Inflation · South Africa
South Africa inflation forecast
Inflation (CPI, annual %)
Model run: Sep 28, 2026 · TimesFM 2.5
12 months forecast · 2026
4.1%
Median forecast · p10-p90: 1.6% to 7.1%
24 months forecast · 2027
4.8%
Median forecast · p10-p90: 1.7% to 8.5%
36 months forecast · 2028
5.0%
Median forecast · p10-p90: 1.3% to 9.2%
Latest observed: 3.2% · 2025
Choose a horizon
The median and range update together.
History & forecast
White: observed history. Blue: forecast median. Shaded: p10-p90 model range.
Display-only line. World Bank data is annual: the forecasts are the yearly points marked on the chart. The monthly line between them is a straight-line interpolation for display, not a monthly forecast.
All values in CPI, annual %
Model validation
| Horizon | Tests | Model errorModel | No-change errorNo change | Band coverageCoverage |
|---|---|---|---|---|
| 12 months | 5 | 1.31 pp | 1.53 pp | 100.0% |
| 24 months | 4 | 2.12 pp | 2.70 pp | 100.0% |
| 36 months | 3 | 2.07 pp | 2.31 pp | 100.0% |
In these tests the model had lower average error than a no-change forecast at all 3 horizons. Rolling backtests of this model on the latest revised history; overlapping forecast errors are not independent. The p10-p90 band is a model range, not a calibrated guarantee. Methodology · Forecast audit
Evidence & context
Source data
Source & observed data
| Year | Value | Change |
|---|---|---|
| 2025 | 3.2% | −1.16 pp |
| 2024 | 4.4% | −1.71 pp |
| 2023 | 6.1% | −0.96 pp |
| 2022 | 7.0% | +2.42 pp |
| 2021 | 4.6% | +1.39 pp |
| 2020 | 3.2% | −0.87 pp |
| 2019 | 4.1% | −0.41 pp |
| 2018 | 4.5% | −0.68 pp |
| 2017 | 5.2% | −1.42 pp |
| 2016 | 6.6% | +2.08 pp |
Sources & limits
Source: World Bank, World Development Indicators · FP.CPI.TOTL.ZG. Methodology · Forecast audit
Historical context
Outside the model
What could change this forecast?
A small, dated reading list. TimesFM sees the data series, not these stories. This context is not a trading signal.
No recent item selected for this series. We would rather leave this quiet than fill it with a weak match.
Selected for relevance, not a claim of causation. Source dates stay visible. Original reporting remains with its publisher.