Machine learning & alternative data
See the ground truth before consensus data releases.
Anatomy of the strategy
Let models discover signals in non-traditional data before the market prices the underlying shift.
Satellite imagery, credit-card panels, social sentiment, and news NLP are transformed into features; deep models rank forward return probabilities.
Information asymmetry — seeing the ground truth before consensus data releases.
Overfitting, data snooping bias, and regulatory limits on data sourcing.
How the desk runs it
A loop, not a tip — the same four steps, every day, without exception.
Source the unconventional
Satellite images of oil tanks and parking lots, credit-card transaction panels, shipping manifests — data that describes reality before the earnings report does.
Engineer features
Raw alternative data is noise-dense. The edge is in transforming it into clean, predictive features — a pipeline, not a download.
Train and validate
Deep models rank forward return probabilities. Validation is ruthless: walk-forward tests, purged cross-validation, and live paper trading before capital commits.
Retire decaying signals
Every signal decays as others discover it. Models are monitored for decay and retired without sentiment.
Wildbull tools
This family's workspaces are in build. Here is what is coming.
Alternative data streams translated into readable market signals.
A transparent lab where model signals are validated out of sample before they reach you.
Who runs this strategy
Two Sigma
Data-driven everything — thousands of independent models fed by alternative data at industrial scale.
D.E. Shaw
Applies machine learning to text analysis and asset modelling across the firm.
Man AHL
Brought deep learning into trend-following and managed futures with published research.