Scenario Desk

Continuous monitoring across news, the Commission, the Parliament, member-state governments and the actors who matter. Anything that could move a reform probability lands in the queue within the hour. A weighted relevance filter surfaces the policy-moving items; an analyst sense-checks every actionable alert before the model re-runs.

31 political actors × 36 reform elements tracked. Probabilities computed by formula: P = Σ (Weight × Score). The model refreshes weekly, with ad-hoc re-runs on bigger policy events.

Every probability change feeds the fundamental S/D model automatically. The model decides what is material; the dashboard shows what moved and by how much, threshold-gated so noise stays in the changelog.

The same balance trajectory anchors a calibrated price framework. Daily UNDER / OVER / FAIR signal with a horizon of 63 trading days: the desk's headline number.

A dedicated monitoring agent sweeps the EU ETS policy landscape every hour across four channel types.

A weighted relevance filter strips ~90% of noise. An LLM classifier scores the rest for policy relevance and signal strength, then splits them into two streams.

An analyst sense-checks each flagged alert before the model runs, and cross-checks the output afterwards. The methodology produces the numbers; the human verifies they make sense before publication.

Affected reform elements get a fresh weighted probability across 31 political actors. Formula-driven, audit-trailed: no analyst guesswork.

The fundamental S/D model refreshes weekly and re-runs ad-hoc on bigger policy events. New TNAC, balance and supply/demand numbers come out; the delta vs the previous run drives the publication threshold.

Model output is compared to the previous run. What gets published depends on how much the numbers actually moved: from silent changelog entries to prominent NEW banners.

• Power & heat: unit-level generation stack, fuel-switching dispatch, coal floor, CHP scaler, weather and renewables overlay.
• Industry: bottom-up at installation level.
• Aviation: EUROCONTROL flight movements.
• Maritime: built on our daily maritime emissions tracking (trade lanes × efficiency × ETS scope).

• Cap: LRF from the ETS Directive.
• Free allocation: installation-level with BM, AL, CBAM, CLEF, conditionality.
• Auctions: residual after all deductions.
• MSR: 24% intake, Sep–Aug cycle.

The framework calibrates on quarterly history. For each historical quarter we estimate a TNAC sensitivity (how much the price moves with the surplus) and curve and participation slopes (how contango and open interest shift the level around the anchor). Calibration is walk-forward: every historical quarter uses only the data that was actually observable at the time, so today's coefficients carry no hindsight bias.

The fair value itself refreshes daily. The TNAC anchor moves with the reform model (weekly, plus ad-hoc on bigger policy events); contango of the futures curve, positioning, open interest and macro signals fold in fresh every session. Slow-moving fundamentals and fast-moving market state combine into one number per day, with an explicit error band around it. The output is a daily fair value for the December front contract and a UNDER / OVER / FAIR signal. Thresholds are deliberately asymmetric (tighter on the buy side, wider on the sell side, set to the empirical noise of each direction), and the signal horizon and cooldown are both 63 trading days, so trades sit one quarter at a time with no overlap.

A live reform view (31 political actors × 36 elements), threshold-gated model updates into the fundamental balance, and a daily fair value with an UNDER / OVER / FAIR signal on a 63 trading-day horizon.

The balance refreshes weekly and re-runs ad-hoc on bigger policy events. The fair value updates daily as positioning, the futures curve and macro signals move.

Formula-driven across the actors who decide: P = Σ (Weight × Score). Human-in-the-loop sense-checks actionable alerts before publication; no analyst guesswork on the weights.