A pass-through formula has four settings, reference index, lag, share and boundaries, and each one decides a different failure mode. With daily feedstock buying against a lagged selling formula, unit margin tracks the change in feedstock cost rather than its level, so a rising market compresses margin at full pass-through. The pre-buy that follows an announced increase is a forecastable event with a date on it, which makes it a planning input rather than forecast error. Basis, meaning the gap between your delivered cost and the published index, is the part no formula passes through, and it is measurable from thirty-six months of history.
The March contract price settles up eighty euro a tonne. The commercial team is comfortable, because every major customer sits on a formula that passes the increase straight through, so margin per tonne is protected by construction. February then closes at 14,400 tonnes against a plan of 12,000, and the plant runs seven days with overtime to cover it. April closes at 9,900. The demand review in early March reads February as a strong month and raises the rest of the quarter. The demand review in early May reads April as a demand problem and cuts it back.
Nothing about demand changed. Customers who could time their orders bought ahead of a price they already knew was coming, and the formula that protected the margin per tonne moved several hundred tonnes across a month boundary and cost more in overtime and idle capacity than the price move earned.
Pass-through terms are written in commercial negotiations and then handed to planning as a fact. They are one of the largest deliberate sources of demand distortion in a chemicals business, and they are almost never modelled anywhere a planner can see them.
The formula has four settings and each one fails differently
Strip any index-linked contract down and you find the same structure.
The reference index. Which published assessment, for which grade, at which location. A European polyolefin contract referencing the monthly ethylene contract price behaves differently from one referencing a naphtha assessment, because the two move together only loosely and the correlation weakens exactly when the market is disorderly. Price reporting agencies have operated under the IOSCO Principles for Oil Price Reporting Agencies since 2012, which set out how an assessment should be constructed and audited. Worth reading once, since several formulas in circulation reference an assessment that no longer reflects a liquid trade.
The lag. This month's settlement, last month's average, or a rolling quarter. The lag decides how much of the margin variability lands on you rather than on the customer, and it interacts with how you actually buy feedstock. A one month lag against daily spot buying is a very different exposure from a one month lag against a matching one month feedstock contract.
The share. Full pass-through of the index move, or a fraction of it, usually with a fixed conversion delta added on top. A seventy percent share means thirty percent of every move stays with you in both directions, which is a deliberate risk position and should be recorded as one.
The boundaries. Caps, floors, collars and trigger thresholds below which the price does not move at all. A trigger threshold of twenty euro means small moves accumulate silently until one crosses the line, at which point the customer sees a large step and behaves accordingly.
Two contracts with the same headline share can produce opposite outcomes in the same quarter if their lags differ. A commercial book described as fully indexed is not a single exposure.
A rising market compresses margin at full pass-through
Take a producer buying feedstock daily at spot and selling on a formula that references last month's average index at full pass-through. In a flat market this works exactly as designed and unit margin sits at the conversion delta.
Now let feedstock rise forty euro a tonne every month for four months. Each month you buy at a cost roughly forty euro above what your selling price is referencing, because the selling price is anchored to a month that has already passed. Unit margin runs forty euro below the delta every month for four months. When the market flattens, margin returns to normal. When it falls forty a month, margin runs forty above the delta.
Full pass-through delivered exactly what it promised across the whole cycle, and it delivered a margin series driven by the first difference of the feedstock price rather than by its level. Annual results look fine. Quarterly results swing, and the swings are perfectly predictable from a series anyone can pull.
Two things follow. A margin miss in a rising market is arithmetic rather than commercial failure, and treating it as failure produces the wrong corrective action, usually a price initiative aimed at customers already paying the formula. And the direction of the market belongs in the financial scenario alongside the level, because a plan built on an average price for the year will be wrong every quarter in a way that averages to right.
A second lag sits underneath, from inventory accounting. With six weeks of feedstock, work in progress and finished goods valued first in first out, the cost reaching the profit and loss account lags the purchase too, so reported margin carries two lags of different lengths and diverges from cash margin for months. How valuation choice moves reported margin is another post's subject (FN1); what matters here is knowing which number you are judging pass-through by.
The pre-buy is a planned event with a date on it
Lee, Padmanabhan and Whang identified price variation as one of four causes of the bullwhip effect in Management Science in 1997, alongside demand signal processing, order batching and rationing behaviour. In consumer goods the price variation they described is largely promotional. In formula-priced chemicals it is contractual, and that difference works in your favour, because a contractual price change has an announcement and an effective date.
Work the example. A book of 12,000 tonnes a month, of which roughly forty percent sits with customers who hold storage and can pull an order forward by two to three weeks without difficulty. An increase of eighty euro a tonne is announced for the first of March. Those customers move perhaps half of their March requirement into February. February closes 2,400 tonnes high, March closes 2,400 tonnes low, and both months miss the plan in opposite directions.
Every element of that is knowable in advance. You know the announcement date, you know which contracts allow order timing, you know roughly how much storage each of those customers has, and you have the last dozen price moves in your own order history as a sample to estimate the response from. A price event calendar sitting alongside the demand plan, holding the effective date and the expected pull-forward volume per customer segment, converts the whole thing from forecast error into a planned movement.
The reporting change that goes with it is to carry two demand series. Consumption is what the customer used. Shipments are what you sent. In a month with a price event those two numbers differ by a quantity you can estimate, and feeding the shipment series into a statistical forecast teaches the model that February is a strong month every time a price move happens to land there. Splitting promotional and baseline demand is a well established practice in consumer goods, and the same logic applies with more precision here, because the event is contractual rather than negotiated store by store.
Downward pass-through moves slower and the destock runs deeper
Prices rise quickly and fall slowly across a wide range of markets. Bacon named the pattern rockets and feathers in Energy Economics in 1991 looking at UK retail petrol; Borenstein, Cameron and Gilbert found the same asymmetry in US gasoline in the Quarterly Journal of Economics in 1997; Peltzman documented it across a large sample of producer and consumer goods in the Journal of Political Economy in 2000. Formula pricing constrains the effect without eliminating it, since the lag structure, the trigger thresholds and the timing of renegotiation all give discretion about how quickly a fall reaches the customer.
The supply consequence is worse than the pre-buy. When a customer expects a price increase, the amount they pull forward is bounded by their storage. When a customer expects a decrease, the amount they defer is bounded by nothing except how long they can run their own operation down, and they will keep waiting as long as the expectation holds. A falling market therefore produces a longer and deeper order trough than the peak that a rising market produced, and it arrives at the moment your reported margin is at its best, which makes it easy to miss.
The practical response is to watch order timing behaviour rather than order volume. Days between orders per customer, and order size relative to their normal cycle, both move before the volume series does. A customer whose reorder interval has stretched from eighteen days to twenty-six is telling you what they expect the index to do.
Basis is the part the formula does not pass through
A published index describes a benchmark grade at a benchmark location on standard terms. Your actual feedstock is a specific grade, delivered to a specific plant, under a specific contract with its own premium or discount. That difference is basis, and no pass-through formula covers it.
The arithmetic is straightforward and rarely done. The index moves up fifty euro a tonne and you pass through fifty. Your delivered cost moves up seventy-two, because the local premium widened by twenty-two on the same day for reasons specific to your supply point. You have passed through one hundred percent of the index and sixty-nine percent of your cost, and the shortfall does not appear anywhere as a pass-through failure because the formula performed exactly as written.
Measure it rather than assuming it. Regress your realised delivered feedstock cost against the reference index over thirty-six months, look at the slope and at the standard deviation of the residuals, and that residual figure is your unhedged exposure per tonne. If the slope is materially below one, the index is a weak proxy and the formula is passing through less than the headline share suggests. Products with no credible published index face a sharper version, since the formula then references a proxy commodity and the relationship is a modelling assumption that nobody has tested.
What to put into the planning model
The contract price rule belongs in the planning model as an object with the four settings, effective dated, so that a scenario which moves the index also moves selling price, margin and the pre-buy volume together. Most businesses hold pass-through terms in contract documents and a spreadsheet, which means no scenario run can see them.
Alongside it, three things earn their place. A portfolio sensitivity, computed as the sum over contracts of volume multiplied by pass-through share, giving tonnes genuinely exposed per unit of index movement, with a second version weighted by lag so you can see when the exposure lands. A reset calendar, because a book where seventy percent of contracts reset on the first of the month moves as one position, and staggering resets is a commercial negotiation with a direct planning payoff. And the split demand series described above, so the statistical forecast never sees a price event as growth.
Where this stops
Pass-through terms are negotiated by commercial teams under competitive pressure, and a planner does not get to change them. What a planner can do is make the consequences visible before signature, which requires being present at contract review with the sensitivity and the pre-buy estimate in hand. Turning up afterwards with an explanation of why the quarter moved is a weaker position and the one most planning teams occupy.
Formula pricing also removes price risk while leaving volume risk untouched, and it weakens the customer's incentive to forecast well. A customer who carries no price consequence for ordering late has less reason to give you visibility, so the improvement in margin stability is partly paid for in forecast quality. That trade is usually still worth making, and it should be made knowingly.
The basis regression has a real limit too. Thirty-six months of monthly observations is a small sample for estimating a slope, and one disorderly period will dominate it. Treat the residual standard deviation as an order of magnitude for the exposure rather than as a number to plan against precisely, and rerun it annually as the sample grows.
Build the table, one row for each of your twenty largest contracts by volume, with five columns: reference index, lag, pass-through share, cap or floor, and reset date. Most commercial organisations cannot produce that in an afternoon, and the ones that can usually find at least one contract referencing an assessment that stopped being representative some years ago.
More on this in industry on the DemandLabs platform.
