How many tonnes of cocoa beans will the Ivory Coast produce in 2026?

TL;DR

My forecast is 1.96 million metric tonnes of cocoa beans for Côte d’Ivoire’s calendar-year production, including uncertainty about the eventual reporting convention (forecast calculation). The central 80% interval is 1.69–2.22 million tonnes, and the central 90% interval is 1.60–2.29 million tonnes (forecast calculation). The recovered preceding crop raises the annual total, but the weaker opening of the next crop keeps the calendar-year forecast below the headline seasonal arrivals figure (arrivals evidence, harvest outlook).

Context

The target is production during the whole calendar year, not exports, processing, or a campaign carrying the same year label. The resolution rules therefore include production before the question was submitted. I forecast the eventual authoritative reported number, including eligible revisions, conditional on a numerical resolution rather than annulment.

The reporting calendars complicate the calculation. Côte d’Ivoire moved its domestic main campaign to September–February and its intermediate campaign to March–August, while international seasonal arrivals continue to be reported on an October–September basis (government announcement, February 27, 2026, international arrivals reporting). ICCO’s bulletin published August 31, 2026 reflects information through early August and temporarily withholds production estimates for the current season; it does not provide a usable calendar-year total (ICCO).

Evidence

The historical backbone is the complete FAOSTAT-derived series below: 64 annual observations covering 1961–2024, measured in metric tonnes. The source is FAO’s 2025 dataset, mirrored on February 25, 2026; these are current-revised observations, not historical first-release vintages (FAOSTAT via Our World in Data).

The modern calibration sample covers 2010–2024: 15 observations and 14 annual changes. Those changes have a sample standard deviation of 214,996 tonnes. I use that volatility to calibrate uncertainty, not to extrapolate a smooth growth trend (calculation from the historical series).

National annual reporting gives a more recent baseline. ANStat’s December 2025 dashboard reports 1,890,774.9 tonnes for the twelve months of 2024 and 1,840,311.5 tonnes for the twelve months of 2025, a decline of 2.7%. These are a separate national reporting vintage; I do not silently substitute them into the FAOSTAT history. The exact figures were corroborated in indexed official table text, but I could not inspect the complete dashboard PDF (ANStat, Table 2).

The first decision-relevant question was whether the earlier part of the target year recovered. ANStat answers yes: its March dashboard, published July 24, 2026, reports first-quarter cocoa production up 43.6% year on year. That is a quarterly comparison, not a full-year growth rate, and I did not verify a corresponding quarterly tonnage (ANStat).

The strongest near-complete seasonal observation is 2.18 million tonnes delivered to ports from October 1, 2025 through September 27, 2026, reported on September 29 and attributed to government data and Bloomberg. This is one cumulative marketed-flow observation, not an audited physical-harvest total. It nevertheless confirms a substantial recovery. Earlier, the regulator had forecast seasonal production of 2.0–2.1 million tonnes, illustrating why arrivals and reported production should not be equated automatically (Barchart reporting, Reuters, May 21, 2026).

The second question was how much the new crop contributes before year-end. Reuters’ August 26 report, based on two regulator sources and five exporters, forecast October–December arrivals of 900,000 tonnes, against a revised 1.1 million tonnes in the same quarter of the preceding year. It also reported an eight-to-ten-week crop delay (Reuters).

The outlook then weakened. Reuters reported on October 5 that industry sources expected the main crop to reach about 1.25 million tonnes, versus 1.5 million previously. Those campaigns cover different month ranges, and the article disclosed no representative survey sample. Farmers expected deliveries to accelerate from mid-October, while identifying vulnerable young pods intended for harvest in the following year. Bloomberg’s October 9 report separately described the state weather agency’s warning of rainfall deficits through early 2027 (Reuters, Bloomberg).

I therefore reduce the fourth-quarter assumption to 850,000 tonnes. This is my judgment, not a published observation. It sits between retaining the earlier quarter forecast and scaling it fully by the deterioration in the main-crop outlook; the latter would wrongly assume that losses are distributed uniformly between the target year and the following year (quarter forecast, later harvest timing).

For consistently defined production quantities, the accounting identity is:

$$P_{\text{calendar}}=P_{\text{Oct–Sep season}}-P_{\text{preceding Oct–Dec}}+P_{\text{target Oct–Dec}}.$$

My arrivals-informed approximation adds a judgmental 10,000 tonnes for the remaining September days, then removes the preceding fourth quarter and adds the forecast fourth quarter:

$$2{,}190{,}000-1{,}100{,}000+850{,}000=1{,}940{,}000\text{ tonnes}.$$

This is a calendar-flow estimate, not an official production figure. The seasonal anchor and preceding-quarter input come from the reports above; the final-days allowance and new-quarter adjustment are forecasting assumptions (seasonal anchor, quarter inputs).

A separate upper-side check comes from the government’s macroeconomic page, updated June 18, 2026. Its projected production growth of 11.4%, applied to the national annual baseline, implies about 2.05 million tonnes. That is a planning forecast predating the later crop deterioration, not measured output (government projection, annual baseline).

My final probability model preserves both agronomic uncertainty and reporting uncertainty. All weights and dispersion parameters below are judgments, not measured frequencies (model calculation).

The main branch’s dispersion reflects uncertainty in the seasonal anchor, the preceding-quarter allocation, the coming quarter, and the conversion from deliveries to production. It is slightly below historical annual-change volatility because much of the relevant season has been observed. It remains broad because the uncompleted quarter is a large harvest period and the errors need not be independent. The proxy branch reflects both the regulator’s production assessment and later arrivals, rather than treating either as a final official total (historical volatility, regulator assessment, arrivals).

The mixture’s mean and median both round to 1.96 million tonnes. It assigns 23% probability below 1.8 million tonnes and 41.9% above 2.0 million tonnes. The accompanying code retains both exterior tails and distinguishes a result exactly on a bucket boundary from one strictly below it (forecast calculation).

What's non-obvious

A strong season is not the same thing as a strong calendar year. The large preceding October–December harvest must be removed from the seasonal headline and replaced with a smaller expected quarter. Allocating production simply by the number of overlapping months misses this concentration of harvesting and deliveries (quarterly timing evidence).

Nor does a FAOSTAT year label prove a particular campaign assignment. FAO assigns crops to the year containing the whole or bulk of harvest; it explicitly warns that this need not mean monthly aggregation across January–December. But there is affirmative national calendar-year evidence: the government’s April 6, 2023 announcement explicitly describes the large annual crop as production from January 1 through December 31. A regional central-bank report uses conflicting campaign language, so I retain a smaller alternative-assignment branch rather than claiming the convention is settled or automatically shifting the target to the next campaign (FAO methodology, explicit government period, BCEAO counter-evidence).

Uncertainties

The largest missing observation is a verified national production subtotal through the end of the completed calendar quarters, accompanied by monthly values and a clear harvest-versus-commercialization definition. The official quarterly growth statement establishes direction, but it cannot close that accounting gap. I also could not verify FAOSTAT’s country-specific period-assignment note or original national submission (ANStat quarterly evidence, FAO reporting methodology).

Deliveries can move between periods without an equivalent change in harvest. The government’s campaign-opening statement describes support for evacuating accumulated stocks and holds the farmgate price at 1,200 CFA francs per kilogram. That price was already in place for the intermediate campaign, so the opening announcement is not a fresh price cut. These marketing conditions justify uncertainty about timing and attribution, not a mechanical deduction from physical production (September 1 government announcement, intermediate-campaign price and stock clearance).

The remaining harvest can improve with timely rain or deteriorate with further drought, disease, and delivery delays. Much of the latest weather warning extends beyond the target calendar year, which limits how much should be deducted now. Final official estimates and revisions can also differ from the arrivals-based bridge; this is why I retain broad intervals rather than impose the inferred completed-period flow as a hard production floor (Reuters harvest timing, Bloomberg weather warning, ICCO revision caution).