Methodology
How CiO Costs builds its numbers
A plain description of every figure on the site: where it comes from, how forecasts are made, how they are tested, and where the limits are. It describes the site as it runs today; it changes when the methods change.
1. Official data
- Uganda — UBOS Construction Input Price Index (monthly, base 2016/17 = 100): overall index, building and civil works, and each material. The index is rebuilt from its own weights to check every release.
- Kenya — KNBS Construction Input Price Indices (quarterly). Six releases are scanned images; their headline values are read by OCR and accepted only when the quarters before them match the text releases exactly.
- South Africa — Stats SA P0151.1 construction materials price indices (monthly). Rebasing is handled by chaining each release's own month-on-month changes.
- Rwanda, Tanzania, Zambia, Nigeria — no official construction price index. Official prices of single items are used where published (Zambia: ERB monthly pump price notices, each checked against the next notice; Rwanda: RURA notices).
- Drivers — Brent crude (EIA/FRED), exchange rates (IMF and a daily reference API), world commodity prices (IMF via FRED: iron ore, copper, aluminium, coal, natural gas; US iron and steel prices), rainfall (ERA5 via Open-Meteo).
- Benchmarks — official monthly indices from the US (BLS), UK (DBT/ONS), Austria (Eurostat), Hong Kong (C&SD), the Philippines (PSA) and Brazil (FGV), used only to show how the method performs where data is good.
2. Forecasts
- Uganda's main model follows the structure of the official index and links each part to its drivers (oil and the exchange rate, rainfall for quarry products, announced tax changes at 50% pass-through). Months since the last UBOS release are filled with the oil and currency readings already known: that is how the gap after a late release is covered.
- Kenya uses a quarterly model with the same kind of drivers; South Africa a model of recent monthly changes anchored on the latest release.
- Each material is then checked separately. A driver model (oil for fuel; the exchange rate for labour; oil and rain for sand and stone) is used only where it clearly beats "no change" in testing; otherwise the forecast is "no change" with a tested range. Each material's choice and test result is shown next to it on the country page.
- Countries without an official index get an import-cost estimate: oil in local currency (weight 0.10) and the exchange rate (0.40), a share similar to imported inputs in the Ugandan and South African baskets. Where we can check it, its three-month direction matched the official index about 6 times in 10. It covers imported inputs only, not local materials or labour.
- Per-item estimates for those countries exist only where testing supports them: fuel from oil (tracks official fuel and asphalt indices well) and steel from world steel prices (only roughly). There is no estimate for cement, sand or blocks, because nothing we tested tracks them; those come from prices people report.
- Likely range = where the real change should land 8 times in 10. Ranges are calibrated from past errors.
3. How forecasts are tested
- Walk-forward tests: every past month, a forecast is made with only the data that existed then, and compared with what was published later. Calibration of the ranges uses only errors that were already known at each point.
- Choose, then judge: where several methods compete, the choice is made on the first half of the test period and judged on the second half, which played no part in the choice. Results are shown even when they are not flattering.
- Live record: every published forecast is saved and checked automatically when the official figure comes out (live track record). Benchmark results are on each benchmark page.
- Data checks run every three hours: complete and ordered ranges, plausible history, fresh oil and currency data, downloads, the published contract index, and the weekly notes.
4. Prices people report
- A typical price is published only from 3 or more different people, pooling the last two months; each person counts once (their own median), so repeated or extreme entries cannot move it.
- A price far from the usual one (more than three times higher or lower) is queried before it is saved.
- The monthly reported-price index for an item is its typical price relative to the first published month (= 100). After 12 months it also gets its own tested forecast.
5. Price adjustment (contracts)
- The IPC tool follows FIDIC (Sub-Clause 13.7 of the 2017 books, 13.8 of 1999): Pn = a + Σ coefficient × In / I0, with I0 at the Base Date and In 49 days before the end of each IPC period.
- Index values are marked official, provisional (a CiO forecast, used only for periods that have ended and until the official figure is published) or CiO estimate. Contracting parties should agree in writing which sources apply and that provisional values are replaced.
6. Limits
- Forecasts are estimates with ranges, not certainties. Many materials move mostly at random month to month; for them the honest forecast is "no change" with a range.
- Estimates for countries without official indices cover imported inputs only.
- Nothing on the site is financial, legal or contractual advice.
Pilot version 0.1, October 2026 · Questions: the CiO Costs WhatsApp channel