Updated week of 24 August 2026 · methodology, sources and comparisons
Lokoja Intelligence is the price intelligence layer for Africa's food trade. We price the trade end to end: what a tonne costs across 67 Nigerian markets, what it costs to move, and what the spread nets after transport — verified weekly by ambassadors inside the markets. This page explains exactly how, including the limits.
Three layers, each labelled for what it is:
| Layer | Nature | Cadence | What it contributes |
|---|---|---|---|
| World Bank Real-Time Food Prices | Model-estimated | Monthly | Continuous coverage of all 67 markets, 8 staples, back to 2007. |
| WFP market monitoring | Observed | Monthly (rotating) | Enumerator-collected observations; also our Kenya layer (lokoja.app/ke/). |
| Lokoja field ambassadors | Observed | Weekly | Per-bag prices in trader units with supply levels, diesel & pump prices, a real transport quote, market disruptions and trader counts — from named people inside the markets. |
Model-estimated feeds are useful but unverified. Our ambassadors are the verification: contracted market insiders — today in Dawanau (Kano), Mile 12 (Lagos) and Onitsha Main Market — who file a structured weekly report under a personal-observation declaration. Reports pass automated range and anomaly checks before they touch the engine; fabricated or gamed reports mean clawback and termination under their signed terms. Expansion is ranked by information value: the sell-side and buy-side of the corridors we publish get ambassadors first.
A price without transport is an illusion. Every corridor we publish is quoted net of haulage: a fixed loading cost plus per-tonne-km trucking rates calibrated from working Nigerian transporters, corrected for road circuity, indexed weekly to the national diesel price, with conflict corridors excluded rather than priced as if they were safe. If a spread does not survive the journey, we do not show it.
Every question asked of our WhatsApp assistant becomes a structured, time-stamped record: commodity, location, trade size, language. We sell that — aggregated and anonymised — as what it honestly is: demand signals, the map of who is asking about what, where, this week. Published prices never come from user queries; they come from the observed and modelled layers above, verified in the field. The two datasets are separate by construction.
Our reading of the public offerings, August 2026 — corrections welcome at hello@lokoja.app:
| Recency | Units | Transport costs | Field verification | Access | |
|---|---|---|---|---|---|
| Lokoja Intelligence | Weekly-stamped | Trader units (50kg/100kg bags, lorry loads) and tonnes | Corridor spreads net of diesel-indexed haulage | Weekly ambassador reports | Free web + free API |
| NBS (official statistics) | Monthly, published with a lag | Metric | — | Enumerator surveys | Free bulletins |
| Tridge | Varies by market | Metric | — | Global network, not Nigeria-market-resident | Enterprise |
| Selina Wamucii | Indicative estimates | Metric | — | — | Free/paid tiers |
| AFEX | Exchange sessions | Contract lots | — | Real transactions (exchange-traded volumes only) | Exchange members + reports |
AFEX's exchange prices are real transactions — the gold standard for the volumes that trade on an exchange. Most of Nigeria's food trade does not: it moves through open markets like Dawanau and Mile 12, which is the trade we price.
For lenders and insurers we publish a 0–100 local price-risk score per commodity, market and month (higher = riskier), banded Low / Moderate / Elevated / High, with a separate A/B/C confidence grade for data quality. It is an input to credit and collateral decisions, never a credit score and never borrower-level. Seven components, each 0–100, combined as a weighted mean: seasonality (what the three months after this calendar month did in earlier years, weight 51), three-month momentum (15), signed dislocation against markets within 300 km (12), 12-month volatility (12), inbound arbitrage pressure — the best net-of-haulage spread traders could earn moving grain into this market (10) — plus demand-signal and trader-verification components at a beta weight of 8 each where the data is thick enough. Conflict-affected markets are flagged, not scored higher: conflict is an access risk, and it showed no power to predict price falls.
Which prices the score uses. For each commodity the score is computed on the observed WFP series — the one the weights and bands were validated on — whenever that series is current, and on the modelled World Bank feed when it is not; every score states its source. This matters more than it sounds: WFP's published Nigeria monitoring for the staple grains stops in early 2023 (only meat and fish continue), so today most staple scores run on the modelled feed and carry a C confidence grade, which is the honest grade for modelled prices. Where a fresher modelled price exists alongside a current observed one, it is attached as a labelled nowcast — its month, the price, and its change against the last observed month — never blended into the score, with a driver when the two diverge by 10% or more. Commodity names resolve to the freshest variant of their family, and a stale series is scored only on request, graded C and flagged. The weights were also re-tested on the World Bank feed itself: 148,070 market-months across all ten staples give an AUC of 0.681, with the High band running 6.4× the Low band's adverse-move rate (31.9% vs 5.0%) and every band in order — so the calibration holds on the modelled prices the staples currently score on.
Validation. We fitted the weights on WFP Nigeria history from 2008 to 2021 and tested them on 2022 onwards, scoring each of 24,319 market-months with only the data available at the time and measuring the worst price move over the following three months. Out-of-sample AUC is 0.705 (0.712 across the full history). Calibration by band across the full history:
| Band | Market-months | Fell 15%+ within 3 months | Median worst 3-month move |
|---|---|---|---|
| Low | 4,547 | 8.0% | +0.0% |
| Moderate | 8,003 | 8.6% | +0.0% |
| Elevated | 5,852 | 16.5% | −0.9% |
| High | 5,917 | 35.1% | −7.7% |
A High-band market is about 4.4 times as likely as a Low-band market to fall 15% or more in the next quarter. The score is strongest on grains (maize AUC 0.84, groundnuts 0.78, cowpeas 0.73) and weakest on gari and fish, which move on supply shocks the monthly series does not see. The first version of the method scored no better than chance (AUC 0.52); it was re-fitted before release, and this note will be updated whenever the method version changes. Method details and the full backtest are in the API documentation; the score is available under a commercial key with the risk scope, with free 90-day validation pilots for lenders.
Upstream feeds are monthly with roughly a month's lag — our weekly stamps reflect the diesel index, corridor recomputation and ambassador layer, not a weekly re-survey of all 67 markets. The market list skews north-east (53 of 67), a legacy of its monitoring origins; the commercial south is covered via Lagos, Aba and Ibadan plus our ambassador markets. Demand-signal history is weeks deep, not years. We would rather you learn these limits here than in a meeting.
Free to cite with attribution: "Lokoja Intelligence (lokoja.app)", linked. Journalists and researchers: the monthly Lokoja Food Trade Index is built for citation, and bespoke data cuts are free on request — hello@lokoja.app.