Metrics and models
A metric summarises observed data. A model estimates, transforms or projects values using assumptions. They should not be presented as the same kind of evidence.
Growth and CAGR
Year-on-year growth compares consecutive observations: (current − previous) / previous × 100%. The compound annual growth rate (CAGR) describes the constant annual rate that would connect a starting and ending value:
CAGR = (Ending value / Starting value)1/n − 1, where n is the number of years between the observations.
CAGR is not the actual growth in each intervening year. It is undefined under some zero or negative starting-value conditions; gaps can make trends misleading.
Partner concentration
Concentration describes how much trade is associated with a small number of partners. Common measures include largest-partner share, top-three partner share, and the Herfindahl–Hirschman Index (HHI).
Conceptually, HHI is the sum of squared partner shares. Whether shares are expressed as fractions (giving a maximum of 1) or percentages (maximum 10,000) matters when interpreting a displayed value. The exact scale must be verified against the implementation before publishing numeric thresholds.
Dependency and exposure
High reliance on one partner can indicate concentration risk, but concentration alone does not demonstrate disruption, financial loss or a causal dependency. Trade structure, alternative suppliers and product importance all influence interpretation.
Forecasts and machine learning
Forecasts are estimates of future outcomes, not observed trade. Their usefulness depends on training data, assumptions, forecasting horizon and validation. Treat projections as scenarios or signals to investigate, not guarantees.
Documentation status: We have not yet reviewed the current forecasting implementation, its exact algorithms or validation results. This page intentionally does not claim a particular ML architecture, accuracy score or confidence interval. We'll add those after checking the backend.
How to read any indicator
- Check the reporter, flow, period and product filter.
- Confirm whether enough observations are available.
- Separate measured values from calculated or forecast values.
- Look for reasonable alternative explanations.