Economic observatory
Methods & interpretation
Last updated
The observatory separates published economic data, heuristic concern, qualitative interpretation and probability research.
Current activity
Jobs, spending, production and income show what the economy has recently been doing.
Warning signals
The signal-strength index brings available readings together. Related measures share weight so one story is not counted repeatedly.
Forecasts
A recession probability answers a separate, dated question. See the model checks and published research.
Why recession indicators can disagree
Different signals answer different questions. A yield curve or expectations survey may warn about future conditions while employment and spending still describe recent growth. That disagreement is a reason to investigate timing and breadth, not to choose whichever series confirms a preferred outlook.
Read levels and changes separately. A low survey balance, a fall in an output index and slower positive employment growth are not interchangeable observations. Compare the exact measure and reporting period before describing activity as contracting.
Use a sequence of checks: identify the warning, look for confirmation in a different evidence family, inspect current activity and then review what is missing. Related measures share exposure to the same economic mechanism; several weak surveys do not necessarily provide several independent forecasts.
Yield-curve warning signals · The Sahm Rule and labour-market deterioration · GDP and observed activity
Detailed definitions
Score Meaning
A 0–100 heuristic concern index, not a recession probability.
Aggregation
Related measures share one family vote; available families are averaged within each group, then covered groups receive equal weight. Group availability requires at least one usable family and 25% of its applicable families, so repeated missing survey components do not count as independent coverage gaps.
Composite Treatment
Composite indices, overlapping national accounts and newly added research signals are reviewed as cross-checks unless explicitly assigned a component role. Adding a chart does not automatically add an aggregate vote.
Vulnerability Treatment
Debt-service, credit-cycle, balance-sheet and policy vulnerabilities are reported separately and excluded from the cyclical concern index. Energy shocks require contextual interpretation.
Missingness
Missing and stale data are excluded, not encoded as zero. Available-case weighting can move when coverage changes; sensitivity bounds and comparable-input movement distinguish that effect. Missing essential groups limit interpretation even where a partial score is shown.
Regional Scope
US and UK require local series. Euro-area analysis may include labelled EU and German corroboration. Relevance weights operate within a family; a singleton proxy still receives a full family vote. Effective weights expose the resulting contribution.
Calibration
Indicator thresholds are transparent monitoring choices. Except for identified published rules, they are not validated recession cut-offs.
Dates, revisions and missingness
Observation periods describe when economic activity occurred. Publisher updates, source checks and brief publication are separate timestamps. A rolling three-month estimate must retain its full range. A successful check can return unchanged data or a source that has stopped advancing.
Current observations can be revised, qualified or withdrawn. Historical snapshots retain what was saved at the time. Coverage and method changes must be distinguished from an economic movement; the same concern number in two regions need not mean the same recession risk.
Evidence strength
The catalogue is curated, not an empirical ranking of the best predictors. Each indicator page gives the economic rationale and limitations. A mechanism or an early-signal label is not evidence of reliable forecasting in every country or regime.
Validation and promotion gates · Source definitions · Editorial controls