Building and Testing Cointegrated Stock Baskets for Mean-Reversion Trading

Summary

This article develops a multivariate statistical-arbitrage workflow intended to reduce dependence on fast order execution. It begins with liquid stocks from a related industry, screens candidates for recent correlation with a reference asset and with one another, then applies Engle–Granger and Johansen cointegration tests. A Johansen eigenvector supplies relative portfolio weights; the weighted spread is checked for stationarity and used to define mean-reversion entries and exits. The sample applies this process to a semiconductor basket associated with Nvidia and discusses testing it in a MetaTrader Expert Advisor.

The article contrasts this approach with a prior correlation-based pairs strategy that performed poorly in a demo account because execution lag erased opportunities. It offers a practical selection heuristic and a sample backtest workflow, rather than proof of durable returns. Correlation screening and visual inspection alone do not establish stationarity, and estimated cointegration relationships and hedge ratios can change. Basket trading also depends on synchronized, liquid price data and remains exposed to execution costs, model updates, and structural breaks.

Key ideas

• The workflow screens a liquid, related group of stocks before testing for pairwise and basket cointegration.
• Johansen eigenvectors provide relative weights for constructing a portfolio spread.
• The weighted spread is assessed for stationarity before it is used in a mean-reversion system.
• The sample aims to reduce sensitivity to execution speed compared with highly timing-dependent pairs trades.
• Correlation and historical stationarity do not guarantee that relationships or hedge ratios will persist.

Tags

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