Back
Correlation Engine

How Laytus reads the way markets move together

A five-layer pipeline that turns prediction market questions into a correlation coefficient the pricing engine can actually use.

Correlation MatrixIllustrative
Fed cuts
BTC > prior high
S&P > target
AI bill passes
Fed cuts
1.00
0.42
0.68
0.11
BTC > prior high
0.42
1.00
0.55
0.18
S&P > target
0.68
0.55
1.00
0.22
AI bill passes
0.11
0.18
0.22
1.00
ρ
lowhigh

Each cell is the correlation between two events. Darker teal means the pair tends to move together.

The Idea

Independence is the default, not the truth

Most platforms price multi-event bets by multiplying individual probabilities, which assumes outcomes are unrelated. They rarely are. When Bitcoin runs, Ethereum usually follows. When the Fed eases, growth stocks rally.

The correlation engine pairs natural language processing with statistical modeling to score how prediction markets actually relate. The output is a coefficient (ρ) the Gaussian copula uses to price the joint outcome fairly.

The Pipeline

Five layers, NLP into statistics

01NLP

Entity Extraction

Parses each prediction market question for the actors, institutions, and assets it references. Overlap between questions becomes a starting signal.

02NLP

Logical Implication

Reads whether one event happening would shift the odds of another, and in which direction. Connected events start with a non-zero link.

03NLP

Category Classification

Questions are clustered by theme. Same-theme pairs share a baseline; clearly unrelated pairs are pinned to zero.

04Statistical

Correlation Matrix

Estimates the correlation (ρ) between every pair of active questions from historical market data. The matrix feeds the pricing engine.

05Statistical

Model Validation

Compares realized outcomes to the model in real time. Fees widen automatically when the model is less confident.

What It Outputs

The signals the rest of the protocol consumes

Pairwise ρ

Correlation between any two events

Joint probability

P(A ∩ B) via Gaussian copula

Confidence score

How much data backs a given pair

Risk flags

Illiquid markets, ambiguous resolution

Regime labels

Easing, risk-on, recession, and others

Brier score

Model accuracy tracked over time