Quantitative Advisory · Now Accepting Engagements

Retail Intelligence
& Commodities
Forecasting.


We transform unstructured spatial data, satellite imagery, and macroeconomic indicators into strategic edge. Access our proprietary signals or leverage our expertise for bespoke infrastructure.

Retail Intelligence
2,500+
Locations Tracked
300+
Chain Registries
0.92
Signal Confidence
Commodities & Energy
2006
Backtest From
+2.40
Sharpe Ratio
53.9%
Win Rate
01Practice Area

Retail Real Estate
Intelligence

Satellite-driven location intelligence covering 2,500+ stores across Northern Europe. We use parking occupancy as a revenue-change proxy — delivering bi-weekly sales-delta signals and competitor positioning without any POS data.

Sales change detection · bi-weekly signals
Competitor occupancy benchmarking
Catchment & Huff-model site scoring
300+ chain registries · DE · NO · SE · DK
SAR · Sentinel-2 · DOP10 tri-layer fusion
Explore data infrastructure
02Practice Area

Commodities &
Energy Forecasting

Multi-parameter predictive and forecasting models for European energy and agricultural commodities. We ingest gas storage, live terminal flows, ENTSO-E power data, weather, satellite flaring, and NDVI crop signals into a systematic ensemble that generates 14-day directional forecasts.

TTF Gas · 14-day forward signals
ENTSO-E generation & interconnect flows
Live LNG terminal delivery monitoring
Satellite gas flaring detection
NDVI crop growth · agricultural momentum
View Momentum Alpha model

Services

Bespoke Engineering
& Advisory

Institutional-grade analytical capability without the overhead. Retainer or project basis, 4–12 week engagements.


Data Architecture & PostGIS

PostgreSQLPostGISPythonETL

Optimizing and building scalable spatial databases for massive geospatial datasets. Multi-tenant schema design, performance tuning, and production PostGIS/PostgreSQL infrastructure.

Discuss this service

Predictive Modelling & AI

CatBoostLightGBMNeural NetworksML Ops

Developing and tuning advanced predictive models — Neural Networks, LightGBM, CatBoost — on complex corporate data. From feature engineering through production deployment.

Discuss this service

Retail & Location Strategy

Huff ModelCatchmentCompetitor AnalysisGIS

Spatial interaction modelling (Huff model), white-space analysis, and competitor tracking. Turn raw location data into defensible site selection and portfolio decisions.

Discuss this service

Data Infrastructure

Three Independent
Imagery Layers

Each layer is validated against the others. Disagreement triggers a confidence downgrade — agreement triggers a lock. Over 2,500 locations active across Northern Europe.


Sentinel-1 SAR

Synthetic Aperture Radar · 10m · All-weather · Biweekly

Parking occupancy index
VH contrast signal
Temporal delta (sales proxy)

C-band SAR (VV+VH polarisation) from ESA Copernicus, processed with Lee speckle filter and terrain correction. Biweekly temporal cadence since 2022 delivers a continuous occupancy baseline — immune to cloud cover and lighting conditions. VH ROI-minus-ring contrast is our primary parking-density proxy.


Sentinel-2 Optical

Multispectral · 10m · Cloud-filtered · B04/B08 fusion

Reflectance occupancy score
S2/SAR Bayesian fusion
Cloud-masked time series

Sentinel-2 B04/B08 (Red + NIR) paired with per-store Static Noise Maps that remove persistent surface artefacts. Bayesian hierarchical fusion with SAR yields cloud-robust occupancy estimates. When S2 and SAR confirm, signal confidence reaches 1.0.


Aerial Orthophotos

10 cm/px · Sub-meter · YOLO ground-truth anchors

YOLO vehicle count anchors
Surface type classification
SAR/S2 calibration baseline

High-resolution national orthophotos — Germany DOP10, Norway NiB (Kartverket), Sweden Lantmäteriet — used as YOLO v8 vehicle-count ground truth. Centimetre-precision car counts calibrate satellite signals and classify parking surface type (asphalt / mixed), improving seasonal model accuracy.


2,500+
Locations tracked
DE · NO · SE · DK
3
Fused imagery layers
SAR · S2 · DOP10
300+
Chain registries
Grocery · DIY · QSR · More
2022–
Data history
Biweekly temporal cadence

Proprietary Signals

Proprietary Signals

Private Beta — Application Required

Model 01 · Commodities

MomentumAlpha TTF Precision Engine

Quantitative fundamental model for the European gas market

Beta

MomentumAlpha TTF Precision Engine operates as a digital twin of Europe's physical energy system. Unlike traditional technical strategies, 70% is driven by physical supply and demand data, while 30% captures the market's psychological momentum. A built-in, dynamic risk management layer automatically reduces exposure when market volatility becomes extreme.

Model Architecture

70% Fundamental
30% Momentum

The Four Pillars

Advanced Temperature & Demand Analysis

HDDNWP TemperaturePopulation Weighting

Rather than relying on linear weather forecasts, the model computes a population- and industry-weighted heating demand (HDD). Specific demographic and industrial hubs are monitored with heavy weight on key regions in Germany and Italy — a cold snap in a densely populated industrial area is assigned proper definitional power over cold fronts in uninhabited regions.


Physical Inventory & Panic Indicators

GIE/AGSITWh Storage LevelPanic Interaction Factor

The model monitors total European storage volume in absolute energy units (TWh) and links storage status to weather data via a proprietary interaction factor. The algorithm identifies and responds to market panic when low storage levels are exposed to sudden cold shocks.


European Logistics & Supply Shocks

LNG ArbitrageNorth Sea FlowLive Terminal Feed

Direct links to physical export terminals for pipeline gas, plus global LNG arbitrage margins (European prices vs. Asia/US). The model detects unexpected supply disruptions in the North Sea in real time via satellite vessel tracking and live terminal data feeds.


Power Sector & Industrial Demand (Spark Spread)

ENTSO-EClean Spark SpreadEUA/CO₂Demand Destruction

Gas is analysed in the context of power and carbon allowances (EUA). The model integrates institutional data from European Day-Ahead power auctions and computes real plant margins (Clean Spark Spread) — registering when prices rise high enough that industry chooses to shut down (Demand Destruction).


Backtest Metrics · Multi-regime (pre- and post-crisis)

+0.153
Information Coefficient
Stable predictive power
53.9%
Win Rate
Directional accuracy
+2.40
Sharpe Ratio
Risk-adjusted return
10-Day
Signal Horizon
Optimised for physical logistics
70/30
Architecture
Fundamental / Trend ensemble

MomentumAlpha gives the investor an emotionless, data-driven “edge” in one of the world’s most volatile commodity markets. By understanding the underlying physical and financial forces, the model positions ahead of large structural price moves — while protecting capital when markets are driven by pure noise.

northstar-api · v1 · signal endpointLIVE
$ curl -X GET api.northstarsignal.eu/v1/signal \
-H "Authorization: Bearer <API_KEY>"
▸ 200 OK
{
  "_tables": ["paper_trading_log", "ttf_fundamentals"],
  "signal_id": "PTL-20260520-TTF-0041",
  "asset": "TTF_NATURAL_GAS",
  "signal_direction": "LONG",
  "confidence": 0.92,
  "horizon_days": 10,
  "regime": "TIGHTENING_SUPPLY",
  "timestamp": "2026-05-20T09:41:00Z",
  "fundamentals": {
    "fundamental_score": 0.71,
    "momentum_score": 0.38,
    "storage_TWh": 612.4,
    "hdd_anomaly": 2.1,
    "spark_spread": "POSITIVE",
    "norwegian_flow_mcm": 318.7
  },
  "status": "BETA_ACCESS_ONLY"
}

Model 02

Retail Real Estate
Intelligence

Beta

Tri-layer remote sensing intelligence covering 2,500+ retail locations across Northern Europe. We track parking occupancy as a real-time proxy for store revenue — delivering weekly sales-change signals and competitor positioning without any POS data.

Sales Change Detection

Bi-weekly parking delta signals serve as a leading proxy for in-store sales change. Calibrated against DOP10 YOLO vehicle anchors, our model tracks revenue momentum across 2,500+ locations with >90% directional accuracy.


Competitor Intelligence

Live occupancy tracking across 300+ chain registries — grocery, DIY, QSR, fashion. Competitive pressure scores, catchment overlap modelling (Huff), and white-space opportunity maps delivered as structured data or API.


Bayesian Fusion Engine

NumPyro hierarchical Bayes fuses SAR, S2, and YOLO anchors. Confidence = 1.0 when SAR + S2 agree. Automatic SAR fallback on cloud-impacted S2 epochs keeps signal continuity year-round.

SAR · S2 · DOP10YOLO v8CatBoostPostGIS2,500+ LocationsDE · NO · SE · DK
Request Data Sample

About

The
Expertise

Northstar Signal is built on a rare combination of strategic business acumen and deep technical craft. With an MSc in Economics and an MBA, we understand the commercial context behind every data problem — not just the mechanics.

Over 15 years of specialised data engineering in Python, PostGIS, and machine learning have been dedicated to building infrastructure that actually works at scale — from processing 3.4 million Overture Places records to fusing multi-source satellite imagery into actionable signals.

The result: advisory that bridges the gap between executive strategy and production-grade quantitative systems. No bloated teams. No off-the-shelf software wrapped in consulting hours. Pure, proprietary engineering.

Academic

  • MSc Economics
  • MBA

Technical

  • Python
  • PostGIS
  • ML

Domain

  • European Retail
  • Energy
Start a conversation