Scoring Cointegrated Stock Baskets for Statistical Arbitrage

Summary

The article proposes a workflow for ranking stock baskets considered for statistical arbitrage. It separates criteria into eliminatory filters, ranking criteria, and strategic choices. Liquidity and transaction costs can disqualify a basket; cointegration strength, cointegrating rank, hedge-weight stability, and spread reversion time can help order the survivors. Data frequency and lookback length are treated as choices aligned with the intended holding horizon, rather than universal measures of quality.

For costs, the author discusses volume, spreads, commissions, fees, and slippage, and uses simplified assumptions for a sample backtest. The article describes applying the workflow to a screened group of semiconductor stocks and says the installment presents backtest results using the first two ranking criteria. It leaves portfolio-weight stability and mean-reversion half-life for a later installment.

The proposed values and cost estimates are approximations that depend on broker, exchange, asset, and market conditions. Cointegration tests and historical rankings do not guarantee that relationships will persist or that a basket will remain tradable. Rolling updates and evaluation across multiple lookback windows are suggested as ways to monitor stability.

Key ideas

• Separate basket selection into strategic choices, eliminatory filters, and criteria used to rank candidates.
• Liquidity and estimated trading costs can rule out baskets that are statistically attractive but difficult or expensive to trade.
• Johansen test statistics and cointegrating rank can characterize the strength and complexity of candidate relationships.
• Hedge-weight stability and spread half-life are proposed as ranking factors, though their implementation is deferred.
• Timeframe and lookback length should fit the trading strategy, and rolling or multiple-window checks can reveal changing relationships.

Tags

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