Energy Crisis Thermometer — Lab Intelligence

1 · In one sentence

A 0–100 index summarises energy pressure, bottlenecks and financial stress while keeping drivers, data quality and recent direction visible.

2 · Why it exists

To separate the macro signal from the narrative: not one price, but a reproducible and auditable composite view.

3 · What it produces

A current gauge and a coordinated history of the drivers.

The gauge separates Calm, Watch, Stress and Crisis. Historical panels show whether Brent, TTF, VSTOXX, BTP-Bund, Supply Shock and Italian funding costs are rising or easing.

Coordinated history of energy, logistics and financial drivers. — Levels and recent direction of the underlying factors.

4 · How it works

From public series to index and publication.

Energy Crisis Thermometer pipeline diagram — Acquisition, robust normalisation, composite index, forecast, state machine and bilingual cards.

1Sources

2Quality

3Factors

40–100 index

5Forecast and state

6Publication

5 · Data used

Sources and role in the calculation
Source / series	Variable	Use	Handled limitation
Yahoo, Stooq, FRED, ECB	Brent, TTF, VSTOXX, BTP-Bund, Italy 10Y, gold	energy and financial stress	provider fallback, cache and freshness
World Bank Pink Sheet	fertiliser and urea	physical supply pressure	monthly frequency and declared forward fill
BDI, Cass Freight or vessel data	shipping and arrivals	logistics stress	if unavailable: 20-day Brent volatility with lower confidence

6 · Models and calculations

Robust normalisation and lightweight ML

Rolling z-scores, robust/sigmoid rescaling, weighted mean normalised over available inputs and span-10 EWM. The three-month forecast uses ridge regression on 1, 5 and 20-day lags and lagged exogenous variables.

7 · Decision logic

Opportunistic state machine

OFF → BUILD_UP → CRISIS → EXIT → RECOVERY → BACK_TO_OFF. Thresholds, index direction, driver confirmation and forecast govern the simulated allocation without conflating it with the communication card.

8 · Validation

Walk-forward and comparison

Out-of-sample forecast metrics, Dynamic-versus-Static backtests, subperiod and stress-window analysis, drawdown, VaR/CVaR, Sortino, Calmar and HHI concentration.

9 · Limits

A proxy, not a physical thermometer

Market series do not directly measure traffic through the Strait or actual availability. Shipping may be proxied; forecast and allocation remain experimental and are not advice.

Possible applications of the pattern

Composite stress indices with quality and operating state.

The pattern can transfer to supply chains, energy, country risk or operational continuity when components, weights, fallbacks and confidence are explicit.

Transferable components

multi-provider adapters and SQLite cache
robust normalisation and available-weight logic
state machine and regime alerts
paper trading and automated reports

Research questions

quality of shipping proxies
out-of-sample weight stability
Dynamic versus Static
recovery robustness

Application map Compare methods and stack

Workflow technical dossier

IMPLEMENTEDDescribes components present in the repository.EXPERIMENTALMarks models or choices still under validation.

AI, ML, RPA and quantitative engineering.

The workflow combines automated acquisition, feature engineering, statistical models, risk control, rendering and bilingual publication.

Architecture and automation

Separate modules for data sources, cache, quality, features, signals, allocations, backtests, paper trading and reporting.
GitHub Actions, cron, artifacts, email and AT Protocol; independent IT/EN posts with duplicate guard.
SQLite for cache and persistent simulation; CSV/HTML/PDF/XLSX for audit.

Features and index

Supply Shock from energy and fertilisers; Financial Stress from volatility, spreads and gold.
Shipping Stress from real series or Brent volatility, down-weighted to 30% confidence.
Availability-aware weights, 0–100 clipping and exponential smoothing.

ML, risk and research

Ridge regression, walk-forward OOS, lagged features and quarterly forecasting.
ATR, volatility targeting, drawdown governor, stop losses, scenarios and attribution.
Universe research and dynamic-portfolio comparison with a static benchmark.

Stack

pandas, NumPy, SciPy, scikit-learn, statsmodels and yfinance.
requests, pandas-datareader, Beautiful Soup and lxml.
Matplotlib, Pillow, Jinja2, ReportLab, openpyxl and PyYAML.

10 · Runtime

Automated cadence and bilingual outputs.

Status: active · experimental
Frequency: several weekly runs with an Italy-time guard
Output: gauge, driver history, reports and paper trade

11 · Technical detail

Compact index formula

Each input is converted into a rolling z-score: zₜ = (xₜ − rolling mean) / rolling standard deviation. Sub-indices are then rescaled to 0–100. The final index is Hₜ = Σ(wᵢ · Sᵢ,ₜ) / Σwᵢ over available components only, clipped to 0–100 and smoothed with span-10 EWM. With real shipping data the weights are 0.42 supply, 0.28 shipping and 0.30 finance; with the Brent proxy they become 0.46, 0.14 and 0.40, with shipping confidence set to 0.30.

Main failure mode

Mistaking prices and proxies for physical availability or an investment signal.

Publication governance

Watermark, #LIwf6 tag, alt text, anti-duplicate state and two independent top-level posts.

See both images from the latest output.

The Publications page lets readers move from the gauge to the driver history within the same post.

Open publications