Partial Cointegration & Statistical Arbitrage

Research Area: Risk & Regime Modeling
Date: March 2026

Overview

This project evaluates a statistical arbitrage strategy based on partial cointegration, a framework allowing financial spreads to combine mean-reverting dynamics with a persistent stochastic trend.

The goal is to assess whether this model produces more robust trading signals than classical Engle–Granger cointegration in equity pairs.

Dataset

S&P 500 equity universe
Daily adjusted prices
Correlation screening for candidate pairs
Cointegration diagnostics

Trading Strategy

Trading signals are generated using the standardized spread:

z_t = (spread_t − μ) / σ

Enter positions when |z| exceeds a threshold
Close positions when the spread reverts toward equilibrium
Transaction costs and slippage included

Experimental Pipeline

Pair selection through correlation filtering
Cointegration testing using Engle–Granger
Partial cointegration estimation via state-space model
Signal generation and trading simulation
Performance evaluation (Sharpe ratio, drawdown)

Results

The strategy compares classical cointegration signals with signals derived from the partial cointegration model.

Preliminary experiments suggest smoother signals and fewer false entries relative to classical cointegration approaches.

Equity Curve

Spread Dynamics

Signal Comparison

Implementation

The pipeline is implemented in R using econometric time-series libraries for cointegration testing, state-space estimation and trading simulation.

CI  <- egcm(X, Y)
PCI <- fit.pci(X, Y)

z = (spread - mean(spread)) / sd(spread)

Technical Background

A theoretical discussion of the partial cointegration model and its state-space representation is available in the research note:

Partial Cointegration: Model and Kalman Representation

Resources

Project Report (PDF)

Code available upon request.