Master nonlinear cointegration trading with robust estimation

Traditional cointegration methods assume linear relationships and Gaussian noise. Markets disagree. Our nonlinear frameworks detect hidden equilibrium dynamics, adapt to regime shifts, and generate signals that survive real-world frictions.

Why linear models fail in real markets

Standard cointegration assumes constant hedge ratios and symmetric adjustment. Empirical evidence shows otherwise. Nonlinear adjustment speeds, threshold effects, and structural breaks destroy the statistical foundations of conventional pairs trading. Our research addresses these failures directly.

Core capabilities of our nonlinear framework

From regime detection to adaptive signal generation, every component is built on peer-reviewed methodology and tested against out-of-sample data.
Asymmetric adjustment

Nonlinear cointegration modeling

We employ threshold autoregressive and Markov-switching frameworks to capture asymmetric adjustment speeds and state-dependent equilibrium dynamics.
State-aware signals Faster convergence
Outlier resistant

Robust estimation methods

Classical least squares collapses under heavy tails and outliers. Our M-estimators and trimmed regression techniques maintain validity when standard assumptions break down.

Heavy-tail robustness Stable coefficients
Real-time states

Regime-switching detection

Markets shift between volatility states without warning. Our hidden Markov models identify regime transitions in real time, enabling dynamic position sizing and risk recalibration.

Dynamic hedging Risk recalibration
Rolling recalibration

Adaptive hedge ratios

Static hedge ratios become stale. We implement rolling robust estimators that recalibrate exposure weights as cointegration parameters evolve, preserving statistical integrity.

Parameter stability Reduced drift
Actionable thresholds

Real-time signal generation

Research means nothing without execution. Our infrastructure translates nonlinear residuals into actionable signals with defined entry thresholds, stop logic, and position limits.
Defined exits Clear logic
Unseen data proof

Out-of-sample validation

Every model must prove itself on unseen data before deployment. Our strict separation of training and test periods prevents the overfitting that invalidates most published strategies.
Overfitting prevention Honest metrics

Research track record

Peer-reviewed methodology with live validation

Our frameworks have been stress-tested through conference presentation, journal submission, and proprietary capital deployment across multiple market regimes.
  • Conference papers
    12+

    Presented at international quantitative finance and econometrics conferences since 2019.

  • Out-of-sample months

    48+

    Strict out-of-sample validation periods required before any strategy sees live capital.

  • Robust estimators

    7

    Distinct M-estimation and trimmed regression techniques implemented and compared.

How practitioners engage with our research

Three pathways from theoretical understanding to practical application, each grounded in rigorous methodology.

1

Foundational quant training

For researchers new to cointegration-based statistical arbitrage, we provide structured analytical reviews covering unit-root testing, error-correction models, and the mathematics of spurious regression. This is not a course. It is a guided walk through the literature that matters, with emphasis on the assumptions that fail and the robust alternatives that replace them.
2

Advanced nonlinear methods

For experienced practitioners, we open our nonlinear cointegration toolkit: threshold autoregressive models, Markov-switching frameworks, and robust estimation under structural breaks. Participants examine real research code, inspect failed model iterations, and understand why certain specifications were abandoned. Intellectual honesty is mandatory.
3

Proprietary strategy review

For teams with existing statistical arbitrage infrastructure, we offer confidential analytical consultations. We audit your current cointegration assumptions, test for nonlinearities you may be missing, and propose robust estimation upgrades. All work remains under strict confidentiality. We do not touch your capital or execution systems.
4

Robust estimation audit

A standalone service for risk-conscious practitioners who suspect their current models are fragile. We stress-test your hedge ratios against heavy-tailed innovations, run structural break diagnostics, and deliver a written assessment of where your assumptions are likely to fail and what robust alternatives exist.

What practitioners say

Verified perspectives from quantitative analysts and portfolio managers

Dr. Henrik Jansen

Head of Quantitative Research, Nordic Alpha Partners

"Their nonlinear cointegration framework exposed regime dependencies we had completely missed. Our linear pairs trading book looked clean in backtests but bled in volatile periods. The threshold autoregressive approach changed how we model reversion speed."

Regime detection Pairs trading
March 2025

Mei-Lin Okafor

Senior Quantitative Analyst, Sydney Statistical Arbitrage

"I appreciated their intellectual honesty. They showed us models that failed, not just cherry-picked winners. The robust estimation module on M-estimators under heavy tails was immediately applicable to our emerging markets book."

Robust methods Heavy tails
January 2025

Thomas Bergström

Portfolio Manager, Baltic Quantitative

"The out-of-sample validation discipline is what sold me. Most vendors show in-sample Sharpe ratios and call it research. These people separate training and test periods religiously. It slows down strategy release but prevents disasters."

Validation Risk control
November 2024

Priya Nandakumar

Quantitative Developer, Melbourne Systematic

"Their adaptive hedge ratio methodology solved a drift problem we had been fighting for months. The rolling robust estimators are computationally heavier than ordinary least squares, but the stability improvement in live trading justified the cost."

Adaptive hedging Live trading
August 2024

From raw data to live deployment

01

Data preprocessing

We clean, align, and inspect price series for splits, dividends, and corporate actions. Outlier detection and stationarity testing happen before any model sees the data. Garbage in, garbage out is not a slogan here. It is a fired researcher.

Clean data
Inspect series
02

Model specification

We test linear versus nonlinear cointegration, compare threshold and Markov-switching specifications, and select robust estimators based on residual diagnostics. Every choice is documented and defensible under peer review.

Specify models
Test specifications
03

Out-of-sample validation

Models are frozen and tested on unseen data. We report performance honestly, including periods where signals underperformed. Only specifications that survive this trial proceed to deployment consideration. No exceptions.

Validate unseen
Report honestly
04

Live deployment

Approved models enter production with strict position limits, stop logic, and real-time monitoring. Regime detection triggers automatic deleveraging when assumptions break. The model does not trade through a crisis it cannot explain.
Deploy signals
Monitor live

Ready to challenge your cointegration assumptions?

Most statistical arbitrage infrastructure is built on linear approximations that markets have already arbitraged away. Our nonlinear frameworks and robust estimators offer a genuine methodological alternative. The first step is a conversation about where your current models might be fragile.

Research-driven

Explore our methodology

Dive into the technical documentation behind our nonlinear cointegration frameworks, robust estimation techniques, and regime-switching detection systems.

View methodology

Meet the research team

Our team emerged from academic econometrics and has presented at international quantitative finance conferences. We publish our assumptions and document our failures.

About us

Start a research conversation

Have questions about nonlinear cointegration, robust estimation, or how our frameworks might apply to your analytical challenges? We respond within two business days.

Contact us
Peer-reviewed methods
Out-of-sample validation
Transparent team