones by as much as the shortage demands.
The two shapes have names. A curve that rises with time is in contango; a curve that falls with time is in backwardation. Both are statements about the present, about storage, financing and scarcity today, rather than predictions that prices will rise or fall.
Valuing a position on the forward curve measures what you could lock in today, not what you expect to receive. A trader who disagrees with the curve has a view, and a view acted on is a position. It is not a correction to the curve.
The curve moves in shape as well as level, and the shape is a position
Because every delivery period has its own price, a curve can steepen, flatten or bend while its average barely changes. A hedge placed in one period against exposure in another holds the difference between the two, and that difference moves.
News about supply now tends to move the nearest periods more than the distant ones. A refinery outage or a cold week is about this month; the far end of the curve is priced on what storage and production are expected to look like by then. Distant periods move too, on different news and usually more slowly. A curve is therefore not one price moving up and down. It is a row of related prices that move together most of the time and apart at the moments that matter.
The consequence shows up in a familiar way of hedging a long-dated commitment. The months where the commitment sits often trade thinly, so the hedge is placed in the nearest liquid month and rolled forward as each contract approaches expiry. Each roll closes a position in one month and opens one in the next, at whatever spread separates them that day. The hedge matches the size of the commitment throughout. What it does not match is the delivery period, and the gap between the two is a series of calendar spreads whose result depends on the shape of the curve at every roll.
Matching the size of an exposure and matching its delivery period are separate jobs, and a rolled hedge does only the first. A position report that adds exposure across the curve into one figure keeps the size and drops the period, which is exactly the part a rolled hedge leaves open.
The same blind spot appears when a book is summed across pricing dates instead of delivery periods. Two legs that cancel in total while diverging day by day are worked through there. Price Exposure: Why What You Own Is Not What You Are Exposed To
The market quotes the curve in blocks, and your book is not held in blocks
Near delivery periods trade month by month. Further out, the market commonly quotes quarters, seasons or whole calendar years, and a single month inside one of those blocks has no quoted price of its own. Its price is a share of the block's price, and a rule decides the share.
Where a month trades as its own period, the curve takes its quote. Where it sits inside a quoted block, the block's price is fixed by the market and the way that price is distributed across the months inside it is not. Any split that averages back to the block's quoted price is consistent with the quote. The months can be priced equally, or follow the pattern of the nearer months, or follow the seasonal shape of past years. Each of these is defensible, and they give the same block different months.
Whether that matters depends on the position. A position spread evenly across the block is valued the same under every split, because it only sees the average. A position concentrated in one month of the block, a winter cargo inside a quoted winter season, or a single month inside a quoted year, is valued on the share the rule gave that month, and two rules give two marks. In power and gas the same question repeats inside the month, where the market quotes blocks of hours and positions are held by the day or by the hour.
Beyond the last quoted block there is nothing left to split. The curve still has to reach as far as the longest position on the book, so the remaining points are extended from the last quoted point by another rule: held flat, held at the last observed spread between periods, or continued along a seasonal pattern. These points have no market behind them at all.
All three routes end in the same place. A point the market quoted, a point split from a block and a point extended past the last quote arrive as numbers on the same line, formatted identically, and a position valued on any of them looks the same on the report.
A point the market never quoted keeps printing after the reasoning behind it has stopped holding. Why that failure arrives without an alarm is worked through for prices in general. Price Discovery: Why Your Book Has a Price for Things Nobody Traded
The front month is a different month after every expiry
Curves, positions and limits are often labelled by their distance from today, such as front month, second month or next quarter, rather than by the delivery period itself. When a contract expires every label moves along by one, and anything stored against a label changes meaning overnight with no trade behind it.
Relative labels are convenient because they are how a desk talks. The front month is where the liquidity is, and "long the second month" is a quick description of a position. The convenience has three costs.
A price history kept by label is a splice. The series called front month is one contract until that contract expires, and the next one afterwards. At each roll it jumps by the spread between the two, and the jump is not a market move. A volatility estimate, a chart or a model calibrated on that series counts every roll as though the market had moved by the spread.
A position report kept by label compares different things across dates. Exposure in the second month this week and in the second month a few weeks ago may be exposure in two different delivery periods. A limit set on a relative bucket can be breached, or cleared, by an expiry that moves a position from one bucket into the next.
And the roll date is not the same for every curve. Exchange contracts expire on their own calendars, and physical delivery months and assessment periods follow theirs. A curve built by label has to roll on the right calendar for each market it covers. Rolled a day early or late, the label points at a period the position does not hold.
The remedy is not to give up labels. It is to store the delivery period as dates and to treat the label as a way of displaying it.
A curve is a dated object with an owner, and yesterday's cannot be rebuilt from today's
Every valuation uses a curve as it stood at one moment, built from particular quotes by particular rules. Revaluing a past date needs that day's curve, kept whole. It cannot be reconstructed from today's curve or from a history of its labels.
Rebuilding a past curve from parts means knowing which quotes were taken and when, which splitting and extension rules applied that day, and which delivery period each label pointed at. If any of these has changed since, the rebuilt curve is a new curve carrying an old date. The difference surfaces on the day an auditor, a counterparty or a colleague asks why a position was valued where it was, and the honest answer turns out to be a version of a workbook that no longer exists.
Ownership raises the same question from another side. The desk builds a curve because it trades against it. A risk or control function often needs one it did not build, so that the people whose results depend on the marks are not the only ones setting them. Two curves of the same market then exist inside one company, and they differ mainly at the points with no quote behind them, which are exactly the points where the rule is a matter of judgement. Which curve values the book is a decision, and it is better written down than inherited from whoever loaded last.
A change to a construction rule moves the marks on every position it touches while nothing in the market has moved. If the change is not recorded beside the curve, the explanation of profit and loss will attribute it to the market. And the curve is read by more than valuation: hedge sizing, credit exposure to counterparties and inventory valuation draw on the same points, so the same unrecorded change reaches all of them at once.
The positions are in the trading system, and the curve is often assembled beside it
The trades a forward curve values are usually held in a trading system. The curve itself is commonly assembled each day in a spreadsheet, from exchange settlement prices, broker quote sheets and published assessments, and the resulting numbers are loaded into the system.
That arrangement exists for good reasons. The inputs arrive in different formats from different places, the rules for splitting blocks and extending past the last quote are judgement calls the curve builder wants to see and adjust, and a spreadsheet makes both easy. The trading system then receives a tidy list of prices by period.
What it receives is the numbers, not the reasoning. The splitting rule, the extension rule, the roll calendar and the time the quotes were taken stay behind in the spreadsheet, in formulas and in the habits of the person who maintains it. The loaded curve tells the system what each point is; only the spreadsheet knows why. When that person is away, or the workbook is copied and edited for a new month, the system goes on valuing the book on whatever arrives, and nothing in it can tell a market move from a change of rule.
One curve carries four kinds of point, and the report shows them as one
| Kind of point | Where its number comes from | The assumption inside it | What can change it without a trade |
|---|---|---|---|
| A period quoted on its own | A settlement price, assessment or firm quote for exactly that period | The choice of source, and the time the quote was taken | A different closing time, or a correction to the published figure |
| A month inside a quoted block | The block's quote, split across its months | The splitting rule, which the market does not set | A change of splitting rule, or a new quote for a neighbouring block |
| A point beyond the last quote | The last quoted point, extended | The extension rule: held flat, held at the last spread, or a seasonal pattern | Any change to the last quote, or to the rule |
| A curve for another grade or location | The reference curve, plus a differential | That the differential still holds that far out | The reference curve moving while the differential is held still |
Read down the second column: only the first row has a market quote for that exact period behind it. Read down the fourth: every row can move without anyone trading, and in the bottom three some of the causes are decisions made inside the company.
A curve point has to state six things before a position can be valued on it
| Field | Why it is load-bearing |
|---|---|
| The delivery period, as dates | A label such as front month points at a different period after every expiry; dates do not |
| The quote it came from, and the period that quote covered | Shows whether the point was quoted itself or is a share of a month, quarter, season or year |
| The rule that split or extended it, and which version of the rule | Two rules consistent with the same quote give the same month two prices |
| The time the quotes were taken | Two curves of one market taken at different closing times are different curves |
| The roll calendar that maps labels to periods for this market | Decides which period a label meant on any past date |
| Who built it, and whether someone else checked it | Decides who can answer for a mark at a point with no quote behind it |
With all six stored beside every point, a valuation from last year can be rebuilt by someone who never saw last year's workbook. Without them, it can be rebuilt only by asking.
Questions people ask about this
Is a forward curve a forecast of future prices?
No. Each point is the price at which delivery in that period can be agreed today. A forecast is an estimate of what the spot price will be when the period arrives. The two are linked, because people who expect higher prices later will buy forward and lift the curve, but the curve also reflects storage costs, financing and present scarcity, which have nothing to do with anyone's view of the future. Valuing a position on the curve measures what could be locked in now.
Why is the forward price different from the spot price?
Because delivery later is a different product from delivery now. When storage is available, a later month can stand above the nearer one by up to the cost of storing and financing goods until then, since anyone holding stock could otherwise buy now, store and sell forward. When the market is short today, nearer months can stand above later ones with no comparable limit, because goods cannot be borrowed from the future. The first shape is called contango and the second backwardation.
What do contango and backwardation mean for a physical trading book?
They decide what holding goods over time earns or costs. In contango, a cargo kept in storage and sold forward can lock in the difference between months, less the cost of carrying it. In backwardation, holding stock costs money relative to selling it now, and a hedge rolled forward month by month gains or loses the spread at each roll. Neither shape says where prices are going. Both say what time is worth in that market today.
Why did our mark-to-market move when the market hardly moved?
Look first at the points with no quote behind them. A change to the rule that splits a quarterly or annual quote into months, or that extends the curve past the last quote, moves every position valued on those points. So does a contract expiry, if positions or price histories are kept by labels such as front month, and so does a change in the time at which closing quotes were taken. Each looks like market movement on a report unless the change is recorded beside the curve.
How do you get a monthly price from a quarterly or calendar-year quote?
By choosing a rule. The quoted block fixes the average across its months, and any set of monthly prices that averages back to it is consistent with the quote. Common rules price the months equally, carry over the pattern of the nearer months, or follow a seasonal shape from past years. Positions spread evenly across the block are valued the same under all of them; a position concentrated in one month is not, which is why the rule in use belongs with the curve.
What happens to the forward curve when the front month expires?
The expiring period drops off the near end, and every relative label moves along by one: what was the second month becomes the front month. Prices do not jump because of it, but any series, report or limit kept by label does, because it now refers to a different delivery period. A front-month price history therefore contains a jump at every expiry equal to the spread between the two contracts, and that jump is not a market move.
Can we build our forward curve in a spreadsheet?
Many companies do, and the arithmetic is rarely what fails. What fails is that the rules, how blocks are split, how the curve is extended, when labels roll and which closing time is used, live in formulas and in the habits of the person who maintains the workbook, while the trading system receives only the resulting numbers. The point at which this becomes a problem is the first time someone needs to rebuild a past valuation and the workbook has changed since.
Should the trading desk or the risk function own the forward curve?
It is a governance decision, and what matters most is that it is made explicitly. The desk is closest to the market and builds a curve to trade against; a control function often needs one it did not build, so that the people whose results depend on the marks are not the only ones setting them. The two curves will differ mainly at points with no quote behind them. Whichever one values the book, the other can check it, and reviewing the differences at those points as a routine is where the second curve earns its keep.
Where this lands in a trading system
Nothing above is an argument for buying software. It is an argument about what a forward curve is made of and which parts of it are decisions. Some of those parts map onto system capabilities and some do not.
The first conclusion is that positions, the hedges against them and the curve that values both have to be held where they can be read together. That is the job 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 parts of it that bear on this article are advanced trading management for futures and derivatives markets and comprehensive risk assessment and monitoring capabilities, held in the same place as the physical trades, which is the condition under which a rolled hedge can be read against the delivery periods it stands in for.
The second conclusion is that the quotes a curve is built from arrive from outside the trading system: settlement files, broker sheets, published assessments, and the workbook that turns them into a curve. That is a different problem, and it is the one X-Ray addresses: a non-invasive data processing and analysis platform, whose collection toolkit XDK performs non-invasive automated data collection from databases, Excel files, and web interfaces, and which is built to collect data without disrupting existing systems or workflows. The last property is the relevant one, because the curve builder's workbook is working, and a plan that begins by taking it away will not survive the first month-end.
The third conclusion is that each day's curve has to be kept whole, with the quotes and rules behind it, if a past valuation is ever to be rebuilt. X-Ray's time series database offers unified storage with unlimited scalability for structured and unstructured data. Whether it is used to keep every day's curve as it was built, rather than only the latest one, is a decision about what you choose to store rather than something a platform decides for you, but a dated series of whole curves is an ordinary shape for a time series store, not an exotic one.
The fourth conclusion is that the same book has to be read by delivery period as dates, by quoted block, and by kind of point, and that these views are usually rebuilt by hand. X-Sheet does one-click generation of personalized reports with Excel-like interface.
And here is where it stops. None of this produces a price. It will not forecast the market, it will not quote a month nobody trades or make the far end of a curve liquid, and it does not choose how a quarter is split or how the curve is extended past the last quote. Those rules are judgement, and the judgement stays with whoever owns the curve; which curve that is, the desk's or the control function's, is not a system's decision either. What a system can settle is narrower and duller than the problem in this article: whether the quotes and rules behind each day's curve are captured when the curve is built, whether a quoted point can be told apart from a split or extended one afterwards, and whether last year's valuation can be rebuilt without finding the person who built last year's workbook. That is a real part of the gap. It is not the whole of it, and it is worth knowing which part you are buying.
A note on the evidence in this article
There are no figures in this article, and that is a choice. The observations underneath it, about where curves are assembled and which of their rules live outside the system that uses them, come from material we are not in a position to publish as a citable figure, and a number nobody could check would make the argument look stronger while making it worth less.
So the argument runs on mechanism, and each claim is meant to be tested against your own curve rather than accepted on our word. Three tests follow from it: mark each position's curve point as quoted, split or extended; pick a valuation from a past date and try to rebuild it from what is stored today; and check whether any report, limit or price history is kept by a label such as front month. If your curve answers those differently from what is written here, your curve is the better evidence.
The curve tells you what each period costs today. Only part of it was quoted by the market.