Multivariate Cointegration Trading with Periodic Hedge Re-estimation

Summary

This document outlines a daily strategy for trading a set of assets using a cointegration vector estimated with the Johansen method on training data. It applies the vector to log prices to form a combined process, then sums its recent changes to determine the signal direction. Assets are divided into long and short groups according to the signs of their vector weights, and position quantities are calculated to target a specified notional exposure. Positions are opened using information through the prior period and closed in the following period.

The strategy periodically re-estimates the vector, with monthly re-estimation offered as an example, and includes an interface for updating prices, generating signals, and tracking trades. The document explains how the signal avoids look-ahead bias by using changes through the prior timestamp, while using the current close for share conversion. It does not report performance, transaction costs, or robustness tests. It assumes the estimated cointegration relationship remains useful between re-estimates and describes a daily strategy that stays invested, so practical risk and execution effects require separate evaluation.

Key ideas

• The strategy estimates a multivariate cointegration vector with the Johansen method using training data.
• It applies the vector to log prices and sums prior changes in the resulting process to set the signal direction.
• Assets are assigned to long and short groups based on the signs of their cointegration weights.
• Position quantities are sized to target a specified notional, and trades are closed on the next timestamp.
• The vector can be re-estimated periodically, but the document reports no performance or transaction-cost analysis.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.