EXPERIMENTAL RESEARCH · VERSION 1.0.0

Policy Divergence Model

A directional outlook for the absolute gap between two official policy rates over the next 90 days. This is TRADE90's heuristic assessment, not market-implied pricing, an estimated number of basis points or a currency-price prediction.

Economic evidence first

Geopolitical events are not scored directly. Their effects must enter through economic observations or the central bank’s official interpretation. Risk-on/risk-off sentiment is separate and is not currently scored by this model.

Exact v1 rules

• Headline year-on-year inflation: compare the latest reading with three months earlier (two quarters for quarterly series). A rise of at least 0.3 percentage points gives +2; a fall of at least 0.3 gives −2; otherwise 0. This is a trend signal, not a comparison with a universal CPI target.
• Unemployment: compare the same intervals. A decrease of at least 0.2 percentage points gives +1; an increase of at least 0.2 gives −1; otherwise 0.
• Real GDP: two consecutive quarters at or above 0.5% quarter-on-quarter growth give +1; two negative quarters give −1; otherwise 0.
• Reviewed official guidance: an explicit conditional tightening lean gives +2; easing gives −2; no explicit forward commitment gives 0. Existing source-linked editorial summaries supply these annotations. They are not inferred from a news headline.
• A score of at least +3, at least two positive evidence groups and no negative groups produces a tightening lean. The inverse gives an easing lean. All neutral groups produce a tentative hold lean. Other combinations remain mixed.

These thresholds are design assumptions, not estimated or calibrated coefficients. Headline CPI is not the Fed’s PCE target measure. Core inflation, wages, neutral-rate estimates, bank forecasts and market pricing are not included in v1; these omissions limit its explanatory and predictive value.

From banks to pairs

Compare the two bank leans while identifying the currently higher-rate currency. A higher-rate bank leaning to tighten while the other holds suggests widening; the inverse suggests narrowing. Two banks leaning in the same non-neutral direction remain unclear because relative move sizes are unknown. Potential narrowing of a gap under one percentage point remains unclear because crossing could reverse the higher-rate currency. Mixed bank evidence withholds a directional pair outlook.

Freshness and missing data

All three macro series need a successful retrieval within seven days, at least four observations and exact comparable prior periods. Monthly observations older than 120 days and quarterly observations older than 240 days are withheld. These deliberately broad ceilings accommodate publication lags; they do not prove the newest release has been captured. Official guidance expires 90 days after publication or 14 days after its editorial review. A newer policy decision also invalidates older guidance. The policy-rate snapshot must be no older than 45 days. The visible assessment expires after 48 hours without a rebuild.

Euro-area inputs must all use the same 21-country aggregate; a fallback to incompatible legacy geography withholds the EUR outlook. UK unemployment is a rolling three-month average: freshness uses its period end rather than its middle-month chart label. Source errors keep saved charts visible with warnings, rather than presenting cached history as a successful update.

Validation and track record

No predictive accuracy or probability is claimed. Historical revised data would introduce hindsight bias into a backtest. Deployment starts a forward assessment archive containing model version, input fingerprint, evidence, starting rates and classifications. New evidence and state changes are recorded; these are a prospective evaluation dataset, not a proven track record.

Before treating this as a validated forecast, compare 90-day realised absolute rate-gap changes with the saved outlooks and a no-change baseline. Account for policy benchmark changes, unavailable evidence and dependent overlapping forecasts. Never rewrite earlier predictions to match later outcomes.

Latest assessment JSON · Assessment archive JSON · Data sources · Explore pairs