EXPERIMENTAL RESEARCH · VERSION 4.0.0

Policy Divergence Model

TRADE90 v4 uses independently verified macroeconomic observations, official central-bank decisions and source-bound policy interpretation to construct conditional 90-day policy-rate divergence research for 28 currency pairs. The outlook is qualitative: it estimates no future policy-rate level, rate-change magnitude, statistical probability or currency-price return.

Three-stage divergence workflow

1. Macroeconomic evidence: date-validated headline and underlying inflation, unemployment and GDP are evaluated against bank-specific objectives and historical comparisons. The resulting economic-pressure scores are descriptive, not numerical policy-rate projections.
2. Central-bank policy evidence: official decisions, source-dated guidance and independent editorial reviews are checked for currency relevance and superseding publications. Automated source monitoring supplies explicitly provisional policy-pressure screens where both banks have sufficient evidence, never invented official guidance.
3. Relative policy-rate gap: the sign and absolute size of today's published policy-rate gap are combined with supported relative policy biases. Widening and Narrowing remain conditional; uncertain or conflicting information withholds a directional assessment. No future rate level or crossing magnitude is calculated.
4. Evidence hierarchy: current source-reviewed policy interpretations take precedence; provisional automated policy-pressure scenarios follow; material independently verified macro-score contrasts remain exploratory. Contradictions between the reviewed/provisional direction and the macro screen are labelled uncertain.
5. Completeness and freshness: missing macro series, invalid benchmarks, superseded guidance and outdated source checks cannot quietly become confident forecasts. The model displays which evidence tier supports each result.

The model labels reviewed policy scenarios, automated provisional scenarios, exploratory macro screens and uncertain or insufficient evidence. These are research hypotheses, not calibrated forecasts, FX-price predictions, probabilities or instructions to trade.

Signed and absolute rate gaps

The signed gap is the first currency’s selected policy rate minus the second currency’s rate. The absolute gap is its magnitude, regardless of sign. Divergence classifications describe this policy-rate difference, never the FX price direction. Widening and narrowing use neutral research styling, not trading-signal colours.

At equal rates, different bank biases support possible divergence in the corresponding direction. A narrowing outlook may involve a reversal of which currency has the higher rate, but the model does not estimate its timing or size. We do not estimate move sizes, so this is an initial directional bias, not a guaranteed absolute gap at day 90.

Inputs and policy objectives

US policy inflation uses BEA headline PCE and core PCE, retrieved through FRED and transformed from price indices into year-on-year changes. Australia uses monthly headline CPI in the model while the comparison charts retain their quarterly history. Supporting underlying inflation comes from OECD CPI excluding food and energy for Australia, Switzerland, Canada, the UK, Japan and New Zealand, and Eurostat HICP excluding energy, food, alcohol and tobacco for the euro area (21 countries).

These underlying measures are not interchangeable with each bank’s preferred measures: Australian exclusion-based CPI is not trimmed-mean CPI; Canadian exclusion-based CPI is not CPI-trim/CPI-median; Japan’s exclusion measure differs from CPI excluding fresh food. Their purpose here is to check headline persistence, not claim a full representation of bank analysis.

• USD: 2% longer-run headline PCE objective — official framework
• EUR: 2% medium-term objective — official framework
• GBP: 2% medium-term objective — official framework
• JPY: 2% medium-term objective — official framework
• CAD: 2% midpoint of the 1–3% control range — official framework
• NZD: 1–3% medium-term range, focused on 2% — official framework
• AUD: 2–3% inflation target, focused on the midpoint — official framework
• CHF: Price stability: non-negative inflation below 2% — official framework

Exact economic-pressure rules

All thresholds below are model design assumptions, not calibrated coefficients or official policy rules. An objective is medium-term context, not an automatic trigger.

• For point objectives, the model uses a ±0.25 percentage-point buffer around 2%; this is not an official tolerance range. Australia uses 2–3%; Switzerland uses non-negative inflation below 2%. Canada and New Zealand retain their 2% focus for model comparisons; their official ranges remain visible above.
• Headline and underlying inflation trends compare the latest observation with three months earlier, or six months for quarterly data. A ±0.3 percentage-point move defines a rising/falling trend. Decimal comparisons are rounded to eight places before threshold tests.
• The single inflation pillar receives +2 when headline inflation is above the model reference zone, neither headline nor underlying inflation is falling materially, and underlying inflation exceeds the objective’s reference (2%; Australia 2.5%; Switzerland 1% as a model reference, not an SNB target). It receives −2 for headline below the zone with non-rising headline/underlying inflation and underlying below the reference. Otherwise, both inflation measures falling gives −1; other combinations give 0. Headline and core are not counted as two independent votes.
• Unemployment compares three months earlier, or six months for quarterly data. A fall of at least 0.2 percentage points contributes +1; a rise contributes −1. An increase of at least 0.5 is a material weakness flag.
• Two negative GDP quarters contribute −1. Two positive quarters exceeding that economy’s median of up to 20 earlier quarters contribute +1, requiring at least 12 historical quarters. The comparison excludes the two assessed quarters. Otherwise GDP contributes 0. This is historical context, not an estimate of potential output.
• The three pillars sum to a descriptive pressure score. Positive inflation pressure and a total of at least +1 are labelled tightening pressure. Negative inflation pressure and a total of at most −1 are easing pressure. Other combinations are balanced or conflicting. The score does not estimate a rate change.

From economic pressure to policy outlook

Each official-statement interpretation has its own publication date, source URL, review date and rationale. It must match the statement it interprets and cannot outlive a newer policy decision. A statement saying that the current meeting held rates is not forward hold guidance.

• Explicit conditional tightening guidance supports “Hike or hold” when inflation pressure and the total score are non-negative, without a material unemployment rise or two-quarter contraction.
• Explicit conditional easing guidance supports “Cut or hold” when inflation pressure and the total score are non-positive.
• Without directional guidance, a +2 inflation pillar and a score of at least +3 support “Hike or hold”, subject to the activity safeguards. A negative inflation pillar and score of at most −3 support “Cut or hold”.
• “Conditional hold” requires an explicit reviewed policy-context rationale, headline and underlying readings within the reference zone, a score between −1 and +1, and no material activity weakness. The SNB’s unchanged-rate conditional forecast is contextual support, not a promise.
• Everything else is uncertain. One minor opposing indicator is visible but does not automatically cancel strong evidence. Invalid economic inputs withhold the policy outlook. Expired or superseded policy commentary preserves validated macro evidence but labels the policy direction uncertain.

From bank outlooks to pair research

Hike-or-hold, hold and cut-or-hold biases are compared in that order. Different biases support an initial signed-gap direction; the existing sign determines whether this means widening or narrowing. The gap can remain unchanged if both banks hold. Identical directional biases remain uncertain because move sizes and timing are unknown. Two supported hold baselines produce a broadly unchanged bias. Unknown policy direction on either side blocks a directional pair result.

Freshness, automation and missing information

Required series must have a successful retrieval within seven days, ordered finite observations and matching comparison periods. Monthly observations expire after 100 days; quarterly observations after 210 days measured from the end of their calendar quarter, not the first-day quarter label. UK unemployment uses the end of its rolling period. These are conservative maximum ages, not a release-calendar guarantee. Failed refreshes remain invalid even when cached points exist.

Guidance expires 90 days after publication or 14 days after its individual interpretation review; newer decisions invalidate earlier guidance. Policy snapshots expire after 45 days. A displayed pair expires at the earliest relevant input/guidance deadline or 48 hours after model generation. Rebuilding HTML alone does not refresh economic evidence or extend an assessment's validity.

Additional inputs refresh with the daily macro workflow. Publication discovery and supported numerical policy updates are automated. Interpretation of new statements still needs a reviewed, source-bound annotation; this version does not claim autonomous policy interpretation. New sources failing checks cannot silently replace verified data.

Geopolitical headlines and risk-on/risk-off sentiment are not scored directly. Their expected economic consequences can enter through official guidance before they appear in released data. Wage histories, complete bank forecast paths, financial-condition measures and release-surprise data are not included.

Additional policy drivers under evaluation

More indicators could improve context, but adding uncalibrated signals must not create numerical probabilities or artificial precision. The independent economic-pressure pillar does not score the following dimensions. They require verified source histories, release-date tracking and genuine out-of-sample validation:

• Wages and services inflation: indicators of domestic inflation persistence, including preferred measures that vary by bank.
• Inflation expectations: consumer, business and professional expectations where official time series are available.
• Policy projections and voting: bank forecasts, member votes, minutes and published conditional rate paths, with decision-time provenance.
• Growth and demand nowcasts: retail spending, PMIs, vacancy data and other early signals; these can be noisy and must not duplicate GDP and labour pillars.
• Financial conditions and credit: credit growth, lending conditions and exchange-rate effects where the bank considers them material.
• Official forward-looking policy material: future bank projections, meeting minutes and official forecasts may contribute source-linked qualitative context only after validation; they are never converted into unsupported future rate levels.
• Fiscal, commodity and geopolitical shocks: interpret through verifiable economic channels instead of ungrounded headline sentiment.

New inputs should remain explanatory and unscored until tests using information available at each historical decision show incremental forecasting benefit. Published probability estimates require calibration and measurement of forecast error.

Validation and version history

Version 4.0.0 removes the market-implied forward-rate assessment entirely and uses independent macroeconomic and central-bank policy evidence. Previous versions explored separate pricing-based signals; those are retired from the current model. Prediction accuracy remains unestablished. Tests verify deterministic safeguards, not economic forecasting skill.

Evaluate saved outlooks prospectively against realised policy gaps and a no-change baseline, including coverage, abstentions and conflicting evidence. Preserve historical inputs to avoid look-ahead bias.

Latest research JSON · Additional input histories · Assessment archive · Explore pairs