The Cocoa Coverage:
Discipline instead of forecast
The cocoa price multiplied within two years and fell again. Whoever wants to predict it rarely finds confirmation in their own backtests. Whoever does not have to predict it needs a rule: how much of the annual requirement is covered, and when minimum coverage is raised. The model recognises market conditions, not prices.
This module is part of a nine-part model calculation on a fictitious mid-sized chocolate manufacturer (€95 million net revenue, 140 items, 4 lines). All euro amounts are derived from the documented assumptions of the sample company. They are not results of real customers and not a performance promise. Which values are achievable on your data is shown only by a pilot.
01 The problem - the largest purchasing item is covered by feel
Every percentage point of dispersion in landed costs is a percentage point of planning risk
The sample company buys 3,600 tonnes of cocoa products per year, liquor, butter and powder, for €21,600,000. Purchasing covers by budget, current price and experience: if the price falls, more is bought; if it rises, they wait. That rule is not a rule but a mood.
The helpful data is public: futures prices, exchange rates, crop reports. Only their combination does not lead to a price forecast - in-house research shows that a volatility model did not see the 2023 cocoa shock in advance and a regime model registered it only as a whisper. The value lies in coverage discipline, not in prediction.
With €21,600,000 of cocoa purchasing, the dispersion of landed costs is the real subject of the module, not the mean.
02 The model - three market conditions and one rule per condition
A regime model on daily price changes estimates the probability of a strained market condition; the coverage rule follows the condition, not the price
In a calm condition purchasing covers three months ahead in tranches. In an elevated condition minimum coverage rises to five months. In a strained condition contract stocks are held and emergency spot purchases below minimum coverage are avoided. The model does not decide, it names the condition; purchasing has laid down the rule in writing beforehand.
Dispersion of annual landed costs (standard deviation, backtest, example values): Buying on demand at spot ±14.2 % Coverage rule per condition ±8.1 %
Two papers with an open calculation and a fixed random seed: The Single-GARCH Limit on Soft Commodities shows that a classic volatility model on cocoa, coffee, sugar and cotton passes the risk discipline but provides no early warning; The Second Layer shows that a regime model registered the 2023 cocoa shock three and a half months before its documented start - as a brief flicker, "a whisper, not an alarm". Both results are negative results against price forecasting. That is exactly why this module does not calculate one.
03 Business impact - avoided dispersion, not savings
The amount stands separately and is not added to the series total
Every assumption comes from the sample company and is stored centrally. With your figures the input changes, not the method.
| Position | Value |
|---|---|
| Cocoa products / year | 3,600 t |
| Blended price liquor, butter, powder per tonne (assumption) | €6,000 |
| Base: Cocoa purchasing / year | €21,600,000 |
| Lever: Avoided dispersion of landed costs (assumption without anchor) | 1,0 % |
| Result: avoided dispersion / year | €216,000 |
Assumption of a sample company, not added up. The blended price includes the semi-finished mark-up; the exchange bean price is no anchor for it.
A coverage rule does not lower the price on average; it lowers the dispersion. One percent of cocoa purchasing is the sample company's assumption for the value of this avoided dispersion: fewer emergency purchases, less recalculation, fewer price talks with retailers mid-year. This amount is a risk measure, not a cost advantage. That is why it stands on this page and nowhere in the series total.
04 Next steps in your factory
From mood to written rule
Purchasing and management fix coverage horizon and minimum coverage per market condition - before the model runs, not after.
The rule is calculated over the last five years against "buying on demand". You see the dispersion of landed costs, not a promised mean.
Every week one line: market condition, probability, applicable minimum coverage, actual coverage. Purchasing decides, the model names.