Build a mean-reversion basket from one correlation path

Membership is part of the design

A mean-reverting basket is constructed so that members generally move together. Nonconforming names can disrupt the intended group behavior.

One correlation path as the basket

A directed acyclic graph can map equities as nodes and pairwise correlation as weighted edges, then return the longest continuous linkage of interrelated names as the candidate basket.

One illustrated pipeline starts from a large universe, measures Pearson correlation on 250 calendar days of end-of-day prices, ranks pairs, loads the top pairs into the graph, and takes the critical path as the output set.

Group co-movement instead of pair cointegration

Correlation analysis is used here because the design goal is co-movement of the whole group, not the two-asset cointegration framing often used for pair selection in Pairs trading.

Mean reversion on group outliers

The intended trade is to fade members that lead away from the group mean and to buy members that lag, using an overbought/oversold metric such as relative strength index to identify outliers.

One illustrated outlier rule computes relative strength index on each member and, each period, selects the highest and lowest readings in the group as the active names.

Construction as a test platform

Comparative backtests of random, near-zero-correlation, negatively correlated, and high-correlation DAG baskets were used to judge whether tighter intercorrelation improved the mean-reversion test platform.

Further testing is still required before treating DAG group selection as superior to other basket-construction mechanisms.