Which data vendors provided the most alpha?

Satellite parking lot data (6% alpha) and credit card transactions (4% alpha) were top performers.

How much did the data cost?

$2M annually across 5 vendors, generating $50M+ PnL—25x ROI.

How does this case study work?

Raise a request, talk to experts, fund the project, expert works, review and approve payment. All remote, all through our platform.

Discovering Alpha with Alternative Data

Executive Summary

A quant fund's traditional factor models showed decaying alpha. By integrating satellite parking lot imagery and credit card transactions, they discovered novel signals predicting retail earnings surprises, generating 12% annual alpha with 0.3 correlation to existing factors.

Key Outcomes

▹ 12% annual alpha from alternative data signals
▹ 0.3 correlation to existing factors (highly diversifying)
▹ 3 new systematic strategies deployed

Client Situation

The fund's existing value/momentum factors had Sharpe ratios declining from 2.5 to 1.2 over 3 years. They needed novel data sources for differentiation.

Key Challenges

⚠ Traditional factors crowded and decaying
⚠ Alternative datasets prohibitively expensive to evaluate
⚠ No infrastructure for processing non-standard data

Existing Architecture

Factor models using price/volume data from Bloomberg. Research in Python notebooks, production in C++. No alternative data pipeline.

No capability to process image or unstructured text data
Manual data vendor integration taking 3+ months
Research-to-production handoff lacking for new data types

Solution Design

Alternative data platform with standardized ingestion, feature extraction, and backtesting for satellite, transaction, and web-scraped datasets.

Key Decisions

✓ Satellite imagery processing with computer vision for parking lot occupancy
✓ Credit card transaction aggregation by merchant category
✓ Unified feature store for all alternative signals

PySparkAWSPyTorchMLflowFeature Store

Implementation

3-month pilot with single data vendor before expanding to 5 sources over 12 months.

Phase 1: Phase 1: Satellite Pilot
Built pipeline for parking lot imagery, discovered strong signal for retail earnings.
Phase 2: Phase 2: Transaction Data
Aggregated anonymized card data at merchant-day level, found consumer spending signal.
Phase 3: Phase 3: Production Integration
Combined signals into multi-strategy portfolio with 1.8 Sharpe ratio.

Technical Challenges

Satellite image processing at scale

Impact: 1TB daily imagery required 200+ GPU hours

Resolution: Pre-computed parking lot features, on-demand inference for new images

Transaction data aggregation lag

Impact: 3-day delay made signals too stale

Resolution: Switched to same-day processing with Spark streaming

Results

Annual alpha (gross): Before0%
After12%
ImprovementNew alpha source
Correlation to existing factors: BeforeN/A
After0.31
ImprovementHighly diversifying
Time to onboard new data vendor: Before3 months
After2 weeks
Improvement86% reduction

Lessons Learned

📘 Parking lot occupancy predicted retail earnings surprises with 3-week lead
📘 Credit card data required careful cleaning for returns/refunds
📘 Alternative data alpha decayed slower than traditional factors (9 months vs 3)

What We Would Do Differently

💡 Build synthetic data for backtesting new signals
💡 Implement automated data quality monitoring earlier

Role Relevance

Quant researchers with alternative data expertise identified signals traditional quants missed, discovering 12% uncorrelated alpha.

Critical Skills Demonstrated

Alternative data analysisSatellite imagery processingTransaction data aggregationFeature engineering

Frequently Asked Questions

Which data vendors provided the most alpha?: Satellite parking lot data (6% alpha) and credit card transactions (4% alpha) were top performers.
How much did the data cost?: $2M annually across 5 vendors, generating $50M+ PnL—25x ROI.